From e74f44d7ae65e37e0165c932b9a8a5def0e5608a Mon Sep 17 00:00:00 2001 From: ethernet Date: Mon, 10 Aug 2026 17:59:16 -0400 Subject: [PATCH 001/227] ci(windows): desktop install/update e2e - published installer to this commit via fake git remote windows sibling of install-e2e-run.yml. no bubblewrap on windows, so the git proxying is git's own transport rewrite: an isolated GIT_CONFIG_GLOBAL with multi-valued url..insteadOf for both hardcoded repo URLs, so the published Hermes-Setup.exe's install.ps1 clone, hermes update's fetch, and the desktop's ls-remote all land on a local bare repo whose main the driver controls - installer and updater run verbatim. one run: seed fake.git from the checkout, force fake main to the newest release tag, drive the real published bootstrap installer with AutoHotkey (GUI, no headless mode), promote fake main to HEAD, then apply the desktop app's builtin update route (scripts/desktop-update.ps1 -NoUi when the installed base ships it, staged hermes-setup.exe --update otherwise) and assert HEAD == target with a working hermes. TODO routes: bare hermes update, and re-running the bootstrap installer over the existing checkout. --- .github/workflows/install-e2e-windows-run.yml | 113 ++++++ .github/workflows/install-e2e.yml | 26 +- tests/install/windows/install-button.png | Bin 0 -> 1200 bytes .../windows/install-hermes-desktop.ahk | 137 +++++++ tests/install/windows/install-update-e2e.ps1 | 383 ++++++++++++++++++ tests/install/windows/launch-button.png | Bin 0 -> 1485 bytes 6 files changed, 654 insertions(+), 5 deletions(-) create mode 100644 .github/workflows/install-e2e-windows-run.yml create mode 100644 tests/install/windows/install-button.png create mode 100644 tests/install/windows/install-hermes-desktop.ahk create mode 100644 tests/install/windows/install-update-e2e.ps1 create mode 100644 tests/install/windows/launch-button.png diff --git a/.github/workflows/install-e2e-windows-run.yml b/.github/workflows/install-e2e-windows-run.yml new file mode 100644 index 000000000000..50b731ab2b8b --- /dev/null +++ b/.github/workflows/install-e2e-windows-run.yml @@ -0,0 +1,113 @@ +name: Install & Update E2E — Windows (reusable) + +# Runs ONE Windows desktop update route against ONE starting point, with a +# real install (the published Hermes-Setup.exe: git, uv, a managed Python, +# Node, the venv, the desktop build) behind it. +# +# The Windows sibling of install-e2e-run.yml. There is no bubblewrap here, so +# tests/install/windows/install-update-e2e.ps1 fakes GitHub with git's own +# transport rewrite (an isolated GIT_CONFIG_GLOBAL carrying +# url..insteadOf for both hardcoded repo URLs) instead of a +# MITM proxy — the installer and updater run verbatim against their real URLs +# and land on a local bare repo the driver controls. The bootstrap installer +# itself is a GUI with no headless mode, so AutoHotkey clicks it. +# +# Call it: +# +# jobs: +# windows-desktop: +# uses: ./.github/workflows/install-e2e-windows-run.yml +# with: +# route: desktop + +on: + workflow_call: + inputs: + route: + description: 'Update path to exercise. desktop = the desktop app''s builtin update hand-off (scripts/desktop-update.ps1). TODO: update (hermes update), installer (re-run the bootstrap installer).' + required: false + type: string + default: desktop + installer-url: + description: 'Bootstrap installer to install with. Default: the latest published one — what a user downloads today.' + required: false + type: string + default: https://hermes-assets.nousresearch.com/Hermes-Setup.exe + timeout-minutes: + description: 'Job timeout. A cold run installs real toolchains and builds the desktop app.' + required: false + type: number + default: 75 + +permissions: + contents: read + +jobs: + e2e: + name: ${{ inputs.route }} from published installer + runs-on: windows-latest + timeout-minutes: ${{ inputs.timeout-minutes }} + + steps: + # Full history + tags: the driver seeds its fake GitHub from this + # checkout (all origin branches for the installer's commit pin, release + # tags for the starting base) and promotes HEAD as the update target. + - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2 + with: + fetch-depth: 0 + + - name: Restore cached test tools + id: test-tools-cache + uses: actions/cache@27d5ce7f107fe9357f9df03efb73ab90386fccae # v5 + with: + path: test-bins + key: test-bins-${{ runner.os }}-v1 + + # AutoHotkey drives the installer GUI; ffmpeg records the screen so a + # failed click is diagnosable from the artifact instead of by guesswork. + - name: Install AutoHotkey v2 and ffmpeg + if: steps.test-tools-cache.outputs.cache-hit != 'true' + shell: pwsh + run: | + New-Item -ItemType Directory -Path test-bins\autohotkey, test-bins\ffmpeg -Force | Out-Null + + # AutoHotkey: copy its whole v2 directory so helper exes/dlls come along. + winget install -e --id AutoHotkey.AutoHotkey --silent --accept-source-agreements --accept-package-agreements --disable-interactivity + $ahkDir = "$env:ProgramW6432\AutoHotkey\v2" + if (-not (Test-Path $ahkDir)) { + throw "AutoHotkey install directory not found: $ahkDir" + } + Copy-Item -Path "$ahkDir\*" -Destination test-bins\autohotkey -Recurse -Force + + winget install -e --id Gyan.FFmpeg --silent --accept-source-agreements --accept-package-agreements --disable-interactivity --location ffmpeg_dir + Copy-Item -Path "ffmpeg_dir\*\*" -Destination test-bins\ffmpeg -Recurse -Force + + - name: Add test tools to PATH + shell: pwsh + run: | + Add-Content -Path $env:GITHUB_PATH -Value "$PWD\test-bins\autohotkey" + Add-Content -Path $env:GITHUB_PATH -Value "$PWD\test-bins\ffmpeg\bin" + + - name: Run install + update E2E + shell: pwsh + run: | + tests/install/windows/install-update-e2e.ps1 ` + -Route '${{ inputs.route }}' ` + -InstallerUrl '${{ inputs.installer-url }}' + env: + # Outside the workspace on purpose: the driver refuses to run on a + # dirty tree, and logs written into the repo would be what makes it + # dirty. + HERMES_E2E_LOG_DIR: ${{ runner.temp }}\e2e-logs + + # The installer's own transcript, the AHK click log, the update hand-off + # log, and the screen recording say far more than the assertion that + # tripped when a real install breaks. + - name: Upload logs and recording + if: always() + uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1 + with: + name: install-e2e-windows-${{ inputs.route }}-${{ github.sha }} + path: ${{ runner.temp }}\e2e-logs + retention-days: 14 + if-no-files-found: ignore diff --git a/.github/workflows/install-e2e.yml b/.github/workflows/install-e2e.yml index 03c9a1d0b810..98b5edc8a051 100644 --- a/.github/workflows/install-e2e.yml +++ b/.github/workflows/install-e2e.yml @@ -12,6 +12,11 @@ name: Install & Update E2E # A hardcoded list would stop covering the newest release the day after it # ships, and would pin an "oldest" that nobody still runs. # +# A separate Windows axis (install-e2e-windows-run.yml) installs with the +# latest PUBLISHED bootstrap installer and applies the desktop app's builtin +# update — no tag matrix there, because the published exe's own build pin is +# the starting point users actually have. +# # Triggers: # * every 12 hours, so upstream drift (a new uv, a Node bump, a PyPI change) # surfaces on a schedule rather than in someone's review cycle; @@ -27,11 +32,11 @@ on: workflow_dispatch: inputs: route: - description: 'Which update route to exercise.' + description: 'Which update route to exercise. all/both include the Windows desktop leg.' required: false type: choice - default: both - options: [both, update, installer] + default: all + options: [all, both, update, installer, windows-desktop] tag-count: description: 'How many release tags to sample (newest, oldest, and a spread between).' required: false @@ -81,7 +86,7 @@ jobs: # `hermes update` -- the route most users take. update: - if: github.event_name != 'workflow_dispatch' || inputs.route != 'installer' + if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "both", "update"]'), inputs.route) needs: pick-releases strategy: # One release breaking is worth knowing about even if another already @@ -97,7 +102,7 @@ jobs: # Re-running the curl one-liner over an existing checkout: autostash + pull # rather than the updater's own git handling. installer: - if: github.event_name != 'workflow_dispatch' || inputs.route != 'update' + if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "both", "installer"]'), inputs.route) needs: pick-releases strategy: fail-fast: false @@ -108,3 +113,14 @@ jobs: with: route: installer install-ref: ${{ matrix.install-ref }} + + # Windows desktop: install with the latest PUBLISHED bootstrap installer + # (Hermes-Setup.exe — the exact bits a user downloads today), then apply the + # desktop app's builtin update. No release-tag matrix: the published exe's + # build pin decides the starting base, which is precisely the "user on the + # current installer" scenario this axis exists to cover. + windows-desktop: + if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "windows-desktop"]'), inputs.route) + uses: ./.github/workflows/install-e2e-windows-run.yml + with: + route: desktop diff --git a/tests/install/windows/install-button.png b/tests/install/windows/install-button.png new file mode 100644 index 0000000000000000000000000000000000000000..feed47bd98fb3ffe464025c6f52553205ed89751 GIT binary patch literal 1200 zcmX9;Ycv!H6rNFQVwr43op!QisWz%piB^c1W3(lqj7@2#u)7|yN|b~R&4`s_WcAQP zsV0vZjgg@envmCE!X}TXd5}p{G45q&cklTg=bm%#x#xU8F3Z!y%}{@-K7l|mbf>v` zX+1|vt9f&@TAIPD)fzF%%WXTcr@`usw$Y2A`9%>3UoZTZL_*%>CE9@k&z(Nh$*Gwu zRp@yRmmWcaH}(wAxF(`qEN(rCc@L4Ah>MwM_A6S1qJJiie*{TC^)KN;6aMuOKB!@26li|{ zM+gsFvAP`|OJGt1j&a!ZHyG~6s&=fC!nH?GBt~Hi#@xis(fH~u49LMD4nKZ|kUZ>u z11m!@?ly#7#)=k{kHfap=#dKf)!-Zt={$Jc0R=U{$VOH%q}MF_aLtUBd(%! z2u>HGUj{z!(JsR9DD;j%N)h(GgM0NeyHc>de}?3X2K%*lXN0l2ECNC2mbqBro^8-rYqMXuvo?d}atT~~FV z&z{X4YpWL&@F_u?%FRQ{mN3ic|G{AB89%ZnbA)JeLiU0i>M-Fu>Ex~WM3(2%&!99I>XmJ$I2a?r}x9-m6nNn4ome4c4RL5 zr1^=mja6vy@s^Xn{7`koQl8mNT0H13|81Z*ajz9KjN_{zXLK>_qq1+l@~QWN=OUdP zsg5KzzMQg2K7Mdr#Av#aYS~j^XmzD=K$||Pja;!0LrFb_vK(=pNU;L$tjCcGE-1PbO7Q;JLf~#Fw>(v1c6!H+q!`M{ft3BwzrCVC>@-Y;#GA!%@pONb9xbfzG zedk-q{cp%sv_N^UO&K+u$}Di$SdnZIlDyOVyASOp$#ZxP9<%$q4?4aT2l49uN267m zX*ylK{H`*whkU^Fb<@~4Yi$l(IUy+PaH}|K6X;zTA(oH^q)h?7Cu#O{dm~-#*CDu5 LJzR^oA4>WMWp}7g literal 0 HcmV?d00001 diff --git a/tests/install/windows/install-hermes-desktop.ahk b/tests/install/windows/install-hermes-desktop.ahk new file mode 100644 index 000000000000..2cd8338c32c4 --- /dev/null +++ b/tests/install/windows/install-hermes-desktop.ahk @@ -0,0 +1,137 @@ +#Requires AutoHotkey v2.0 +#SingleInstance Force + +; Drives the Hermes bootstrap installer (Hermes-Setup.exe) through a real +; install: waits for the window, clicks Install, waits for the Launch button +; to appear (install finished), then closes the window WITHOUT launching the +; app -- the E2E driver verifies the install and runs the update route itself. +; +; Args: +; 1: log file path (default ahk.log in the working dir) +; +; Button images live next to this script. They are literal screenshots of the +; installer's buttons; ImageSearch runs with *10 shade tolerance so minor +; rendering differences (ClearType, DPI rounding) still match. + +logPath := A_Args.Length >= 1 ? A_Args[1] : "ahk.log" + +Log(text) { + msg := Format("[autohotkey] {}`n", text) + ToolTip(text) + FileAppend(msg, '*') + FileAppend(msg, logPath) +} + +OnError(LogError) + +LogError(err, mode) { + Log(Format("Unhandled error: {}", err.Message)) + ExitApp(1) + return -1 ; suppress the standard error dialog +} + +SetWorkingDir(A_ScriptDir) +CoordMode("Pixel", "Screen") +CoordMode("Mouse", "Screen") + +ClickWithMarker(x, y, button := "Left") { + Click(x, y, button) + + Sleep(10) + MouseMove(30, 30) + Log(Format("Clicking at {1}, {2}", x, y)) + ; Draw a short-lived red dot where we clicked so the screen recording + ; shows WHERE the automation acted, not just what happened after. + size := 20 + g := Gui("-Caption +AlwaysOnTop +ToolWindow") + g.BackColor := "Red" + g.Show(Format( + "x{} y{} w{} h{} NoActivate" + , x - size // 2 + , y - size // 2 + , size + , size + )) + hRegion := DllCall( + "CreateEllipticRgn" + , "Int", 0 + , "Int", 0 + , "Int", size + , "Int", size + , "Ptr" + ) + DllCall("SetWindowRgn", "Ptr", g.Hwnd, "Ptr", hRegion, "Int", true) + WinSetTransparent(255, g.Hwnd) + SetTimer(() => g.Destroy(), -500) +} + +FindImageInWindow(winTitle, imageFile, &outX, &outY, timeoutMs := 10000, intervalMs := 250) +{ + WinGetPos(&wx, &wy, &ww, &wh, winTitle) + + hBitmap := LoadPicture(imageFile) + + if !hBitmap { + throw Error("LoadPicture failed: " imageFile) + } + bm := Buffer(32, 0) ; BITMAP structure on x64 + DllCall("GetObject", "Ptr", hBitmap, "Int", bm.Size, "Ptr", bm) + + width := NumGet(bm, 4, "Int") + height := NumGet(bm, 8, "Int") + + startTime := A_TickCount + + timeLeft := 1 + + Log(Format("Searching for button file {} in window {}...", imageFile, winTitle)) + searchImage := Format("*10 {}", imageFile) + while (timeLeft > 0) + { + if ImageSearch(&x, &y, wx, wy, wx + ww, wy + wh, searchImage) + { + outX := x + Floor(width / 2) + outY := y + Floor(height / 2) + Log("Found button!") + return + } + + Sleep intervalMs + timeLeft := timeoutMs - (A_TickCount - startTime) + ToolTip(Format("Searching for button {} in window {}... {}s left", imageFile, winTitle, Round(timeLeft / 1000, 2))) + } + + throw Error(Format("Failed to find button {} in window {}", imageFile, winTitle)) +} + +ClickCenterOfImageInWindow(winTitle, imageFile, timeoutMs := 10000, intervalMs := 250) +{ + FindImageInWindow(winTitle, imageFile, &x, &y, timeoutMs, intervalMs) + ClickWithMarker(x, y) +} + +Log("Waiting for the installer window to appear...") +winTitle := "Hermes" +try { + WinWait(winTitle, , 30) +} catch { + throw Error("Hermes installer window did not appear within 30s") +} +WinGetPos(&x, &y, &w, &h, winTitle) +Log(Format("Window found at x={1} y={2} w={3} h={4}", x, y, w, h)) + +ClickCenterOfImageInWindow(winTitle, A_ScriptDir "\install-button.png") + +; Wait for the install to finish. The Launch button only renders when every +; stage (git, uv, Python, Node, venv, desktop build) has completed, so its +; appearance IS the success signal. A real install takes many minutes. +FindImageInWindow(winTitle, A_ScriptDir "\launch-button.png", &launchX, &launchY, 1000 * 60 * 25) + +; Close instead of clicking Launch: the E2E driver owns everything after the +; install, and a launched desktop would hold the venv shim open and block the +; update route. +WinClose(winTitle) + +Sleep(2000) + +ExitApp(0) diff --git a/tests/install/windows/install-update-e2e.ps1 b/tests/install/windows/install-update-e2e.ps1 new file mode 100644 index 000000000000..e7e2bec60d39 --- /dev/null +++ b/tests/install/windows/install-update-e2e.ps1 @@ -0,0 +1,383 @@ +# Prove a Windows desktop user on the published bootstrap installer can reach +# this commit through the desktop's builtin update. +# +# The Windows analog of tests/install/install-update-e2e.sh. Linux gets its +# fake GitHub from the bubblewrap sandbox's MITM proxy + git-upload-pack shim; +# there is no such sandbox on Windows, so the git proxying here is git's own +# transport rewrite instead: a throwaway GIT_CONFIG_GLOBAL carrying multi- +# valued url..insteadOf entries for BOTH hardcoded repo URL +# forms (SSH and HTTPS). Every git child of this process -- install.ps1's +# clone, `hermes update`'s fetch, the desktop's ls-remote -- resolves +# github.com/NousResearch/hermes-agent to a local bare repo whose `main` this +# driver controls, while the tools themselves run VERBATIM with their real +# URLs. scripts/fake_remote_update_probe.sh (hermes-install-update-testing +# skill) is the verified reference for the mechanism, including why --add is +# load-bearing (a second plain `git config` REPLACES the first rewrite and one +# URL silently reaches real GitHub). +# +# What one run does: +# 1. Seed fake.git from this checkout (all origin branches + tags), then +# force fake `main` to the NEWEST release tag -- so an installer with a +# branch pin lands on a released base, not on the target. +# uploadpack.allowAnySHA1InWant covers installers with a -Commit pin. +# 2. Run the real published Hermes-Setup.exe, driven by AutoHotkey (the +# installer is a GUI with no headless mode). The exe downloads its +# pinned install.ps1 from raw.githubusercontent for real -- the same +# fixture-miss passthrough posture as the Linux sandbox -- and that +# script's git clone rides the rewrite onto fake.git. +# 3. Promote fake `main` to this checkout's HEAD (the --from-main dance). +# 4. Apply ONE update route and require HEAD == target with a working +# `hermes`. +# +# Routes: +# desktop the desktop app's builtin update, minus only the Electron +# process around it, following applyUpdates' own preference +# order (apps/desktop/electron/main.ts): the repo-owned +# scripts/desktop-update.ps1 hand-off when the installed base +# ships it, else the staged hermes-setup.exe --update (which +# auto-runs: update mode is a hand-off, not a click-through). +# Either way the update engine is the installed release's own +# `hermes update`, so old CLIs meet their contemporaneous flags. +# TODO update bare `hermes update` from the installed venv (the route +# the Linux matrix calls `update`). +# TODO installer re-run the bootstrap installer over the existing +# checkout (the Linux `installer` route; needs the AHK +# flow to handle the repair/reinstall UI). +# +# Requires: git, AutoHotkey64.exe on PATH, network (real toolchain download), +# a clean full-history checkout with release tags fetched. ffmpeg on PATH is +# optional -- when present the run is screen-recorded for the artifact. + +#Requires -Version 7 + +param( + [ValidateSet("desktop")] + [string]$Route = "desktop", + # Latest published installer -- "what a user downloads today". + [string]$InstallerUrl = "https://hermes-assets.nousresearch.com/Hermes-Setup.exe", + # Local exe override (skips the download; for iterating on this driver). + [string]$InstallerPath = "" +) + +$ErrorActionPreference = "Stop" + +$RepoRoot = (Resolve-Path (Join-Path $PSScriptRoot "..\..\..")).Path +$RepoUrlSsh = "git@github.com:NousResearch/hermes-agent.git" +$RepoUrlHttps = "https://github.com/NousResearch/hermes-agent.git" + +# Everything lives OUTSIDE the checkout: an untracked dir inside the repo +# would make later verification steps lie about a dirty tree, and RUNNER_TEMP +# is wiped with the runner. +$WorkRoot = Join-Path ($env:RUNNER_TEMP ?? [System.IO.Path]::GetTempPath()) "hermes-install-e2e" +$LogDir = if ($env:HERMES_E2E_LOG_DIR) { $env:HERMES_E2E_LOG_DIR } else { Join-Path $WorkRoot "logs" } +$FakeRepo = Join-Path $WorkRoot "fake.git" + +function Step([string]$Message) { Write-Host "`n=== $Message ===" } +function Ok([string]$Message) { Write-Host " OK $Message" } +function Fail([string]$Message) { + Write-Host "FAIL: $Message" -ForegroundColor Red + exit 1 +} + +function Invoke-Git { + param([string[]]$GitArgs, [string]$Cwd = $RepoRoot) + $out = & git -C $Cwd @GitArgs 2>&1 + if ($LASTEXITCODE -ne 0) { + Fail "git $($GitArgs -join ' ') failed (exit $LASTEXITCODE): $out" + } + return ($out | Out-String).Trim() +} + +# --- preflight --------------------------------------------------------------- + +if (-not (Get-Command git -ErrorAction SilentlyContinue)) { Fail "git not on PATH" } +if (-not (Get-Command AutoHotkey64.exe -ErrorAction SilentlyContinue)) { + Fail "AutoHotkey64.exe not on PATH (winget install AutoHotkey.AutoHotkey)" +} + +# The promote step pushes this worktree's HEAD as the update target, so a +# dirty tree means the tested commit is not the commit anyone can review. +$dirty = & git -C $RepoRoot status --porcelain +if ($dirty) { + Write-Host ($dirty | Out-String) + Fail "working tree is dirty; the update target must be a real commit" +} + +Remove-Item -Recurse -Force $WorkRoot -ErrorAction SilentlyContinue +New-Item -ItemType Directory -Force -Path $WorkRoot, $LogDir | Out-Null + +# Isolated HERMES_HOME so the real install never touches the runner's (or a +# developer's) profile. Both install.ps1 and the Tauri installer honor it. +if (-not $env:HERMES_HOME) { + $env:HERMES_HOME = Join-Path $WorkRoot "hermes-home" +} +$InstallRoot = Join-Path $env:HERMES_HOME "hermes-agent" +$TargetSha = Invoke-Git @("rev-parse", "HEAD") + +# --- fake GitHub ------------------------------------------------------------- + +Step "seeding fake remote at $FakeRepo" +Invoke-Git @("init", "--bare", "--initial-branch=main", $FakeRepo) $WorkRoot | Out-Null +# Published installers carry a -Commit pin and fetch that raw SHA; a bare +# repo refuses SHA wants unless told otherwise. +Invoke-Git @("config", "uploadpack.allowAnySHA1InWant", "true") $FakeRepo | Out-Null + +# All origin branches + tags: the installer's build pin may be any commit on +# any branch that existed when the exe was built. +Invoke-Git @("push", "--quiet", $FakeRepo, "refs/remotes/origin/*:refs/heads/*") +Invoke-Git @("push", "--quiet", "--force", $FakeRepo, "refs/tags/*:refs/tags/*") + +# Fake main starts at the newest release tag: a released base a real user +# could be installed on, and never the update target itself. Major capped at +# three digits, matching _parse_release_tag (hermes_cli/update_cmd.py) and +# latestReleaseFromLsRemote (apps/desktop/electron/bundled-runtime.ts): the +# repo's historical CalVer tags (v2026.7.20) would otherwise win every +# numeric sort forever. +$releaseTags = @(& git -C $RepoRoot tag --list | + Where-Object { $_ -match '^v\d{1,3}\.\d+\.\d+(\.\d+)?$' } | + Sort-Object { [version]($_.Substring(1)) }) +if ($releaseTags.Count -eq 0) { + Fail "no release tags in this checkout -- fetch with tags (fetch-depth: 0 + fetch-tags)" +} +$newestTag = $releaseTags[-1] +$baseMainSha = Invoke-Git @("rev-parse", "$newestTag^{commit}") +Invoke-Git @("push", "--quiet", "--force", $FakeRepo, "${baseMainSha}:refs/heads/main") +Ok "fake main = $newestTag ($($baseMainSha.Substring(0,12))); target is $($TargetSha.Substring(0,12))" + +# --- git transport rewrite --------------------------------------------------- + +Step "redirecting github.com/NousResearch/hermes-agent to the fake remote" +# Process-scoped global config: every git spawned below this point (installer, +# hermes update, desktop hand-off) inherits it; nothing on the machine does. +$env:GIT_CONFIG_GLOBAL = Join-Path $WorkRoot "gitconfig" +Set-Content -Path $env:GIT_CONFIG_GLOBAL -Value "" -NoNewline +# Fail loudly if anything still reaches a URL that wants credentials. +$env:GIT_TERMINAL_PROMPT = "0" + +$fakeUrl = "file:///" + $FakeRepo.Replace("\", "/") +foreach ($url in @($RepoUrlSsh, $RepoUrlHttps)) { + Invoke-Git @("config", "--global", "--add", "url.$fakeUrl.insteadOf", $url) $WorkRoot +} +$rewrites = @(& git config --global --get-all "url.$fakeUrl.insteadOf") +if ($rewrites.Count -ne 2) { + Fail "expected 2 insteadOf rewrites, got $($rewrites.Count) -- one URL would reach real GitHub" +} +Ok "both repo URL forms rewritten (SSH clone attempts ride the file transport)" + +# --- fetch the installer ----------------------------------------------------- + +if (-not $InstallerPath) { + Step "downloading published installer" + $InstallerPath = Join-Path $WorkRoot "Hermes-Setup.exe" + Invoke-WebRequest -Uri $InstallerUrl -OutFile $InstallerPath +} +Ok "installer: $InstallerPath ($([math]::Round((Get-Item $InstallerPath).Length / 1MB, 1)) MB)" + +# --- screen recording (optional) ---------------------------------------------- + +# ffmpeg must be started, fed, and stopped from THIS process: the graceful +# stop is the character 'q' on its LIVE stdin pipe, which only +# System.Diagnostics.Process exposes (Start-Process -RedirectStandardInput +# hands it a file handle already at EOF). +$ffmpeg = $null +if (Get-Command ffmpeg -ErrorAction SilentlyContinue) { + $psi = New-Object System.Diagnostics.ProcessStartInfo + $psi.FileName = "ffmpeg" + $psi.Arguments = "-y -f gdigrab -framerate 15 -i desktop " + + "-hide_banner -loglevel error " + + "-c:v libx264 -preset ultrafast -pix_fmt yuv420p `"$LogDir\recording.mkv`"" + $psi.RedirectStandardInput = $true + $psi.UseShellExecute = $false + $ffmpeg = [System.Diagnostics.Process]::Start($psi) + Ok "screen recording started (pid $($ffmpeg.Id))" +} else { + Write-Host " (ffmpeg not on PATH; skipping screen recording)" +} + +function Stop-Recording { + if ($script:ffmpeg -and -not $script:ffmpeg.HasExited) { + try { + $script:ffmpeg.StandardInput.Write("q") + $script:ffmpeg.StandardInput.Close() + } catch {} + if (-not $script:ffmpeg.WaitForExit(15000)) { $script:ffmpeg.Kill() } + } +} + +# --- run the real installer under AutoHotkey ---------------------------------- + +Step "installing via Hermes-Setup.exe (real toolchains: git, uv, Python, Node, venv, desktop)" +$installerOk = $false +try { + $proc = Start-Process -FilePath $InstallerPath -PassThru + $ahkLog = Join-Path $LogDir "ahk.log" + $ahkProc = Start-Process -FilePath "AutoHotkey64.exe" ` + -ArgumentList "`"$PSScriptRoot\install-hermes-desktop.ahk`"", "`"$ahkLog`"" -PassThru + + # Tail the bootstrap log into the job log while we wait: the install IS + # the substance of this test, and a failure explanation should not need + # an artifact download. FileShare.ReadWrite because the installer still + # has the file open for writing. + $logReader = $null + $logStream = $null + $bootstrapLog = Join-Path $env:HERMES_HOME "logs\bootstrap-installer.log" + $deadline = (Get-Date).AddMinutes(30) + try { + while ((Get-Date) -lt $deadline -and -not $ahkProc.HasExited) { + if (-not $logReader) { + if (Test-Path $bootstrapLog) { + $logStream = [System.IO.File]::Open($bootstrapLog, 'Open', 'Read', 'ReadWrite') + $logReader = New-Object System.IO.StreamReader($logStream) + } + } else { + $line = $logReader.ReadLine() + while ($null -ne $line) { + Write-Host "[bootstrap] $line" + $line = $logReader.ReadLine() + } + } + Start-Sleep -Milliseconds 500 + } + # Drain what was written in the final tick. + if ($logReader) { + $line = $logReader.ReadLine() + while ($null -ne $line) { + Write-Host "[bootstrap] $line" + $line = $logReader.ReadLine() + } + } + } finally { + if ($logReader) { $logReader.Dispose() } + if ($logStream) { $logStream.Dispose() } + } + + if (-not $ahkProc.HasExited) { + Stop-Process -Id $ahkProc.Id -Force -ErrorAction SilentlyContinue + Fail "AutoHotkey helper still running at the deadline -- install never finished. See ahk.log + recording." + } + if ($ahkProc.ExitCode -ne 0) { + Fail "AutoHotkey helper failed (exit $($ahkProc.ExitCode)) -- see ahk.log + recording" + } + + # The AHK helper closes the window after the Launch button appears; a + # still-running installer means the close did not land. + if (-not $proc.WaitForExit(30000)) { + Stop-Process -Id $proc.Id -Force -ErrorAction SilentlyContinue + Fail "installer process still running after the window was closed" + } + $installerOk = $true +} finally { + Stop-Recording + if (Test-Path (Join-Path $LogDir "ahk.log")) { + Write-Host "--- ahk.log ---" + Get-Content (Join-Path $LogDir "ahk.log") | ForEach-Object { Write-Host $_ } + Write-Host "--- end ahk.log ---" + } + if (-not $installerOk -and (Test-Path (Join-Path $env:HERMES_HOME "logs\bootstrap-installer.log"))) { + Copy-Item (Join-Path $env:HERMES_HOME "logs\bootstrap-installer.log") $LogDir -Force + } +} + +# --- verify the install ------------------------------------------------------ + +Step "verifying the installed checkout" +if (-not (Test-Path (Join-Path $InstallRoot ".git"))) { + Fail "no git checkout at $InstallRoot -- the installer's clone did not ride the rewrite?" +} +$BaseSha = Invoke-Git @("rev-parse", "HEAD") $InstallRoot +if ($BaseSha -eq $TargetSha) { + Fail "install landed on the update target ($BaseSha); base and target must differ" +} +Ok "installed $($BaseSha.Substring(0,12)); update target is $($TargetSha.Substring(0,12))" + +$HermesExe = Join-Path $InstallRoot "venv\Scripts\hermes.exe" +if (-not (Test-Path $HermesExe)) { Fail "venv shim missing: $HermesExe" } + +# The real smoke test: goes through the venv launcher and imports the app. +$version = & $HermesExe --version 2>&1 +if ($LASTEXITCODE -ne 0) { Fail "hermes --version failed after install: $version" } +Write-Host " $version" +Ok "hermes runs after install" + +# --- promote fake main to this checkout -------------------------------------- + +Step "promoting fake main to this checkout (the state a user sees when an update is waiting)" +Invoke-Git @("push", "--quiet", "--force", $FakeRepo, "HEAD:refs/heads/main") +Ok "fake main advanced to $($TargetSha.Substring(0,12))" + +# --- apply exactly one update route ------------------------------------------- + +switch ($Route) { + "desktop" { + Step "ROUTE: desktop builtin update" + # Mirror the PATH contract the desktop passes the hand-off + # (pathWithHermesManagedNode in apps/desktop/electron/main.ts): + # managed node first, then the venv scripts dir. + $managedNode = Join-Path $env:HERMES_HOME "node" + $env:PATH = ((@( + $managedNode, + (Join-Path $managedNode "bin"), + (Join-Path $InstallRoot "venv\Scripts") + ) | Where-Object { Test-Path $_ }) + @($env:PATH)) -join ";" + + # applyUpdates' preference order: the repo-owned hand-off script when + # the INSTALLED checkout ships it, else the staged Tauri binary. Run + # whichever the Update button would actually spawn against this base. + $handoff = Join-Path $InstallRoot "scripts\desktop-update.ps1" + $stagedExe = Join-Path $env:HERMES_HOME "hermes-setup.exe" + + if (Test-Path $handoff) { + # powershell.exe (5.1), not pwsh: that is what the desktop spawns. + # -NoUi is the script's own headless switch; -DesktopPid 0 skips + # the wait-for-desktop gate (no desktop is running); no + # -RelaunchExe so nothing is launched afterwards. + $handoffLog = Join-Path $LogDir "desktop-update.log" + & powershell -NoProfile -ExecutionPolicy Bypass -File $handoff ` + -InstallRoot $InstallRoot -Branch main -DesktopPid 0 -NoUi 2>&1 | + Tee-Object -FilePath $handoffLog + $handoffExit = $LASTEXITCODE + + $handoffInternalLog = Join-Path $env:HERMES_HOME "logs\desktop-update-handoff.log" + if (Test-Path $handoffInternalLog) { Copy-Item $handoffInternalLog $LogDir -Force } + if ($handoffExit -ne 0) { + Fail "desktop-update.ps1 failed (exit $handoffExit) -- see desktop-update.log + desktop-update-handoff.log" + } + } elseif (Test-Path $stagedExe) { + # Update mode is a hand-off, not a click-through: --update jumps + # straight to progress and start_update runs unattended, exiting + # when done -- no AHK needed. On success it auto-launches the + # desktop; kill that below rather than letting it hold the venv. + Write-Host " installed base predates desktop-update.ps1; using staged hermes-setup.exe --update" + $upd = Start-Process -FilePath $stagedExe -ArgumentList "--update" -PassThru + if (-not $upd.WaitForExit(45 * 60 * 1000)) { + Stop-Process -Id $upd.Id -Force -ErrorAction SilentlyContinue + Fail "hermes-setup.exe --update still running after 45 minutes" + } + $updLog = Join-Path $env:HERMES_HOME "logs\update.log" + if (Test-Path $updLog) { Copy-Item $updLog $LogDir -Force } + if ($upd.ExitCode -ne 0) { + Fail "hermes-setup.exe --update failed (exit $($upd.ExitCode)) -- see update.log" + } + # The successful updater relaunches Hermes; a live desktop locks + # the venv shim and would poison later assertions. + Get-Process -Name "Hermes" -ErrorAction SilentlyContinue | Stop-Process -Force -ErrorAction SilentlyContinue + } else { + Fail "neither scripts/desktop-update.ps1 (in the installed base) nor a staged hermes-setup.exe exists -- no desktop update path to exercise" + } + + $After = Invoke-Git @("rev-parse", "HEAD") $InstallRoot + if ($After -ne $TargetSha) { + Fail "desktop update left HEAD at $After, wanted $TargetSha" + } + Ok "desktop update landed on $($After.Substring(0,12))" + + $version = & $HermesExe --version 2>&1 + if ($LASTEXITCODE -ne 0) { Fail "hermes --version failed after update: $version" } + Write-Host " $version" + Ok "hermes runs after desktop update" + } +} + +Write-Host "" +Write-Host "PASS: Windows install/update E2E (route: $Route, base: $($BaseSha.Substring(0,12)) -> $($TargetSha.Substring(0,12)))" -ForegroundColor Green +exit 0 diff --git a/tests/install/windows/launch-button.png b/tests/install/windows/launch-button.png new file mode 100644 index 0000000000000000000000000000000000000000..6ab89a75ca32480987264e3ee359063cb11f6446 GIT binary patch literal 1485 zcmWkuc{G#@9Q{-xm5TO7#jA*VQxr+Ba?sdHQkJLX*++UaDn=?Klp@|JDU((qktI~3 zv1A|XC}SBzgc(T^!}t5md-Lu&_ug~=y64RZ1a)Ow#)W*urR5B%fUWTV`ODX4Z zg&TGbNRLIMRRmgJ!zUcHkH#&&xXufAT|kR)T%GuQJaD7{bV(t+;a&TNx10@+TFx2V)9cjjE;j(B63<_LIP1)@T>$412LbA zQQ7$U3p5JIvR2Sx;Mgyy?I3?nz%&UvnRqk`Jrgi44|+wgnT|XB@oGA1{*ApNvak_W zxuelVX#p30jKkD4RJDo<5* z4E#PApMbVL(DjqPZGiX!=pP2*5U6{jS*WzRo2=^q|0kGSii0B%n+pwH&^-t#Wl+;j z7Sv;E8Azs}vJC>0AuJu)g)lM(lM=|N#1bAP6~THMHVeszMZkQDL*K#m4hAM;S~=K6 z%666(H)CZRc_$xkWZ|z#SvC33Ylz8#lPsV=z=08H7D8h;zO2CaO?W#WPO$J+4qQ%= zo_mCOb!cz_)jaW46|@LRb|DV`fH(CJl!E*|vTFcn_ki08nU&DdPi{SnjRLT}flb}y zg=ZL(3mV>7+6rc&m{TJ?&cfCn=~^0UoJPSQ&I?>OLJ%`;%uV*&1+c%hounQ)yjo%O zXD;`i$MdaAmhd;WY(L{|n#obK(=)zdd)MLK;qg0}3+xl2nFoh=`XSGw=qvE@OaAybf+m>9iCpe?z~FB-U5aQV$ZY0)I%;u}Ka_%!u| zgwd|=r_5>Yc|)n!t1$7&iZA_9kr0tEZ(u>mmQUNc6x9*>z6lL0?YeS*#n>`Ri;`~7 z{29HJXi8#}Jo}C#$1!m#Dpy0lt7KQ%a?hQeDqUX)udmInv5t&-Uar!-4!;I1ufV!5 zM%RaR$J89!O&3Ur(ig$fM}#mgsAJi_E3wNvUgIJ5zY=w0F@)Q!3wpKIZO+!z?#TO1 zOvQ{T&7j)torda1B&`~+ulN}0mWI#Ni(ncm z@3wr5UVre8OP#8|f_}JHR;{MUc|=bqk-`IEAqm2l*Y3d#YFO z*`i2KZOHXk*)%po-rgbhrAi)^$CP&&?LDvfO2pWs+I^nBn=0`hJ{-N_g}(agF7fO} zVahkOK}K!TDtgF-I}zC^nSyIe%we(5@#sIZ}Frpd1=*Iqvkxj4W(|uPZm-w z!)`sk7cb1|F>~=y?)-Yqv#vZj&nIa01^HT0`IVt%V;g;G1^K*MZcnf1*sK%k!4Mq)=rnhkV|2X<_Pn&Y8)$Fv{Z=$k)A563t&_XmW^%wh_x4v*HR=XCSv1^LA z!&B*VJpagR>xAGj_O^rO%ebOHVwQPZGICPO6xj975up! zEbWq9o zi|RJRl|2PUl|D-upFHF|&$Qig{bj72ewS0j+@3*+V;PKj)HENle|ObMH8UjO`TC8v h%6~e3;57(9p257kTi97NOLobGx#>ZZOyg6b{{c0JSjYeX literal 0 HcmV?d00001 From f5cd0b7a6da25c5f4cf445cad2795e6d64153661 Mon Sep 17 00:00:00 2001 From: ethernet Date: Mon, 10 Aug 2026 19:37:55 -0400 Subject: [PATCH 002/227] ci(windows): install test tools under RUNNER_TEMP - untracked dirs in the checkout trip the driver's dirty-tree guard --- .github/workflows/install-e2e-windows-run.yml | 18 +++++++++++------- 1 file changed, 11 insertions(+), 7 deletions(-) diff --git a/.github/workflows/install-e2e-windows-run.yml b/.github/workflows/install-e2e-windows-run.yml index 50b731ab2b8b..962878f9b1fd 100644 --- a/.github/workflows/install-e2e-windows-run.yml +++ b/.github/workflows/install-e2e-windows-run.yml @@ -60,16 +60,20 @@ jobs: id: test-tools-cache uses: actions/cache@27d5ce7f107fe9357f9df03efb73ab90386fccae # v5 with: - path: test-bins + path: ${{ runner.temp }}\test-bins key: test-bins-${{ runner.os }}-v1 # AutoHotkey drives the installer GUI; ffmpeg records the screen so a # failed click is diagnosable from the artifact instead of by guesswork. + # Everything lands under RUNNER_TEMP: an untracked dir inside the + # checkout makes the tree dirty, and the driver refuses a dirty tree + # (the update target must be a reviewable commit). - name: Install AutoHotkey v2 and ffmpeg if: steps.test-tools-cache.outputs.cache-hit != 'true' shell: pwsh run: | - New-Item -ItemType Directory -Path test-bins\autohotkey, test-bins\ffmpeg -Force | Out-Null + $bins = "$env:RUNNER_TEMP\test-bins" + New-Item -ItemType Directory -Path $bins\autohotkey, $bins\ffmpeg -Force | Out-Null # AutoHotkey: copy its whole v2 directory so helper exes/dlls come along. winget install -e --id AutoHotkey.AutoHotkey --silent --accept-source-agreements --accept-package-agreements --disable-interactivity @@ -77,16 +81,16 @@ jobs: if (-not (Test-Path $ahkDir)) { throw "AutoHotkey install directory not found: $ahkDir" } - Copy-Item -Path "$ahkDir\*" -Destination test-bins\autohotkey -Recurse -Force + Copy-Item -Path "$ahkDir\*" -Destination $bins\autohotkey -Recurse -Force - winget install -e --id Gyan.FFmpeg --silent --accept-source-agreements --accept-package-agreements --disable-interactivity --location ffmpeg_dir - Copy-Item -Path "ffmpeg_dir\*\*" -Destination test-bins\ffmpeg -Recurse -Force + winget install -e --id Gyan.FFmpeg --silent --accept-source-agreements --accept-package-agreements --disable-interactivity --location "$env:RUNNER_TEMP\ffmpeg_dir" + Copy-Item -Path "$env:RUNNER_TEMP\ffmpeg_dir\*\*" -Destination $bins\ffmpeg -Recurse -Force - name: Add test tools to PATH shell: pwsh run: | - Add-Content -Path $env:GITHUB_PATH -Value "$PWD\test-bins\autohotkey" - Add-Content -Path $env:GITHUB_PATH -Value "$PWD\test-bins\ffmpeg\bin" + Add-Content -Path $env:GITHUB_PATH -Value "$env:RUNNER_TEMP\test-bins\autohotkey" + Add-Content -Path $env:GITHUB_PATH -Value "$env:RUNNER_TEMP\test-bins\ffmpeg\bin" - name: Run install + update E2E shell: pwsh From b685efed678d9f3be6bd56b3401256eed05e0229 Mon Sep 17 00:00:00 2001 From: ethernet Date: Mon, 10 Aug 2026 20:12:37 -0400 Subject: [PATCH 003/227] fix(windows-e2e): AHK helper wedged on invalid stdout handle before clicking Install AutoHotkey64 is a GUI-subsystem exe: spawned without -NoNewWindow it has no console, FileAppend('*') throws '(6) The handle is invalid' on the first Log call, and OnError's own Log rethrows inside the handler - the script hangs with the error tooltip painted over the installer and Install is never clicked (confirmed from the run 31443096241 screen recording; ahk.log was never created because the stdout write preceded the file write). Wrap the stdout append in try (the log file is the record) and spawn the helper with -NoNewWindow so its live lines reach the job log. --- tests/install/windows/install-hermes-desktop.ahk | 8 +++++++- tests/install/windows/install-update-e2e.ps1 | 4 +++- 2 files changed, 10 insertions(+), 2 deletions(-) diff --git a/tests/install/windows/install-hermes-desktop.ahk b/tests/install/windows/install-hermes-desktop.ahk index 2cd8338c32c4..dedf4ce752ef 100644 --- a/tests/install/windows/install-hermes-desktop.ahk +++ b/tests/install/windows/install-hermes-desktop.ahk @@ -18,7 +18,13 @@ logPath := A_Args.Length >= 1 ? A_Args[1] : "ahk.log" Log(text) { msg := Format("[autohotkey] {}`n", text) ToolTip(text) - FileAppend(msg, '*') + ; stdout only exists when the launcher attached a console (Start-Process + ; -NoNewWindow). AutoHotkey64 is a GUI-subsystem exe, so a bare spawn has + ; an invalid stdout handle and FileAppend('*') throws "(6) The handle is + ; invalid" -- recursively, from inside OnError's own Log call, which + ; wedges the script instead of exiting. The log FILE is the record; + ; stdout is best-effort. + try FileAppend(msg, '*') FileAppend(msg, logPath) } diff --git a/tests/install/windows/install-update-e2e.ps1 b/tests/install/windows/install-update-e2e.ps1 index e7e2bec60d39..2b4884e5744b 100644 --- a/tests/install/windows/install-update-e2e.ps1 +++ b/tests/install/windows/install-update-e2e.ps1 @@ -211,7 +211,9 @@ $installerOk = $false try { $proc = Start-Process -FilePath $InstallerPath -PassThru $ahkLog = Join-Path $LogDir "ahk.log" - $ahkProc = Start-Process -FilePath "AutoHotkey64.exe" ` + # -NoNewWindow attaches our console as the GUI-subsystem exe's stdout so + # the helper's live lines land in the job log as they happen. + $ahkProc = Start-Process -FilePath "AutoHotkey64.exe" -NoNewWindow ` -ArgumentList "`"$PSScriptRoot\install-hermes-desktop.ahk`"", "`"$ahkLog`"" -PassThru # Tail the bootstrap log into the job log while we wait: the install IS From 746aca265472f582869248ffd3c9922d36972ba2 Mon Sep 17 00:00:00 2001 From: ethernet Date: Mon, 10 Aug 2026 20:24:00 -0400 Subject: [PATCH 004/227] fix(windows-e2e): button images matched a dev build, not the published installer Run 31445244722's recording shows the published Hermes-Setup.exe renders '[ INSTALL ]' as flat blue text on off-white - nothing like the solid-blue 'Install Hermes ->' reference from the dev-build era, so ImageSearch never matched. Replace the reference with a crop of the real button taken from that recording (tolerance *60 to absorb H.264 drift, click retried across animation frames), and drop launch-button.png entirely: completion now polls the installer's own bootstrap-complete marker (.hermes-bootstrap-complete, see paths.rs likely_bootstrap_marker), which cannot go stale with a UI restyle. --- tests/install/windows/install-button.png | Bin 1200 -> 2034 bytes .../windows/install-hermes-desktop.ahk | 75 +++++++++++++----- tests/install/windows/install-update-e2e.ps1 | 5 +- tests/install/windows/launch-button.png | Bin 1485 -> 0 bytes 4 files changed, 59 insertions(+), 21 deletions(-) delete mode 100644 tests/install/windows/launch-button.png diff --git a/tests/install/windows/install-button.png b/tests/install/windows/install-button.png index feed47bd98fb3ffe464025c6f52553205ed89751..4a7705fcca7b8796da06004de2d341113ffbec8c 100644 GIT binary patch literal 2034 zcmVSpMy{*h-<2qLwslX@qWxjaW^nkyH~*p-qVP$w+_)QX2wd1Rn5- z7!wmnhza4PA}WaqK@4hYv+}2@$?|05Q_qQ@TmuaIF=zlwa|2?!RYZ-0IT1K0)meHoHWwa@48Ewj1Mr^#A zGgV~{+&LEjrwWLK_x+@xz-dsW3}%>MMmJRvF^@osxD#^*vIG$XbN4O_?#z|Q83A{? z6!)6vnE;>}Kom1N1$U67s!1HkIRItJ?j9_-HVMcMva;8yA*+UQk>;v*r zRWqBLMul-XgUpV&a(a9 zW0&uI@(Vj3y#BVw$0jF}U}A)P=oNPC`r(=zhc4NEXyXl!U$y=5y}$Zh1%y&g;&hCi z@GEYeTD4>3$>-jll9<&jxLd&EPY*Ki?fpX|Z_E{pn?)x&POhfPwtL5R{Unu)k}Z`M z?xGsIdVKRWR}5TrS=4g?)dlurlPLxF#R0e@DvSNS=QMWjzH-;DE5G%FsV!T2@A}4F z?x+_)M9l1&XQp;UTOUD$;Py{XvRfwgOS^wB$?`YZR`^XKp0 zvVGGgF1AKgu0f6+JJ#3t*;T80pMCbpO`EPO;xV%rg9v7(szo}5(28JEN`;5w?NTzU zMGqn(g1f6~vtFB2CYV`$1LT9~s-1>NSe00n-U>EW7FHx%nN3X8RUC27TXFI9oc-*N zvtxQ*C5tI0$s~I9rWu=MW zkGEqkTh_gH?Xp9M4$aJ*vEJRywr$&nhlh9VI&kZ)4^B=Z!qVmC44gjAOE0}NFwo!M z-+S`p$>YZnk(~c?4;S=3Z=1g5uK0&=YHFrQEvVs|TvIhzH5dYtE=mV){t5@)xbU|} za5Yy<)I?X<%-61zp}iZwdDZK`pV+eb=HcHwd-CmuPwz1AFfckp>JAAlA`>&Q7sog< zebHIV=vmH0?0@<&B+6P+R8bdPQXmBgG9P5#T+KBrSt+tBTPj;xGNtCM1!S>hT>OEL z8f(_DVZ)~e2Tz>Wv{>t9&h58fzxp-qyKi)KbQG5TD-Nfo7#|PC+2`$OfdUd?#bLkQEM3wT=rW=~?LM%G+>L1(~xLE|N~#Oe7`nlI6v4RDJhJpZS6 zui5Z@;rqX2-@aGhn7p*FFLWdB9*r`Uzw0N<$~j&Z;ew(dq3+oz8ywX#ureV+ zDLSAz14abOjtB`cvw0x}vYR)Z#{y+VcQCUM7A6{ZXNl7@GZA^pw z?fv!c-Mg>dm_lv*K}6iyuz_{!*1i7P2@#4v&CGEB{SQ6&N_O|@&hgETsTG@i;<(20rD2VboY z44l6_0z_GQdI3XKnKRr83XmNUIu&=9EcIg`=Nv+qS5zsbEF#sQIb%xBbQKy%ImR$A z0MP-mVn(0{T8g~fi*@e`06B|09uNV`K-EHwoHjzJ-AT$`yTiD9Rf9W4aViNp(^1N1 z7JcEYOp06-_{TfE{POhX_2K;U&o1X08)pm|A~HKWyKmo{hY!EHpt2=U&{5n0wC*Nn za?VY&K0u5irPR=!h!k^w@e>Mtg+*I6%9E;?oe@b0QXid~jRgn+_nKO3_mg72s$D`F znIb}rwUCQ{q{x7jm{)@|XXEy3eu@4-|YTB_58nc`f+5w0b zfuxDLzG}mcuwZZHhPPlwLp6o7;ps^Tf&t=%at&}gsH>*- z+@j{QP^C4Z?Jxt9F09a2*}nm4{~xnev{lwJ+LX17Hf1fNOi_@%07*qoM6N<$f@+52<^TWy literal 1200 zcmX9;Ycv!H6rNFQVwr43op!QisWz%piB^c1W3(lqj7@2#u)7|yN|b~R&4`s_WcAQP zsV0vZjgg@envmCE!X}TXd5}p{G45q&cklTg=bm%#x#xU8F3Z!y%}{@-K7l|mbf>v` zX+1|vt9f&@TAIPD)fzF%%WXTcr@`usw$Y2A`9%>3UoZTZL_*%>CE9@k&z(Nh$*Gwu zRp@yRmmWcaH}(wAxF(`qEN(rCc@L4Ah>MwM_A6S1qJJiie*{TC^)KN;6aMuOKB!@26li|{ zM+gsFvAP`|OJGt1j&a!ZHyG~6s&=fC!nH?GBt~Hi#@xis(fH~u49LMD4nKZ|kUZ>u z11m!@?ly#7#)=k{kHfap=#dKf)!-Zt={$Jc0R=U{$VOH%q}MF_aLtUBd(%! z2u>HGUj{z!(JsR9DD;j%N)h(GgM0NeyHc>de}?3X2K%*lXN0l2ECNC2mbqBro^8-rYqMXuvo?d}atT~~FV z&z{X4YpWL&@F_u?%FRQ{mN3ic|G{AB89%ZnbA)JeLiU0i>M-Fu>Ex~WM3(2%&!99I>XmJ$I2a?r}x9-m6nNn4ome4c4RL5 zr1^=mja6vy@s^Xn{7`koQl8mNT0H13|81Z*ajz9KjN_{zXLK>_qq1+l@~QWN=OUdP zsg5KzzMQg2K7Mdr#Av#aYS~j^XmzD=K$||Pja;!0LrFb_vK(=pNU;L$tjCcGE-1PbO7Q;JLf~#Fw>(v1c6!H+q!`M{ft3BwzrCVC>@-Y;#GA!%@pONb9xbfzG zedk-q{cp%sv_N^UO&K+u$}Di$SdnZIlDyOVyASOp$#ZxP9<%$q4?4aT2l49uN267m zX*ylK{H`*whkU^Fb<@~4Yi$l(IUy+PaH}|K6X;zTA(oH^q)h?7Cu#O{dm~-#*CDu5 LJzR^oA4>WMWp}7g diff --git a/tests/install/windows/install-hermes-desktop.ahk b/tests/install/windows/install-hermes-desktop.ahk index dedf4ce752ef..eeda2f02d5d5 100644 --- a/tests/install/windows/install-hermes-desktop.ahk +++ b/tests/install/windows/install-hermes-desktop.ahk @@ -2,18 +2,25 @@ #SingleInstance Force ; Drives the Hermes bootstrap installer (Hermes-Setup.exe) through a real -; install: waits for the window, clicks Install, waits for the Launch button -; to appear (install finished), then closes the window WITHOUT launching the -; app -- the E2E driver verifies the install and runs the update route itself. +; install: waits for the window, clicks Install, waits for the +; bootstrap-complete marker file (the installer writes it after every stage +; has finished, before the Launch screen renders), then closes the window +; WITHOUT launching the app -- the E2E driver verifies the install and runs +; the update route itself. ; ; Args: ; 1: log file path (default ahk.log in the working dir) +; 2: bootstrap-complete marker path to poll for (required) ; -; Button images live next to this script. They are literal screenshots of the -; installer's buttons; ImageSearch runs with *10 shade tolerance so minor -; rendering differences (ClearType, DPI rounding) still match. +; The Install button image lives next to this script. It is a literal crop +; of the published installer's button from a CI screen recording; +; ImageSearch runs with a generous shade tolerance because the reference +; passed through video compression. Completion deliberately does NOT use a +; second button image: the marker file is the installer's own completion +; signal and cannot go stale with a UI restyle. logPath := A_Args.Length >= 1 ? A_Args[1] : "ahk.log" +markerPath := A_Args.Length >= 2 ? A_Args[2] : "" Log(text) { msg := Format("[autohotkey] {}`n", text) @@ -91,7 +98,9 @@ FindImageInWindow(winTitle, imageFile, &outX, &outY, timeoutMs := 10000, interva timeLeft := 1 Log(Format("Searching for button file {} in window {}...", imageFile, winTitle)) - searchImage := Format("*10 {}", imageFile) + ; *60: the reference crop survived H.264 video compression, so per-channel + ; drift up to ~60 shades must still count as a match. + searchImage := Format("*60 {}", imageFile) while (timeLeft > 0) { if ImageSearch(&x, &y, wx, wy, wx + ww, wy + wh, searchImage) @@ -126,18 +135,44 @@ try { WinGetPos(&x, &y, &w, &h, winTitle) Log(Format("Window found at x={1} y={2} w={3} h={4}", x, y, w, h)) -ClickCenterOfImageInWindow(winTitle, A_ScriptDir "\install-button.png") - -; Wait for the install to finish. The Launch button only renders when every -; stage (git, uv, Python, Node, venv, desktop build) has completed, so its -; appearance IS the success signal. A real install takes many minutes. -FindImageInWindow(winTitle, A_ScriptDir "\launch-button.png", &launchX, &launchY, 1000 * 60 * 25) - -; Close instead of clicking Launch: the E2E driver owns everything after the -; install, and a launched desktop would hold the venv shim open and block the -; update route. -WinClose(winTitle) +if (markerPath = "") { + throw Error("marker path argument is required") +} -Sleep(2000) +; The reference crop went through H.264 compression, so allow a wide shade +; tolerance; retry the click a few times in case the first lands during a +; window animation frame. +clicked := false +attempts := 0 +while (!clicked && attempts < 5) { + attempts += 1 + try { + ClickCenterOfImageInWindow(winTitle, A_ScriptDir "\install-button.png", 20000, 250) + clicked := true + } catch as err { + Log(Format("Install click attempt {} failed: {}", attempts, err.Message)) + Sleep(2000) + } +} +if (!clicked) { + throw Error("could not find/click the Install button after " attempts " attempts") +} -ExitApp(0) +; Wait for the installer's own completion signal: the bootstrap-complete +; marker is written after the last stage succeeds. A real install takes many +; minutes (git, uv, Python, Node, venv, desktop build). +Log(Format("Waiting for bootstrap-complete marker: {}", markerPath)) +deadline := A_TickCount + 1000 * 60 * 25 +while (A_TickCount < deadline) { + if FileExist(markerPath) { + Log("Marker found -- install complete") + ; Close instead of clicking Launch: the E2E driver owns everything + ; after the install, and a launched desktop would hold the venv shim + ; open and block the update route. + try WinClose(winTitle) + Sleep(2000) + ExitApp(0) + } + Sleep(5000) +} +throw Error("bootstrap-complete marker never appeared within 25 minutes") diff --git a/tests/install/windows/install-update-e2e.ps1 b/tests/install/windows/install-update-e2e.ps1 index 2b4884e5744b..77fbd4406e14 100644 --- a/tests/install/windows/install-update-e2e.ps1 +++ b/tests/install/windows/install-update-e2e.ps1 @@ -211,10 +211,13 @@ $installerOk = $false try { $proc = Start-Process -FilePath $InstallerPath -PassThru $ahkLog = Join-Path $LogDir "ahk.log" + # The helper polls for the installer's own completion signal instead of a + # second button screenshot (see paths.rs likely_bootstrap_marker). + $bootstrapMarker = Join-Path $InstallRoot ".hermes-bootstrap-complete" # -NoNewWindow attaches our console as the GUI-subsystem exe's stdout so # the helper's live lines land in the job log as they happen. $ahkProc = Start-Process -FilePath "AutoHotkey64.exe" -NoNewWindow ` - -ArgumentList "`"$PSScriptRoot\install-hermes-desktop.ahk`"", "`"$ahkLog`"" -PassThru + -ArgumentList "`"$PSScriptRoot\install-hermes-desktop.ahk`"", "`"$ahkLog`"", "`"$bootstrapMarker`"" -PassThru # Tail the bootstrap log into the job log while we wait: the install IS # the substance of this test, and a failure explanation should not need diff --git a/tests/install/windows/launch-button.png b/tests/install/windows/launch-button.png deleted file mode 100644 index 6ab89a75ca32480987264e3ee359063cb11f6446..0000000000000000000000000000000000000000 GIT binary patch literal 0 HcmV?d00001 literal 1485 zcmWkuc{G#@9Q{-xm5TO7#jA*VQxr+Ba?sdHQkJLX*++UaDn=?Klp@|JDU((qktI~3 zv1A|XC}SBzgc(T^!}t5md-Lu&_ug~=y64RZ1a)Ow#)W*urR5B%fUWTV`ODX4Z zg&TGbNRLIMRRmgJ!zUcHkH#&&xXufAT|kR)T%GuQJaD7{bV(t+;a&TNx10@+TFx2V)9cjjE;j(B63<_LIP1)@T>$412LbA zQQ7$U3p5JIvR2Sx;Mgyy?I3?nz%&UvnRqk`Jrgi44|+wgnT|XB@oGA1{*ApNvak_W zxuelVX#p30jKkD4RJDo<5* z4E#PApMbVL(DjqPZGiX!=pP2*5U6{jS*WzRo2=^q|0kGSii0B%n+pwH&^-t#Wl+;j z7Sv;E8Azs}vJC>0AuJu)g)lM(lM=|N#1bAP6~THMHVeszMZkQDL*K#m4hAM;S~=K6 z%666(H)CZRc_$xkWZ|z#SvC33Ylz8#lPsV=z=08H7D8h;zO2CaO?W#WPO$J+4qQ%= zo_mCOb!cz_)jaW46|@LRb|DV`fH(CJl!E*|vTFcn_ki08nU&DdPi{SnjRLT}flb}y zg=ZL(3mV>7+6rc&m{TJ?&cfCn=~^0UoJPSQ&I?>OLJ%`;%uV*&1+c%hounQ)yjo%O zXD;`i$MdaAmhd;WY(L{|n#obK(=)zdd)MLK;qg0}3+xl2nFoh=`XSGw=qvE@OaAybf+m>9iCpe?z~FB-U5aQV$ZY0)I%;u}Ka_%!u| zgwd|=r_5>Yc|)n!t1$7&iZA_9kr0tEZ(u>mmQUNc6x9*>z6lL0?YeS*#n>`Ri;`~7 z{29HJXi8#}Jo}C#$1!m#Dpy0lt7KQ%a?hQeDqUX)udmInv5t&-Uar!-4!;I1ufV!5 zM%RaR$J89!O&3Ur(ig$fM}#mgsAJi_E3wNvUgIJ5zY=w0F@)Q!3wpKIZO+!z?#TO1 zOvQ{T&7j)torda1B&`~+ulN}0mWI#Ni(ncm z@3wr5UVre8OP#8|f_}JHR;{MUc|=bqk-`IEAqm2l*Y3d#YFO z*`i2KZOHXk*)%po-rgbhrAi)^$CP&&?LDvfO2pWs+I^nBn=0`hJ{-N_g}(agF7fO} zVahkOK}K!TDtgF-I}zC^nSyIe%we(5@#sIZ}Frpd1=*Iqvkxj4W(|uPZm-w z!)`sk7cb1|F>~=y?)-Yqv#vZj&nIa01^HT0`IVt%V;g;G1^K*MZcnf1*sK%k!4Mq)=rnhkV|2X<_Pn&Y8)$Fv{Z=$k)A563t&_XmW^%wh_x4v*HR=XCSv1^LA z!&B*VJpagR>xAGj_O^rO%ebOHVwQPZGICPO6xj975up! zEbWq9o zi|RJRl|2PUl|D-upFHF|&$Qig{bj72ewS0j+@3*+V;PKj)HENle|ObMH8UjO`TC8v h%6~e3;57(9p257kTi97NOLobGx#>ZZOyg6b{{c0JSjYeX From 931a9e9d5663ea62962d61d7a731d650ff6ed266 Mon Sep 17 00:00:00 2001 From: ethernet Date: Mon, 10 Aug 2026 20:30:28 -0400 Subject: [PATCH 005/227] fix(windows-e2e): raise the installer window before each ImageSearch attempt Run 31445907233's recording shows the runner session's maximized console covering the installer for the whole run: WinWait matches by title regardless of z-order, but ImageSearch reads screen pixels, so the Install button was never visible to it. WinActivate + WinMoveTop before every attempt; run 31443096241 already proved the same reference crop renders match-ably when the window is frontmost. --- tests/install/windows/install-hermes-desktop.ahk | 7 +++++++ 1 file changed, 7 insertions(+) diff --git a/tests/install/windows/install-hermes-desktop.ahk b/tests/install/windows/install-hermes-desktop.ahk index eeda2f02d5d5..f98fd7387b42 100644 --- a/tests/install/windows/install-hermes-desktop.ahk +++ b/tests/install/windows/install-hermes-desktop.ahk @@ -147,6 +147,13 @@ attempts := 0 while (!clicked && attempts < 5) { attempts += 1 try { + ; ImageSearch reads SCREEN pixels: anything covering the installer + ; (the runner session keeps a maximized console in front) makes the + ; button invisible even though WinWait's title match succeeded. + ; Force the installer to the foreground before every attempt. + WinActivate(winTitle) + WinMoveTop(winTitle) + Sleep(500) ClickCenterOfImageInWindow(winTitle, A_ScriptDir "\install-button.png", 20000, 250) clicked := true } catch as err { From 5e6eab0cd38b8ecf6bcfc08d01d42b4f9ad7209b Mon Sep 17 00:00:00 2001 From: ethernet Date: Mon, 10 Aug 2026 20:37:29 -0400 Subject: [PATCH 006/227] fix(windows-e2e): fall back to PixelSearch for the Install button Run 31446292343 disproved the z-order theory: the recording shows the installer frontmost, red click-marker dots painting on it, and the button rendered - yet ImageSearch missed on all 5 attempts. The reference crop is the problem: it came from an H.264/yuv420p recording whose chroma subsampling smears glyph edges. Diffing the crop against run 5's OWN recording of the same screen gives max 8 shades/channel (matches easily), so the crop is video-faithful but not screen-faithful, and no tolerance fixes that reliably. Keep ImageSearch as the first try, but fall back to PixelSearch for the button text's saturated blue (~0x3B82F6, variation 90) in the window's lower half - the only blue there (the title sits in the upper third). Verify the click landed by the blue vanishing (the progress view replaces the button); retry up to 10 times. --- .../windows/install-hermes-desktop.ahk | 71 ++++++++++++++----- 1 file changed, 55 insertions(+), 16 deletions(-) diff --git a/tests/install/windows/install-hermes-desktop.ahk b/tests/install/windows/install-hermes-desktop.ahk index f98fd7387b42..5c1a9f381fc5 100644 --- a/tests/install/windows/install-hermes-desktop.ahk +++ b/tests/install/windows/install-hermes-desktop.ahk @@ -139,27 +139,66 @@ if (markerPath = "") { throw Error("marker path argument is required") } -; The reference crop went through H.264 compression, so allow a wide shade -; tolerance; retry the click a few times in case the first lands during a -; window animation frame. +; Find the Install button and click it. +; +; ImageSearch against the reference crop is attempted first, but it is +; expected to fail on a real screen: the crop came from an H.264/yuv420p +; recording, whose chroma subsampling smears the glyph edges -- the same +; crop matches a RECORDING of this screen within 8 shades/channel while +; missing the live screen entirely. The reliable path is PixelSearch for +; the saturated blue of the "[ INSTALL ]" text, which is the only blue in +; the lower half of the window (the title is in the upper third). +FindInstallClickPoint(winTitle, &outX, &outY) { + WinGetPos(&wx, &wy, &ww, &wh, winTitle) + try { + FindImageInWindow(winTitle, A_ScriptDir "\install-button.png", &outX, &outY, 3000, 250) + Log("Install button located via ImageSearch") + return true + } catch { + } + lowerY := wy + Floor(wh * 0.45) + if PixelSearch(&px, &py, wx, lowerY, wx + ww, wy + wh, 0x3B82F6, 90) { + outX := px + outY := py + Log(Format("Install button located via PixelSearch (blue text at {1}, {2})", px, py)) + return true + } + return false +} + +; Did the click land? The button's blue text vanishes when the UI flips to +; the progress view, so lingering blue in the lower half means it did not. +InstallButtonStillVisible(winTitle) { + WinGetPos(&wx, &wy, &ww, &wh, winTitle) + lowerY := wy + Floor(wh * 0.45) + return PixelSearch(&px, &py, wx, lowerY, wx + ww, wy + wh, 0x3B82F6, 90) +} + clicked := false attempts := 0 -while (!clicked && attempts < 5) { +while (!clicked && attempts < 10) { attempts += 1 - try { - ; ImageSearch reads SCREEN pixels: anything covering the installer - ; (the runner session keeps a maximized console in front) makes the - ; button invisible even though WinWait's title match succeeded. - ; Force the installer to the foreground before every attempt. - WinActivate(winTitle) - WinMoveTop(winTitle) - Sleep(500) - ClickCenterOfImageInWindow(winTitle, A_ScriptDir "\install-button.png", 20000, 250) - clicked := true - } catch as err { - Log(Format("Install click attempt {} failed: {}", attempts, err.Message)) + ; ImageSearch/PixelSearch read SCREEN pixels: anything covering the + ; installer (the runner session keeps a maximized console in front) + ; hides the button even though WinWait's title match succeeded. Force + ; the installer to the foreground before every attempt. + WinActivate(winTitle) + WinMoveTop(winTitle) + Sleep(500) + x := 0 + y := 0 + if (!FindInstallClickPoint(winTitle, &x, &y)) { + Log(Format("Install click attempt {}: button not found on screen", attempts)) Sleep(2000) + continue + } + ClickWithMarker(x, y) + Sleep(3000) + if (InstallButtonStillVisible(winTitle)) { + Log(Format("Install click attempt {}: UI did not advance; retrying", attempts)) + continue } + clicked := true } if (!clicked) { throw Error("could not find/click the Install button after " attempts " attempts") From f4243c0c7c64c9650aac92ee51e4b7cc82c417f7 Mon Sep 17 00:00:00 2001 From: ethernet Date: Mon, 10 Aug 2026 20:44:05 -0400 Subject: [PATCH 007/227] fix(windows-e2e): PixelSearch scan band clipped the blue title, not the button Run 31446691812: every attempt logged 'blue text at 220, 330' - exactly the 45%-height scan boundary, which lands inside the HERMES AGENT title (title bottom ~47% of window height; button ~62%, measured from the run 2/5 recordings). The click-landed check then correctly reported no advance, ten times. Raise the boundary to 55%, between the two. --- tests/install/windows/install-hermes-desktop.ahk | 10 ++++++---- 1 file changed, 6 insertions(+), 4 deletions(-) diff --git a/tests/install/windows/install-hermes-desktop.ahk b/tests/install/windows/install-hermes-desktop.ahk index 5c1a9f381fc5..92195e00ae10 100644 --- a/tests/install/windows/install-hermes-desktop.ahk +++ b/tests/install/windows/install-hermes-desktop.ahk @@ -146,8 +146,10 @@ if (markerPath = "") { ; recording, whose chroma subsampling smears the glyph edges -- the same ; crop matches a RECORDING of this screen within 8 shades/channel while ; missing the live screen entirely. The reliable path is PixelSearch for -; the saturated blue of the "[ INSTALL ]" text, which is the only blue in -; the lower half of the window (the title is in the upper third). +; the saturated blue of the "[ INSTALL ]" text. Scan only BELOW 55% of the +; window height: the HERMES AGENT title is blue too, and its bottom rows +; reach ~47% -- a boundary that clips them makes every attempt click the +; title (run 31446691812 clicked "blue text at 220, 330" ten times). FindInstallClickPoint(winTitle, &outX, &outY) { WinGetPos(&wx, &wy, &ww, &wh, winTitle) try { @@ -156,7 +158,7 @@ FindInstallClickPoint(winTitle, &outX, &outY) { return true } catch { } - lowerY := wy + Floor(wh * 0.45) + lowerY := wy + Floor(wh * 0.55) if PixelSearch(&px, &py, wx, lowerY, wx + ww, wy + wh, 0x3B82F6, 90) { outX := px outY := py @@ -170,7 +172,7 @@ FindInstallClickPoint(winTitle, &outX, &outY) { ; the progress view, so lingering blue in the lower half means it did not. InstallButtonStillVisible(winTitle) { WinGetPos(&wx, &wy, &ww, &wh, winTitle) - lowerY := wy + Floor(wh * 0.45) + lowerY := wy + Floor(wh * 0.55) return PixelSearch(&px, &py, wx, lowerY, wx + ww, wy + wh, 0x3B82F6, 90) } From 91490032a82c75b16b17b9558886b343f1877b60 Mon Sep 17 00:00:00 2001 From: ethernet Date: Mon, 10 Aug 2026 20:50:37 -0400 Subject: [PATCH 008/227] fix(windows-e2e): pre-seed managed uv - astral receipt hijacks UV_INSTALL_DIR on runners Run 31447045981: the click landed (manifest received, stages ran) but Stage-Uv failed with 'uv installed but not found at ...\bin\uv.exe'. GitHub windows runners ship uv preinstalled WITH an astral install receipt; astral's cargo-dist installer then updates the receipt's location in place and ignores UV_INSTALL_DIR, so Install-Uv's managed copy never appears. Seed HERMES_HOME\bin\uv.exe from the runner's uv before launching the installer - Install-Uv short-circuits on an existing managed uv, and 'user already has a managed uv' is a legitimate install state, not a bypass. Also abort the run the moment the tailed bootstrap log says 'bootstrap FAILED': the failure screen waits on a human Retry, and the AHK helper would otherwise idle out its whole 25-minute marker deadline (and its blue-text retry loop hammers the Retry button, re-running doomed installs - observed in run 7). --- tests/install/windows/install-update-e2e.ps1 | 34 ++++++++++++++++++++ 1 file changed, 34 insertions(+) diff --git a/tests/install/windows/install-update-e2e.ps1 b/tests/install/windows/install-update-e2e.ps1 index 77fbd4406e14..748104e56cba 100644 --- a/tests/install/windows/install-update-e2e.ps1 +++ b/tests/install/windows/install-update-e2e.ps1 @@ -114,6 +114,32 @@ if (-not $env:HERMES_HOME) { $InstallRoot = Join-Path $env:HERMES_HOME "hermes-agent" $TargetSha = Invoke-Git @("rev-parse", "HEAD") +# Pre-seed the managed uv. install.ps1's Install-Uv runs astral's installer +# with UV_INSTALL_DIR pointing into HERMES_HOME\bin and discards its output -- +# but when an astral install RECEIPT exists (GitHub runners ship uv +# preinstalled with one), the cargo-dist installer updates the receipt's +# location in place and ignores UV_INSTALL_DIR, so the managed path stays +# empty and the stage fails blind ("uv installed but not found", run +# 31447045981). Install-Uv short-circuits on an existing managed uv, so +# seeding it is a legitimate user state, not a bypass. +$managedBin = Join-Path $env:HERMES_HOME "bin" +New-Item -ItemType Directory -Force -Path $managedBin | Out-Null +$uvOnRunner = Get-Command uv.exe -ErrorAction SilentlyContinue +if ($uvOnRunner) { + Copy-Item $uvOnRunner.Source (Join-Path $managedBin "uv.exe") -Force + Ok "seeded managed uv from runner: $($uvOnRunner.Source)" +} else { + # No preinstalled uv means no receipt, so the plain astral path works -- + # with output visible, unlike Install-Uv's. + $env:UV_INSTALL_DIR = $managedBin + Invoke-RestMethod https://astral.sh/uv/install.ps1 | Invoke-Expression + Remove-Item Env:\UV_INSTALL_DIR + if (-not (Test-Path (Join-Path $managedBin "uv.exe"))) { + Fail "could not seed managed uv into $managedBin" + } + Ok "seeded managed uv via astral installer" +} + # --- fake GitHub ------------------------------------------------------------- Step "seeding fake remote at $FakeRepo" @@ -238,6 +264,14 @@ try { $line = $logReader.ReadLine() while ($null -ne $line) { Write-Host "[bootstrap] $line" + # The installer's failure screen waits for a human (Retry + # button); the AHK helper would idle out its full marker + # deadline. Abort as soon as the log says the run is dead. + if ($line -match "bootstrap FAILED") { + Stop-Process -Id $ahkProc.Id -Force -ErrorAction SilentlyContinue + Stop-Process -Id $proc.Id -Force -ErrorAction SilentlyContinue + Fail "installer reported: $line" + } $line = $logReader.ReadLine() } } From 7b3f197c459256a4023f4b2e85e45d896042dff2 Mon Sep 17 00:00:00 2001 From: ethernet Date: Mon, 10 Aug 2026 20:56:06 -0400 Subject: [PATCH 009/227] fix(windows-e2e): AHK loop killed a healthy install on its own flawed heuristic Run 31447405319 is the big win and the bug in one log: uv seeding worked, the Install click LANDED, and 6 stages ran (clone off the fake remote via SSH rewrite, venv, all Python dependencies) - then the AHK loop, still hunting for 'the button', PixelSearch-matched the PROGRESS view's own blue stage text (left column, x~106-115), decided the UI 'did not advance' 10 times, threw, and the driver killed a healthy install mid-node-deps. Fixes, all sourced from that recording: narrow the scan band to the center third so the left-column stage text can never match; verify a click by the blue vanishing AT THE CLICK POINT (a 24px box) instead of anywhere in the band; and never throw from the click loop - the authoritative failure signal is the driver's 'bootstrap FAILED' log abort, and the marker deadline caps a wedged UI. --- .../windows/install-hermes-desktop.ahk | 56 +++++++++++-------- 1 file changed, 34 insertions(+), 22 deletions(-) diff --git a/tests/install/windows/install-hermes-desktop.ahk b/tests/install/windows/install-hermes-desktop.ahk index 92195e00ae10..838804cd2437 100644 --- a/tests/install/windows/install-hermes-desktop.ahk +++ b/tests/install/windows/install-hermes-desktop.ahk @@ -146,10 +146,11 @@ if (markerPath = "") { ; recording, whose chroma subsampling smears the glyph edges -- the same ; crop matches a RECORDING of this screen within 8 shades/channel while ; missing the live screen entirely. The reliable path is PixelSearch for -; the saturated blue of the "[ INSTALL ]" text. Scan only BELOW 55% of the -; window height: the HERMES AGENT title is blue too, and its bottom rows -; reach ~47% -- a boundary that clips them makes every attempt click the -; title (run 31446691812 clicked "blue text at 220, 330" ten times). +; the saturated blue of the "[ INSTALL ]" text. Scan only the CENTER THIRD +; horizontally and BELOW 55% of the window height: the HERMES AGENT title +; is blue too (bottom rows ~47% -- run 31446691812 clicked it ten times), +; and the post-click PROGRESS view has blue stage text on the left side +; (run 31447405319 clicked that while the install was running fine). FindInstallClickPoint(winTitle, &outX, &outY) { WinGetPos(&wx, &wy, &ww, &wh, winTitle) try { @@ -158,8 +159,10 @@ FindInstallClickPoint(winTitle, &outX, &outY) { return true } catch { } - lowerY := wy + Floor(wh * 0.55) - if PixelSearch(&px, &py, wx, lowerY, wx + ww, wy + wh, 0x3B82F6, 90) { + x1 := wx + Floor(ww * 0.33) + x2 := wx + Floor(ww * 0.67) + y1 := wy + Floor(wh * 0.55) + if PixelSearch(&px, &py, x1, y1, x2, wy + wh, 0x3B82F6, 90) { outX := px outY := py Log(Format("Install button located via PixelSearch (blue text at {1}, {2})", px, py)) @@ -168,17 +171,22 @@ FindInstallClickPoint(winTitle, &outX, &outY) { return false } -; Did the click land? The button's blue text vanishes when the UI flips to -; the progress view, so lingering blue in the lower half means it did not. -InstallButtonStillVisible(winTitle) { - WinGetPos(&wx, &wy, &ww, &wh, winTitle) - lowerY := wy + Floor(wh * 0.55) - return PixelSearch(&px, &py, wx, lowerY, wx + ww, wy + wh, 0x3B82F6, 90) +; Did the click land? Check whether the blue we clicked is still at THAT +; SPOT (small box around the click point): the button vanishes when the UI +; flips to the progress view, and a point check cannot false-positive on +; the progress view's own blue elsewhere in the window. +BlueStillAt(x, y) { + return PixelSearch(&px, &py, x - 12, y - 12, x + 12, y + 12, 0x3B82F6, 90) } -clicked := false +; Best-effort clicking: NEVER throw here. The authoritative signals are +; owned elsewhere -- the driver aborts on "bootstrap FAILED" in the +; installer log, and the marker wait below caps the run. Run 31447405319 +; killed a healthy mid-install run because this loop threw on its own +; flawed UI heuristic; that class of failure must stay impossible. attempts := 0 -while (!clicked && attempts < 10) { +everClicked := false +while (attempts < 10) { attempts += 1 ; ImageSearch/PixelSearch read SCREEN pixels: anything covering the ; installer (the runner session keeps a maximized console in front) @@ -190,20 +198,24 @@ while (!clicked && attempts < 10) { x := 0 y := 0 if (!FindInstallClickPoint(winTitle, &x, &y)) { - Log(Format("Install click attempt {}: button not found on screen", attempts)) + if (everClicked) { + ; Click already landed and this is the progress view. + Log(Format("Attempt {}: button gone after click; proceeding to marker wait", attempts)) + break + } + ; Welcome screen may still be rendering. + Log(Format("Attempt {}: no Install button in scan band yet; waiting", attempts)) Sleep(2000) continue } ClickWithMarker(x, y) + everClicked := true Sleep(3000) - if (InstallButtonStillVisible(winTitle)) { - Log(Format("Install click attempt {}: UI did not advance; retrying", attempts)) - continue + if (!BlueStillAt(x, y)) { + Log("Install click landed (button gone from click point)") + break } - clicked := true -} -if (!clicked) { - throw Error("could not find/click the Install button after " attempts " attempts") + Log(Format("Install click attempt {}: blue persists at click point; retrying", attempts)) } ; Wait for the installer's own completion signal: the bootstrap-complete From caf0e6ac428072bebdc70cbe7291d7587e3e64a3 Mon Sep 17 00:00:00 2001 From: ethernet Date: Mon, 10 Aug 2026 21:18:06 -0400 Subject: [PATCH 010/227] ci(windows-e2e): capture a lossless welcome-screen PNG before the AHK helper runs The ImageSearch reference must be cropped from a lossless capture of the real screen: the ffmpeg recording is H.264/yuv420p and its chroma subsampling shifts glyph pixels enough that a video-sourced crop never matches live rendering. Taken before the helper starts so no tooltip or click marker contaminates it; lands in the log-dir artifact. --- tests/install/windows/install-update-e2e.ps1 | 17 +++++++++++++++++ 1 file changed, 17 insertions(+) diff --git a/tests/install/windows/install-update-e2e.ps1 b/tests/install/windows/install-update-e2e.ps1 index 748104e56cba..8742c9e96986 100644 --- a/tests/install/windows/install-update-e2e.ps1 +++ b/tests/install/windows/install-update-e2e.ps1 @@ -236,6 +236,23 @@ Step "installing via Hermes-Setup.exe (real toolchains: git, uv, Python, Node, v $installerOk = $false try { $proc = Start-Process -FilePath $InstallerPath -PassThru + + # Lossless PNG of the welcome screen, taken BEFORE the AHK helper starts + # so no tooltip or click marker contaminates it. This is the artifact the + # ImageSearch reference crop is made from: the ffmpeg recording is + # H.264/yuv420p, whose chroma subsampling shifts glyph pixels enough that + # a crop from video never matches the live screen. + Start-Sleep -Seconds 10 + Add-Type -AssemblyName System.Windows.Forms, System.Drawing + $bounds = [System.Windows.Forms.Screen]::PrimaryScreen.Bounds + $bmp = New-Object System.Drawing.Bitmap $bounds.Width, $bounds.Height + $gfx = [System.Drawing.Graphics]::FromImage($bmp) + $gfx.CopyFromScreen($bounds.Location, [System.Drawing.Point]::Empty, $bounds.Size) + $gfx.Dispose() + $bmp.Save((Join-Path $LogDir "welcome-screen.png"), [System.Drawing.Imaging.ImageFormat]::Png) + $bmp.Dispose() + Ok "welcome screen captured (lossless) to welcome-screen.png" + $ahkLog = Join-Path $LogDir "ahk.log" # The helper polls for the installer's own completion signal instead of a # second button screenshot (see paths.rs likely_bootstrap_marker). From cba48161b9bfc0edca15a9e2473923bf9decab89 Mon Sep 17 00:00:00 2001 From: ethernet Date: Mon, 10 Aug 2026 21:19:46 -0400 Subject: [PATCH 011/227] ci(windows-e2e): TEMP - stop after the welcome screenshot, skip the AHK path --- tests/install/windows/install-update-e2e.ps1 | 9 +++++++++ 1 file changed, 9 insertions(+) diff --git a/tests/install/windows/install-update-e2e.ps1 b/tests/install/windows/install-update-e2e.ps1 index 8742c9e96986..0fd0431caedb 100644 --- a/tests/install/windows/install-update-e2e.ps1 +++ b/tests/install/windows/install-update-e2e.ps1 @@ -253,6 +253,15 @@ try { $bmp.Dispose() Ok "welcome screen captured (lossless) to welcome-screen.png" + # TEMPORARY: stop here. The ImageSearch reference crop has to be cut from + # this capture before the AHK click path can work, so running it now only + # burns a doomed 30-minute leg. Remove this exit once install-button.png + # is regenerated from welcome-screen.png. + Stop-Process -Id $proc.Id -Force -ErrorAction SilentlyContinue + Stop-Recording + Write-Host "STOPPING EARLY: welcome-screen.png captured; AHK path disabled until the button crop is regenerated from it." + exit 0 + $ahkLog = Join-Path $LogDir "ahk.log" # The helper polls for the installer's own completion signal instead of a # second button screenshot (see paths.rs likely_bootstrap_marker). From 455016b82d99ce851e37693e333293f7ba85dece Mon Sep 17 00:00:00 2001 From: ethernet Date: Mon, 10 Aug 2026 21:24:15 -0400 Subject: [PATCH 012/227] ci(windows-e2e): poll for a rendered UI before the welcome screenshot - 10s was blank --- tests/install/windows/install-update-e2e.ps1 | 39 ++++++++++++++++---- 1 file changed, 32 insertions(+), 7 deletions(-) diff --git a/tests/install/windows/install-update-e2e.ps1 b/tests/install/windows/install-update-e2e.ps1 index 0fd0431caedb..a8af3d342157 100644 --- a/tests/install/windows/install-update-e2e.ps1 +++ b/tests/install/windows/install-update-e2e.ps1 @@ -242,15 +242,40 @@ try { # ImageSearch reference crop is made from: the ffmpeg recording is # H.264/yuv420p, whose chroma subsampling shifts glyph pixels enough that # a crop from video never matches the live screen. - Start-Sleep -Seconds 10 + # + # Poll instead of a fixed sleep: WebView2's cold start left the window + # pure white past 17s on a runner (run 31448964917's capture was blank). + # "Rendered" = colored (non-grayscale) pixels in the window content area, + # which the blue HERMES AGENT title guarantees. Add-Type -AssemblyName System.Windows.Forms, System.Drawing $bounds = [System.Windows.Forms.Screen]::PrimaryScreen.Bounds - $bmp = New-Object System.Drawing.Bitmap $bounds.Width, $bounds.Height - $gfx = [System.Drawing.Graphics]::FromImage($bmp) - $gfx.CopyFromScreen($bounds.Location, [System.Drawing.Point]::Empty, $bounds.Size) - $gfx.Dispose() - $bmp.Save((Join-Path $LogDir "welcome-screen.png"), [System.Drawing.Imaging.ImageFormat]::Png) - $bmp.Dispose() + $shotPath = Join-Path $LogDir "welcome-screen.png" + $renderDeadline = (Get-Date).AddSeconds(120) + $rendered = $false + while ((Get-Date) -lt $renderDeadline) { + Start-Sleep -Seconds 5 + $bmp = New-Object System.Drawing.Bitmap $bounds.Width, $bounds.Height + $gfx = [System.Drawing.Graphics]::FromImage($bmp) + $gfx.CopyFromScreen($bounds.Location, [System.Drawing.Point]::Empty, $bounds.Size) + $gfx.Dispose() + $colored = 0 + for ($y = 100; $y -lt 600; $y += 7) { + for ($x = 100; $x -lt 900; $x += 7) { + $p = $bmp.GetPixel($x, $y) + $mx = [Math]::Max($p.R, [Math]::Max($p.G, $p.B)) + $mn = [Math]::Min($p.R, [Math]::Min($p.G, $p.B)) + if (($mx - $mn) -gt 60) { $colored++ } + } + } + $bmp.Save($shotPath, [System.Drawing.Imaging.ImageFormat]::Png) + $bmp.Dispose() + Write-Host " screen poll: $colored colored samples" + if ($colored -gt 20) { $rendered = $true; break } + } + if (-not $rendered) { + Stop-Recording + Fail "installer UI never rendered within 120s (last capture saved to welcome-screen.png)" + } Ok "welcome screen captured (lossless) to welcome-screen.png" # TEMPORARY: stop here. The ImageSearch reference crop has to be cut from From bde4b59f2fcf0eb1a48b764bbb52fa408e650714 Mon Sep 17 00:00:00 2001 From: ethernet Date: Mon, 10 Aug 2026 21:29:24 -0400 Subject: [PATCH 013/227] fix(windows-e2e): screen-native button crop, ImageSearch only - PixelSearch removed install-button.png is now cut from run 31449192962's LOSSLESS welcome-screen.png (blue-text extents x465-558 y449-459 plus 3px margin, verified complete '[ INSTALL ]' with no foreign pixels) instead of the H.264 recording whose chroma subsampling made every video-sourced crop miss the live screen. Tolerance drops from *60 to *20 accordingly. PixelSearch is deleted outright: color-hunting matched the blue title (run 31446691812) and the progress view's stage text (run 31447405319) before it ever matched the button. Click-landed detection reuses the same ImageSearch (button visible = not clicked). The TEMP stop-after-screenshot exit is removed - the AHK path is live again. --- tests/install/windows/install-button.png | Bin 2034 -> 680 bytes .../windows/install-hermes-desktop.ahk | 64 +++++++----------- tests/install/windows/install-update-e2e.ps1 | 9 --- 3 files changed, 24 insertions(+), 49 deletions(-) diff --git a/tests/install/windows/install-button.png b/tests/install/windows/install-button.png index 4a7705fcca7b8796da06004de2d341113ffbec8c..19c478dc81023bf3d01c1e68acd33ebdd50b7aab 100644 GIT binary patch literal 680 zcmV;Z0$2TsP)0#>VXpn36XfKIXH|{uT@)o-`ftz-b5I=l{ zs~1E%x>o?;+X{7e2~{Kg`1*h%aa_5}?q8=FtFS>N4wD`h69QrUu2mET&#V)N@x6#g zubDueFJUJm6iR1X+Q6>ro>$o201`u3_z@_dk+?b+wpK)YfxCi$o zUqWylb}KFpQZj=W_h(Js2W7LNXNo>15sV7uGgk0^nj57=8hvh&(`PSY9FiFrdvUB6 z1O%hauh1q?HdrJ5xI9G9LgFs1y7A^V^&G+mC)H;zi|sVV7g?X1CDMp-e>H0>UT4{@ z*iIu?6xVWD#ug(b(hw91xgy3FMM{}6Pu^ElHg7Y@fvR`ItrS9E*z_>z zI%SXYUy{n<l4n)L8`{kKZ!2g0HXO>ml{8gLKr3v<%c>$>1ldSpMy{*h-<2qLwslX@qWxjaW^nkyH~*p-qVP$w+_)QX2wd1Rn5- z7!wmnhza4PA}WaqK@4hYv+}2@$?|05Q_qQ@TmuaIF=zlwa|2?!RYZ-0IT1K0)meHoHWwa@48Ewj1Mr^#A zGgV~{+&LEjrwWLK_x+@xz-dsW3}%>MMmJRvF^@osxD#^*vIG$XbN4O_?#z|Q83A{? z6!)6vnE;>}Kom1N1$U67s!1HkIRItJ?j9_-HVMcMva;8yA*+UQk>;v*r zRWqBLMul-XgUpV&a(a9 zW0&uI@(Vj3y#BVw$0jF}U}A)P=oNPC`r(=zhc4NEXyXl!U$y=5y}$Zh1%y&g;&hCi z@GEYeTD4>3$>-jll9<&jxLd&EPY*Ki?fpX|Z_E{pn?)x&POhfPwtL5R{Unu)k}Z`M z?xGsIdVKRWR}5TrS=4g?)dlurlPLxF#R0e@DvSNS=QMWjzH-;DE5G%FsV!T2@A}4F z?x+_)M9l1&XQp;UTOUD$;Py{XvRfwgOS^wB$?`YZR`^XKp0 zvVGGgF1AKgu0f6+JJ#3t*;T80pMCbpO`EPO;xV%rg9v7(szo}5(28JEN`;5w?NTzU zMGqn(g1f6~vtFB2CYV`$1LT9~s-1>NSe00n-U>EW7FHx%nN3X8RUC27TXFI9oc-*N zvtxQ*C5tI0$s~I9rWu=MW zkGEqkTh_gH?Xp9M4$aJ*vEJRywr$&nhlh9VI&kZ)4^B=Z!qVmC44gjAOE0}NFwo!M z-+S`p$>YZnk(~c?4;S=3Z=1g5uK0&=YHFrQEvVs|TvIhzH5dYtE=mV){t5@)xbU|} za5Yy<)I?X<%-61zp}iZwdDZK`pV+eb=HcHwd-CmuPwz1AFfckp>JAAlA`>&Q7sog< zebHIV=vmH0?0@<&B+6P+R8bdPQXmBgG9P5#T+KBrSt+tBTPj;xGNtCM1!S>hT>OEL z8f(_DVZ)~e2Tz>Wv{>t9&h58fzxp-qyKi)KbQG5TD-Nfo7#|PC+2`$OfdUd?#bLkQEM3wT=rW=~?LM%G+>L1(~xLE|N~#Oe7`nlI6v4RDJhJpZS6 zui5Z@;rqX2-@aGhn7p*FFLWdB9*r`Uzw0N<$~j&Z;ew(dq3+oz8ywX#ureV+ zDLSAz14abOjtB`cvw0x}vYR)Z#{y+VcQCUM7A6{ZXNl7@GZA^pw z?fv!c-Mg>dm_lv*K}6iyuz_{!*1i7P2@#4v&CGEB{SQ6&N_O|@&hgETsTG@i;<(20rD2VboY z44l6_0z_GQdI3XKnKRr83XmNUIu&=9EcIg`=Nv+qS5zsbEF#sQIb%xBbQKy%ImR$A z0MP-mVn(0{T8g~fi*@e`06B|09uNV`K-EHwoHjzJ-AT$`yTiD9Rf9W4aViNp(^1N1 z7JcEYOp06-_{TfE{POhX_2K;U&o1X08)pm|A~HKWyKmo{hY!EHpt2=U&{5n0wC*Nn za?VY&K0u5irPR=!h!k^w@e>Mtg+*I6%9E;?oe@b0QXid~jRgn+_nKO3_mg72s$D`F znIb}rwUCQ{q{x7jm{)@|XXEy3eu@4-|YTB_58nc`f+5w0b zfuxDLzG}mcuwZZHhPPlwLp6o7;ps^Tf&t=%at&}gsH>*- z+@j{QP^C4Z?Jxt9F09a2*}nm4{~xnev{lwJ+LX17Hf1fNOi_@%07*qoM6N<$f@+52<^TWy diff --git a/tests/install/windows/install-hermes-desktop.ahk b/tests/install/windows/install-hermes-desktop.ahk index 838804cd2437..46efd2b9f7c0 100644 --- a/tests/install/windows/install-hermes-desktop.ahk +++ b/tests/install/windows/install-hermes-desktop.ahk @@ -13,11 +13,13 @@ ; 2: bootstrap-complete marker path to poll for (required) ; ; The Install button image lives next to this script. It is a literal crop -; of the published installer's button from a CI screen recording; -; ImageSearch runs with a generous shade tolerance because the reference -; passed through video compression. Completion deliberately does NOT use a -; second button image: the marker file is the installer's own completion -; signal and cannot go stale with a UI restyle. +; of the published installer's "[ INSTALL ]" button, cut from the LOSSLESS +; welcome-screen.png the driver captures in CI (run 31449192962) -- never +; from the ffmpeg recording, whose H.264/yuv420p chroma subsampling shifts +; glyph pixels enough that a video-sourced crop misses the live screen. +; Completion deliberately does NOT use a second button image: the marker +; file is the installer's own completion signal and cannot go stale with a +; UI restyle. logPath := A_Args.Length >= 1 ? A_Args[1] : "ahk.log" markerPath := A_Args.Length >= 2 ? A_Args[2] : "" @@ -98,9 +100,9 @@ FindImageInWindow(winTitle, imageFile, &outX, &outY, timeoutMs := 10000, interva timeLeft := 1 Log(Format("Searching for button file {} in window {}...", imageFile, winTitle)) - ; *60: the reference crop survived H.264 video compression, so per-channel - ; drift up to ~60 shades must still count as a match. - searchImage := Format("*60 {}", imageFile) + ; *20: the reference is a lossless screen crop, so only minor rendering + ; drift (ClearType phase, sub-shade rounding) needs absorbing. + searchImage := Format("*20 {}", imageFile) while (timeLeft > 0) { if ImageSearch(&x, &y, wx, wy, wx + ww, wy + wh, searchImage) @@ -139,44 +141,26 @@ if (markerPath = "") { throw Error("marker path argument is required") } -; Find the Install button and click it. -; -; ImageSearch against the reference crop is attempted first, but it is -; expected to fail on a real screen: the crop came from an H.264/yuv420p -; recording, whose chroma subsampling smears the glyph edges -- the same -; crop matches a RECORDING of this screen within 8 shades/channel while -; missing the live screen entirely. The reliable path is PixelSearch for -; the saturated blue of the "[ INSTALL ]" text. Scan only the CENTER THIRD -; horizontally and BELOW 55% of the window height: the HERMES AGENT title -; is blue too (bottom rows ~47% -- run 31446691812 clicked it ten times), -; and the post-click PROGRESS view has blue stage text on the left side -; (run 31447405319 clicked that while the install was running fine). +; Find the Install button via ImageSearch against the lossless screen crop. +; No PixelSearch fallback: color-hunting matched the blue HERMES AGENT title +; (run 31446691812) and the progress view's blue stage text (run 31447405319) +; before it ever matched the button. FindInstallClickPoint(winTitle, &outX, &outY) { - WinGetPos(&wx, &wy, &ww, &wh, winTitle) try { FindImageInWindow(winTitle, A_ScriptDir "\install-button.png", &outX, &outY, 3000, 250) Log("Install button located via ImageSearch") return true } catch { + return false } - x1 := wx + Floor(ww * 0.33) - x2 := wx + Floor(ww * 0.67) - y1 := wy + Floor(wh * 0.55) - if PixelSearch(&px, &py, x1, y1, x2, wy + wh, 0x3B82F6, 90) { - outX := px - outY := py - Log(Format("Install button located via PixelSearch (blue text at {1}, {2})", px, py)) - return true - } - return false } -; Did the click land? Check whether the blue we clicked is still at THAT -; SPOT (small box around the click point): the button vanishes when the UI -; flips to the progress view, and a point check cannot false-positive on -; the progress view's own blue elsewhere in the window. -BlueStillAt(x, y) { - return PixelSearch(&px, &py, x - 12, y - 12, x + 12, y + 12, 0x3B82F6, 90) +; Did the click land? The button vanishes when the UI flips to the progress +; view, so finding it again means the click did not take. +InstallButtonStillVisible(winTitle) { + x := 0 + y := 0 + return FindInstallClickPoint(winTitle, &x, &y) } ; Best-effort clicking: NEVER throw here. The authoritative signals are @@ -211,11 +195,11 @@ while (attempts < 10) { ClickWithMarker(x, y) everClicked := true Sleep(3000) - if (!BlueStillAt(x, y)) { - Log("Install click landed (button gone from click point)") + if (!InstallButtonStillVisible(winTitle)) { + Log("Install click landed (button no longer on screen)") break } - Log(Format("Install click attempt {}: blue persists at click point; retrying", attempts)) + Log(Format("Install click attempt {}: button still visible; retrying", attempts)) } ; Wait for the installer's own completion signal: the bootstrap-complete diff --git a/tests/install/windows/install-update-e2e.ps1 b/tests/install/windows/install-update-e2e.ps1 index a8af3d342157..a252a8cd832c 100644 --- a/tests/install/windows/install-update-e2e.ps1 +++ b/tests/install/windows/install-update-e2e.ps1 @@ -278,15 +278,6 @@ try { } Ok "welcome screen captured (lossless) to welcome-screen.png" - # TEMPORARY: stop here. The ImageSearch reference crop has to be cut from - # this capture before the AHK click path can work, so running it now only - # burns a doomed 30-minute leg. Remove this exit once install-button.png - # is regenerated from welcome-screen.png. - Stop-Process -Id $proc.Id -Force -ErrorAction SilentlyContinue - Stop-Recording - Write-Host "STOPPING EARLY: welcome-screen.png captured; AHK path disabled until the button crop is regenerated from it." - exit 0 - $ahkLog = Join-Path $LogDir "ahk.log" # The helper polls for the installer's own completion signal instead of a # second button screenshot (see paths.rs likely_bootstrap_marker). From bdb1b53c088ca7006e32ac15d6070770f085dcd4 Mon Sep 17 00:00:00 2001 From: ethernet Date: Mon, 10 Aug 2026 22:55:48 -0400 Subject: [PATCH 014/227] ci(windows-e2e): 45-min handoff deadline + uv debug tracing - diagnose the silent dep-install hang The first full AHK run (31449642122) got all the way through install, verify, and the desktop hand-off's git leg (fetch from the fake remote, reset to target - the proxying works), then sat 65 minutes inside 'Updating Python dependencies' with zero uv output until the job timeout cancelled it. Cancellation kills any chance of a post-mortem: no process table, no partial log. Run the hand-off through Start-Process with a driver-owned 45-minute deadline (the same budget the staged-exe branch already gets), poll-tail its log into the console, and on deadline dump the live uv/python/git process table plus stderr tail before killing the tree - so a hang diagnoses itself instead of burning another silent 75 minutes. RUST_LOG=uv=debug is scoped to the update leg so uv says what it is doing (or waiting on: cache lock, network, resolution). --- tests/install/windows/install-update-e2e.ps1 | 55 ++++++++++++++++++-- 1 file changed, 51 insertions(+), 4 deletions(-) diff --git a/tests/install/windows/install-update-e2e.ps1 b/tests/install/windows/install-update-e2e.ps1 index a252a8cd832c..a5f1a76ad045 100644 --- a/tests/install/windows/install-update-e2e.ps1 +++ b/tests/install/windows/install-update-e2e.ps1 @@ -413,10 +413,57 @@ switch ($Route) { # the wait-for-desktop gate (no desktop is running); no # -RelaunchExe so nothing is launched afterwards. $handoffLog = Join-Path $LogDir "desktop-update.log" - & powershell -NoProfile -ExecutionPolicy Bypass -File $handoff ` - -InstallRoot $InstallRoot -Branch main -DesktopPid 0 -NoUi 2>&1 | - Tee-Object -FilePath $handoffLog - $handoffExit = $LASTEXITCODE + $handoffErrLog = Join-Path $LogDir "desktop-update.err.log" + # Run 31449642122 hung 65 minutes inside "Updating Python + # dependencies" with zero output from uv, and the job-level + # timeout killed the run before anything could say why. Two + # countermeasures: uv debug tracing (uv reads RUST_LOG), and a + # driver-owned deadline (same 45 minutes the staged-exe branch + # gets) so THIS script outlives the hang and can dump the live + # process table before GitHub cancels the job. + $env:RUST_LOG = "uv=debug" + $hp = Start-Process -FilePath "powershell" -ArgumentList @( + "-NoProfile", "-ExecutionPolicy", "Bypass", "-File", $handoff, + "-InstallRoot", $InstallRoot, "-Branch", "main", + "-DesktopPid", "0", "-NoUi" + ) -RedirectStandardOutput $handoffLog -RedirectStandardError $handoffErrLog ` + -PassThru -NoNewWindow + $handoffDeadline = (Get-Date).AddMinutes(45) + $tailPos = 0 + # Poll-tail the log into the job console so progress stays live. + while (-not $hp.HasExited -and (Get-Date) -lt $handoffDeadline) { + Start-Sleep -Seconds 15 + if (Test-Path $handoffLog) { + $content = Get-Content $handoffLog -Raw -ErrorAction SilentlyContinue + if ($content -and $content.Length -gt $tailPos) { + Write-Host ($content.Substring($tailPos)) -NoNewline + $tailPos = $content.Length + } + } + } + $env:RUST_LOG = $null + if (-not $hp.HasExited) { + Write-Host "--- handoff still running at the 45-minute deadline ---" + Write-Host "--- live processes (who is actually stuck): ---" + Get-CimInstance Win32_Process | + Where-Object { $_.Name -match "uv|python|git|node|hermes|powershell|pip" } | + Select-Object ProcessId, ParentProcessId, CreationDate, Name, CommandLine | + Format-Table -AutoSize -Wrap | Out-String -Width 4096 | Write-Host + if (Test-Path $handoffErrLog) { + Write-Host "--- desktop-update.err.log (last 100 lines) ---" + Get-Content $handoffErrLog -Tail 100 | ForEach-Object { Write-Host $_ } + } + taskkill /PID $hp.Id /T /F 2>$null | Out-Null + Fail "desktop-update.ps1 hung past 45 minutes -- process table above, full logs in artifacts" + } + # Flush whatever the poll loop had not printed yet. + if (Test-Path $handoffLog) { + $content = Get-Content $handoffLog -Raw -ErrorAction SilentlyContinue + if ($content -and $content.Length -gt $tailPos) { + Write-Host ($content.Substring($tailPos)) -NoNewline + } + } + $handoffExit = $hp.ExitCode $handoffInternalLog = Join-Path $env:HERMES_HOME "logs\desktop-update-handoff.log" if (Test-Path $handoffInternalLog) { Copy-Item $handoffInternalLog $LogDir -Force } From e3d22b5b2966319f51822cbf5e61a52b46d0995f Mon Sep 17 00:00:00 2001 From: ethernet Date: Tue, 11 Aug 2026 00:03:47 -0400 Subject: [PATCH 015/227] fix(update): drain uv/pip stderr - undrained pipe deadlocked windows desktop updates Two consecutive Windows E2E runs hung inside dependency install until the job timeout: 31449642122 65 minutes in uv pip install .[all], 31453853006 43 minutes in the SQLite-repair uv sync (RUST_LOG=uv=debug made THAT run hang earlier and with more stderr - the tell). Root cause: scripts/desktop-update.ps1 redirects both child pipes but only pumps stdout while the child runs; stderr is ReadToEnd()'d after exit. uv and pip write progress to stderr. Once that pipe hits the ~64KB buffer, uv blocks on write, hermes update blocks on uv, the hand-off blocks on hermes update: deadlock. Slower stderr producers survive by finishing before the buffer fills, which is why the linux sandbox never sees this. Fix both sides of the class: - managed_uv.py candidate sync + main.py _run_install_with_heartbeat: merge stderr into stdout (the pipe that IS drained). This arm heals EXISTING installs, whose old hand-off script drives the NEW python after the git reset. - desktop-update.ps1: drain stderr concurrently via ReadToEndAsync so future bases never block regardless of what a child writes there. _run_logged_subprocess and _run_npm_install_deterministic already merge or capture both pipes; the two fixed sites were the only update- path spawns that redirect stderr without draining it live. --- hermes_cli/main.py | 9 +++++++++ hermes_cli/managed_uv.py | 9 +++++++++ scripts/desktop-update.ps1 | 11 ++++++++--- 3 files changed, 26 insertions(+), 3 deletions(-) diff --git a/hermes_cli/main.py b/hermes_cli/main.py index b1047c44c379..59b19684e15c 100644 --- a/hermes_cli/main.py +++ b/hermes_cli/main.py @@ -8060,11 +8060,20 @@ def _heartbeat() -> None: t = threading.Thread(target=_heartbeat, daemon=True) t.start() try: + # stderr=STDOUT: uv/pip write progress to stderr. The desktop + # hand-off (scripts/desktop-update.ps1) only drains the child's + # stdout while the child runs, so a full stderr pipe (~64KB) + # blocks the installer forever — run 31449642122 hung 65 minutes + # inside this call. Merged into stdout, the output rides the pipe + # that IS drained. The fix lives here (not only in the hand-off + # script) because old installed bases run their OLD copy of the + # hand-off, but import THIS file fresh after the git reset. subprocess.run( cmd, cwd=PROJECT_ROOT, check=True, env=env, + stderr=subprocess.STDOUT, ) finally: done.set() diff --git a/hermes_cli/managed_uv.py b/hermes_cli/managed_uv.py index 8c42cdcb41c8..db9563ab398b 100644 --- a/hermes_cli/managed_uv.py +++ b/hermes_cli/managed_uv.py @@ -780,6 +780,14 @@ def _stage_candidate_venv( # UV_NO_CONFIG drops it and uv 0.12+ refuses --locked. sync_env = dict(env) sync_env.pop("UV_NO_CONFIG", None) + # stderr=STDOUT: uv writes progress to stderr. When the desktop + # hand-off (scripts/desktop-update.ps1) runs `hermes update`, it only + # drains the child's stdout while the child runs; a full stderr pipe + # (~64KB) blocks uv forever. Run 31453853006 hung 43 minutes in this + # exact call. Merging into stdout keeps the output streaming through + # the pipe that IS drained. Old installed bases run their old copy of + # the hand-off script, so the fix must live here, on the Python side + # the update refreshes before dependencies are installed. synced = subprocess.run( [ uv_bin, @@ -792,6 +800,7 @@ def _stage_candidate_venv( ], cwd=project_root, env=sync_env, + stderr=subprocess.STDOUT, check=False, ) if synced.returncode != 0: diff --git a/scripts/desktop-update.ps1 b/scripts/desktop-update.ps1 index d45484501db3..fdcb3a426db8 100644 --- a/scripts/desktop-update.ps1 +++ b/scripts/desktop-update.ps1 @@ -268,8 +268,13 @@ function Invoke-StreamedHermes([string]$Exe, [string[]]$HermesArgs, [string]$Tag $proc = [System.Diagnostics.Process]::Start($psi) $outWriter = [System.IO.File]::CreateText($outFile) $errWriter = [System.IO.File]::CreateText($errFile) - # Pump synchronously in small reads so the UI stays alive; stderr is - # drained at the end (hermes update is stdout-dominant). + # Drain stderr from the START, asynchronously. uv writes its progress + # to stderr; a redirected pipe nobody reads fills at ~64KB and blocks + # the child mid-update (E2E run 31453853006 deadlocked 43 minutes in + # `uv sync` exactly this way). ReadToEndAsync keeps the pipe empty + # while the loop below pumps stdout. + $errTask = $proc.StandardError.ReadToEndAsync() + # Pump stdout synchronously in small reads so the UI stays alive. while (-not $proc.HasExited) { while (-not $proc.StandardOutput.EndOfStream) { $ln = $proc.StandardOutput.ReadLine() @@ -289,7 +294,7 @@ function Invoke-StreamedHermes([string]$Exe, [string[]]$HermesArgs, [string]$Tag if ($ln.Trim()) { Write-HandoffLog ("{0}| {1}" -f $Tag, $ln) } } } - $errText = $proc.StandardError.ReadToEnd() + $errText = $errTask.GetAwaiter().GetResult() if ($errText) { $errWriter.Write($errText) foreach ($ln in ($errText -split "`r?`n")) { From f6f9f89a77f3b757cb458384e987c45be1cc004c Mon Sep 17 00:00:00 2001 From: ethernet Date: Tue, 11 Aug 2026 01:10:57 -0400 Subject: [PATCH 016/227] ci(windows-e2e): drop RUST_LOG from the update leg - it manufactured the deadlock it was diagnosing Run 31457301901 proved both halves of the stderr fix and then hung anyway, in uv pip install -e .[all] (process table caught it live): - The SQLite-repair uv sync that deadlocked run 2 now streams its whole package list and completes in ~80s WITH debug tracing on - managed_uv.py is imported lazily after the git reset, so even this old base ran the fixed copy. - The .[all] install runs _run_install_with_heartbeat from main.py, which was imported when hermes update STARTED - the v0.20.1 copy, which pipes uv's stderr undrained. RUST_LOG=uv=debug guaranteed >64KB of stderr, so the driver's own diagnostic manufactured the deadlock. Comments in both fixed call sites now state the real import-time reach of each arm. This also explains run 1 failing WITHOUT tracing: the pipe budget is cumulative across every child of the update sharing it. The old-code sync burned ~30KB of it on the package list; the .[all] leg finished the job. With the sync leg now on stdout, the old .[all] leg's natural output should fit - which is the real-world story too: old bases hang or survive on stderr luck, new bases are safe by construction. --- hermes_cli/main.py | 9 ++++++--- hermes_cli/managed_uv.py | 7 ++++--- tests/install/windows/install-update-e2e.ps1 | 15 ++++++++------- 3 files changed, 18 insertions(+), 13 deletions(-) diff --git a/hermes_cli/main.py b/hermes_cli/main.py index 59b19684e15c..90277c5215fb 100644 --- a/hermes_cli/main.py +++ b/hermes_cli/main.py @@ -8065,9 +8065,12 @@ def _heartbeat() -> None: # stdout while the child runs, so a full stderr pipe (~64KB) # blocks the installer forever — run 31449642122 hung 65 minutes # inside this call. Merged into stdout, the output rides the pipe - # that IS drained. The fix lives here (not only in the hand-off - # script) because old installed bases run their OLD copy of the - # hand-off, but import THIS file fresh after the git reset. + # that IS drained. Note the reach of this arm: main.py is imported + # when `hermes update` STARTS, so an update running from an old + # base executes the old copy regardless of the git reset — this + # protects updates initiated from bases that already ship it. + # (managed_uv.py gets the same fix AND is imported lazily after + # the reset, so its sync is protected even on old bases.) subprocess.run( cmd, cwd=PROJECT_ROOT, diff --git a/hermes_cli/managed_uv.py b/hermes_cli/managed_uv.py index db9563ab398b..4072da6976c8 100644 --- a/hermes_cli/managed_uv.py +++ b/hermes_cli/managed_uv.py @@ -785,9 +785,10 @@ def _stage_candidate_venv( # drains the child's stdout while the child runs; a full stderr pipe # (~64KB) blocks uv forever. Run 31453853006 hung 43 minutes in this # exact call. Merging into stdout keeps the output streaming through - # the pipe that IS drained. Old installed bases run their old copy of - # the hand-off script, so the fix must live here, on the Python side - # the update refreshes before dependencies are installed. + # the pipe that IS drained. This module is imported lazily by + # update_cmd AFTER the git reset, so even an update running from an + # old base executes THIS copy — unlike main.py, which is imported at + # startup and only protects bases that already ship its twin fix. synced = subprocess.run( [ uv_bin, diff --git a/tests/install/windows/install-update-e2e.ps1 b/tests/install/windows/install-update-e2e.ps1 index a5f1a76ad045..6a7b56cf4b38 100644 --- a/tests/install/windows/install-update-e2e.ps1 +++ b/tests/install/windows/install-update-e2e.ps1 @@ -416,12 +416,14 @@ switch ($Route) { $handoffErrLog = Join-Path $LogDir "desktop-update.err.log" # Run 31449642122 hung 65 minutes inside "Updating Python # dependencies" with zero output from uv, and the job-level - # timeout killed the run before anything could say why. Two - # countermeasures: uv debug tracing (uv reads RUST_LOG), and a - # driver-owned deadline (same 45 minutes the staged-exe branch - # gets) so THIS script outlives the hang and can dump the live - # process table before GitHub cancels the job. - $env:RUST_LOG = "uv=debug" + # timeout killed the run before anything could say why. So the + # hand-off runs under a driver-owned deadline (same 45 minutes + # the staged-exe branch gets): THIS script outlives a hang and + # dumps the live process table before GitHub cancels the job. + # Do NOT set RUST_LOG here: the installed base's Python pipes + # uv's stderr without draining it (the very bug this axis + # exposed), so debug tracing FLOODS that pipe and manufactures + # the deadlock it was meant to diagnose (run 31457301901). $hp = Start-Process -FilePath "powershell" -ArgumentList @( "-NoProfile", "-ExecutionPolicy", "Bypass", "-File", $handoff, "-InstallRoot", $InstallRoot, "-Branch", "main", @@ -441,7 +443,6 @@ switch ($Route) { } } } - $env:RUST_LOG = $null if (-not $hp.HasExited) { Write-Host "--- handoff still running at the 45-minute deadline ---" Write-Host "--- live processes (who is actually stuck): ---" From e64853a37333c44ac2d2ca4b01a9b6786345cf6a Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Mon, 10 Aug 2026 22:02:38 -0700 Subject: [PATCH 017/227] ci(windows): desktop install + update E2E on a real Windows runner Every commit on main now proves, on a real Windows machine, that: 1. the PRIOR commit (HEAD~1) installs from scratch through its own scripts/install.ps1 (-IncludeDesktop: uv, managed Python, Node, venv, packaged Electron Hermes.exe), 2. that install updates TO this commit through the real Desktop GUI update path (scripts/desktop-update.ps1, the exact hand-off the Update button spawns -- fail-closed gates, marker lifecycle, hermes update, result JSON), and 3. this commit updates FORWARD to a synthetic next commit, proving the updater code shipping in this commit is not the one that strands users when the next commit lands. Staging: the driver bare-clones the checkout into serve.git and redirects the canonical GitHub URLs at it with git insteadOf env config, then advances the served main ref BASE -> CURRENT -> NEXT between legs. Installer and updater run byte-for-byte unmodified. Supersedes the AutoHotkey pixel-driving approach (#68183): the GUI Update button's entire effect is spawning desktop-update.ps1 with documented flags, so driving that contract directly tests the same production code deterministically. --- .github/workflows/desktop-windows-e2e.yml | 99 +++++++ tests/install/windows-desktop-e2e.ps1 | 322 ++++++++++++++++++++++ 2 files changed, 421 insertions(+) create mode 100644 .github/workflows/desktop-windows-e2e.yml create mode 100644 tests/install/windows-desktop-e2e.ps1 diff --git a/.github/workflows/desktop-windows-e2e.yml b/.github/workflows/desktop-windows-e2e.yml new file mode 100644 index 000000000000..d2ddfe57e0f9 --- /dev/null +++ b/.github/workflows/desktop-windows-e2e.yml @@ -0,0 +1,99 @@ +name: Desktop Windows Install/Update E2E + +# Can a real Windows user (a) install the commit BEFORE this one from +# scratch, (b) update that install TO this commit through the Desktop GUI +# update path, and (c) update FROM this commit to a future main? +# +# The driver (tests/install/windows-desktop-e2e.ps1) stages a local bare +# repo serving BASE (HEAD~1) -> CURRENT (HEAD) -> NEXT (synthetic same-tree +# child of HEAD) and redirects the canonical GitHub URLs at it via git +# insteadOf env config. The install runs BASE's own scripts/install.ps1 +# with -IncludeDesktop (real uv/Python/Node/venv/Electron build); each +# update leg runs the installed checkout's scripts/desktop-update.ps1 -- +# the exact hand-off the Desktop Update button spawns -- headless (-NoUi), +# including the desktop-exit and venv-lock fail-closed gates, the real +# `hermes update`, marker lifecycle, and the result JSON the relaunched +# Desktop surfaces. +# +# before -> current proves this commit can be updated TO. +# current -> next proves this commit can update FROM (its updater code is +# the one that runs when the NEXT commit ships). +# +# Triggers: every push to main (concurrency-coalesced), release tags, +# nightly, and manual dispatch. Deliberately NOT on pull_request: a leg +# does ~30-60 minutes of real toolchain + Electron work on a 2x-cost +# Windows runner; breakage on main pages within one commit either way. + +on: + workflow_dispatch: + schedule: + # Nightly, off the hour to dodge the top-of-hour runner crunch. + - cron: '40 6 * * *' + push: + branches: [main] + tags: + - 'v[0-9]+.[0-9]+.[0-9]+' + - 'v[0-9]+.[0-9]+.[0-9]+.[0-9]+' + +permissions: + contents: read + +concurrency: + group: desktop-windows-e2e-${{ github.ref }} + cancel-in-progress: true + +jobs: + e2e: + name: install BASE, update to CURRENT, update to NEXT + runs-on: windows-latest + timeout-minutes: 180 + + env: + HERMES_E2E_WORKROOT: ${{ runner.temp }}\hermes-desktop-e2e + + steps: + # Full history: the driver resolves HEAD~1 and bare-clones this + # checkout as the repo the installer/updater talk to. A shallow + # checkout cannot be served as a clone source. + - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2 + with: + fetch-depth: 0 + + - name: Stage serve repo (BASE / CURRENT / NEXT) + shell: powershell + run: powershell -NoProfile -ExecutionPolicy Bypass -File tests\install\windows-desktop-e2e.ps1 -Phase stage + + - name: Install BASE from scratch (real install.ps1, with Desktop) + shell: powershell + run: powershell -NoProfile -ExecutionPolicy Bypass -File tests\install\windows-desktop-e2e.ps1 -Phase install + + - name: Update BASE -> CURRENT (Desktop GUI update path) + shell: powershell + run: powershell -NoProfile -ExecutionPolicy Bypass -File tests\install\windows-desktop-e2e.ps1 -Phase update-to-current + + - name: Update CURRENT -> NEXT (Desktop GUI update path) + shell: powershell + run: powershell -NoProfile -ExecutionPolicy Bypass -File tests\install\windows-desktop-e2e.ps1 -Phase update-to-next + + - name: Collect logs + if: always() + shell: powershell + run: | + $out = "e2e-logs" + New-Item -ItemType Directory -Path $out -Force | Out-Null + $home_ = Join-Path $env:HERMES_E2E_WORKROOT "hermes-home" + foreach ($p in @("logs", ".hermes-update-result.json")) { + $src = Join-Path $home_ $p + if (Test-Path $src) { Copy-Item $src (Join-Path $out $p) -Recurse -Force } + } + $shas = Join-Path $env:HERMES_E2E_WORKROOT "shas.json" + if (Test-Path $shas) { Copy-Item $shas $out -Force } + + - name: Upload logs + if: always() + uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1 + with: + name: desktop-windows-e2e-logs-${{ github.sha }} + path: e2e-logs + retention-days: 7 + if-no-files-found: ignore diff --git a/tests/install/windows-desktop-e2e.ps1 b/tests/install/windows-desktop-e2e.ps1 new file mode 100644 index 000000000000..436cb7df71d8 --- /dev/null +++ b/tests/install/windows-desktop-e2e.ps1 @@ -0,0 +1,322 @@ +# ============================================================================ +# Windows Desktop install + update E2E driver +# ============================================================================ +# Proves, on a real Windows machine, that: +# +# 1. INSTALL - the commit-prior-to-HEAD ("BASE") installs from scratch +# through the real scripts/install.ps1 (uv, managed Python, +# Node, venv, packaged Desktop Hermes.exe). +# 2. UPDATE 1 - that BASE install updates to HEAD ("CURRENT") through the +# real Desktop GUI update path: scripts/desktop-update.ps1, +# the exact hand-off script the Update button spawns +# (see apps/desktop/electron/main.ts applyUpdates()). +# 3. UPDATE 2 - the now-CURRENT install updates once more to a synthetic +# "NEXT" commit staged on top of HEAD, proving the *outgoing* +# update path of the commit under test is not broken either. +# (before -> current proves we can be updated TO; +# current -> next proves we can update FROM.) +# +# HOW THE STAGING WORKS (no MITM proxy, no network fakery): +# We bare-clone the checkout into \serve.git and point every git +# process at it with url..insteadOf rewrites for the two +# canonical repo URLs, injected via GIT_CONFIG_{COUNT,KEY_n,VALUE_n} +# environment variables. install.ps1's `git clone` and `hermes update`'s +# `git fetch origin` therefore transparently hit OUR bare repo, whose +# `main` ref we advance between legs: BASE -> CURRENT -> NEXT. The +# installer and updater code paths run unmodified, byte-for-byte as in +# production. Everything else (uv, PyPI, npm, PortableGit download) uses +# the real network, same as a user install. +# +# NEXT is a same-tree child of CURRENT (git commit-tree), so it exists +# only in serve.git and can never leak a content change; leg 3 verifies +# by commit SHA. +# +# WHY NOT AutoHotkey / pixel-driving the installer window (the prior +# attempt, PR #68183): image-search button clicking is resolution- and +# theme-fragile and never stabilized. The GUI Update button's entire effect +# is spawning desktop-update.ps1 with documented flags (its "CONTRACT" +# header); driving that contract directly tests the same production code +# deterministically. -NoUi and the DesktopPid wait gate exist in that script +# precisely for tests. +# +# USAGE (local Windows box or CI): +# powershell -File tests\install\windows-desktop-e2e.ps1 # all +# powershell -File tests\install\windows-desktop-e2e.ps1 -Phase stage +# ... -Phase install / update-to-current / update-to-next +# Phases share state via \shas.json, so CI can run them as +# separate steps for readable logs. +# +# The workroot defaults to $env:HERMES_E2E_WORKROOT or %TEMP%. Pass -Keep +# to leave everything on disk for inspection. +# ============================================================================ + +param( + [ValidateSet("stage", "install", "update-to-current", "update-to-next", "all")] + [string]$Phase = "all", + + # Repo checkout whose HEAD is the commit under test. + [string]$RepoRoot = "", + + # Where the bare serve repo + isolated HERMES_HOME live. + [string]$WorkRoot = $(if ($env:HERMES_E2E_WORKROOT) { $env:HERMES_E2E_WORKROOT } else { Join-Path $env:TEMP "hermes-desktop-e2e" }), + + # Keep the workroot after a successful `-Phase all` run. + [switch]$Keep +) + +$ErrorActionPreference = "Stop" +$ProgressPreference = "SilentlyContinue" + +if (-not $RepoRoot) { + $RepoRoot = (Resolve-Path (Join-Path $PSScriptRoot "..\..")).Path +} + +$ServeRepo = Join-Path $WorkRoot "serve.git" +$HermesHome = Join-Path $WorkRoot "hermes-home" +$InstallDir = Join-Path $HermesHome "hermes-agent" +$StatePath = Join-Path $WorkRoot "shas.json" + +# The two URL spellings install.ps1 clones from and `hermes update` fetches +# from. insteadOf is prefix-based; we register the exact full forms only, so +# nothing else can accidentally rewrite to a path + ".git" suffix. +$RepoUrlHttps = "https://github.com/NousResearch/hermes-agent.git" +$RepoUrlSsh = "git@github.com:NousResearch/hermes-agent.git" + +function Write-Step([string]$Message) { + Write-Host "" + Write-Host ("=" * 74) + Write-Host "== $Message" + Write-Host ("=" * 74) +} + +function Assert-True([bool]$Condition, [string]$Message) { + if (-not $Condition) { + throw "E2E ASSERTION FAILED: $Message" + } + Write-Host " [ok] $Message" +} + +function Invoke-Git([string[]]$GitArgs, [string]$WorkDir = $null) { + $prev = if ($WorkDir) { Get-Location } else { $null } + # PS 5.1 trap: under $ErrorActionPreference = "Stop", a native command + # that writes ANYTHING to stderr while merged via 2>&1 throws a + # NativeCommandError even when it exits 0 (git loves stderr for + # progress/notices). Relax EAP around the native call only; exit-code + # checking below is the real error gate. + $prevEap = $ErrorActionPreference + $ErrorActionPreference = "Continue" + try { + if ($WorkDir) { Set-Location $WorkDir } + $output = & git @GitArgs 2>&1 + if ($LASTEXITCODE -ne 0) { + throw "git $($GitArgs -join ' ') failed (exit $LASTEXITCODE): $output" + } + return ($output | Out-String).Trim() + } finally { + $ErrorActionPreference = $prevEap + if ($prev) { Set-Location $prev } + } +} + +function Set-GitRedirect { + # Route the canonical repo URLs to the local bare repo for THIS process + # and every child (install.ps1's git, hermes update's git). Environment- + # based config beats `git config --global`: nothing leaks onto the + # machine if the driver dies, and local dev runs stay clean. + $fileUrl = "file:///" + ($ServeRepo -replace "\\", "/") + $env:GIT_CONFIG_COUNT = "2" + $env:GIT_CONFIG_KEY_0 = "url.$fileUrl.insteadOf" + $env:GIT_CONFIG_VALUE_0 = $RepoUrlHttps + $env:GIT_CONFIG_KEY_1 = "url.$fileUrl.insteadOf" + $env:GIT_CONFIG_VALUE_1 = $RepoUrlSsh + Write-Host " git URL redirect: $RepoUrlHttps -> $fileUrl" +} + +function Read-State { + if (-not (Test-Path -LiteralPath $StatePath)) { + throw "State file not found: $StatePath -- run '-Phase stage' first." + } + return Get-Content -LiteralPath $StatePath -Raw | ConvertFrom-Json +} + +function Get-InstalledHead { + return Invoke-Git @("-C", $InstallDir, "rev-parse", "HEAD") +} + +function Get-DesktopExe { + $candidates = @( + (Join-Path $InstallDir "apps\desktop\release\win-unpacked\Hermes.exe"), + (Join-Path $InstallDir "apps\desktop\release\win-arm64-unpacked\Hermes.exe") + ) + foreach ($c in $candidates) { + if (Test-Path -LiteralPath $c) { return $c } + } + return $null +} + +function Test-HermesRuns([string]$Label) { + $hermesExe = Join-Path $InstallDir "venv\Scripts\hermes.exe" + Assert-True (Test-Path -LiteralPath $hermesExe) "$Label -- venv\Scripts\hermes.exe exists" + & $hermesExe --version 2>&1 | ForEach-Object { Write-Host " hermes --version| $_" } + Assert-True ($LASTEXITCODE -eq 0) "$Label -- hermes --version exits 0" +} + +# ---------------------------------------------------------------------------- +# Phase: stage +# ---------------------------------------------------------------------------- +function Invoke-PhaseStage { + Write-Step "STAGE: bare serve repo + BASE/CURRENT/NEXT refs" + + if (Test-Path -LiteralPath $WorkRoot) { + Remove-Item -LiteralPath $WorkRoot -Recurse -Force + } + New-Item -ItemType Directory -Path $WorkRoot -Force | Out-Null + + $current = Invoke-Git @("-C", $RepoRoot, "rev-parse", "HEAD") + $base = Invoke-Git @("-C", $RepoRoot, "rev-parse", "HEAD~1") + Write-Host " CURRENT (commit under test): $current" + Write-Host " BASE (its parent): $base" + + # Bare-clone the checkout: this is the repo the installer and updater + # will actually talk to. Local-path clone hardlinks objects, so it's + # fast even for full history. + Invoke-Git @("clone", "--bare", "--quiet", $RepoRoot, $ServeRepo) | Out-Null + + # Synthesize NEXT inside the bare repo: a same-tree child of CURRENT. + # It exists nowhere else, which is the point -- leg 3 proves the commit + # under test can update to a future main it has never seen. + $env:GIT_AUTHOR_NAME = "Hermes E2E"; $env:GIT_AUTHOR_EMAIL = "e2e@nousresearch.com" + $env:GIT_COMMITTER_NAME = "Hermes E2E"; $env:GIT_COMMITTER_EMAIL = "e2e@nousresearch.com" + $next = Invoke-Git @("-C", $ServeRepo, "commit-tree", "$current^{tree}", "-p", $current, "-m", "e2e: synthetic next commit (same tree as CURRENT)") + Write-Host " NEXT (synthetic): $next" + + # Serve BASE as `main` first; update legs advance this ref. + Invoke-Git @("-C", $ServeRepo, "update-ref", "refs/heads/main", $base) | Out-Null + Invoke-Git @("-C", $ServeRepo, "symbolic-ref", "HEAD", "refs/heads/main") | Out-Null + + @{ base = $base; current = $current; next = $next } | + ConvertTo-Json | Set-Content -LiteralPath $StatePath -Encoding UTF8 + Write-Host " state written: $StatePath" +} + +# ---------------------------------------------------------------------------- +# Phase: install (BASE, via BASE's own install.ps1) +# ---------------------------------------------------------------------------- +function Invoke-PhaseInstall { + $state = Read-State + Write-Step "INSTALL: BASE ($($state.base)) via its own install.ps1 -IncludeDesktop" + + # The honest test runs the installer *as it existed at BASE* -- a user + # installing yesterday used yesterday's script. + $baseInstaller = Join-Path $WorkRoot "install-base.ps1" + & git -C $ServeRepo show "$($state.base):scripts/install.ps1" | + Set-Content -LiteralPath $baseInstaller -Encoding UTF8 + if ($LASTEXITCODE -ne 0) { throw "could not extract scripts/install.ps1 at BASE" } + + $env:HERMES_HOME = $HermesHome + New-Item -ItemType Directory -Path $HermesHome -Force | Out-Null + + # Windows PowerShell 5.1 on purpose: it is what the production + # `irm | iex` one-liner and the desktop bootstrap both run under. + & powershell.exe -NoProfile -ExecutionPolicy Bypass -File $baseInstaller ` + -NonInteractive -SkipSetup -IncludeDesktop ` + -HermesHome $HermesHome -InstallDir $InstallDir + Assert-True ($LASTEXITCODE -eq 0) "install.ps1 (BASE) exited 0" + + Assert-True ((Get-InstalledHead) -eq $state.base) "installed checkout is at BASE" + Test-HermesRuns "post-install" + $desktopExe = Get-DesktopExe + Assert-True ($null -ne $desktopExe) "packaged Desktop Hermes.exe exists ($desktopExe)" +} + +# ---------------------------------------------------------------------------- +# Update legs (shared): the real Desktop GUI update path +# ---------------------------------------------------------------------------- +function Invoke-DesktopUpdateLeg([string]$TargetSha, [string]$LegName) { + Write-Step "UPDATE ($LegName): advance served main -> $TargetSha, run desktop-update.ps1" + + $env:HERMES_HOME = $HermesHome + Invoke-Git @("-C", $ServeRepo, "update-ref", "refs/heads/main", $TargetSha) | Out-Null + + # The hand-off script from the INSTALLED checkout drives the update -- + # exactly the production contract: each `hermes update` refreshes the + # script that drives the NEXT update. + $handoff = Join-Path $InstallDir "scripts\desktop-update.ps1" + Assert-True (Test-Path -LiteralPath $handoff) "installed checkout ships scripts/desktop-update.ps1" + + # Stand in for the exiting Electron main process: desktop-update.ps1 + # FAIL-CLOSED waits for this pid to exit before touching the install. + # A short-lived real process exercises that gate. + $dummy = Start-Process -FilePath "powershell.exe" ` + -ArgumentList "-NoProfile", "-Command", "Start-Sleep -Seconds 3" ` + -WindowStyle Hidden -PassThru + + $resultPath = Join-Path $HermesHome ".hermes-update-result.json" + $markerPath = Join-Path $HermesHome ".hermes-update-in-progress" + Remove-Item -LiteralPath $resultPath -Force -ErrorAction SilentlyContinue + + # -NoUi: headless (no WinForms in CI). No -RelaunchExe: contract says + # omit = no relaunch, so no orphaned Electron process on the runner. + & powershell.exe -NoProfile -ExecutionPolicy Bypass -File $handoff ` + -InstallRoot $InstallDir -Branch main -DesktopPid $dummy.Id -NoUi + $handoffExit = $LASTEXITCODE + + # Surface the hand-off log before asserting, so failures are debuggable + # straight from the CI step output. + $logPath = Join-Path $HermesHome "logs\desktop-update-handoff.log" + if (Test-Path -LiteralPath $logPath) { + Write-Host " --- desktop-update-handoff.log (tail) ---" + Get-Content -LiteralPath $logPath -Tail 40 | ForEach-Object { Write-Host " | $_" } + } + + Assert-True ($handoffExit -eq 0) "$LegName -- desktop-update.ps1 exited 0" + + Assert-True (Test-Path -LiteralPath $resultPath) "$LegName -- update result JSON written" + $result = Get-Content -LiteralPath $resultPath -Raw | ConvertFrom-Json + Assert-True ([bool]$result.ok) "$LegName -- result JSON reports ok=true ('$($result.message)')" + + Assert-True (-not (Test-Path -LiteralPath $markerPath)) "$LegName -- update marker cleaned up" + Assert-True ((Get-InstalledHead) -eq $TargetSha) "$LegName -- checkout landed on target commit" + Test-HermesRuns $LegName + Assert-True ($null -ne (Get-DesktopExe)) "$LegName -- Desktop Hermes.exe still present after update" +} + +function Invoke-PhaseUpdateToCurrent { + $state = Read-State + Invoke-DesktopUpdateLeg $state.current "BASE -> CURRENT" +} + +function Invoke-PhaseUpdateToNext { + $state = Read-State + Invoke-DesktopUpdateLeg $state.next "CURRENT -> NEXT" +} + +# ---------------------------------------------------------------------------- +# Dispatch +# ---------------------------------------------------------------------------- +Write-Host "Windows Desktop E2E driver" +Write-Host " phase: $Phase" +Write-Host " repo: $RepoRoot" +Write-Host " workroot: $WorkRoot" + +Set-GitRedirect + +switch ($Phase) { + "stage" { Invoke-PhaseStage } + "install" { Invoke-PhaseInstall } + "update-to-current" { Invoke-PhaseUpdateToCurrent } + "update-to-next" { Invoke-PhaseUpdateToNext } + "all" { + Invoke-PhaseStage + Invoke-PhaseInstall + Invoke-PhaseUpdateToCurrent + Invoke-PhaseUpdateToNext + if (-not $Keep) { + Write-Step "CLEANUP" + Remove-Item -LiteralPath $WorkRoot -Recurse -Force -ErrorAction SilentlyContinue + } + } +} + +Write-Host "" +Write-Host "Phase '$Phase' completed successfully." From f7f7d2f99475a7e5e50f055add4482fdf9873821 Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Mon, 10 Aug 2026 22:20:10 -0700 Subject: [PATCH 018/227] fix(ci): runner.temp is not a valid context in job-level env The push-triggered validation run failed with zero jobs ('workflow file issue'): job-level env only allows github/inputs/matrix/needs/secrets/ strategy/vars. Use a sibling of github.workspace for the E2E workroot instead. actionlint now passes clean. --- .github/workflows/desktop-windows-e2e.yml | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-) diff --git a/.github/workflows/desktop-windows-e2e.yml b/.github/workflows/desktop-windows-e2e.yml index d2ddfe57e0f9..74a0509aaa31 100644 --- a/.github/workflows/desktop-windows-e2e.yml +++ b/.github/workflows/desktop-windows-e2e.yml @@ -49,7 +49,12 @@ jobs: timeout-minutes: 180 env: - HERMES_E2E_WORKROOT: ${{ runner.temp }}\hermes-desktop-e2e + # Sibling of the checkout (D:\a\hermes-agent\hermes-desktop-e2e): outside + # the repo so the staged bare clone and the install never collide with + # the checkout itself. NOTE: ${{ runner.temp }} is NOT available in + # job-level env (only github/inputs/matrix/needs/secrets/strategy/vars); + # that exact mistake made the whole workflow file invalid on first push. + HERMES_E2E_WORKROOT: ${{ github.workspace }}\..\hermes-desktop-e2e steps: # Full history: the driver resolves HEAD~1 and bare-clones this From 82f48a2de5114601f12a5774d347b9fb16badb47 Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Mon, 10 Aug 2026 22:28:25 -0700 Subject: [PATCH 019/227] ci(temp): trigger the Windows E2E on this branch for pre-merge validation Will be reverted before merge; workflow_dispatch only becomes available once the workflow file exists on the default branch. --- .github/workflows/desktop-windows-e2e.yml | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-) diff --git a/.github/workflows/desktop-windows-e2e.yml b/.github/workflows/desktop-windows-e2e.yml index 74a0509aaa31..3701e033e46e 100644 --- a/.github/workflows/desktop-windows-e2e.yml +++ b/.github/workflows/desktop-windows-e2e.yml @@ -30,7 +30,10 @@ on: # Nightly, off the hour to dodge the top-of-hour runner crunch. - cron: '40 6 * * *' push: - branches: [main] + branches: + - main + # TEMPORARY pre-merge validation: remove before merge. + - hermes/hermes-76e68435 tags: - 'v[0-9]+.[0-9]+.[0-9]+' - 'v[0-9]+.[0-9]+.[0-9]+.[0-9]+' From 3aea9781f6e29289ac12f3d72f15390bc1f16c1c Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Mon, 10 Aug 2026 22:37:11 -0700 Subject: [PATCH 020/227] fix(e2e): redirect via GIT_CONFIG_GLOBAL, not GIT_CONFIG_COUNT env config First CI run's install leg cloned real GitHub main instead of the staged BASE (caught by the HEAD-at-BASE assert): install.ps1 sets GIT_CONFIG_COUNT=1 / windows.appendAtomically itself, silently clobbering the driver's env-config insteadOf rewrites. A driver-owned gitconfig file selected via GIT_CONFIG_GLOBAL survives that (and install.ps1's own --global writes land harmlessly in the same file). Verified locally by cloning with the clobber vars set: clone lands on staged BASE. --- tests/install/windows-desktop-e2e.ps1 | 35 ++++++++++++++++++++------- 1 file changed, 26 insertions(+), 9 deletions(-) diff --git a/tests/install/windows-desktop-e2e.ps1 b/tests/install/windows-desktop-e2e.ps1 index 436cb7df71d8..959576bb25ec 100644 --- a/tests/install/windows-desktop-e2e.ps1 +++ b/tests/install/windows-desktop-e2e.ps1 @@ -120,16 +120,30 @@ function Invoke-Git([string[]]$GitArgs, [string]$WorkDir = $null) { function Set-GitRedirect { # Route the canonical repo URLs to the local bare repo for THIS process - # and every child (install.ps1's git, hermes update's git). Environment- - # based config beats `git config --global`: nothing leaks onto the - # machine if the driver dies, and local dev runs stay clean. + # and every child (install.ps1's git, hermes update's git). + # + # MECHANISM: a driver-owned global gitconfig selected via + # GIT_CONFIG_GLOBAL. Do NOT use GIT_CONFIG_COUNT/KEY_n/VALUE_n env + # config here -- install.ps1 SETS those itself (GIT_CONFIG_COUNT=1, + # windows.appendAtomically), silently clobbering any redirect we put + # there. That exact clobber made the first CI run clone real GitHub + # main instead of the staged BASE (caught by the HEAD-at-BASE assert). + # install.ps1's own `git config --global` writes simply land in our + # file, so its compat settings still apply. Nothing leaks onto the + # machine: the file lives in the workroot and dies with it. $fileUrl = "file:///" + ($ServeRepo -replace "\\", "/") - $env:GIT_CONFIG_COUNT = "2" - $env:GIT_CONFIG_KEY_0 = "url.$fileUrl.insteadOf" - $env:GIT_CONFIG_VALUE_0 = $RepoUrlHttps - $env:GIT_CONFIG_KEY_1 = "url.$fileUrl.insteadOf" - $env:GIT_CONFIG_VALUE_1 = $RepoUrlSsh - Write-Host " git URL redirect: $RepoUrlHttps -> $fileUrl" + $gitCfg = Join-Path $WorkRoot "e2e-gitconfig" + if (-not (Test-Path -LiteralPath $WorkRoot)) { + New-Item -ItemType Directory -Path $WorkRoot -Force | Out-Null + } + @" +[url "$fileUrl"] + insteadOf = $RepoUrlHttps + insteadOf = $RepoUrlSsh +"@ | Set-Content -LiteralPath $gitCfg -Encoding ASCII + $env:GIT_CONFIG_GLOBAL = $gitCfg + Write-Host " git URL redirect via GIT_CONFIG_GLOBAL=$gitCfg" + Write-Host " $RepoUrlHttps -> $fileUrl" } function Read-State { @@ -171,6 +185,9 @@ function Invoke-PhaseStage { Remove-Item -LiteralPath $WorkRoot -Recurse -Force } New-Item -ItemType Directory -Path $WorkRoot -Force | Out-Null + # The purge above deleted the redirect gitconfig; re-arm it so the + # bare-clone below (and everything after) sees the redirect file. + Set-GitRedirect $current = Invoke-Git @("-C", $RepoRoot, "rev-parse", "HEAD") $base = Invoke-Git @("-C", $RepoRoot, "rev-parse", "HEAD~1") From 3e265f4a4397733d9b3ffdf9c3e8da165d516923 Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Mon, 10 Aug 2026 22:46:32 -0700 Subject: [PATCH 021/227] ci: drop the temporary pre-merge branch trigger The Windows E2E ran end-to-end green on this branch (run 31462244593): install at BASE, update BASE->CURRENT, update CURRENT->NEXT, all asserts passing. Back to main/nightly/tags/dispatch triggers only. --- .github/workflows/desktop-windows-e2e.yml | 5 +---- 1 file changed, 1 insertion(+), 4 deletions(-) diff --git a/.github/workflows/desktop-windows-e2e.yml b/.github/workflows/desktop-windows-e2e.yml index 3701e033e46e..74a0509aaa31 100644 --- a/.github/workflows/desktop-windows-e2e.yml +++ b/.github/workflows/desktop-windows-e2e.yml @@ -30,10 +30,7 @@ on: # Nightly, off the hour to dodge the top-of-hour runner crunch. - cron: '40 6 * * *' push: - branches: - - main - # TEMPORARY pre-merge validation: remove before merge. - - hermes/hermes-76e68435 + branches: [main] tags: - 'v[0-9]+.[0-9]+.[0-9]+' - 'v[0-9]+.[0-9]+.[0-9]+.[0-9]+' From 381cc65c78be58805bf7d535c3131a016284ada5 Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Mon, 10 Aug 2026 23:52:20 -0700 Subject: [PATCH 022/227] =?UTF-8?q?ci(windows):=20REAL-flow=20GUI=20E2E=20?= =?UTF-8?q?=E2=80=94=20website=20setup.exe,=20headed=20clicks,=20GUI=20upd?= =?UTF-8?q?ater?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Second job on the Windows E2E workflow covering the surfaces a user actually touches, per Teknium's requirement: * INSTALL: downloads the production Hermes-Setup.exe from hermes-assets.nousresearch.com, launches it HEADED, and AutoHotkey clicks Install -> waits -> clicks Launch (button templates + ImageSearch approach from @ethernet8023's #68183, retargeted by process name and extended to exercise the Launch hand-off). The real Electron Hermes.exe window must appear. * UPDATE x2: the installed Hermes.exe is launched under Playwright's Electron driver and the test CLICKS Settings -> About -> Update now. The production hand-off chain runs untouched: app quits, detached updater (repo script or staged binary) runs hermes update, rebuilds the desktop, relaunches Hermes.exe. Asserts: target sha, marker cleanup, result JSON when the script path wrote one, working hermes, and the RELAUNCHED app window. Leg 1 -> CURRENT, leg 2 -> synthetic NEXT. Proof artifacts: per-step renderer screenshots (booted app, settings, About panel, update-available, updating overlay), full-desktop frames every 3s across the whole run, ahk.log, bootstrap-installer.log, desktop-update-handoff.log — uploaded on success AND failure. The website exe runs exactly as shipped (its own pinned install.ps1, its baked release-pin commit); the only environmental deltas are the serve.git URL redirect, uploadpack.allowAnySHA1InWant for the commit pin fetch, and a placeholder provider key so the update legs meet the app shell instead of onboarding. The contract job from the previous commits is unchanged and independent — it remains the rollback position if the GUI job proves flaky. --- .github/workflows/desktop-windows-e2e.yml | 75 ++- tests/install/e2e-assets/drive-update.cjs | 203 +++++++ .../install/e2e-assets/install-and-launch.ahk | 126 +++++ tests/install/e2e-assets/install-button.png | Bin 0 -> 1200 bytes tests/install/e2e-assets/launch-button.png | Bin 0 -> 1485 bytes tests/install/windows-desktop-gui-e2e.ps1 | 499 ++++++++++++++++++ 6 files changed, 902 insertions(+), 1 deletion(-) create mode 100644 tests/install/e2e-assets/drive-update.cjs create mode 100644 tests/install/e2e-assets/install-and-launch.ahk create mode 100644 tests/install/e2e-assets/install-button.png create mode 100644 tests/install/e2e-assets/launch-button.png create mode 100644 tests/install/windows-desktop-gui-e2e.ps1 diff --git a/.github/workflows/desktop-windows-e2e.yml b/.github/workflows/desktop-windows-e2e.yml index 74a0509aaa31..26243856dd70 100644 --- a/.github/workflows/desktop-windows-e2e.yml +++ b/.github/workflows/desktop-windows-e2e.yml @@ -30,7 +30,10 @@ on: # Nightly, off the hour to dodge the top-of-hour runner crunch. - cron: '40 6 * * *' push: - branches: [main] + branches: + - main + # TEMPORARY pre-merge validation of the gui-e2e job: remove before merge. + - hermes/hermes-76e68435 tags: - 'v[0-9]+.[0-9]+.[0-9]+' - 'v[0-9]+.[0-9]+.[0-9]+.[0-9]+' @@ -102,3 +105,73 @@ jobs: path: e2e-logs retention-days: 7 if-no-files-found: ignore + + # ── The REAL user flow ────────────────────────────────────────────────── + # Same staging, but every leg is the surface a user touches: the website's + # Hermes-Setup.exe run HEADED with AutoHotkey clicking Install/Launch, then + # the installed Electron app driven by Playwright clicking Settings → + # About → "Update now" twice (→ CURRENT, → synthetic NEXT), asserting the + # detached updater chain end-to-end (result JSON, marker, sha, relaunch). + # Proof artifacts: per-step renderer screenshots, full-desktop frames every + # 3s, ahk.log, bootstrap-installer.log, desktop-update-handoff.log. + # + # GitHub's windows-latest runners have an interactive desktop session, so + # headed GUI automation works without RDP tricks (same substrate the prior + # AutoHotkey attempt in #68183 targeted). + gui-e2e: + name: "REAL flow: website setup.exe → GUI update → GUI update" + runs-on: windows-latest + timeout-minutes: 180 + # Deliberately NOT `needs: e2e` — the jobs run independently, so each + # provides rollback coverage for the other and a GUI-layer flake cannot + # mask a machinery regression. + + env: + HERMES_E2E_WORKROOT: ${{ github.workspace }}\..\hermes-desktop-gui-e2e + + steps: + - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2 + with: + fetch-depth: 0 + + - name: Stage serve repo (BASE / CURRENT / NEXT) + shell: powershell + run: powershell -NoProfile -ExecutionPolicy Bypass -File tests\install\windows-desktop-gui-e2e.ps1 -Phase stage + + - name: Install via website Hermes-Setup.exe (headed, AHK-clicked) + shell: powershell + run: powershell -NoProfile -ExecutionPolicy Bypass -File tests\install\windows-desktop-gui-e2e.ps1 -Phase install-gui + + - name: GUI update to CURRENT (real Update-now click) + shell: powershell + run: powershell -NoProfile -ExecutionPolicy Bypass -File tests\install\windows-desktop-gui-e2e.ps1 -Phase update-gui-to-current + + - name: GUI update to NEXT (real Update-now click) + shell: powershell + run: powershell -NoProfile -ExecutionPolicy Bypass -File tests\install\windows-desktop-gui-e2e.ps1 -Phase update-gui-to-next + + - name: Collect proof + logs + if: always() + shell: powershell + run: | + $out = "gui-e2e-proof" + New-Item -ItemType Directory -Path $out -Force | Out-Null + $work = $env:HERMES_E2E_WORKROOT + $home_ = Join-Path $work "hermes-home" + foreach ($pair in @( + @{ src = (Join-Path $work "proof"); dst = "proof" }, + @{ src = (Join-Path $work "shas.json"); dst = "shas.json" }, + @{ src = (Join-Path $home_ "logs"); dst = "logs" }, + @{ src = (Join-Path $home_ ".hermes-update-result.json"); dst = ".hermes-update-result.json" } + )) { + if (Test-Path $pair.src) { Copy-Item $pair.src (Join-Path $out $pair.dst) -Recurse -Force } + } + + - name: Upload proof + logs + if: always() + uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1 + with: + name: desktop-windows-gui-e2e-proof-${{ github.sha }} + path: gui-e2e-proof + retention-days: 14 + if-no-files-found: ignore diff --git a/tests/install/e2e-assets/drive-update.cjs b/tests/install/e2e-assets/drive-update.cjs new file mode 100644 index 000000000000..c328ff89380b --- /dev/null +++ b/tests/install/e2e-assets/drive-update.cjs @@ -0,0 +1,203 @@ +// drive-update.cjs — launch the INSTALLED Hermes.exe (real Electron desktop +// app) under Playwright's Electron driver and perform the update the way a +// user does: Settings -> About -> "Update now". Screenshots at every step. +// +// Run from the installed checkout's apps/desktop directory (so +// @playwright/test resolves from ITS node_modules — the same deps the +// installed app was built with): +// +// node +// +// Exit codes: 0 = update hand-off started and the app quit (the detached +// updater takes it from there — the PowerShell driver polls for the result); +// 1 = any step failed. The driver treats nonzero as leg failure. +// +// This intentionally does NOT call any store/bridge function directly: only +// real clicks on the real UI, so a regression in the button, the About +// panel, the overlay, or the renderer->main bridge fails the test. + +const path = require('node:path') +const fs = require('node:fs') + +const { _electron } = require('@playwright/test') + +const exePath = process.argv[2] +const proofDir = process.argv[3] + +if (!exePath || !proofDir) { + console.error('usage: node drive-update.cjs ') + process.exit(1) +} + +fs.mkdirSync(proofDir, { recursive: true }) + +function log(msg) { + console.log(`[drive-update] ${new Date().toISOString()} ${msg}`) +} + +async function shot(page, name) { + const file = path.join(proofDir, `${name}.png`) + + try { + await page.screenshot({ path: file }) + log(`screenshot: ${file}`) + } catch (err) { + log(`screenshot ${name} failed: ${err.message}`) + } +} + +// Hard ceiling so a hung renderer can't wedge the CI job; the driver's own +// step timeout is the real guard, this is belt-and-braces. +const KILL_AFTER_MS = 15 * 60 * 1000 +const killer = setTimeout(() => { + console.error('[drive-update] global timeout — aborting') + process.exit(1) +}, KILL_AFTER_MS) +killer.unref() + +async function clickFirstVisible(page, locators, description, timeoutMs) { + const deadline = Date.now() + timeoutMs + + for (;;) { + for (const make of locators) { + const locator = make(page).first() + + try { + if (await locator.isVisible()) { + await locator.click() + log(`clicked: ${description}`) + + return true + } + } catch { + // locator invalid in this state; try the next + } + } + if (Date.now() > deadline) { + return false + } + await page.waitForTimeout(500) + } +} + +async function main() { + log(`launching ${exePath}`) + + const app = await _electron.launch({ + executablePath: exePath, + args: ['--disable-gpu', '--no-sandbox'], + // Inherit the driver's env: HERMES_HOME (isolated install) and + // GIT_CONFIG_GLOBAL (URL redirect to the staged serve repo) MUST reach + // the main process so its update check fetches from the staged repo. + env: { ...process.env }, + timeout: 120_000 + }) + + const page = await app.firstWindow({ timeout: 120_000 }) + log('first window acquired') + + // Boot: wait for the composer to exist — the shell is mounted by then. + // The real backend (`hermes serve`) is booting underneath; give it time. + await page.waitForSelector('textarea, [contenteditable="true"]', { + state: 'attached', + timeout: 300_000 + }) + log('renderer booted (composer attached)') + await page.waitForTimeout(3000) + await shot(page, '01-app-booted') + + // ── Open Settings (titlebar gear) ───────────────────────────────────── + const openedSettings = await clickFirstVisible( + page, + [ + p => p.getByLabel('Open settings'), + p => p.locator('[aria-label="Open settings"]'), + p => p.locator('[title="Open settings"]'), + p => p.getByRole('button', { name: 'Open settings' }) + ], + 'Open settings', + 60_000 + ) + + if (!openedSettings) { + await shot(page, 'ERROR-no-settings-button') + throw new Error('could not find the Open settings control') + } + await page.waitForTimeout(1500) + await shot(page, '02-settings-open') + + // ── Go to the About section ─────────────────────────────────────────── + const openedAbout = await clickFirstVisible( + page, + [ + p => p.getByRole('tab', { name: 'About' }), + p => p.getByRole('button', { name: 'About' }), + p => p.getByText('About', { exact: true }) + ], + 'About section', + 30_000 + ) + + if (!openedAbout) { + await shot(page, 'ERROR-no-about-tab') + throw new Error('could not find the About section in Settings') + } + await page.waitForTimeout(1500) + await shot(page, '03-about-panel') + + // ── Wait for "Update now" (appears when behind > 0) ────────────────── + // checkUpdates() runs at boot; if its result hasn't landed yet, press + // "Check now" like an impatient user would. + const updateNow = page.getByRole('button', { name: 'Update now' }).first() + let visible = await updateNow.isVisible().catch(() => false) + + if (!visible) { + log('Update now not visible yet — clicking Check now') + await clickFirstVisible(page, [p => p.getByRole('button', { name: 'Check now' })], 'Check now', 20_000) + + try { + await updateNow.waitFor({ state: 'visible', timeout: 120_000 }) + visible = true + } catch { + visible = false + } + } + + if (!visible) { + await shot(page, 'ERROR-no-update-now') + throw new Error('"Update now" never appeared — update check did not report behind > 0') + } + await shot(page, '04-update-available') + + // ── The click under test ────────────────────────────────────────────── + await updateNow.click() + log('clicked: Update now') + + // The "Updating Hermes — this window will close" overlay should appear, + // then the app quits (hand-off dwell). Screenshot the overlay while the + // window is still alive. + await page.waitForTimeout(1200) + await shot(page, '05-updating-overlay') + + // ── Wait for the app to quit for the hand-off ───────────────────────── + await new Promise((resolve, reject) => { + const t = setTimeout( + () => reject(new Error('app did not quit within 120s of Update now — hand-off did not start')), + 120_000 + ) + + app.on('close', () => { + clearTimeout(t) + resolve() + }) + }) + + log('app quit for updater hand-off — success, the detached updater owns the rest') +} + +main() + .then(() => process.exit(0)) + .catch(err => { + console.error(`[drive-update] FAILED: ${err.message}`) + process.exit(1) + }) diff --git a/tests/install/e2e-assets/install-and-launch.ahk b/tests/install/e2e-assets/install-and-launch.ahk new file mode 100644 index 000000000000..0b6ec05aa5fd --- /dev/null +++ b/tests/install/e2e-assets/install-and-launch.ahk @@ -0,0 +1,126 @@ +#Requires AutoHotkey v2.0 +#SingleInstance Force + +; Drive the REAL Hermes-Setup.exe (Tauri bootstrap installer) window: +; click Install, wait for the install to finish, click Launch, and wait for +; the real Hermes.exe (Electron desktop) window to appear. +; +; Adapted from @ethernet8023's e2e/windows/install-hermes-desktop.ahk +; (PR #68183) — same ImageSearch approach and button templates; this +; variant targets windows by process name (ahk_exe) so the installer +; window and the launched app window (both titled "Hermes") can't be +; confused, and it actually clicks Launch instead of closing the window, +; because the Launch hand-off is part of the flow under test. +; +; Args: [1] log path [2] setup exe name (default Hermes-Setup.exe) + +logPath := A_Args.Length >= 1 ? A_Args[1] : "ahk.log" +setupExe := A_Args.Length >= 2 ? A_Args[2] : "Hermes-Setup.exe" + +Log(text) { + msg := Format("[autohotkey] {}`n", text) + ToolTip(text) + FileAppend(msg, '*') + FileAppend(msg, logPath) +} + +OnError(LogError) + +LogError(err, mode) { + Log(Format("Unhandled error: {}", err.Message)) + ExitApp(1) + return -1 ; suppress the standard error dialog +} + +SetWorkingDir(A_ScriptDir) +CoordMode("Pixel", "Screen") +CoordMode("Mouse", "Screen") + +ClickWithMarker(x, y, button := "Left") { + Click(x, y, button) + Sleep(10) + MouseMove(30, 30) + Log(Format("Clicked at {1}, {2}", x, y)) +} + +FindImageInWindow(winTitle, imageFile, &outX, &outY, timeoutMs := 10000, intervalMs := 250) +{ + WinGetPos(&wx, &wy, &ww, &wh, winTitle) + + hBitmap := LoadPicture(imageFile) + if !hBitmap { + throw Error("LoadPicture failed: " imageFile) + } + bm := Buffer(32, 0) ; BITMAP structure on x64 + DllCall("GetObject", "Ptr", hBitmap, "Int", bm.Size, "Ptr", bm) + width := NumGet(bm, 4, "Int") + height := NumGet(bm, 8, "Int") + + startTime := A_TickCount + timeLeft := 1 + Log(Format("Searching for {} in {} ...", imageFile, winTitle)) + searchImage := Format("*10 {}", imageFile) + while (timeLeft > 0) + { + ; Refresh the window rect each pass — the installer window can move + ; or resize between stages. + try WinGetPos(&wx, &wy, &ww, &wh, winTitle) + if ImageSearch(&x, &y, wx, wy, wx + ww, wy + wh, searchImage) + { + outX := x + Floor(width / 2) + outY := y + Floor(height / 2) + Log("Found " imageFile) + return + } + Sleep intervalMs + timeLeft := timeoutMs - (A_TickCount - startTime) + ToolTip(Format("Searching {} in {} ... {}s left", imageFile, winTitle, Round(timeLeft / 1000, 2))) + } + throw Error(Format("Failed to find {} in window {}", imageFile, winTitle)) +} + +ClickCenterOfImageInWindow(winTitle, imageFile, timeoutMs := 10000, intervalMs := 250) +{ + FindImageInWindow(winTitle, imageFile, &x, &y, timeoutMs, intervalMs) + ClickWithMarker(x, y) +} + +installerWin := "ahk_exe " setupExe +appWin := "ahk_exe Hermes.exe" + +Log("Waiting for the installer window (" installerWin ") ...") +try { + WinWait(installerWin, , 60) +} catch { + throw Error("installer window did not appear within 60s") +} +WinGetPos(&x, &y, &w, &h, installerWin) +Log(Format("Window found at x={1} y={2} w={3} h={4}", x, y, w, h)) + +; ── Step 1: click Install ─────────────────────────────────────────────── +ClickCenterOfImageInWindow(installerWin, A_ScriptDir "\install-button.png", 60000) +Log("Install clicked; waiting for the Launch button (install can take a while)") + +; ── Step 2: wait for install to finish (Launch button appears) ────────── +FindImageInWindow(installerWin, A_ScriptDir "\launch-button.png", &launchX, &launchY, 1000 * 60 * 45) +Log("Install finished (Launch button visible)") + +; ── Step 3: click Launch — the hand-off under test ────────────────────── +ClickWithMarker(launchX, launchY) +Log("Launch clicked; waiting for the Hermes desktop app window") + +; The installer spawns Hermes.exe detached and exits itself. +try { + WinWait(appWin, , 120) +} catch { + throw Error("Hermes.exe window did not appear within 120s of clicking Launch") +} +WinGetPos(&ax, &ay, &aw, &ah, appWin) +Log(Format("App window appeared at x={1} y={2} w={3} h={4}", ax, ay, aw, ah)) + +; Give the renderer a few seconds on screen (recorded as proof), then hand +; control back to the PowerShell driver, which closes the app and re-launches +; it under Playwright for the update legs. +Sleep(8000) +Log("done") +ExitApp(0) diff --git a/tests/install/e2e-assets/install-button.png b/tests/install/e2e-assets/install-button.png new file mode 100644 index 0000000000000000000000000000000000000000..feed47bd98fb3ffe464025c6f52553205ed89751 GIT binary patch literal 1200 zcmX9;Ycv!H6rNFQVwr43op!QisWz%piB^c1W3(lqj7@2#u)7|yN|b~R&4`s_WcAQP zsV0vZjgg@envmCE!X}TXd5}p{G45q&cklTg=bm%#x#xU8F3Z!y%}{@-K7l|mbf>v` zX+1|vt9f&@TAIPD)fzF%%WXTcr@`usw$Y2A`9%>3UoZTZL_*%>CE9@k&z(Nh$*Gwu zRp@yRmmWcaH}(wAxF(`qEN(rCc@L4Ah>MwM_A6S1qJJiie*{TC^)KN;6aMuOKB!@26li|{ zM+gsFvAP`|OJGt1j&a!ZHyG~6s&=fC!nH?GBt~Hi#@xis(fH~u49LMD4nKZ|kUZ>u z11m!@?ly#7#)=k{kHfap=#dKf)!-Zt={$Jc0R=U{$VOH%q}MF_aLtUBd(%! z2u>HGUj{z!(JsR9DD;j%N)h(GgM0NeyHc>de}?3X2K%*lXN0l2ECNC2mbqBro^8-rYqMXuvo?d}atT~~FV z&z{X4YpWL&@F_u?%FRQ{mN3ic|G{AB89%ZnbA)JeLiU0i>M-Fu>Ex~WM3(2%&!99I>XmJ$I2a?r}x9-m6nNn4ome4c4RL5 zr1^=mja6vy@s^Xn{7`koQl8mNT0H13|81Z*ajz9KjN_{zXLK>_qq1+l@~QWN=OUdP zsg5KzzMQg2K7Mdr#Av#aYS~j^XmzD=K$||Pja;!0LrFb_vK(=pNU;L$tjCcGE-1PbO7Q;JLf~#Fw>(v1c6!H+q!`M{ft3BwzrCVC>@-Y;#GA!%@pONb9xbfzG zedk-q{cp%sv_N^UO&K+u$}Di$SdnZIlDyOVyASOp$#ZxP9<%$q4?4aT2l49uN267m zX*ylK{H`*whkU^Fb<@~4Yi$l(IUy+PaH}|K6X;zTA(oH^q)h?7Cu#O{dm~-#*CDu5 LJzR^oA4>WMWp}7g literal 0 HcmV?d00001 diff --git a/tests/install/e2e-assets/launch-button.png b/tests/install/e2e-assets/launch-button.png new file mode 100644 index 0000000000000000000000000000000000000000..6ab89a75ca32480987264e3ee359063cb11f6446 GIT binary patch literal 1485 zcmWkuc{G#@9Q{-xm5TO7#jA*VQxr+Ba?sdHQkJLX*++UaDn=?Klp@|JDU((qktI~3 zv1A|XC}SBzgc(T^!}t5md-Lu&_ug~=y64RZ1a)Ow#)W*urR5B%fUWTV`ODX4Z zg&TGbNRLIMRRmgJ!zUcHkH#&&xXufAT|kR)T%GuQJaD7{bV(t+;a&TNx10@+TFx2V)9cjjE;j(B63<_LIP1)@T>$412LbA zQQ7$U3p5JIvR2Sx;Mgyy?I3?nz%&UvnRqk`Jrgi44|+wgnT|XB@oGA1{*ApNvak_W zxuelVX#p30jKkD4RJDo<5* z4E#PApMbVL(DjqPZGiX!=pP2*5U6{jS*WzRo2=^q|0kGSii0B%n+pwH&^-t#Wl+;j z7Sv;E8Azs}vJC>0AuJu)g)lM(lM=|N#1bAP6~THMHVeszMZkQDL*K#m4hAM;S~=K6 z%666(H)CZRc_$xkWZ|z#SvC33Ylz8#lPsV=z=08H7D8h;zO2CaO?W#WPO$J+4qQ%= zo_mCOb!cz_)jaW46|@LRb|DV`fH(CJl!E*|vTFcn_ki08nU&DdPi{SnjRLT}flb}y zg=ZL(3mV>7+6rc&m{TJ?&cfCn=~^0UoJPSQ&I?>OLJ%`;%uV*&1+c%hounQ)yjo%O zXD;`i$MdaAmhd;WY(L{|n#obK(=)zdd)MLK;qg0}3+xl2nFoh=`XSGw=qvE@OaAybf+m>9iCpe?z~FB-U5aQV$ZY0)I%;u}Ka_%!u| zgwd|=r_5>Yc|)n!t1$7&iZA_9kr0tEZ(u>mmQUNc6x9*>z6lL0?YeS*#n>`Ri;`~7 z{29HJXi8#}Jo}C#$1!m#Dpy0lt7KQ%a?hQeDqUX)udmInv5t&-Uar!-4!;I1ufV!5 zM%RaR$J89!O&3Ur(ig$fM}#mgsAJi_E3wNvUgIJ5zY=w0F@)Q!3wpKIZO+!z?#TO1 zOvQ{T&7j)torda1B&`~+ulN}0mWI#Ni(ncm z@3wr5UVre8OP#8|f_}JHR;{MUc|=bqk-`IEAqm2l*Y3d#YFO z*`i2KZOHXk*)%po-rgbhrAi)^$CP&&?LDvfO2pWs+I^nBn=0`hJ{-N_g}(agF7fO} zVahkOK}K!TDtgF-I}zC^nSyIe%we(5@#sIZ}Frpd1=*Iqvkxj4W(|uPZm-w z!)`sk7cb1|F>~=y?)-Yqv#vZj&nIa01^HT0`IVt%V;g;G1^K*MZcnf1*sK%k!4Mq)=rnhkV|2X<_Pn&Y8)$Fv{Z=$k)A563t&_XmW^%wh_x4v*HR=XCSv1^LA z!&B*VJpagR>xAGj_O^rO%ebOHVwQPZGICPO6xj975up! zEbWq9o zi|RJRl|2PUl|D-upFHF|&$Qig{bj72ewS0j+@3*+V;PKj)HENle|ObMH8UjO`TC8v h%6~e3;57(9p257kTi97NOLobGx#>ZZOyg6b{{c0JSjYeX literal 0 HcmV?d00001 diff --git a/tests/install/windows-desktop-gui-e2e.ps1 b/tests/install/windows-desktop-gui-e2e.ps1 new file mode 100644 index 000000000000..57cbe2bf24b1 --- /dev/null +++ b/tests/install/windows-desktop-gui-e2e.ps1 @@ -0,0 +1,499 @@ +# ============================================================================ +# Windows Desktop GUI install + update E2E driver (the REAL user flow) +# ============================================================================ +# Same before -> current -> next staging as windows-desktop-e2e.ps1, but every +# leg goes through the surfaces a real user touches: +# +# INSTALL - downloads the production Hermes-Setup.exe from the website, +# launches it HEADED, and AutoHotkey clicks Install, waits, +# then clicks Launch. The real Electron Hermes.exe must appear. +# The exe runs EXACTLY as shipped: it downloads its own pinned +# install.ps1 from GitHub raw and installs its baked +# BUILD_PIN_COMMIT — so the "before" version in this harness is +# the website release pin, the literal starting point of every +# real GUI user. (The strict HEAD~1 -> HEAD guarantee is the +# contract harness's job; this one covers the production pin.) +# UPDATE x2 - launches the installed Hermes.exe under Playwright's Electron +# driver and CLICKS Settings -> About -> "Update now". That +# spawns the production hand-off (desktop-update.ps1 via +# cmd start, or the staged hermes-setup.exe on old checkouts), +# the app quits, the detached updater runs `hermes update`, +# rebuilds the desktop, and RELAUNCHES Hermes.exe. We assert +# the whole chain: target sha, marker cleanup, working hermes, +# and the relaunched app window. Leg 1 lands on CURRENT, leg 2 +# on the synthetic NEXT. +# +# PROOF: screenshots at every renderer step (Playwright), full-desktop +# screenshots around the installer/AHK phases, a rolling desktop capture +# (every 3s) for the whole run, ahk.log, and the hand-off log. All uploaded +# as CI artifacts. +# +# DEVIATIONS FROM PRODUCTION (each one deliberate and small): +# * git URL redirect (GIT_CONFIG_GLOBAL) routes the canonical repo URLs +# to the staged serve.git — the staging requirement itself. The +# installer's raw.githubusercontent install.ps1 download and the pinned +# ZIP fallback are NOT redirected (real network, as shipped). +# * serve.git gets uploadpack.allowAnySHA1InWant=true so the installer's +# baked -Commit pin can be fetched from the redirected clone the same +# way GitHub's upload-pack allows it. +# * A dummy provider key is seeded after install so the update legs see +# the ready app shell instead of the onboarding overlay (a real +# updating user has a configured provider). +# +# Usage mirrors windows-desktop-e2e.ps1: +# powershell -File tests\install\windows-desktop-gui-e2e.ps1 -Phase all +# ... -Phase stage / install-gui / update-gui-to-current / update-gui-to-next +# ============================================================================ + +param( + [ValidateSet("stage", "install-gui", "update-gui-to-current", "update-gui-to-next", "all")] + [string]$Phase = "all", + [string]$RepoRoot = "", + [string]$WorkRoot = $(if ($env:HERMES_E2E_WORKROOT) { $env:HERMES_E2E_WORKROOT } else { Join-Path $env:TEMP "hermes-desktop-gui-e2e" }), + [string]$SetupExeUrl = "https://hermes-assets.nousresearch.com/Hermes-Setup.exe" +) + +$ErrorActionPreference = "Stop" +$ProgressPreference = "SilentlyContinue" + +if (-not $RepoRoot) { + $RepoRoot = (Resolve-Path (Join-Path $PSScriptRoot "..\..")).Path +} + +$ServeRepo = Join-Path $WorkRoot "serve.git" +$HermesHome = Join-Path $WorkRoot "hermes-home" +$InstallDir = Join-Path $HermesHome "hermes-agent" +$StatePath = Join-Path $WorkRoot "shas.json" +$ProofRoot = Join-Path $WorkRoot "proof" +$AhkDir = Join-Path $WorkRoot "ahk" +$AssetsDir = Join-Path $PSScriptRoot "e2e-assets" + +$RepoUrlHttps = "https://github.com/NousResearch/hermes-agent.git" +$RepoUrlSsh = "git@github.com:NousResearch/hermes-agent.git" + +function Write-Step([string]$Message) { + Write-Host "" + Write-Host ("=" * 74) + Write-Host "== $Message" + Write-Host ("=" * 74) +} + +function Assert-True([bool]$Condition, [string]$Message) { + if (-not $Condition) { + throw "E2E ASSERTION FAILED: $Message" + } + Write-Host " [ok] $Message" +} + +function Invoke-Git([string[]]$GitArgs) { + $prevEap = $ErrorActionPreference + $ErrorActionPreference = "Continue" + try { + $output = & git @GitArgs 2>&1 + if ($LASTEXITCODE -ne 0) { + throw "git $($GitArgs -join ' ') failed (exit $LASTEXITCODE): $output" + } + return ($output | Out-String).Trim() + } finally { + $ErrorActionPreference = $prevEap + } +} + +function Set-GitRedirect { + # Same mechanism (and same install.ps1-clobber rationale) as + # windows-desktop-e2e.ps1: a driver-owned gitconfig via GIT_CONFIG_GLOBAL. + $fileUrl = "file:///" + ($ServeRepo -replace "\\", "/") + $gitCfg = Join-Path $WorkRoot "e2e-gitconfig" + if (-not (Test-Path -LiteralPath $WorkRoot)) { + New-Item -ItemType Directory -Path $WorkRoot -Force | Out-Null + } + @" +[url "$fileUrl"] + insteadOf = $RepoUrlHttps + insteadOf = $RepoUrlSsh +"@ | Set-Content -LiteralPath $gitCfg -Encoding ASCII + $env:GIT_CONFIG_GLOBAL = $gitCfg + Write-Host " git URL redirect via GIT_CONFIG_GLOBAL=$gitCfg" +} + +function Read-State { + if (-not (Test-Path -LiteralPath $StatePath)) { + throw "State file not found: $StatePath -- run '-Phase stage' first." + } + return Get-Content -LiteralPath $StatePath -Raw | ConvertFrom-Json +} + +function Get-InstalledHead { + return Invoke-Git @("-C", $InstallDir, "rev-parse", "HEAD") +} + +function Get-DesktopExe { + foreach ($c in @( + (Join-Path $InstallDir "apps\desktop\release\win-unpacked\Hermes.exe"), + (Join-Path $InstallDir "apps\desktop\release\win-arm64-unpacked\Hermes.exe") + )) { + if (Test-Path -LiteralPath $c) { return $c } + } + return $null +} + +function Test-HermesRuns([string]$Label) { + $hermesExe = Join-Path $InstallDir "venv\Scripts\hermes.exe" + Assert-True (Test-Path -LiteralPath $hermesExe) "$Label -- venv\Scripts\hermes.exe exists" + & $hermesExe --version 2>&1 | ForEach-Object { Write-Host " hermes --version| $_" } + Assert-True ($LASTEXITCODE -eq 0) "$Label -- hermes --version exits 0" +} + +function Save-DesktopScreenshot([string]$OutFile) { + # Single full-desktop screenshot (all monitors' primary screen). + try { + Add-Type -AssemblyName System.Windows.Forms, System.Drawing + $bounds = [System.Windows.Forms.Screen]::PrimaryScreen.Bounds + $bmp = New-Object System.Drawing.Bitmap($bounds.Width, $bounds.Height) + $gfx = [System.Drawing.Graphics]::FromImage($bmp) + $gfx.CopyFromScreen($bounds.Location, [System.Drawing.Point]::Empty, $bounds.Size) + $bmp.Save($OutFile, [System.Drawing.Imaging.ImageFormat]::Png) + $gfx.Dispose(); $bmp.Dispose() + Write-Host " desktop screenshot: $OutFile" + } catch { + Write-Host " WARNING: desktop screenshot failed: $($_.Exception.Message)" + } +} + +function Start-DesktopRecorder([string]$OutDir) { + # Rolling desktop capture: one PNG every 3s from a detached PowerShell, + # capped at 800 frames (~40 min). Proof that survives any step failure. + New-Item -ItemType Directory -Path $OutDir -Force | Out-Null + $script = Join-Path $WorkRoot "recorder.ps1" + @' +param([string]$OutDir) +Add-Type -AssemblyName System.Windows.Forms, System.Drawing +for ($i = 0; $i -lt 800; $i++) { + if (Test-Path (Join-Path $OutDir "STOP")) { break } + try { + $bounds = [System.Windows.Forms.Screen]::PrimaryScreen.Bounds + $bmp = New-Object System.Drawing.Bitmap($bounds.Width, $bounds.Height) + $gfx = [System.Drawing.Graphics]::FromImage($bmp) + $gfx.CopyFromScreen($bounds.Location, [System.Drawing.Point]::Empty, $bounds.Size) + $bmp.Save((Join-Path $OutDir ("frame-{0:D4}.png" -f $i)), [System.Drawing.Imaging.ImageFormat]::Png) + $gfx.Dispose(); $bmp.Dispose() + } catch {} + Start-Sleep -Seconds 3 +} +'@ | Set-Content -LiteralPath $script -Encoding UTF8 + $proc = Start-Process -FilePath "powershell.exe" ` + -ArgumentList "-NoProfile", "-ExecutionPolicy", "Bypass", "-File", $script, "-OutDir", $OutDir ` + -WindowStyle Hidden -PassThru + Write-Host " desktop recorder started (pid $($proc.Id)) -> $OutDir" + return $proc +} + +function Stop-DesktopRecorder($proc, [string]$OutDir) { + try { Set-Content -LiteralPath (Join-Path $OutDir "STOP") -Value "stop" } catch {} + if ($proc) { + try { $proc.WaitForExit(8000) | Out-Null } catch {} + try { if (-not $proc.HasExited) { Stop-Process -Id $proc.Id -Force } } catch {} + } +} + +function Stop-HermesAppProcesses([string]$Label) { + # Close the desktop app the blunt way between legs (a user quitting). + # Only Hermes.exe (Electron) — never hermes.exe (the venv CLI shim). + $procs = @(Get-Process -Name "Hermes" -ErrorAction SilentlyContinue) + foreach ($p in $procs) { + try { Stop-Process -Id $p.Id -Force -ErrorAction SilentlyContinue } catch {} + } + if ($procs.Count -gt 0) { + Write-Host " [$Label] stopped $($procs.Count) Hermes.exe process(es)" + Start-Sleep -Seconds 3 + } +} + +function Get-ManagedNode { + # `hermes update`/desktop builds use the Hermes-managed Node; use the same + # one to run the Playwright driver so no system Node is required. + $candidates = @( + (Join-Path $HermesHome "node\node.exe"), + (Join-Path $HermesHome "bin\node\node.exe"), + (Join-Path $InstallDir "node\node.exe") + ) + foreach ($c in $candidates) { + if (Test-Path -LiteralPath $c) { return $c } + } + $fromPath = Get-Command node -ErrorAction SilentlyContinue + if ($fromPath) { return $fromPath.Source } + throw "No node.exe found (managed or on PATH)" +} + +# ---------------------------------------------------------------------------- +# Phase: stage — reuse the contract driver's stage (identical staging) +# ---------------------------------------------------------------------------- +function Invoke-PhaseStage { + Write-Step "STAGE (gui): delegating to windows-desktop-e2e.ps1 -Phase stage" + & powershell.exe -NoProfile -ExecutionPolicy Bypass ` + -File (Join-Path $PSScriptRoot "windows-desktop-e2e.ps1") ` + -Phase stage -RepoRoot $RepoRoot -WorkRoot $WorkRoot + if ($LASTEXITCODE -ne 0) { throw "stage phase failed" } + # The website installer pins a specific release commit (-Commit ). + # That sha is in serve.git's history but not at a ref tip, so the + # redirected fetch needs any-SHA1 upload-pack permission (GitHub grants + # the equivalent for archive/fetch of reachable commits). + Invoke-Git @("-C", $ServeRepo, "config", "uploadpack.allowAnySHA1InWant", "true") | Out-Null + Write-Host " serve.git: uploadpack.allowAnySHA1InWant=true (installer commit pin)" + New-Item -ItemType Directory -Path $ProofRoot -Force | Out-Null +} + +# ---------------------------------------------------------------------------- +# Phase: install-gui — website Hermes-Setup.exe, headed, AHK-driven +# ---------------------------------------------------------------------------- +function Invoke-PhaseInstallGui { + $state = Read-State + Write-Step "INSTALL (GUI): Hermes-Setup.exe from the website, headed, AHK clicks" + $proof = Join-Path $ProofRoot "install-gui" + New-Item -ItemType Directory -Path $proof -Force | Out-Null + + # The production installer, from the website. This is the binary users + # double-click, run EXACTLY as shipped: its own pinned install.ps1, its + # own baked BUILD_PIN_COMMIT. The only environmental difference is the + # git URL redirect to serve.git. + $setupExe = Join-Path $WorkRoot "Hermes-Setup.exe" + if (-not (Test-Path -LiteralPath $setupExe)) { + Write-Host " downloading $SetupExeUrl" + Invoke-WebRequest -Uri $SetupExeUrl -OutFile $setupExe + } + Assert-True ((Get-Item $setupExe).Length -gt 1MB) "Hermes-Setup.exe downloaded ($([math]::Round((Get-Item $setupExe).Length / 1MB, 1)) MB)" + + # AutoHotkey v2, portable zip (no installer, no winget flakes). + $ahkExe = Join-Path $AhkDir "AutoHotkey64.exe" + if (-not (Test-Path -LiteralPath $ahkExe)) { + $zip = Join-Path $WorkRoot "ahk.zip" + Invoke-WebRequest -Uri "https://github.com/AutoHotkey/AutoHotkey/releases/download/v2.0.19/AutoHotkey_2.0.19.zip" -OutFile $zip + Expand-Archive -Path $zip -DestinationPath $AhkDir -Force + } + Assert-True (Test-Path -LiteralPath $ahkExe) "AutoHotkey64.exe available" + + # AHK script + button templates side by side (ImageSearch resolves + # relative to the script dir). + Copy-Item -Path (Join-Path $AssetsDir "install-and-launch.ahk"), (Join-Path $AssetsDir "install-button.png"), (Join-Path $AssetsDir "launch-button.png") -Destination $AhkDir -Force + + $env:HERMES_HOME = $HermesHome + # As shipped: NO dev-root override, no pin override. Ensure a stray + # local dev checkout can't hijack resolution. + Remove-Item Env:HERMES_SETUP_DEV_REPO_ROOT -ErrorAction SilentlyContinue + New-Item -ItemType Directory -Path $HermesHome -Force | Out-Null + + $recorder = Start-DesktopRecorder (Join-Path $proof "desktop-frames") + $ahkLog = Join-Path $proof "ahk.log" + try { + Save-DesktopScreenshot (Join-Path $proof "00-before-installer.png") + + # Launch the REAL installer, headed — exactly a double-click. + $installer = Start-Process -FilePath $setupExe -PassThru + Write-Host " Hermes-Setup.exe launched (pid $($installer.Id))" + + # Drive it: Install click -> wait -> Launch click -> Hermes.exe window. + $ahk = Start-Process -FilePath $ahkExe ` + -ArgumentList (Join-Path $AhkDir "install-and-launch.ahk"), $ahkLog ` + -PassThru + # Install on a cold runner takes a while; the AHK script's own inner + # timeout (45 min on the Launch wait) is the effective budget. + if (-not $ahk.WaitForExit(50 * 60 * 1000)) { + Stop-Process -Id $ahk.Id -Force -ErrorAction SilentlyContinue + throw "AutoHotkey driver did not finish within 50 minutes" + } + if (Test-Path -LiteralPath $ahkLog) { + Get-Content -LiteralPath $ahkLog | ForEach-Object { Write-Host " ahk| $_" } + } + Assert-True ($ahk.ExitCode -eq 0) "AutoHotkey driver exited 0 (Install clicked, Launch clicked, app window seen)" + + Save-DesktopScreenshot (Join-Path $proof "01-app-launched.png") + + # The Launch hand-off under test: the app the installer spawned must + # actually be running. + Assert-True ($null -ne (Get-Process -Name "Hermes" -ErrorAction SilentlyContinue)) "Hermes.exe process is running (installer Launch hand-off worked)" + + # Installer should have exited after Launch. + if (-not $installer.HasExited) { + Start-Sleep -Seconds 10 + } + Assert-True $installer.HasExited "Hermes-Setup.exe exited after Launch" + } + finally { + Stop-DesktopRecorder $recorder (Join-Path $proof "desktop-frames") + # Surface the installer's own log win or lose. + $bootLog = Join-Path $HermesHome "logs\bootstrap-installer.log" + if (Test-Path -LiteralPath $bootLog) { + Write-Host " --- bootstrap-installer.log (tail) ---" + Get-Content -LiteralPath $bootLog -Tail 40 | ForEach-Object { Write-Host " | $_" } + Copy-Item $bootLog $proof -Force -ErrorAction SilentlyContinue + } + } + + # Close the freshly launched app (user quits after first look). + Stop-HermesAppProcesses "post-install" + + # The website exe installs its baked release pin — record it as the + # "before" version. Must be an ancestor of CURRENT (i.e. genuinely + # "before" this commit) and must not already BE CURRENT. + $installedSha = Get-InstalledHead + Write-Host " installer landed on: $installedSha (website release pin)" + Assert-True ($installedSha -ne $state.current) "installed pin differs from CURRENT (an update is genuinely available)" + $prevEap = $ErrorActionPreference; $ErrorActionPreference = "Continue" + & git -C $InstallDir merge-base --is-ancestor $installedSha $state.current 2>&1 | Out-Null + $isAncestor = ($LASTEXITCODE -eq 0) + $ErrorActionPreference = $prevEap + Assert-True $isAncestor "installed pin is an ancestor of CURRENT (before -> current is a forward update)" + Test-HermesRuns "post-install-gui" + Assert-True ($null -ne (Get-DesktopExe)) "packaged Desktop Hermes.exe exists" + + # Seed a provider so the update legs meet the ready app shell, not the + # onboarding overlay (an updating user has a configured provider). + $envFile = Join-Path $HermesHome ".env" + if (-not (Test-Path -LiteralPath $envFile) -or -not ((Get-Content $envFile -Raw -ErrorAction SilentlyContinue) -match "OPENROUTER_API_KEY")) { + Add-Content -LiteralPath $envFile -Value "OPENROUTER_API_KEY=sk-or-e2e-placeholder-not-a-real-key" + } + Write-Host " seeded placeholder provider key for update legs" +} + +# ---------------------------------------------------------------------------- +# Update legs: real app, real clicks (Playwright Electron driver) +# ---------------------------------------------------------------------------- +function Invoke-GuiUpdateLeg([string]$TargetSha, [string]$LegName, [string]$LegSlug) { + Write-Step "UPDATE GUI ($LegName): advance served main -> $TargetSha, click Update now" + $proof = Join-Path $ProofRoot $LegSlug + New-Item -ItemType Directory -Path $proof -Force | Out-Null + + $env:HERMES_HOME = $HermesHome + Invoke-Git @("-C", $ServeRepo, "update-ref", "refs/heads/main", $TargetSha) | Out-Null + + $desktopExe = Get-DesktopExe + Assert-True ($null -ne $desktopExe) "$LegName -- packaged Hermes.exe present before update" + + $resultPath = Join-Path $HermesHome ".hermes-update-result.json" + $markerPath = Join-Path $HermesHome ".hermes-update-in-progress" + Remove-Item -LiteralPath $resultPath -Force -ErrorAction SilentlyContinue + + $node = Get-ManagedNode + $appsDesktop = Join-Path $InstallDir "apps\desktop" + Assert-True (Test-Path -LiteralPath (Join-Path $appsDesktop "node_modules\@playwright\test")) "$LegName -- @playwright/test present in installed checkout" + + $recorder = Start-DesktopRecorder (Join-Path $proof "desktop-frames") + try { + # Launch the installed app and click through Settings -> About -> + # Update now. Exit 0 = the app quit for the updater hand-off. + # Copy the driver INTO the installed apps/desktop first: Node resolves + # require('@playwright/test') from the SCRIPT's own directory upward, + # so running it from the CI checkout would resolve the wrong (or no) + # node_modules. + $driver = Join-Path $appsDesktop "e2e-drive-update.cjs" + Copy-Item (Join-Path $AssetsDir "drive-update.cjs") $driver -Force + Push-Location $appsDesktop + try { + & $node $driver $desktopExe $proof 2>&1 | + ForEach-Object { Write-Host " $_" } + $driveExit = $LASTEXITCODE + } finally { + Pop-Location + Remove-Item -LiteralPath $driver -Force -ErrorAction SilentlyContinue + } + Assert-True ($driveExit -eq 0) "$LegName -- GUI driver clicked Update now and the app quit for hand-off" + + # The detached updater (spawned by the app, NOT by us) now runs + # `hermes update` + desktop rebuild + relaunch. Which updater depends + # on the installed checkout, and BOTH are production paths: + # * checkouts shipping scripts/desktop-update.ps1 -> that script, + # which writes .hermes-update-result.json on every exit; + # * older checkouts -> the staged hermes-setup.exe --update flow, + # which does NOT write the result JSON. + # So: poll for COMPLETION = (result JSON) OR (checkout reached the + # target sha AND the marker is gone). The sha/marker/hermes/relaunch + # asserts below are the hard gate either way; the JSON is asserted + # only when the script path produced it. + Write-Host " waiting for the detached updater to finish ..." + $deadline = (Get-Date).AddMinutes(40) + while ((Get-Date) -lt $deadline) { + if (Test-Path -LiteralPath $resultPath) { break } + $head = "" + try { $head = Get-InstalledHead } catch {} + if ($head -eq $TargetSha -and -not (Test-Path -LiteralPath $markerPath)) { break } + Start-Sleep -Seconds 10 + } + if (Test-Path -LiteralPath $resultPath) { + $result = Get-Content -LiteralPath $resultPath -Raw | ConvertFrom-Json + Write-Host " updater result: ok=$($result.ok) code=$($result.exit_code) msg=$($result.message)" + Assert-True ([bool]$result.ok) "$LegName -- updater result ok=true" + } else { + Write-Host " (no result JSON -- staged-binary updater path; relying on sha/marker/relaunch asserts)" + } + + # Marker may briefly outlive the result write; allow it a moment. + $mDeadline = (Get-Date).AddMinutes(2) + while ((Get-Date) -lt $mDeadline -and (Test-Path -LiteralPath $markerPath)) { Start-Sleep -Seconds 5 } + Assert-True (-not (Test-Path -LiteralPath $markerPath)) "$LegName -- update marker cleaned up" + + Assert-True ((Get-InstalledHead) -eq $TargetSha) "$LegName -- checkout landed on target commit" + Test-HermesRuns $LegName + Assert-True ($null -ne (Get-DesktopExe)) "$LegName -- Hermes.exe still present after update" + + # The production hand-off relaunches the desktop (RelaunchExe). + # A relaunched window is the user-visible proof the update loop closed. + Write-Host " waiting for the relaunched Hermes.exe ..." + $rDeadline = (Get-Date).AddMinutes(5) + $relaunched = $null + while ((Get-Date) -lt $rDeadline) { + $relaunched = Get-Process -Name "Hermes" -ErrorAction SilentlyContinue + if ($relaunched) { break } + Start-Sleep -Seconds 5 + } + Assert-True ($null -ne $relaunched) "$LegName -- updater relaunched the desktop app" + Start-Sleep -Seconds 12 # let the window paint for the screenshot + Save-DesktopScreenshot (Join-Path $proof "99-relaunched-desktop.png") + } + finally { + Stop-DesktopRecorder $recorder (Join-Path $proof "desktop-frames") + $handoffLog = Join-Path $HermesHome "logs\desktop-update-handoff.log" + if (Test-Path -LiteralPath $handoffLog) { + Write-Host " --- desktop-update-handoff.log (tail) ---" + Get-Content -LiteralPath $handoffLog -Tail 60 | ForEach-Object { Write-Host " | $_" } + Copy-Item $handoffLog (Join-Path $proof "desktop-update-handoff.log") -Force -ErrorAction SilentlyContinue + } + # Quit the relaunched app so the next leg (or job teardown) is clean. + Stop-HermesAppProcesses $LegName + } +} + +function Invoke-PhaseUpdateGuiToCurrent { + $state = Read-State + Invoke-GuiUpdateLeg $state.current "BASE -> CURRENT (GUI)" "update-gui-to-current" +} + +function Invoke-PhaseUpdateGuiToNext { + $state = Read-State + Invoke-GuiUpdateLeg $state.next "CURRENT -> NEXT (GUI)" "update-gui-to-next" +} + +# ---------------------------------------------------------------------------- +# Dispatch +# ---------------------------------------------------------------------------- +Write-Host "Windows Desktop GUI E2E driver (real user flow)" +Write-Host " phase: $Phase" +Write-Host " repo: $RepoRoot" +Write-Host " workroot: $WorkRoot" + +Set-GitRedirect + +switch ($Phase) { + "stage" { Invoke-PhaseStage } + "install-gui" { Invoke-PhaseInstallGui } + "update-gui-to-current" { Invoke-PhaseUpdateGuiToCurrent } + "update-gui-to-next" { Invoke-PhaseUpdateGuiToNext } + "all" { + Invoke-PhaseStage + Invoke-PhaseInstallGui + Invoke-PhaseUpdateGuiToCurrent + Invoke-PhaseUpdateGuiToNext + } +} + +Write-Host "" +Write-Host "Phase '$Phase' completed successfully." From af7b3d3af4d2dcf5dca5c428994214aa670a2751 Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Tue, 11 Aug 2026 00:47:15 -0700 Subject: [PATCH 023/227] =?UTF-8?q?fix(e2e):=20AHK=20driver=20died=20on=20?= =?UTF-8?q?first=20Log()=20=E2=80=94=20no-console=20stdout=20write=20throw?= =?UTF-8?q?s?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Frame-0005 of the proof capture showed the exact failure: 'Unhandled error: (6) The handle is invalid' rendered over the installer within seconds of launch. AutoHotkey started via Start-Process has no console, so FileAppend to '*' (stdout) throws — and the throw fired inside Log(), killing the script before it clicked anything. The installer then sat untouched at the INSTALL screen for 50 minutes. * Log() now try-wraps the stdout write (file log is the real record) * Install/Launch clicks fall back to the button's relative window position when the #68183-era PNG templates don't match the restyled UI ('[ INSTALL ]' bracket style visible in the same frame) * install-finished has a second signal: 'bootstrap complete' in bootstrap-installer.log (read with write-sharing), so a template miss can't strand the wait * driver passes the bootstrap log path as arg 3 --- .../install/e2e-assets/install-and-launch.ahk | 153 +++++++++++++----- tests/install/windows-desktop-gui-e2e.ps1 | 4 +- 2 files changed, 116 insertions(+), 41 deletions(-) diff --git a/tests/install/e2e-assets/install-and-launch.ahk b/tests/install/e2e-assets/install-and-launch.ahk index 0b6ec05aa5fd..98dbed0c3241 100644 --- a/tests/install/e2e-assets/install-and-launch.ahk +++ b/tests/install/e2e-assets/install-and-launch.ahk @@ -9,18 +9,32 @@ ; (PR #68183) — same ImageSearch approach and button templates; this ; variant targets windows by process name (ahk_exe) so the installer ; window and the launched app window (both titled "Hermes") can't be -; confused, and it actually clicks Launch instead of closing the window, -; because the Launch hand-off is part of the flow under test. +; confused, and it clicks Launch instead of closing the window, because +; the Launch hand-off is part of the flow under test. ; -; Args: [1] log path [2] setup exe name (default Hermes-Setup.exe) +; Robustness beyond the original: +; * Log() survives a missing stdout (GUI-subsystem AHK started without a +; console throws "(6) The handle is invalid" on FileAppend to '*' — +; that single throw killed the whole first CI attempt). +; * If a button template doesn't match (installer UI restyled), falls +; back to clicking the button's known relative position, and the +; install-finished signal falls back to "bootstrap complete" in +; bootstrap-installer.log. Every fallback is logged loudly. +; +; Args: [1] log path [2] setup exe name [3] bootstrap-installer.log path logPath := A_Args.Length >= 1 ? A_Args[1] : "ahk.log" setupExe := A_Args.Length >= 2 ? A_Args[2] : "Hermes-Setup.exe" +bootstrapLog := A_Args.Length >= 3 ? A_Args[3] : "" Log(text) { msg := Format("[autohotkey] {}`n", text) ToolTip(text) - FileAppend(msg, '*') + ; stdout only exists when AHK was launched from a console. Under + ; Start-Process (no console) FileAppend to '*' throws "(6) The handle + ; is invalid" — and a Log() that throws kills the whole script from + ; inside OnError. The file log below is the real record. + try FileAppend(msg, '*') FileAppend(msg, logPath) } @@ -43,10 +57,13 @@ ClickWithMarker(x, y, button := "Left") { Log(Format("Clicked at {1}, {2}", x, y)) } -FindImageInWindow(winTitle, imageFile, &outX, &outY, timeoutMs := 10000, intervalMs := 250) -{ - WinGetPos(&wx, &wy, &ww, &wh, winTitle) - +; Single-pass image search inside a window. Returns true + center coords. +TryFindImage(winTitle, imageFile, &outX, &outY) { + try { + WinGetPos(&wx, &wy, &ww, &wh, winTitle) + } catch { + return false + } hBitmap := LoadPicture(imageFile) if !hBitmap { throw Error("LoadPicture failed: " imageFile) @@ -55,39 +72,49 @@ FindImageInWindow(winTitle, imageFile, &outX, &outY, timeoutMs := 10000, interva DllCall("GetObject", "Ptr", hBitmap, "Int", bm.Size, "Ptr", bm) width := NumGet(bm, 4, "Int") height := NumGet(bm, 8, "Int") - - startTime := A_TickCount - timeLeft := 1 - Log(Format("Searching for {} in {} ...", imageFile, winTitle)) - searchImage := Format("*10 {}", imageFile) - while (timeLeft > 0) - { - ; Refresh the window rect each pass — the installer window can move - ; or resize between stages. - try WinGetPos(&wx, &wy, &ww, &wh, winTitle) - if ImageSearch(&x, &y, wx, wy, wx + ww, wy + wh, searchImage) - { - outX := x + Floor(width / 2) - outY := y + Floor(height / 2) - Log("Found " imageFile) - return - } - Sleep intervalMs - timeLeft := timeoutMs - (A_TickCount - startTime) - ToolTip(Format("Searching {} in {} ... {}s left", imageFile, winTitle, Round(timeLeft / 1000, 2))) + if ImageSearch(&x, &y, wx, wy, wx + ww, wy + wh, Format("*10 {}", imageFile)) { + outX := x + Floor(width / 2) + outY := y + Floor(height / 2) + return true } - throw Error(Format("Failed to find {} in window {}", imageFile, winTitle)) + return false } -ClickCenterOfImageInWindow(winTitle, imageFile, timeoutMs := 10000, intervalMs := 250) -{ - FindImageInWindow(winTitle, imageFile, &x, &y, timeoutMs, intervalMs) - ClickWithMarker(x, y) +; Fractional window position -> screen coords (fallback click target). +WindowRelPoint(winTitle, fx, fy, &outX, &outY) { + WinGetPos(&wx, &wy, &ww, &wh, winTitle) + outX := wx + Floor(ww * fx) + outY := wy + Floor(wh * fy) +} + +BootstrapLogContains(needle) { + global bootstrapLog + if (bootstrapLog = "" or !FileExist(bootstrapLog)) { + return false + } + try { + ; Read-share open: the installer still holds the file for writing. + f := FileOpen(bootstrapLog, "r-d") + if !f { + return false + } + content := f.Read() + f.Close() + return InStr(content, needle) > 0 + } catch { + return false + } } installerWin := "ahk_exe " setupExe appWin := "ahk_exe Hermes.exe" +; The Install/Launch button sits centered horizontally near the bottom of +; the installer window (measured from production screenshots; used only +; when the image template fails to match a restyled UI). +BTN_FX := 0.50 +BTN_FY := 0.87 + Log("Waiting for the installer window (" installerWin ") ...") try { WinWait(installerWin, , 60) @@ -97,16 +124,62 @@ try { WinGetPos(&x, &y, &w, &h, installerWin) Log(Format("Window found at x={1} y={2} w={3} h={4}", x, y, w, h)) -; ── Step 1: click Install ─────────────────────────────────────────────── -ClickCenterOfImageInWindow(installerWin, A_ScriptDir "\install-button.png", 60000) -Log("Install clicked; waiting for the Launch button (install can take a while)") +; ── Step 1: click Install (template first, relative-position fallback) ── +installClicked := false +deadline := A_TickCount + 60000 +while (A_TickCount < deadline) { + if TryFindImage(installerWin, A_ScriptDir "\install-button.png", &ix, &iy) { + ClickWithMarker(ix, iy) + Log("Install clicked (template match)") + installClicked := true + break + } + Sleep(500) +} +if !installClicked { + WindowRelPoint(installerWin, BTN_FX, BTN_FY, &ix, &iy) + ClickWithMarker(ix, iy) + Log("FALLBACK: install template never matched; clicked relative position") +} -; ── Step 2: wait for install to finish (Launch button appears) ────────── -FindImageInWindow(installerWin, A_ScriptDir "\launch-button.png", &launchX, &launchY, 1000 * 60 * 45) -Log("Install finished (Launch button visible)") +; ── Step 2: wait for the install to finish ────────────────────────────── +; Primary signal: the Launch button template appears. Secondary signal: +; "bootstrap complete" in bootstrap-installer.log (the installer's own +; completion line) — after which we give the template 2 more minutes and +; then fall back to the relative-position click. +launchX := 0, launchY := 0 +launchFound := false +completeSince := 0 +waitDeadline := A_TickCount + 1000 * 60 * 45 +Log("Waiting for install to finish (Launch template or bootstrap log) ...") +while (A_TickCount < waitDeadline) { + if TryFindImage(installerWin, A_ScriptDir "\launch-button.png", &launchX, &launchY) { + launchFound := true + Log("Install finished (Launch template visible)") + break + } + if (completeSince = 0 and BootstrapLogContains("bootstrap complete")) { + completeSince := A_TickCount + Log("bootstrap-installer.log reports completion; giving the Launch template 120s") + } + if (completeSince > 0 and A_TickCount - completeSince > 120000) { + Log("FALLBACK: log says complete but Launch template never matched") + break + } + Sleep(2000) +} +if (!launchFound and completeSince = 0) { + throw Error("install did not finish within 45 minutes (no Launch button, no completion log line)") +} ; ── Step 3: click Launch — the hand-off under test ────────────────────── -ClickWithMarker(launchX, launchY) +if launchFound { + ClickWithMarker(launchX, launchY) +} else { + WindowRelPoint(installerWin, BTN_FX, BTN_FY, &lx, &ly) + ClickWithMarker(lx, ly) + Log("FALLBACK: clicked Launch at relative position") +} Log("Launch clicked; waiting for the Hermes desktop app window") ; The installer spawns Hermes.exe detached and exits itself. diff --git a/tests/install/windows-desktop-gui-e2e.ps1 b/tests/install/windows-desktop-gui-e2e.ps1 index 57cbe2bf24b1..d5cda4b694f7 100644 --- a/tests/install/windows-desktop-gui-e2e.ps1 +++ b/tests/install/windows-desktop-gui-e2e.ps1 @@ -292,8 +292,10 @@ function Invoke-PhaseInstallGui { Write-Host " Hermes-Setup.exe launched (pid $($installer.Id))" # Drive it: Install click -> wait -> Launch click -> Hermes.exe window. + # Arg 3 lets the AHK script use the installer's own log as the + # install-finished fallback signal. $ahk = Start-Process -FilePath $ahkExe ` - -ArgumentList (Join-Path $AhkDir "install-and-launch.ahk"), $ahkLog ` + -ArgumentList (Join-Path $AhkDir "install-and-launch.ahk"), $ahkLog, "Hermes-Setup.exe", (Join-Path $HermesHome "logs\bootstrap-installer.log") ` -PassThru # Install on a cold runner takes a while; the AHK script's own inner # timeout (45 min on the Launch wait) is the effective budget. From 100abc953e477fb275444bd0d0f7bbee494dff9a Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Tue, 11 Aug 2026 01:47:10 -0700 Subject: [PATCH 024/227] fix(e2e): re-capture install-button template + fix phantom window geometry Attempt 2's proof frames showed two bugs, both now fixed from the live evidence: 1. The #68183 install-button.png predated the installer UI restyle to the '[ INSTALL ]' bracket look, so the template never matched and we fell through to the position fallback. Re-captured install-button.png from a real CI desktop frame (the actual rendered button). 2. The fallback then clicked the WRONG spot: ahk_exe's first WinGetPos matched a hidden 16x16 helper window ('Window found at w=16 h=16' in ahk.log), and BTN_FY=0.87 aimed below the real button anyway. The button center measured at ~(0.50, 0.59) of the ~full-screen window. Rewrite: * WaitForRealWindow() skips phantom/hidden matches (requires w>400,h>300) and returns the true rect; the installer window is then activated before any click. * Install-finished is now driven primarily by the authoritative 'bootstrap complete' line in bootstrap-installer.log (matches BootstrapEvent::Complete), with the Launch template as a secondary signal and a window-relative fallback click. * Fallback clicks use the corrected (0.50, 0.59) window fraction. --- .../install/e2e-assets/install-and-launch.ahk | 164 ++++++++++-------- tests/install/e2e-assets/install-button.png | Bin 1200 -> 911 bytes 2 files changed, 88 insertions(+), 76 deletions(-) diff --git a/tests/install/e2e-assets/install-and-launch.ahk b/tests/install/e2e-assets/install-and-launch.ahk index 98dbed0c3241..873b3a5d6580 100644 --- a/tests/install/e2e-assets/install-and-launch.ahk +++ b/tests/install/e2e-assets/install-and-launch.ahk @@ -6,20 +6,23 @@ ; the real Hermes.exe (Electron desktop) window to appear. ; ; Adapted from @ethernet8023's e2e/windows/install-hermes-desktop.ahk -; (PR #68183) — same ImageSearch approach and button templates; this -; variant targets windows by process name (ahk_exe) so the installer -; window and the launched app window (both titled "Hermes") can't be -; confused, and it clicks Launch instead of closing the window, because -; the Launch hand-off is part of the flow under test. +; (PR #68183) — same ImageSearch approach; the install-button template was +; re-captured from a live CI frame (the #68183 templates predated the +; installer UI restyle to the "[ INSTALL ]" bracket look and never matched). ; -; Robustness beyond the original: -; * Log() survives a missing stdout (GUI-subsystem AHK started without a -; console throws "(6) The handle is invalid" on FileAppend to '*' — -; that single throw killed the whole first CI attempt). -; * If a button template doesn't match (installer UI restyled), falls -; back to clicking the button's known relative position, and the -; install-finished signal falls back to "bootstrap complete" in -; bootstrap-installer.log. Every fallback is logged loudly. +; Robustness beyond the original — every one earned from a real CI failure: +; * Log() survives a missing stdout. GUI-subsystem AHK started via +; Start-Process has no console, so FileAppend to '*' throws +; "(6) The handle is invalid" — that single throw killed attempt 1. +; * We wait for a REAL-sized installer window before reading geometry. +; ahk_exe matched a hidden 16x16 helper window first (attempt 2's +; "Window found at ... w=16 h=16"), so relative-position math was +; computed against a phantom rect. +; * Clicks fall back to a SCREEN-fraction position when the template +; doesn't match (the Tauri window is effectively full-screen on the +; runner, so screen ~= window). Button center measured at ~0.50, 0.59. +; * Install-finished has a second signal: "bootstrap complete" in +; bootstrap-installer.log, so a Launch-template miss can't strand us. ; ; Args: [1] log path [2] setup exe name [3] bootstrap-installer.log path @@ -30,49 +33,59 @@ bootstrapLog := A_Args.Length >= 3 ? A_Args[3] : "" Log(text) { msg := Format("[autohotkey] {}`n", text) ToolTip(text) - ; stdout only exists when AHK was launched from a console. Under - ; Start-Process (no console) FileAppend to '*' throws "(6) The handle - ; is invalid" — and a Log() that throws kills the whole script from - ; inside OnError. The file log below is the real record. - try FileAppend(msg, '*') + try FileAppend(msg, '*') ; stdout may not exist (no console) FileAppend(msg, logPath) } OnError(LogError) - LogError(err, mode) { Log(Format("Unhandled error: {}", err.Message)) ExitApp(1) - return -1 ; suppress the standard error dialog + return -1 } SetWorkingDir(A_ScriptDir) CoordMode("Pixel", "Screen") CoordMode("Mouse", "Screen") -ClickWithMarker(x, y, button := "Left") { - Click(x, y, button) +ClickWithMarker(x, y) { + Click(x, y) Sleep(10) MouseMove(30, 30) Log(Format("Clicked at {1}, {2}", x, y)) } -; Single-pass image search inside a window. Returns true + center coords. -TryFindImage(winTitle, imageFile, &outX, &outY) { - try { - WinGetPos(&wx, &wy, &ww, &wh, winTitle) - } catch { - return false +; Wait until an installer window exists AND has a real (non-phantom) size, +; then return its rect. ahk_exe can transiently match a hidden helper. +WaitForRealWindow(winTitle, timeoutMs) { + deadline := A_TickCount + timeoutMs + while (A_TickCount < deadline) { + for hwnd in WinGetList(winTitle) { + try { + WinGetPos(&wx, &wy, &ww, &wh, "ahk_id " hwnd) + if (ww > 400 && wh > 300) { + return { hwnd: hwnd, x: wx, y: wy, w: ww, h: wh } + } + } catch { + continue + } + } + Sleep(500) } + throw Error(Format("no real-sized window matched {} within {}ms", winTitle, timeoutMs)) +} + +; Image search inside a rect. Returns true + center coords. +TryFindImage(rect, imageFile, &outX, &outY) { hBitmap := LoadPicture(imageFile) if !hBitmap { throw Error("LoadPicture failed: " imageFile) } - bm := Buffer(32, 0) ; BITMAP structure on x64 + bm := Buffer(32, 0) DllCall("GetObject", "Ptr", hBitmap, "Int", bm.Size, "Ptr", bm) width := NumGet(bm, 4, "Int") height := NumGet(bm, 8, "Int") - if ImageSearch(&x, &y, wx, wy, wx + ww, wy + wh, Format("*10 {}", imageFile)) { + if ImageSearch(&x, &y, rect.x, rect.y, rect.x + rect.w, rect.y + rect.h, Format("*20 {}", imageFile)) { outX := x + Floor(width / 2) outY := y + Floor(height / 2) return true @@ -80,21 +93,13 @@ TryFindImage(winTitle, imageFile, &outX, &outY) { return false } -; Fractional window position -> screen coords (fallback click target). -WindowRelPoint(winTitle, fx, fy, &outX, &outY) { - WinGetPos(&wx, &wy, &ww, &wh, winTitle) - outX := wx + Floor(ww * fx) - outY := wy + Floor(wh * fy) -} - BootstrapLogContains(needle) { global bootstrapLog if (bootstrapLog = "" or !FileExist(bootstrapLog)) { return false } try { - ; Read-share open: the installer still holds the file for writing. - f := FileOpen(bootstrapLog, "r-d") + f := FileOpen(bootstrapLog, "r-d") ; read, share read+write if !f { return false } @@ -109,26 +114,26 @@ BootstrapLogContains(needle) { installerWin := "ahk_exe " setupExe appWin := "ahk_exe Hermes.exe" -; The Install/Launch button sits centered horizontally near the bottom of -; the installer window (measured from production screenshots; used only -; when the image template fails to match a restyled UI). +; Button center as a fraction of the window rect (installer is ~full-screen +; on the runner). Measured from a live CI frame: center ~ (0.50, 0.59). BTN_FX := 0.50 -BTN_FY := 0.87 +BTN_FY := 0.59 -Log("Waiting for the installer window (" installerWin ") ...") +Log("Waiting for a real-sized installer window (" installerWin ") ...") +rect := WaitForRealWindow(installerWin, 90000) +Log(Format("Installer window: x={1} y={2} w={3} h={4}", rect.x, rect.y, rect.w, rect.h)) try { - WinWait(installerWin, , 60) + WinActivate("ahk_id " rect.hwnd) + Sleep(500) } catch { - throw Error("installer window did not appear within 60s") + Log("WARNING: could not activate installer window") } -WinGetPos(&x, &y, &w, &h, installerWin) -Log(Format("Window found at x={1} y={2} w={3} h={4}", x, y, w, h)) -; ── Step 1: click Install (template first, relative-position fallback) ── +; ── Step 1: click Install ─────────────────────────────────────────────── installClicked := false -deadline := A_TickCount + 60000 +deadline := A_TickCount + 30000 while (A_TickCount < deadline) { - if TryFindImage(installerWin, A_ScriptDir "\install-button.png", &ix, &iy) { + if TryFindImage(rect, A_ScriptDir "\install-button.png", &ix, &iy) { ClickWithMarker(ix, iy) Log("Install clicked (template match)") installClicked := true @@ -137,48 +142,58 @@ while (A_TickCount < deadline) { Sleep(500) } if !installClicked { - WindowRelPoint(installerWin, BTN_FX, BTN_FY, &ix, &iy) + ix := rect.x + Floor(rect.w * BTN_FX) + iy := rect.y + Floor(rect.h * BTN_FY) ClickWithMarker(ix, iy) - Log("FALLBACK: install template never matched; clicked relative position") + Log("FALLBACK: install template never matched; clicked window-relative position") } ; ── Step 2: wait for the install to finish ────────────────────────────── -; Primary signal: the Launch button template appears. Secondary signal: -; "bootstrap complete" in bootstrap-installer.log (the installer's own -; completion line) — after which we give the template 2 more minutes and -; then fall back to the relative-position click. +; Primary: the "bootstrap complete" line in the installer's own log — the +; authoritative done signal. Secondary: the Launch button template. launchX := 0, launchY := 0 launchFound := false -completeSince := 0 +complete := false waitDeadline := A_TickCount + 1000 * 60 * 45 -Log("Waiting for install to finish (Launch template or bootstrap log) ...") +Log("Waiting for install to finish (bootstrap log or Launch template) ...") while (A_TickCount < waitDeadline) { - if TryFindImage(installerWin, A_ScriptDir "\launch-button.png", &launchX, &launchY) { - launchFound := true - Log("Install finished (Launch template visible)") + if BootstrapLogContains("bootstrap complete") { + complete := true + Log("bootstrap-installer.log reports completion") break } - if (completeSince = 0 and BootstrapLogContains("bootstrap complete")) { - completeSince := A_TickCount - Log("bootstrap-installer.log reports completion; giving the Launch template 120s") - } - if (completeSince > 0 and A_TickCount - completeSince > 120000) { - Log("FALLBACK: log says complete but Launch template never matched") + ; refresh the rect (window can move/resize between stages) + try rect := WaitForRealWindow(installerWin, 2000) + if TryFindImage(rect, A_ScriptDir "\launch-button.png", &launchX, &launchY) { + launchFound := true + Log("Install finished (Launch template visible)") break } Sleep(2000) } -if (!launchFound and completeSince = 0) { - throw Error("install did not finish within 45 minutes (no Launch button, no completion log line)") +if (!launchFound and !complete) { + throw Error("install did not finish within 45 minutes (no completion log line, no Launch button)") +} + +; Give the button a moment to swap Install -> Launch after completion. +if (complete and !launchFound) { + Sleep(2000) + try rect := WaitForRealWindow(installerWin, 10000) + if TryFindImage(rect, A_ScriptDir "\launch-button.png", &launchX, &launchY) { + launchFound := true + Log("Launch template matched after completion") + } } ; ── Step 3: click Launch — the hand-off under test ────────────────────── if launchFound { ClickWithMarker(launchX, launchY) + Log("Launch clicked (template)") } else { - WindowRelPoint(installerWin, BTN_FX, BTN_FY, &lx, &ly) + lx := rect.x + Floor(rect.w * BTN_FX) + ly := rect.y + Floor(rect.h * BTN_FY) ClickWithMarker(lx, ly) - Log("FALLBACK: clicked Launch at relative position") + Log("FALLBACK: clicked Launch at window-relative position") } Log("Launch clicked; waiting for the Hermes desktop app window") @@ -191,9 +206,6 @@ try { WinGetPos(&ax, &ay, &aw, &ah, appWin) Log(Format("App window appeared at x={1} y={2} w={3} h={4}", ax, ay, aw, ah)) -; Give the renderer a few seconds on screen (recorded as proof), then hand -; control back to the PowerShell driver, which closes the app and re-launches -; it under Playwright for the update legs. -Sleep(8000) +Sleep(8000) ; let the renderer paint (recorded as proof) Log("done") ExitApp(0) diff --git a/tests/install/e2e-assets/install-button.png b/tests/install/e2e-assets/install-button.png index feed47bd98fb3ffe464025c6f52553205ed89751..7d86dbc4c0e67252abe3fc37f630707c94dd73c1 100644 GIT binary patch literal 911 zcmeAS@N?(olHy`uVBq!ia0vp^vw>KLgAGU)tvlt&z`z{l>EaktG3V_aUymt{GRHr@ z*I72rW5QmCsCFY)tq%*_@?9iDCM;+Vut^ChxwWVKmGjF+<71jny)Lwr zFPu;rGR1>4TP5d^6zA*>Za~cwH{jx`JzmEzeUJa(-@3gW+Y))-++Fl1$InmdKil_t zyL;5X9{SOkb@bKKe_M`zpSN~}KvbFLdB^P6ZHL0jZ#~Yt|HUX)@NZSP3%j*cb645) zjGU^S8(FTD`&|vb`R>t?<H=t_7*f7cF$#m~!%u{r(9xZ*Phn+G;Q-KlF|D>!KX}IeDjKZms5j_+s9f($@7a zp6iE;9Q8h%nsxf&=Tuh{ws!p;F;*)x))-IkJ+@#C%c@ffm)wsPIX+Fr*bI5uhSU$1!uI{f}g__!dIk%U*mh;xh_!_BhkQg1#bX$ADbJk<>dn+{OSS_je z#TF)O&cE=p;j|0KvsN|#jTUWIHh566USFI=Er;Q4yNEvf!sYk>)FsQR(hb_QH9IPzsdiNl8~DptP{Vi-AR9r zMU70`-SilP!ERC;S>)4Aei@aa7&kX>vuQi%aZ(g`=4X z7k}K&H@9=rgRW0ey#M^>vB?FWJGHerx>agf5Bl%>mZMZY(`d@q=Q&E%9yeM{h0GFr zTYMvw6NRr~;U3%2I%QwzQh_*d!tC+h{Ijw?@4=t!jk8sYF7C;D!1M3DfWS`ob2q^T boz4Fqq4?@TW5|7AMq}`F^>bP0l+XkKhMBWX literal 1200 zcmX9;Ycv!H6rNFQVwr43op!QisWz%piB^c1W3(lqj7@2#u)7|yN|b~R&4`s_WcAQP zsV0vZjgg@envmCE!X}TXd5}p{G45q&cklTg=bm%#x#xU8F3Z!y%}{@-K7l|mbf>v` zX+1|vt9f&@TAIPD)fzF%%WXTcr@`usw$Y2A`9%>3UoZTZL_*%>CE9@k&z(Nh$*Gwu zRp@yRmmWcaH}(wAxF(`qEN(rCc@L4Ah>MwM_A6S1qJJiie*{TC^)KN;6aMuOKB!@26li|{ zM+gsFvAP`|OJGt1j&a!ZHyG~6s&=fC!nH?GBt~Hi#@xis(fH~u49LMD4nKZ|kUZ>u z11m!@?ly#7#)=k{kHfap=#dKf)!-Zt={$Jc0R=U{$VOH%q}MF_aLtUBd(%! z2u>HGUj{z!(JsR9DD;j%N)h(GgM0NeyHc>de}?3X2K%*lXN0l2ECNC2mbqBro^8-rYqMXuvo?d}atT~~FV z&z{X4YpWL&@F_u?%FRQ{mN3ic|G{AB89%ZnbA)JeLiU0i>M-Fu>Ex~WM3(2%&!99I>XmJ$I2a?r}x9-m6nNn4ome4c4RL5 zr1^=mja6vy@s^Xn{7`koQl8mNT0H13|81Z*ajz9KjN_{zXLK>_qq1+l@~QWN=OUdP zsg5KzzMQg2K7Mdr#Av#aYS~j^XmzD=K$||Pja;!0LrFb_vK(=pNU;L$tjCcGE-1PbO7Q;JLf~#Fw>(v1c6!H+q!`M{ft3BwzrCVC>@-Y;#GA!%@pONb9xbfzG zedk-q{cp%sv_N^UO&K+u$}Di$SdnZIlDyOVyASOp$#ZxP9<%$q4?4aT2l49uN267m zX*ylK{H`*whkU^Fb<@~4Yi$l(IUy+PaH}|K6X;zTA(oH^q)h?7Cu#O{dm~-#*CDu5 LJzR^oA4>WMWp}7g From 60493c9e7efc9fc8b0d98d0674c7d1dfde13320e Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Tue, 11 Aug 2026 02:00:44 -0700 Subject: [PATCH 025/227] fix(e2e): re-capture launch-button template + correct fallback fraction MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Attempt 3's proof frames showed the install SUCCEEDED end-to-end (bootstrap complete, installer self-copied to HERMES_HOME) and the window advanced to 'HERMES IS READY' with a [ LAUNCH ] button at the same centered CTA spot the [ INSTALL ] button occupied — screen (511,454) inside window x=64 y=34 w=896 h=659. Two Launch-step bugs, both fixed from that evidence: * launch-button.png was the stale #68183 template and never matched the restyled '[ LAUNCH ]' button. Re-captured from the live frame. * the window-relative fallback used fy=0.59, clicking y=422 — above the real button. Correct fraction is (454-34)/659 = 0.637. With the template now matching, the fallback is belt-and-braces anyway. Install click, completion detection, and the app-window wait were all already correct in attempt 3; only the Launch click missed. --- .../install/e2e-assets/install-and-launch.ahk | 6 ++++-- tests/install/e2e-assets/launch-button.png | Bin 1485 -> 882 bytes 2 files changed, 4 insertions(+), 2 deletions(-) diff --git a/tests/install/e2e-assets/install-and-launch.ahk b/tests/install/e2e-assets/install-and-launch.ahk index 873b3a5d6580..03079212099b 100644 --- a/tests/install/e2e-assets/install-and-launch.ahk +++ b/tests/install/e2e-assets/install-and-launch.ahk @@ -115,9 +115,11 @@ installerWin := "ahk_exe " setupExe appWin := "ahk_exe Hermes.exe" ; Button center as a fraction of the window rect (installer is ~full-screen -; on the runner). Measured from a live CI frame: center ~ (0.50, 0.59). +; on the runner). Measured from a live CI frame where the template match +; landed at screen (511,454) inside window x=64 y=34 w=896 h=659: +; fx = (511-64)/896 = 0.50 ; fy = (454-34)/659 = 0.637 BTN_FX := 0.50 -BTN_FY := 0.59 +BTN_FY := 0.637 Log("Waiting for a real-sized installer window (" installerWin ") ...") rect := WaitForRealWindow(installerWin, 90000) diff --git a/tests/install/e2e-assets/launch-button.png b/tests/install/e2e-assets/launch-button.png index 6ab89a75ca32480987264e3ee359063cb11f6446..7b2d9fd55c29973367910d9616d5cf39ec2187ac 100644 GIT binary patch literal 882 zcmeAS@N?(olHy`uVBq!ia0vp^i-1^3l(tCaHo)+2_R4lX@y=S{wenfAT#u^WnMH|Ep*G{5Ze(z3t;R`zJPr zJoQbNjuzh2JgVns=2dqtz$MMiQAAk8#n2IjO1xOZpJ0_cciy~xA8cY@N9@n4J3iUu zeqG&e^8@SR@BZP-5bAu*Dcg$=uKadw4!Hb6Gxx)m6t()lS;afp z4W_hy5j`)z{h0jvEr})jzvbPy^dj@UT>NHL=B2#qZpsS;-%@`H5fucXwe=j5A1S|69qUas9_JNwbq=wQy_}rsN-ekDuxu-zD05eY2aap}thh z|6dQcpI-d%br+ZQ;X9i-``(q@sW!5^;C=c~jfoFq-5PB%RriN zHUHNet5|OKY^Q|%sf4(<^Q~^wga{ zoBLMbrimSWH!ic=*2a~GCH?qz&g;|%*WAytfs@?pMJF5y5;)4j8t4;%gKD|qs8cxM z^x0RZ1a)Ow#)W*urR5B%fUWTV`ODX4Z zg&TGbNRLIMRRmgJ!zUcHkH#&&xXufAT|kR)T%GuQJaD7{bV(t+;a&TNx10@+TFx2V)9cjjE;j(B63<_LIP1)@T>$412LbA zQQ7$U3p5JIvR2Sx;Mgyy?I3?nz%&UvnRqk`Jrgi44|+wgnT|XB@oGA1{*ApNvak_W zxuelVX#p30jKkD4RJDo<5* z4E#PApMbVL(DjqPZGiX!=pP2*5U6{jS*WzRo2=^q|0kGSii0B%n+pwH&^-t#Wl+;j z7Sv;E8Azs}vJC>0AuJu)g)lM(lM=|N#1bAP6~THMHVeszMZkQDL*K#m4hAM;S~=K6 z%666(H)CZRc_$xkWZ|z#SvC33Ylz8#lPsV=z=08H7D8h;zO2CaO?W#WPO$J+4qQ%= zo_mCOb!cz_)jaW46|@LRb|DV`fH(CJl!E*|vTFcn_ki08nU&DdPi{SnjRLT}flb}y zg=ZL(3mV>7+6rc&m{TJ?&cfCn=~^0UoJPSQ&I?>OLJ%`;%uV*&1+c%hounQ)yjo%O zXD;`i$MdaAmhd;WY(L{|n#obK(=)zdd)MLK;qg0}3+xl2nFoh=`XSGw=qvE@OaAybf+m>9iCpe?z~FB-U5aQV$ZY0)I%;u}Ka_%!u| zgwd|=r_5>Yc|)n!t1$7&iZA_9kr0tEZ(u>mmQUNc6x9*>z6lL0?YeS*#n>`Ri;`~7 z{29HJXi8#}Jo}C#$1!m#Dpy0lt7KQ%a?hQeDqUX)udmInv5t&-Uar!-4!;I1ufV!5 zM%RaR$J89!O&3Ur(ig$fM}#mgsAJi_E3wNvUgIJ5zY=w0F@)Q!3wpKIZO+!z?#TO1 zOvQ{T&7j)torda1B&`~+ulN}0mWI#Ni(ncm z@3wr5UVre8OP#8|f_}JHR;{MUc|=bqk-`IEAqm2l*Y3d#YFO z*`i2KZOHXk*)%po-rgbhrAi)^$CP&&?LDvfO2pWs+I^nBn=0`hJ{-N_g}(agF7fO} zVahkOK}K!TDtgF-I}zC^nSyIe%we(5@#sIZ}Frpd1=*Iqvkxj4W(|uPZm-w z!)`sk7cb1|F>~=y?)-Yqv#vZj&nIa01^HT0`IVt%V;g;G1^K*MZcnf1*sK%k!4Mq)=rnhkV|2X<_Pn&Y8)$Fv{Z=$k)A563t&_XmW^%wh_x4v*HR=XCSv1^LA z!&B*VJpagR>xAGj_O^rO%ebOHVwQPZGICPO6xj975up! zEbWq9o zi|RJRl|2PUl|D-upFHF|&$Qig{bj72ewS0j+@3*+V;PKj)HENle|ObMH8UjO`TC8v h%6~e3;57(9p257kTi97NOLobGx#>ZZOyg6b{{c0JSjYeX From f95506e665a00d0ad4980a07b65d685dae09744b Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Tue, 11 Aug 2026 02:12:37 -0700 Subject: [PATCH 026/227] fix(e2e): don't require installer pin to be an ancestor of CURRENT MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The full GUI install flow now works end-to-end (attempt 4 proof: Install clicked, bootstrap complete, Launch clicked, real Hermes.exe window appeared 1024x720, installer exited, 5 Hermes processes running). The only failure was an over-strict staging assertion. The website Hermes-Setup.exe pins a main release commit. On a real push-to-main run CURRENT is main's tip, so that pin is its ancestor and the check holds. On a diverged feature branch CURRENT is a branch commit the release pin is not an ancestor of — a legitimate topology, not a bug. The update leg resets the checkout to serve.git's main ref (= CURRENT) regardless of ancestry and asserts it lands there, which is the actual forward-update proof. Downgrade the ancestor check to an informational note so branch validation can exercise the update legs. --- tests/install/windows-desktop-gui-e2e.ps1 | 15 ++++++++++++--- 1 file changed, 12 insertions(+), 3 deletions(-) diff --git a/tests/install/windows-desktop-gui-e2e.ps1 b/tests/install/windows-desktop-gui-e2e.ps1 index d5cda4b694f7..f466d716d18e 100644 --- a/tests/install/windows-desktop-gui-e2e.ps1 +++ b/tests/install/windows-desktop-gui-e2e.ps1 @@ -335,8 +335,13 @@ function Invoke-PhaseInstallGui { Stop-HermesAppProcesses "post-install" # The website exe installs its baked release pin — record it as the - # "before" version. Must be an ancestor of CURRENT (i.e. genuinely - # "before" this commit) and must not already BE CURRENT. + # "before" version. It must differ from CURRENT (an update is genuinely + # available). We do NOT require it to be an ancestor of CURRENT: on a + # real `push: main` run CURRENT is main's tip and the release pin is + # behind it (ancestor), but on a feature branch CURRENT has diverged + # from main, so the release pin legitimately isn't in its ancestry. The + # update leg resets the checkout to serve.git's main ref regardless, and + # asserts it lands on CURRENT — that is the real forward-update proof. $installedSha = Get-InstalledHead Write-Host " installer landed on: $installedSha (website release pin)" Assert-True ($installedSha -ne $state.current) "installed pin differs from CURRENT (an update is genuinely available)" @@ -344,7 +349,11 @@ function Invoke-PhaseInstallGui { & git -C $InstallDir merge-base --is-ancestor $installedSha $state.current 2>&1 | Out-Null $isAncestor = ($LASTEXITCODE -eq 0) $ErrorActionPreference = $prevEap - Assert-True $isAncestor "installed pin is an ancestor of CURRENT (before -> current is a forward update)" + if ($isAncestor) { + Write-Host " [ok] installed pin is an ancestor of CURRENT (linear before -> current)" + } else { + Write-Host " [note] installed pin is NOT an ancestor of CURRENT — expected on a diverged feature branch; the update leg still resets to CURRENT" + } Test-HermesRuns "post-install-gui" Assert-True ($null -ne (Get-DesktopExe)) "packaged Desktop Hermes.exe exists" From f83548420a16b3a334df8e9b4681f833713dc9b5 Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Tue, 11 Aug 2026 02:24:01 -0700 Subject: [PATCH 027/227] fix(e2e): pure-ASCII PowerShell/AHK (PS 5.1 parser choke on non-ASCII) MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Windows PowerShell 5.1 reads .ps1 without a BOM under the legacy OEM codepage, mis-decoding UTF-8 bytes. Em-dashes/box-drawing survived in comments through attempts 2-4, but the previous commit added an em-dash INSIDE a double-quoted Write-Host string — the misdecode there ate the quote boundary and cascaded into a whole-file parse failure at the Stage step ('Unexpected token', 'string is missing the terminator'). scripts/install.ps1 documents this exact constraint ('pure ASCII for PS 5.1 parser compatibility'). Strip all non-ASCII from the .ps1 and .ahk files (em-dash->--, arrows->->, box-drawing->-). drive-update.cjs keeps UTF-8 (Node decodes it natively). Both PowerShell files parse clean. --- .../install/e2e-assets/install-and-launch.ahk | 14 +++++++------- tests/install/windows-desktop-gui-e2e.ps1 | 18 +++++++++--------- 2 files changed, 16 insertions(+), 16 deletions(-) diff --git a/tests/install/e2e-assets/install-and-launch.ahk b/tests/install/e2e-assets/install-and-launch.ahk index 03079212099b..960a740bb30f 100644 --- a/tests/install/e2e-assets/install-and-launch.ahk +++ b/tests/install/e2e-assets/install-and-launch.ahk @@ -6,14 +6,14 @@ ; the real Hermes.exe (Electron desktop) window to appear. ; ; Adapted from @ethernet8023's e2e/windows/install-hermes-desktop.ahk -; (PR #68183) — same ImageSearch approach; the install-button template was +; (PR #68183) -- same ImageSearch approach; the install-button template was ; re-captured from a live CI frame (the #68183 templates predated the ; installer UI restyle to the "[ INSTALL ]" bracket look and never matched). ; -; Robustness beyond the original — every one earned from a real CI failure: +; Robustness beyond the original -- every one earned from a real CI failure: ; * Log() survives a missing stdout. GUI-subsystem AHK started via ; Start-Process has no console, so FileAppend to '*' throws -; "(6) The handle is invalid" — that single throw killed attempt 1. +; "(6) The handle is invalid" -- that single throw killed attempt 1. ; * We wait for a REAL-sized installer window before reading geometry. ; ahk_exe matched a hidden 16x16 helper window first (attempt 2's ; "Window found at ... w=16 h=16"), so relative-position math was @@ -131,7 +131,7 @@ try { Log("WARNING: could not activate installer window") } -; ── Step 1: click Install ─────────────────────────────────────────────── +; -- Step 1: click Install ----------------------------------------------- installClicked := false deadline := A_TickCount + 30000 while (A_TickCount < deadline) { @@ -150,8 +150,8 @@ if !installClicked { Log("FALLBACK: install template never matched; clicked window-relative position") } -; ── Step 2: wait for the install to finish ────────────────────────────── -; Primary: the "bootstrap complete" line in the installer's own log — the +; -- Step 2: wait for the install to finish ------------------------------ +; Primary: the "bootstrap complete" line in the installer's own log -- the ; authoritative done signal. Secondary: the Launch button template. launchX := 0, launchY := 0 launchFound := false @@ -187,7 +187,7 @@ if (complete and !launchFound) { } } -; ── Step 3: click Launch — the hand-off under test ────────────────────── +; -- Step 3: click Launch -- the hand-off under test ---------------------- if launchFound { ClickWithMarker(launchX, launchY) Log("Launch clicked (template)") diff --git a/tests/install/windows-desktop-gui-e2e.ps1 b/tests/install/windows-desktop-gui-e2e.ps1 index f466d716d18e..2a7b80e53f72 100644 --- a/tests/install/windows-desktop-gui-e2e.ps1 +++ b/tests/install/windows-desktop-gui-e2e.ps1 @@ -9,7 +9,7 @@ # then clicks Launch. The real Electron Hermes.exe must appear. # The exe runs EXACTLY as shipped: it downloads its own pinned # install.ps1 from GitHub raw and installs its baked -# BUILD_PIN_COMMIT — so the "before" version in this harness is +# BUILD_PIN_COMMIT -- so the "before" version in this harness is # the website release pin, the literal starting point of every # real GUI user. (The strict HEAD~1 -> HEAD guarantee is the # contract harness's job; this one covers the production pin.) @@ -30,7 +30,7 @@ # # DEVIATIONS FROM PRODUCTION (each one deliberate and small): # * git URL redirect (GIT_CONFIG_GLOBAL) routes the canonical repo URLs -# to the staged serve.git — the staging requirement itself. The +# to the staged serve.git -- the staging requirement itself. The # installer's raw.githubusercontent install.ps1 download and the pinned # ZIP fallback are NOT redirected (real network, as shipped). # * serve.git gets uploadpack.allowAnySHA1InWant=true so the installer's @@ -198,7 +198,7 @@ function Stop-DesktopRecorder($proc, [string]$OutDir) { function Stop-HermesAppProcesses([string]$Label) { # Close the desktop app the blunt way between legs (a user quitting). - # Only Hermes.exe (Electron) — never hermes.exe (the venv CLI shim). + # Only Hermes.exe (Electron) -- never hermes.exe (the venv CLI shim). $procs = @(Get-Process -Name "Hermes" -ErrorAction SilentlyContinue) foreach ($p in $procs) { try { Stop-Process -Id $p.Id -Force -ErrorAction SilentlyContinue } catch {} @@ -226,7 +226,7 @@ function Get-ManagedNode { } # ---------------------------------------------------------------------------- -# Phase: stage — reuse the contract driver's stage (identical staging) +# Phase: stage -- reuse the contract driver's stage (identical staging) # ---------------------------------------------------------------------------- function Invoke-PhaseStage { Write-Step "STAGE (gui): delegating to windows-desktop-e2e.ps1 -Phase stage" @@ -244,7 +244,7 @@ function Invoke-PhaseStage { } # ---------------------------------------------------------------------------- -# Phase: install-gui — website Hermes-Setup.exe, headed, AHK-driven +# Phase: install-gui -- website Hermes-Setup.exe, headed, AHK-driven # ---------------------------------------------------------------------------- function Invoke-PhaseInstallGui { $state = Read-State @@ -287,7 +287,7 @@ function Invoke-PhaseInstallGui { try { Save-DesktopScreenshot (Join-Path $proof "00-before-installer.png") - # Launch the REAL installer, headed — exactly a double-click. + # Launch the REAL installer, headed -- exactly a double-click. $installer = Start-Process -FilePath $setupExe -PassThru Write-Host " Hermes-Setup.exe launched (pid $($installer.Id))" @@ -334,14 +334,14 @@ function Invoke-PhaseInstallGui { # Close the freshly launched app (user quits after first look). Stop-HermesAppProcesses "post-install" - # The website exe installs its baked release pin — record it as the + # The website exe installs its baked release pin -- record it as the # "before" version. It must differ from CURRENT (an update is genuinely # available). We do NOT require it to be an ancestor of CURRENT: on a # real `push: main` run CURRENT is main's tip and the release pin is # behind it (ancestor), but on a feature branch CURRENT has diverged # from main, so the release pin legitimately isn't in its ancestry. The # update leg resets the checkout to serve.git's main ref regardless, and - # asserts it lands on CURRENT — that is the real forward-update proof. + # asserts it lands on CURRENT -- that is the real forward-update proof. $installedSha = Get-InstalledHead Write-Host " installer landed on: $installedSha (website release pin)" Assert-True ($installedSha -ne $state.current) "installed pin differs from CURRENT (an update is genuinely available)" @@ -352,7 +352,7 @@ function Invoke-PhaseInstallGui { if ($isAncestor) { Write-Host " [ok] installed pin is an ancestor of CURRENT (linear before -> current)" } else { - Write-Host " [note] installed pin is NOT an ancestor of CURRENT — expected on a diverged feature branch; the update leg still resets to CURRENT" + Write-Host " [note] installed pin is NOT an ancestor of CURRENT -- expected on a diverged feature branch; the update leg still resets to CURRENT" } Test-HermesRuns "post-install-gui" Assert-True ($null -ne (Get-DesktopExe)) "packaged Desktop Hermes.exe exists" From 256c9b350cf7c92f3833271225eafbe3be89fcad Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Tue, 11 Aug 2026 02:35:02 -0700 Subject: [PATCH 028/227] fix(e2e): resolve @playwright/test via Node, not a hoist-blind path check MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Install + GUI update leg now reached (attempt 6): full install passes, first update leg begins. It tripped a preflight assert checking apps/desktop/node_modules/@playwright/test — but the root npm ci HOISTS workspace devDependencies to the REPO-ROOT node_modules, so that path is empty by design. Node's own resolution walks up from apps/desktop and finds it (which is exactly how the copied-in drive-update.cjs will load it), so assert via 'node -e require.resolve(...)' from apps/desktop instead of a hardcoded nested path. --- tests/install/windows-desktop-gui-e2e.ps1 | 10 +++++++++- 1 file changed, 9 insertions(+), 1 deletion(-) diff --git a/tests/install/windows-desktop-gui-e2e.ps1 b/tests/install/windows-desktop-gui-e2e.ps1 index 2a7b80e53f72..6f0a23af8486 100644 --- a/tests/install/windows-desktop-gui-e2e.ps1 +++ b/tests/install/windows-desktop-gui-e2e.ps1 @@ -386,7 +386,15 @@ function Invoke-GuiUpdateLeg([string]$TargetSha, [string]$LegName, [string]$LegS $node = Get-ManagedNode $appsDesktop = Join-Path $InstallDir "apps\desktop" - Assert-True (Test-Path -LiteralPath (Join-Path $appsDesktop "node_modules\@playwright\test")) "$LegName -- @playwright/test present in installed checkout" + # @playwright/test is a workspace devDependency; the root `npm ci` + # HOISTS it to the repo-root node_modules, not apps/desktop's. Resolve + # it the way Node will (walk up from apps/desktop) instead of asserting + # a hardcoded path that hoisting makes wrong. + $prevEap = $ErrorActionPreference; $ErrorActionPreference = "Continue" + & $node -e "require.resolve('@playwright/test', { paths: [process.argv[1]] })" $appsDesktop 2>&1 | Out-Null + $pwResolved = ($LASTEXITCODE -eq 0) + $ErrorActionPreference = $prevEap + Assert-True $pwResolved "$LegName -- @playwright/test resolvable from installed apps/desktop" $recorder = Start-DesktopRecorder (Join-Path $proof "desktop-frames") try { From 31b97289aeb74ae067ded3c701755647084ace85 Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Tue, 11 Aug 2026 02:47:01 -0700 Subject: [PATCH 029/227] fix(e2e): dismiss onboarding before opening Settings in the update driver Attempt 7 got the whole way into the GUI update leg: the installed Electron app launched under Playwright, booted, composer attached, first screenshot captured. It then couldn't find the settings gear -- the ERROR screenshot showed why: a fresh install with no CONFIGURED provider (the seeded .env key isn't read as model.provider) shows the onboarding card ('Let''s get you setup with Hermes Agent'), which covers the shell and its settings gear. The update path needs no provider, so the driver now clicks 'I'll choose a provider later' (with skip fallbacks) to dismiss onboarding and reach the shell before looking for the gear. Harmless no-op when onboarding isn't shown. Gear (aria-label 'Open settings') and About nav ('About') selectors already match the real components. --- tests/install/e2e-assets/drive-update.cjs | 22 ++++++++++++++++++++++ 1 file changed, 22 insertions(+) diff --git a/tests/install/e2e-assets/drive-update.cjs b/tests/install/e2e-assets/drive-update.cjs index c328ff89380b..5eecdb6a5947 100644 --- a/tests/install/e2e-assets/drive-update.cjs +++ b/tests/install/e2e-assets/drive-update.cjs @@ -106,6 +106,28 @@ async function main() { await page.waitForTimeout(3000) await shot(page, '01-app-booted') + // ── Dismiss onboarding if present ───────────────────────────────────── + // A fresh install with no configured provider shows the onboarding card + // ("Let's get you setup..."). The update path needs no provider, so skip + // it via "I'll choose a provider later" to reach the app shell (which + // has the settings gear). Harmless no-op if onboarding isn't shown. + const dismissedOnboarding = await clickFirstVisible( + page, + [ + p => p.getByRole('button', { name: /choose a provider later/i }), + p => p.getByText(/choose a provider later/i), + p => p.getByRole('button', { name: /skip/i }) + ], + 'skip onboarding', + 8_000 + ) + + if (dismissedOnboarding) { + log('dismissed onboarding overlay') + await page.waitForTimeout(2500) + await shot(page, '01b-onboarding-dismissed') + } + // ── Open Settings (titlebar gear) ───────────────────────────────────── const openedSettings = await clickFirstVisible( page, From 7ffc26ec563de453f7e827b873a4d959e3327d82 Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Tue, 11 Aug 2026 03:01:52 -0700 Subject: [PATCH 030/227] fix(e2e): detect update hand-off via marker file, not Playwright close event Attempt 8 drove the ENTIRE GUI update click-path successfully: onboarding dismissed, Settings opened, About opened, Update now clicked, updating overlay shown. The hand-off log proves the real update then ran: desktop (pid 8880) exited, venv unlocked, 'hermes update --yes --gateway --force --branch main' fetched from serve.git, found 1 new commit, pulled, and restored. Everything worked. The only failure was the driver waiting on Playwright's app 'close' event, which doesn't fire reliably when the Electron app self-quits for the hand-off. Switch to the authoritative signal: poll for the HERMES_HOME/.hermes-update-in-progress marker (or the result JSON, or a genuine window-gone), which the hand-off writes ~4s after the click. The PowerShell driver still owns asserting the OUTCOME (target sha, marker cleanup, working hermes, relaunched app) after the driver returns. --- tests/install/e2e-assets/drive-update.cjs | 68 +++++++++++++++++++---- 1 file changed, 56 insertions(+), 12 deletions(-) diff --git a/tests/install/e2e-assets/drive-update.cjs b/tests/install/e2e-assets/drive-update.cjs index 5eecdb6a5947..25d436072d27 100644 --- a/tests/install/e2e-assets/drive-update.cjs +++ b/tests/install/e2e-assets/drive-update.cjs @@ -201,20 +201,64 @@ async function main() { await page.waitForTimeout(1200) await shot(page, '05-updating-overlay') - // ── Wait for the app to quit for the hand-off ───────────────────────── - await new Promise((resolve, reject) => { - const t = setTimeout( - () => reject(new Error('app did not quit within 120s of Update now — hand-off did not start')), - 120_000 - ) - - app.on('close', () => { - clearTimeout(t) - resolve() - }) + // ── Wait for the hand-off to take over ──────────────────────────────── + // Clicking Update now spawns the detached updater (desktop-update.ps1 or + // the staged binary), which claims HERMES_HOME/.hermes-update-in-progress + // and then the desktop quits. We do NOT rely on Playwright's app 'close' + // event: when the app self-quits for the hand-off that event is + // unreliable (attempt 8 timed out on it even though the hand-off log + // proved the desktop had exited and `hermes update` was already running). + // + // The authoritative "hand-off started" signal is the marker file (or the + // result JSON, if the whole update finished fast). Poll for either, and + // also accept a genuine app close. Any one is success — the PowerShell + // driver owns asserting the update's OUTCOME (sha, marker cleanup, + // relaunch) after we return. + const hermesHome = process.env.HERMES_HOME + const markerPath = hermesHome ? path.join(hermesHome, '.hermes-update-in-progress') : null + const resultPath = hermesHome ? path.join(hermesHome, '.hermes-update-result.json') : null + + let appClosed = false + app.on('close', () => { + appClosed = true }) - log('app quit for updater hand-off — success, the detached updater owns the rest') + const handoffDeadline = Date.now() + 150_000 + let handoffStarted = false + + while (Date.now() < handoffDeadline) { + if (markerPath && fs.existsSync(markerPath)) { + log('hand-off marker present — updater has taken over') + handoffStarted = true + break + } + if (resultPath && fs.existsSync(resultPath)) { + log('update result JSON already present — updater finished fast') + handoffStarted = true + break + } + if (appClosed) { + log('app closed — hand-off in progress') + handoffStarted = true + break + } + // Secondary: if the renderer window is gone, evaluate throws. + try { + await page.evaluate(() => true) + } catch { + log('renderer window gone — app quit for hand-off') + handoffStarted = true + break + } + await new Promise(r => setTimeout(r, 2000)) + } + + if (!handoffStarted) { + await shot(page, 'ERROR-no-handoff') + throw new Error('no hand-off within 150s of Update now (no marker, no result, app still alive)') + } + + log('hand-off confirmed — detached updater owns the rest') } main() From a263a7c670b6aef662dc782ba9e35978dacb1d3c Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Tue, 11 Aug 2026 03:56:38 -0700 Subject: [PATCH 031/227] fix(e2e): allow time for the release->CURRENT desktop rebuild + tail update.log Attempt 9 drove the full real update through the hand-off: marker detected, desktop exited, hermes update fetched from serve.git, found the commit, pulled, restored -- all correct. It then timed out because the updater legitimately runs LONG here: the website release we install (v0.20.0) is weeks of main behind CURRENT, so the update pulls a large diff AND does a full Electron desktop rebuild (vite + electron-builder) plus uv sync. The contract job's BASE->CURRENT is a 1-commit tests-only diff that skips the rebuild, which is why it finishes in ~1 min; the GUI job's release->CURRENT does not. * wait window 40 -> 90 min per leg; job timeout 180 -> 240 min * tail logs/update.log during the wait so the desktop-rebuild phase is visible in CI output instead of tens of minutes of silence (the rebuild streams there, not to the handoff log) The CURRENT->NEXT leg stays fast (NEXT is a same-tree child of CURRENT, no rebuild), so total stays well within 240 min. --- .github/workflows/desktop-windows-e2e.yml | 2 +- tests/install/windows-desktop-gui-e2e.ps1 | 26 ++++++++++++++++++++--- 2 files changed, 24 insertions(+), 4 deletions(-) diff --git a/.github/workflows/desktop-windows-e2e.yml b/.github/workflows/desktop-windows-e2e.yml index 26243856dd70..3d3c6a8adac4 100644 --- a/.github/workflows/desktop-windows-e2e.yml +++ b/.github/workflows/desktop-windows-e2e.yml @@ -121,7 +121,7 @@ jobs: gui-e2e: name: "REAL flow: website setup.exe → GUI update → GUI update" runs-on: windows-latest - timeout-minutes: 180 + timeout-minutes: 240 # Deliberately NOT `needs: e2e` — the jobs run independently, so each # provides rollback coverage for the other and a GUI-layer flake cannot # mask a machinery regression. diff --git a/tests/install/windows-desktop-gui-e2e.ps1 b/tests/install/windows-desktop-gui-e2e.ps1 index 6f0a23af8486..8b4b0adb9239 100644 --- a/tests/install/windows-desktop-gui-e2e.ps1 +++ b/tests/install/windows-desktop-gui-e2e.ps1 @@ -428,14 +428,34 @@ function Invoke-GuiUpdateLeg([string]$TargetSha, [string]$LegName, [string]$LegS # target sha AND the marker is gone). The sha/marker/hermes/relaunch # asserts below are the hard gate either way; the JSON is asserted # only when the script path produced it. - Write-Host " waiting for the detached updater to finish ..." - $deadline = (Get-Date).AddMinutes(40) + # + # This can take a LONG time: the website release we installed is weeks + # of main behind CURRENT, so the update pulls a large diff AND does a + # full Electron desktop rebuild (vite + electron-builder) plus a uv + # sync. The desktop-build output goes to logs/update.log (not the + # streamed handoff log), so we tail update.log here to show progress + # instead of going silent for tens of minutes. + Write-Host " waiting for the detached updater to finish (up to 90 min; large release->CURRENT rebuild) ..." + $updateLog = Join-Path $HermesHome "logs\update.log" + $updateLogPos = 0 + $deadline = (Get-Date).AddMinutes(90) while ((Get-Date) -lt $deadline) { if (Test-Path -LiteralPath $resultPath) { break } $head = "" try { $head = Get-InstalledHead } catch {} if ($head -eq $TargetSha -and -not (Test-Path -LiteralPath $markerPath)) { break } - Start-Sleep -Seconds 10 + # Tail any new update.log lines so the desktop-rebuild phase is + # visible in the CI step output. + if (Test-Path -LiteralPath $updateLog) { + try { + $lines = Get-Content -LiteralPath $updateLog -ErrorAction SilentlyContinue + if ($lines.Count -gt $updateLogPos) { + $lines[$updateLogPos..($lines.Count - 1)] | ForEach-Object { Write-Host " update.log| $_" } + $updateLogPos = $lines.Count + } + } catch {} + } + Start-Sleep -Seconds 20 } if (Test-Path -LiteralPath $resultPath) { $result = Get-Content -LiteralPath $resultPath -Raw | ConvertFrom-Json From cf2692418cb436f255f8a467f10b0f7a56e59382 Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Tue, 11 Aug 2026 05:43:35 -0700 Subject: [PATCH 032/227] fix(e2e): pre-set .skip_upstream_prompt so the fork-only input() can't hang Attempt 10 ran 1h40m and the diagnosis is precise: the GUI update hung on a bare input() in hermes update's _sync_with_upstream_if_needed. Our serve.git origin is a file:// URL, so _is_fork() is true and the updater asks 'Add official repo as upstream? [Y/n]' via raw input(). When the Desktop spawns the hand-off through 'cmd start /min' that child has a real but EMPTY console, so input() blocks forever (no EOF, no keystroke). The contract job spawns the hand-off with inherited non-interactive stdin, so input() hits EOF and defaults immediately -- which is why it never hung. The proof chain confirmed everything else worked: backend exited, venv unlocked, git pull found the commit and applied it; the process then just sat in input(). update.log was never created because the hang is BEFORE the desktop-build step. Fix: create HERMES_HOME/.skip_upstream_prompt after install -- the product's own 'don't ask about upstream' marker (_should_skip_upstream_ prompt). Real GUI users install from the official github origin where _is_fork() is false and this prompt never fires, so this only neutralizes a staging artifact of the file:// serve repo, not real behavior. (Noted for a separate product follow-up: hermes update --gateway should route this input() through _gateway_prompt like its other prompts, so a fork-origin GUI update can't hang even without the marker.) --- tests/install/windows-desktop-gui-e2e.ps1 | 13 +++++++++++++ 1 file changed, 13 insertions(+) diff --git a/tests/install/windows-desktop-gui-e2e.ps1 b/tests/install/windows-desktop-gui-e2e.ps1 index 8b4b0adb9239..8d6f3b54bf71 100644 --- a/tests/install/windows-desktop-gui-e2e.ps1 +++ b/tests/install/windows-desktop-gui-e2e.ps1 @@ -364,6 +364,19 @@ function Invoke-PhaseInstallGui { Add-Content -LiteralPath $envFile -Value "OPENROUTER_API_KEY=sk-or-e2e-placeholder-not-a-real-key" } Write-Host " seeded placeholder provider key for update legs" + + # Suppress the interactive "add upstream remote?" prompt during the GUI + # update legs. Our serve.git origin (file://) looks like a fork to + # `hermes update`, so `_sync_with_upstream_if_needed` would call bare + # input() -- which HANGS FOREVER when the Desktop spawns the hand-off via + # `cmd start /min` (a real but empty console; input() blocks waiting for a + # keystroke that never comes). Real GUI users on the official github + # origin never hit this path (_is_fork is false). The skip marker + # (.skip_upstream_prompt in HERMES_HOME) is the product's own mechanism + # for "don't ask about upstream", so setting it keeps the update flow + # faithful while avoiding the fork-only prompt. + New-Item -ItemType File -Path (Join-Path $HermesHome ".skip_upstream_prompt") -Force | Out-Null + Write-Host " set .skip_upstream_prompt (serve.git origin looks like a fork; avoids the fork-only input() hang)" } # ---------------------------------------------------------------------------- From 0ce3d187d7020b6b657843791129dea7e9d21d35 Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Tue, 11 Aug 2026 06:02:04 -0700 Subject: [PATCH 033/227] =?UTF-8?q?ci(windows):=20GUI=20E2E=20fully=20gree?= =?UTF-8?q?n=20=E2=80=94=20foreground=20relaunched=20window,=20drop=20temp?= =?UTF-8?q?=20trigger?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The real-user-flow job passed end to end (run 31492613931, 10m57s): website Hermes-Setup.exe installed headed (AHK Install+Launch, real app window), then TWO GUI updates driven by real Settings -> About -> 'Update now' clicks, each carried through the detached hand-off to a relaunched desktop on the target commit. Every assertion green on both legs (marker cleanup, checkout on target sha, working hermes, relaunch). Two finishing touches: * Foreground the relaunched Hermes window before the 99-relaunched proof screenshot — the full-desktop grab is z-order dependent and one run caught VS Code on top. The relaunch ASSERT already passed on the process signal; this is purely to make the proof image show Hermes. * Remove the temporary branch push trigger used for pre-merge validation; back to main + nightly + release tags + manual dispatch only. --- .github/workflows/desktop-windows-e2e.yml | 5 +---- tests/install/windows-desktop-gui-e2e.ps1 | 16 ++++++++++++++++ 2 files changed, 17 insertions(+), 4 deletions(-) diff --git a/.github/workflows/desktop-windows-e2e.yml b/.github/workflows/desktop-windows-e2e.yml index 3d3c6a8adac4..4d547a4ec9f0 100644 --- a/.github/workflows/desktop-windows-e2e.yml +++ b/.github/workflows/desktop-windows-e2e.yml @@ -30,10 +30,7 @@ on: # Nightly, off the hour to dodge the top-of-hour runner crunch. - cron: '40 6 * * *' push: - branches: - - main - # TEMPORARY pre-merge validation of the gui-e2e job: remove before merge. - - hermes/hermes-76e68435 + branches: [main] tags: - 'v[0-9]+.[0-9]+.[0-9]+' - 'v[0-9]+.[0-9]+.[0-9]+.[0-9]+' diff --git a/tests/install/windows-desktop-gui-e2e.ps1 b/tests/install/windows-desktop-gui-e2e.ps1 index 8d6f3b54bf71..799ba5ea5501 100644 --- a/tests/install/windows-desktop-gui-e2e.ps1 +++ b/tests/install/windows-desktop-gui-e2e.ps1 @@ -499,6 +499,22 @@ function Invoke-GuiUpdateLeg([string]$TargetSha, [string]$LegName, [string]$LegS } Assert-True ($null -ne $relaunched) "$LegName -- updater relaunched the desktop app" Start-Sleep -Seconds 12 # let the window paint for the screenshot + # Foreground the relaunched Hermes window so the proof screenshot + # captures IT, not whatever else is on top (the full-desktop grab is + # otherwise at the mercy of z-order -- an earlier run caught VS Code). + try { + $mainProc = Get-Process -Name "Hermes" -ErrorAction SilentlyContinue | + Where-Object { $_.MainWindowHandle -ne 0 } | Select-Object -First 1 + if ($mainProc) { + Add-Type -Namespace HdE2E -Name Win -MemberDefinition @' +[System.Runtime.InteropServices.DllImport("user32.dll")] public static extern bool SetForegroundWindow(System.IntPtr h); +[System.Runtime.InteropServices.DllImport("user32.dll")] public static extern bool ShowWindow(System.IntPtr h, int n); +'@ -ErrorAction SilentlyContinue + [HdE2E.Win]::ShowWindow($mainProc.MainWindowHandle, 9) | Out-Null # SW_RESTORE + [HdE2E.Win]::SetForegroundWindow($mainProc.MainWindowHandle) | Out-Null + Start-Sleep -Seconds 2 + } + } catch {} Save-DesktopScreenshot (Join-Path $proof "99-relaunched-desktop.png") } finally { From 1f81c63dd0f782eb2390898540775c8ead1adc6c Mon Sep 17 00:00:00 2001 From: ethernet Date: Tue, 11 Aug 2026 13:29:46 -0400 Subject: [PATCH 034/227] ci(windows-e2e): retire the AHK-only axis - superseded by desktop-windows-e2e.yml The cherry-picked desktop-windows-e2e.yml covers everything the install-e2e-windows-run.yml axis did and more: the contract job drives the same desktop-update.ps1 hand-off (plus a CURRENT->NEXT forward leg), and the GUI job replaces AHK-only driving with the full real user flow - website Hermes-Setup.exe, clicked Install/Launch, then Playwright clicking Settings -> About -> 'Update now' in the packaged app, through the detached hand-off to a relaunched window. Remove the superseded workflow, its driver, and the AHK/button assets under tests/install/windows/ (the GUI job's e2e-assets carry the re-captured templates), and drop the windows-desktop route from install-e2e.yml's dispatch options. --- .github/workflows/install-e2e-windows-run.yml | 117 ---- .github/workflows/install-e2e.yml | 29 +- tests/install/windows/install-button.png | Bin 680 -> 0 bytes .../windows/install-hermes-desktop.ahk | 222 -------- tests/install/windows/install-update-e2e.ps1 | 512 ------------------ 5 files changed, 13 insertions(+), 867 deletions(-) delete mode 100644 .github/workflows/install-e2e-windows-run.yml delete mode 100644 tests/install/windows/install-button.png delete mode 100644 tests/install/windows/install-hermes-desktop.ahk delete mode 100644 tests/install/windows/install-update-e2e.ps1 diff --git a/.github/workflows/install-e2e-windows-run.yml b/.github/workflows/install-e2e-windows-run.yml deleted file mode 100644 index 962878f9b1fd..000000000000 --- a/.github/workflows/install-e2e-windows-run.yml +++ /dev/null @@ -1,117 +0,0 @@ -name: Install & Update E2E — Windows (reusable) - -# Runs ONE Windows desktop update route against ONE starting point, with a -# real install (the published Hermes-Setup.exe: git, uv, a managed Python, -# Node, the venv, the desktop build) behind it. -# -# The Windows sibling of install-e2e-run.yml. There is no bubblewrap here, so -# tests/install/windows/install-update-e2e.ps1 fakes GitHub with git's own -# transport rewrite (an isolated GIT_CONFIG_GLOBAL carrying -# url..insteadOf for both hardcoded repo URLs) instead of a -# MITM proxy — the installer and updater run verbatim against their real URLs -# and land on a local bare repo the driver controls. The bootstrap installer -# itself is a GUI with no headless mode, so AutoHotkey clicks it. -# -# Call it: -# -# jobs: -# windows-desktop: -# uses: ./.github/workflows/install-e2e-windows-run.yml -# with: -# route: desktop - -on: - workflow_call: - inputs: - route: - description: 'Update path to exercise. desktop = the desktop app''s builtin update hand-off (scripts/desktop-update.ps1). TODO: update (hermes update), installer (re-run the bootstrap installer).' - required: false - type: string - default: desktop - installer-url: - description: 'Bootstrap installer to install with. Default: the latest published one — what a user downloads today.' - required: false - type: string - default: https://hermes-assets.nousresearch.com/Hermes-Setup.exe - timeout-minutes: - description: 'Job timeout. A cold run installs real toolchains and builds the desktop app.' - required: false - type: number - default: 75 - -permissions: - contents: read - -jobs: - e2e: - name: ${{ inputs.route }} from published installer - runs-on: windows-latest - timeout-minutes: ${{ inputs.timeout-minutes }} - - steps: - # Full history + tags: the driver seeds its fake GitHub from this - # checkout (all origin branches for the installer's commit pin, release - # tags for the starting base) and promotes HEAD as the update target. - - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2 - with: - fetch-depth: 0 - - - name: Restore cached test tools - id: test-tools-cache - uses: actions/cache@27d5ce7f107fe9357f9df03efb73ab90386fccae # v5 - with: - path: ${{ runner.temp }}\test-bins - key: test-bins-${{ runner.os }}-v1 - - # AutoHotkey drives the installer GUI; ffmpeg records the screen so a - # failed click is diagnosable from the artifact instead of by guesswork. - # Everything lands under RUNNER_TEMP: an untracked dir inside the - # checkout makes the tree dirty, and the driver refuses a dirty tree - # (the update target must be a reviewable commit). - - name: Install AutoHotkey v2 and ffmpeg - if: steps.test-tools-cache.outputs.cache-hit != 'true' - shell: pwsh - run: | - $bins = "$env:RUNNER_TEMP\test-bins" - New-Item -ItemType Directory -Path $bins\autohotkey, $bins\ffmpeg -Force | Out-Null - - # AutoHotkey: copy its whole v2 directory so helper exes/dlls come along. - winget install -e --id AutoHotkey.AutoHotkey --silent --accept-source-agreements --accept-package-agreements --disable-interactivity - $ahkDir = "$env:ProgramW6432\AutoHotkey\v2" - if (-not (Test-Path $ahkDir)) { - throw "AutoHotkey install directory not found: $ahkDir" - } - Copy-Item -Path "$ahkDir\*" -Destination $bins\autohotkey -Recurse -Force - - winget install -e --id Gyan.FFmpeg --silent --accept-source-agreements --accept-package-agreements --disable-interactivity --location "$env:RUNNER_TEMP\ffmpeg_dir" - Copy-Item -Path "$env:RUNNER_TEMP\ffmpeg_dir\*\*" -Destination $bins\ffmpeg -Recurse -Force - - - name: Add test tools to PATH - shell: pwsh - run: | - Add-Content -Path $env:GITHUB_PATH -Value "$env:RUNNER_TEMP\test-bins\autohotkey" - Add-Content -Path $env:GITHUB_PATH -Value "$env:RUNNER_TEMP\test-bins\ffmpeg\bin" - - - name: Run install + update E2E - shell: pwsh - run: | - tests/install/windows/install-update-e2e.ps1 ` - -Route '${{ inputs.route }}' ` - -InstallerUrl '${{ inputs.installer-url }}' - env: - # Outside the workspace on purpose: the driver refuses to run on a - # dirty tree, and logs written into the repo would be what makes it - # dirty. - HERMES_E2E_LOG_DIR: ${{ runner.temp }}\e2e-logs - - # The installer's own transcript, the AHK click log, the update hand-off - # log, and the screen recording say far more than the assertion that - # tripped when a real install breaks. - - name: Upload logs and recording - if: always() - uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1 - with: - name: install-e2e-windows-${{ inputs.route }}-${{ github.sha }} - path: ${{ runner.temp }}\e2e-logs - retention-days: 14 - if-no-files-found: ignore diff --git a/.github/workflows/install-e2e.yml b/.github/workflows/install-e2e.yml index 98b5edc8a051..ecce5c51718e 100644 --- a/.github/workflows/install-e2e.yml +++ b/.github/workflows/install-e2e.yml @@ -12,10 +12,11 @@ name: Install & Update E2E # A hardcoded list would stop covering the newest release the day after it # ships, and would pin an "oldest" that nobody still runs. # -# A separate Windows axis (install-e2e-windows-run.yml) installs with the -# latest PUBLISHED bootstrap installer and applies the desktop app's builtin -# update — no tag matrix there, because the published exe's own build pin is -# the starting point users actually have. +# A separate Windows workflow (desktop-windows-e2e.yml) covers the Windows +# desktop surfaces: a headless desktop-update.ps1 contract job and a real +# user-flow GUI job (website installer, clicked Install/Launch/Update now). +# No tag matrix there — the published exe's own build pin is the starting +# point users actually have. # # Triggers: # * every 12 hours, so upstream drift (a new uv, a Node bump, a PyPI change) @@ -32,11 +33,11 @@ on: workflow_dispatch: inputs: route: - description: 'Which update route to exercise. all/both include the Windows desktop leg.' + description: 'Which update route to exercise.' required: false type: choice default: all - options: [all, both, update, installer, windows-desktop] + options: [all, both, update, installer] tag-count: description: 'How many release tags to sample (newest, oldest, and a spread between).' required: false @@ -114,13 +115,9 @@ jobs: route: installer install-ref: ${{ matrix.install-ref }} - # Windows desktop: install with the latest PUBLISHED bootstrap installer - # (Hermes-Setup.exe — the exact bits a user downloads today), then apply the - # desktop app's builtin update. No release-tag matrix: the published exe's - # build pin decides the starting base, which is precisely the "user on the - # current installer" scenario this axis exists to cover. - windows-desktop: - if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "windows-desktop"]'), inputs.route) - uses: ./.github/workflows/install-e2e-windows-run.yml - with: - route: desktop + # Windows desktop coverage moved to desktop-windows-e2e.yml: two jobs — + # a headless desktop-update.ps1 contract job (BASE -> CURRENT -> NEXT) + # and a REAL user-flow GUI job (website Hermes-Setup.exe, AutoHotkey + # Install/Launch clicks, Playwright-driven Settings -> About -> Update + # now, relaunched-window assert). Triggered on push-to-main + nightly + + # release tags + dispatch there, so no leg here. diff --git a/tests/install/windows/install-button.png b/tests/install/windows/install-button.png deleted file mode 100644 index 19c478dc81023bf3d01c1e68acd33ebdd50b7aab..0000000000000000000000000000000000000000 GIT binary patch literal 0 HcmV?d00001 literal 680 zcmV;Z0$2TsP)0#>VXpn36XfKIXH|{uT@)o-`ftz-b5I=l{ zs~1E%x>o?;+X{7e2~{Kg`1*h%aa_5}?q8=FtFS>N4wD`h69QrUu2mET&#V)N@x6#g zubDueFJUJm6iR1X+Q6>ro>$o201`u3_z@_dk+?b+wpK)YfxCi$o zUqWylb}KFpQZj=W_h(Js2W7LNXNo>15sV7uGgk0^nj57=8hvh&(`PSY9FiFrdvUB6 z1O%hauh1q?HdrJ5xI9G9LgFs1y7A^V^&G+mC)H;zi|sVV7g?X1CDMp-e>H0>UT4{@ z*iIu?6xVWD#ug(b(hw91xgy3FMM{}6Pu^ElHg7Y@fvR`ItrS9E*z_>z zI%SXYUy{n<l4n)L8`{kKZ!2g0HXO>ml{8gLKr3v<%c>$>1ld= 1 ? A_Args[1] : "ahk.log" -markerPath := A_Args.Length >= 2 ? A_Args[2] : "" - -Log(text) { - msg := Format("[autohotkey] {}`n", text) - ToolTip(text) - ; stdout only exists when the launcher attached a console (Start-Process - ; -NoNewWindow). AutoHotkey64 is a GUI-subsystem exe, so a bare spawn has - ; an invalid stdout handle and FileAppend('*') throws "(6) The handle is - ; invalid" -- recursively, from inside OnError's own Log call, which - ; wedges the script instead of exiting. The log FILE is the record; - ; stdout is best-effort. - try FileAppend(msg, '*') - FileAppend(msg, logPath) -} - -OnError(LogError) - -LogError(err, mode) { - Log(Format("Unhandled error: {}", err.Message)) - ExitApp(1) - return -1 ; suppress the standard error dialog -} - -SetWorkingDir(A_ScriptDir) -CoordMode("Pixel", "Screen") -CoordMode("Mouse", "Screen") - -ClickWithMarker(x, y, button := "Left") { - Click(x, y, button) - - Sleep(10) - MouseMove(30, 30) - Log(Format("Clicking at {1}, {2}", x, y)) - ; Draw a short-lived red dot where we clicked so the screen recording - ; shows WHERE the automation acted, not just what happened after. - size := 20 - g := Gui("-Caption +AlwaysOnTop +ToolWindow") - g.BackColor := "Red" - g.Show(Format( - "x{} y{} w{} h{} NoActivate" - , x - size // 2 - , y - size // 2 - , size - , size - )) - hRegion := DllCall( - "CreateEllipticRgn" - , "Int", 0 - , "Int", 0 - , "Int", size - , "Int", size - , "Ptr" - ) - DllCall("SetWindowRgn", "Ptr", g.Hwnd, "Ptr", hRegion, "Int", true) - WinSetTransparent(255, g.Hwnd) - SetTimer(() => g.Destroy(), -500) -} - -FindImageInWindow(winTitle, imageFile, &outX, &outY, timeoutMs := 10000, intervalMs := 250) -{ - WinGetPos(&wx, &wy, &ww, &wh, winTitle) - - hBitmap := LoadPicture(imageFile) - - if !hBitmap { - throw Error("LoadPicture failed: " imageFile) - } - bm := Buffer(32, 0) ; BITMAP structure on x64 - DllCall("GetObject", "Ptr", hBitmap, "Int", bm.Size, "Ptr", bm) - - width := NumGet(bm, 4, "Int") - height := NumGet(bm, 8, "Int") - - startTime := A_TickCount - - timeLeft := 1 - - Log(Format("Searching for button file {} in window {}...", imageFile, winTitle)) - ; *20: the reference is a lossless screen crop, so only minor rendering - ; drift (ClearType phase, sub-shade rounding) needs absorbing. - searchImage := Format("*20 {}", imageFile) - while (timeLeft > 0) - { - if ImageSearch(&x, &y, wx, wy, wx + ww, wy + wh, searchImage) - { - outX := x + Floor(width / 2) - outY := y + Floor(height / 2) - Log("Found button!") - return - } - - Sleep intervalMs - timeLeft := timeoutMs - (A_TickCount - startTime) - ToolTip(Format("Searching for button {} in window {}... {}s left", imageFile, winTitle, Round(timeLeft / 1000, 2))) - } - - throw Error(Format("Failed to find button {} in window {}", imageFile, winTitle)) -} - -ClickCenterOfImageInWindow(winTitle, imageFile, timeoutMs := 10000, intervalMs := 250) -{ - FindImageInWindow(winTitle, imageFile, &x, &y, timeoutMs, intervalMs) - ClickWithMarker(x, y) -} - -Log("Waiting for the installer window to appear...") -winTitle := "Hermes" -try { - WinWait(winTitle, , 30) -} catch { - throw Error("Hermes installer window did not appear within 30s") -} -WinGetPos(&x, &y, &w, &h, winTitle) -Log(Format("Window found at x={1} y={2} w={3} h={4}", x, y, w, h)) - -if (markerPath = "") { - throw Error("marker path argument is required") -} - -; Find the Install button via ImageSearch against the lossless screen crop. -; No PixelSearch fallback: color-hunting matched the blue HERMES AGENT title -; (run 31446691812) and the progress view's blue stage text (run 31447405319) -; before it ever matched the button. -FindInstallClickPoint(winTitle, &outX, &outY) { - try { - FindImageInWindow(winTitle, A_ScriptDir "\install-button.png", &outX, &outY, 3000, 250) - Log("Install button located via ImageSearch") - return true - } catch { - return false - } -} - -; Did the click land? The button vanishes when the UI flips to the progress -; view, so finding it again means the click did not take. -InstallButtonStillVisible(winTitle) { - x := 0 - y := 0 - return FindInstallClickPoint(winTitle, &x, &y) -} - -; Best-effort clicking: NEVER throw here. The authoritative signals are -; owned elsewhere -- the driver aborts on "bootstrap FAILED" in the -; installer log, and the marker wait below caps the run. Run 31447405319 -; killed a healthy mid-install run because this loop threw on its own -; flawed UI heuristic; that class of failure must stay impossible. -attempts := 0 -everClicked := false -while (attempts < 10) { - attempts += 1 - ; ImageSearch/PixelSearch read SCREEN pixels: anything covering the - ; installer (the runner session keeps a maximized console in front) - ; hides the button even though WinWait's title match succeeded. Force - ; the installer to the foreground before every attempt. - WinActivate(winTitle) - WinMoveTop(winTitle) - Sleep(500) - x := 0 - y := 0 - if (!FindInstallClickPoint(winTitle, &x, &y)) { - if (everClicked) { - ; Click already landed and this is the progress view. - Log(Format("Attempt {}: button gone after click; proceeding to marker wait", attempts)) - break - } - ; Welcome screen may still be rendering. - Log(Format("Attempt {}: no Install button in scan band yet; waiting", attempts)) - Sleep(2000) - continue - } - ClickWithMarker(x, y) - everClicked := true - Sleep(3000) - if (!InstallButtonStillVisible(winTitle)) { - Log("Install click landed (button no longer on screen)") - break - } - Log(Format("Install click attempt {}: button still visible; retrying", attempts)) -} - -; Wait for the installer's own completion signal: the bootstrap-complete -; marker is written after the last stage succeeds. A real install takes many -; minutes (git, uv, Python, Node, venv, desktop build). -Log(Format("Waiting for bootstrap-complete marker: {}", markerPath)) -deadline := A_TickCount + 1000 * 60 * 25 -while (A_TickCount < deadline) { - if FileExist(markerPath) { - Log("Marker found -- install complete") - ; Close instead of clicking Launch: the E2E driver owns everything - ; after the install, and a launched desktop would hold the venv shim - ; open and block the update route. - try WinClose(winTitle) - Sleep(2000) - ExitApp(0) - } - Sleep(5000) -} -throw Error("bootstrap-complete marker never appeared within 25 minutes") diff --git a/tests/install/windows/install-update-e2e.ps1 b/tests/install/windows/install-update-e2e.ps1 deleted file mode 100644 index 6a7b56cf4b38..000000000000 --- a/tests/install/windows/install-update-e2e.ps1 +++ /dev/null @@ -1,512 +0,0 @@ -# Prove a Windows desktop user on the published bootstrap installer can reach -# this commit through the desktop's builtin update. -# -# The Windows analog of tests/install/install-update-e2e.sh. Linux gets its -# fake GitHub from the bubblewrap sandbox's MITM proxy + git-upload-pack shim; -# there is no such sandbox on Windows, so the git proxying here is git's own -# transport rewrite instead: a throwaway GIT_CONFIG_GLOBAL carrying multi- -# valued url..insteadOf entries for BOTH hardcoded repo URL -# forms (SSH and HTTPS). Every git child of this process -- install.ps1's -# clone, `hermes update`'s fetch, the desktop's ls-remote -- resolves -# github.com/NousResearch/hermes-agent to a local bare repo whose `main` this -# driver controls, while the tools themselves run VERBATIM with their real -# URLs. scripts/fake_remote_update_probe.sh (hermes-install-update-testing -# skill) is the verified reference for the mechanism, including why --add is -# load-bearing (a second plain `git config` REPLACES the first rewrite and one -# URL silently reaches real GitHub). -# -# What one run does: -# 1. Seed fake.git from this checkout (all origin branches + tags), then -# force fake `main` to the NEWEST release tag -- so an installer with a -# branch pin lands on a released base, not on the target. -# uploadpack.allowAnySHA1InWant covers installers with a -Commit pin. -# 2. Run the real published Hermes-Setup.exe, driven by AutoHotkey (the -# installer is a GUI with no headless mode). The exe downloads its -# pinned install.ps1 from raw.githubusercontent for real -- the same -# fixture-miss passthrough posture as the Linux sandbox -- and that -# script's git clone rides the rewrite onto fake.git. -# 3. Promote fake `main` to this checkout's HEAD (the --from-main dance). -# 4. Apply ONE update route and require HEAD == target with a working -# `hermes`. -# -# Routes: -# desktop the desktop app's builtin update, minus only the Electron -# process around it, following applyUpdates' own preference -# order (apps/desktop/electron/main.ts): the repo-owned -# scripts/desktop-update.ps1 hand-off when the installed base -# ships it, else the staged hermes-setup.exe --update (which -# auto-runs: update mode is a hand-off, not a click-through). -# Either way the update engine is the installed release's own -# `hermes update`, so old CLIs meet their contemporaneous flags. -# TODO update bare `hermes update` from the installed venv (the route -# the Linux matrix calls `update`). -# TODO installer re-run the bootstrap installer over the existing -# checkout (the Linux `installer` route; needs the AHK -# flow to handle the repair/reinstall UI). -# -# Requires: git, AutoHotkey64.exe on PATH, network (real toolchain download), -# a clean full-history checkout with release tags fetched. ffmpeg on PATH is -# optional -- when present the run is screen-recorded for the artifact. - -#Requires -Version 7 - -param( - [ValidateSet("desktop")] - [string]$Route = "desktop", - # Latest published installer -- "what a user downloads today". - [string]$InstallerUrl = "https://hermes-assets.nousresearch.com/Hermes-Setup.exe", - # Local exe override (skips the download; for iterating on this driver). - [string]$InstallerPath = "" -) - -$ErrorActionPreference = "Stop" - -$RepoRoot = (Resolve-Path (Join-Path $PSScriptRoot "..\..\..")).Path -$RepoUrlSsh = "git@github.com:NousResearch/hermes-agent.git" -$RepoUrlHttps = "https://github.com/NousResearch/hermes-agent.git" - -# Everything lives OUTSIDE the checkout: an untracked dir inside the repo -# would make later verification steps lie about a dirty tree, and RUNNER_TEMP -# is wiped with the runner. -$WorkRoot = Join-Path ($env:RUNNER_TEMP ?? [System.IO.Path]::GetTempPath()) "hermes-install-e2e" -$LogDir = if ($env:HERMES_E2E_LOG_DIR) { $env:HERMES_E2E_LOG_DIR } else { Join-Path $WorkRoot "logs" } -$FakeRepo = Join-Path $WorkRoot "fake.git" - -function Step([string]$Message) { Write-Host "`n=== $Message ===" } -function Ok([string]$Message) { Write-Host " OK $Message" } -function Fail([string]$Message) { - Write-Host "FAIL: $Message" -ForegroundColor Red - exit 1 -} - -function Invoke-Git { - param([string[]]$GitArgs, [string]$Cwd = $RepoRoot) - $out = & git -C $Cwd @GitArgs 2>&1 - if ($LASTEXITCODE -ne 0) { - Fail "git $($GitArgs -join ' ') failed (exit $LASTEXITCODE): $out" - } - return ($out | Out-String).Trim() -} - -# --- preflight --------------------------------------------------------------- - -if (-not (Get-Command git -ErrorAction SilentlyContinue)) { Fail "git not on PATH" } -if (-not (Get-Command AutoHotkey64.exe -ErrorAction SilentlyContinue)) { - Fail "AutoHotkey64.exe not on PATH (winget install AutoHotkey.AutoHotkey)" -} - -# The promote step pushes this worktree's HEAD as the update target, so a -# dirty tree means the tested commit is not the commit anyone can review. -$dirty = & git -C $RepoRoot status --porcelain -if ($dirty) { - Write-Host ($dirty | Out-String) - Fail "working tree is dirty; the update target must be a real commit" -} - -Remove-Item -Recurse -Force $WorkRoot -ErrorAction SilentlyContinue -New-Item -ItemType Directory -Force -Path $WorkRoot, $LogDir | Out-Null - -# Isolated HERMES_HOME so the real install never touches the runner's (or a -# developer's) profile. Both install.ps1 and the Tauri installer honor it. -if (-not $env:HERMES_HOME) { - $env:HERMES_HOME = Join-Path $WorkRoot "hermes-home" -} -$InstallRoot = Join-Path $env:HERMES_HOME "hermes-agent" -$TargetSha = Invoke-Git @("rev-parse", "HEAD") - -# Pre-seed the managed uv. install.ps1's Install-Uv runs astral's installer -# with UV_INSTALL_DIR pointing into HERMES_HOME\bin and discards its output -- -# but when an astral install RECEIPT exists (GitHub runners ship uv -# preinstalled with one), the cargo-dist installer updates the receipt's -# location in place and ignores UV_INSTALL_DIR, so the managed path stays -# empty and the stage fails blind ("uv installed but not found", run -# 31447045981). Install-Uv short-circuits on an existing managed uv, so -# seeding it is a legitimate user state, not a bypass. -$managedBin = Join-Path $env:HERMES_HOME "bin" -New-Item -ItemType Directory -Force -Path $managedBin | Out-Null -$uvOnRunner = Get-Command uv.exe -ErrorAction SilentlyContinue -if ($uvOnRunner) { - Copy-Item $uvOnRunner.Source (Join-Path $managedBin "uv.exe") -Force - Ok "seeded managed uv from runner: $($uvOnRunner.Source)" -} else { - # No preinstalled uv means no receipt, so the plain astral path works -- - # with output visible, unlike Install-Uv's. - $env:UV_INSTALL_DIR = $managedBin - Invoke-RestMethod https://astral.sh/uv/install.ps1 | Invoke-Expression - Remove-Item Env:\UV_INSTALL_DIR - if (-not (Test-Path (Join-Path $managedBin "uv.exe"))) { - Fail "could not seed managed uv into $managedBin" - } - Ok "seeded managed uv via astral installer" -} - -# --- fake GitHub ------------------------------------------------------------- - -Step "seeding fake remote at $FakeRepo" -Invoke-Git @("init", "--bare", "--initial-branch=main", $FakeRepo) $WorkRoot | Out-Null -# Published installers carry a -Commit pin and fetch that raw SHA; a bare -# repo refuses SHA wants unless told otherwise. -Invoke-Git @("config", "uploadpack.allowAnySHA1InWant", "true") $FakeRepo | Out-Null - -# All origin branches + tags: the installer's build pin may be any commit on -# any branch that existed when the exe was built. -Invoke-Git @("push", "--quiet", $FakeRepo, "refs/remotes/origin/*:refs/heads/*") -Invoke-Git @("push", "--quiet", "--force", $FakeRepo, "refs/tags/*:refs/tags/*") - -# Fake main starts at the newest release tag: a released base a real user -# could be installed on, and never the update target itself. Major capped at -# three digits, matching _parse_release_tag (hermes_cli/update_cmd.py) and -# latestReleaseFromLsRemote (apps/desktop/electron/bundled-runtime.ts): the -# repo's historical CalVer tags (v2026.7.20) would otherwise win every -# numeric sort forever. -$releaseTags = @(& git -C $RepoRoot tag --list | - Where-Object { $_ -match '^v\d{1,3}\.\d+\.\d+(\.\d+)?$' } | - Sort-Object { [version]($_.Substring(1)) }) -if ($releaseTags.Count -eq 0) { - Fail "no release tags in this checkout -- fetch with tags (fetch-depth: 0 + fetch-tags)" -} -$newestTag = $releaseTags[-1] -$baseMainSha = Invoke-Git @("rev-parse", "$newestTag^{commit}") -Invoke-Git @("push", "--quiet", "--force", $FakeRepo, "${baseMainSha}:refs/heads/main") -Ok "fake main = $newestTag ($($baseMainSha.Substring(0,12))); target is $($TargetSha.Substring(0,12))" - -# --- git transport rewrite --------------------------------------------------- - -Step "redirecting github.com/NousResearch/hermes-agent to the fake remote" -# Process-scoped global config: every git spawned below this point (installer, -# hermes update, desktop hand-off) inherits it; nothing on the machine does. -$env:GIT_CONFIG_GLOBAL = Join-Path $WorkRoot "gitconfig" -Set-Content -Path $env:GIT_CONFIG_GLOBAL -Value "" -NoNewline -# Fail loudly if anything still reaches a URL that wants credentials. -$env:GIT_TERMINAL_PROMPT = "0" - -$fakeUrl = "file:///" + $FakeRepo.Replace("\", "/") -foreach ($url in @($RepoUrlSsh, $RepoUrlHttps)) { - Invoke-Git @("config", "--global", "--add", "url.$fakeUrl.insteadOf", $url) $WorkRoot -} -$rewrites = @(& git config --global --get-all "url.$fakeUrl.insteadOf") -if ($rewrites.Count -ne 2) { - Fail "expected 2 insteadOf rewrites, got $($rewrites.Count) -- one URL would reach real GitHub" -} -Ok "both repo URL forms rewritten (SSH clone attempts ride the file transport)" - -# --- fetch the installer ----------------------------------------------------- - -if (-not $InstallerPath) { - Step "downloading published installer" - $InstallerPath = Join-Path $WorkRoot "Hermes-Setup.exe" - Invoke-WebRequest -Uri $InstallerUrl -OutFile $InstallerPath -} -Ok "installer: $InstallerPath ($([math]::Round((Get-Item $InstallerPath).Length / 1MB, 1)) MB)" - -# --- screen recording (optional) ---------------------------------------------- - -# ffmpeg must be started, fed, and stopped from THIS process: the graceful -# stop is the character 'q' on its LIVE stdin pipe, which only -# System.Diagnostics.Process exposes (Start-Process -RedirectStandardInput -# hands it a file handle already at EOF). -$ffmpeg = $null -if (Get-Command ffmpeg -ErrorAction SilentlyContinue) { - $psi = New-Object System.Diagnostics.ProcessStartInfo - $psi.FileName = "ffmpeg" - $psi.Arguments = "-y -f gdigrab -framerate 15 -i desktop " + - "-hide_banner -loglevel error " + - "-c:v libx264 -preset ultrafast -pix_fmt yuv420p `"$LogDir\recording.mkv`"" - $psi.RedirectStandardInput = $true - $psi.UseShellExecute = $false - $ffmpeg = [System.Diagnostics.Process]::Start($psi) - Ok "screen recording started (pid $($ffmpeg.Id))" -} else { - Write-Host " (ffmpeg not on PATH; skipping screen recording)" -} - -function Stop-Recording { - if ($script:ffmpeg -and -not $script:ffmpeg.HasExited) { - try { - $script:ffmpeg.StandardInput.Write("q") - $script:ffmpeg.StandardInput.Close() - } catch {} - if (-not $script:ffmpeg.WaitForExit(15000)) { $script:ffmpeg.Kill() } - } -} - -# --- run the real installer under AutoHotkey ---------------------------------- - -Step "installing via Hermes-Setup.exe (real toolchains: git, uv, Python, Node, venv, desktop)" -$installerOk = $false -try { - $proc = Start-Process -FilePath $InstallerPath -PassThru - - # Lossless PNG of the welcome screen, taken BEFORE the AHK helper starts - # so no tooltip or click marker contaminates it. This is the artifact the - # ImageSearch reference crop is made from: the ffmpeg recording is - # H.264/yuv420p, whose chroma subsampling shifts glyph pixels enough that - # a crop from video never matches the live screen. - # - # Poll instead of a fixed sleep: WebView2's cold start left the window - # pure white past 17s on a runner (run 31448964917's capture was blank). - # "Rendered" = colored (non-grayscale) pixels in the window content area, - # which the blue HERMES AGENT title guarantees. - Add-Type -AssemblyName System.Windows.Forms, System.Drawing - $bounds = [System.Windows.Forms.Screen]::PrimaryScreen.Bounds - $shotPath = Join-Path $LogDir "welcome-screen.png" - $renderDeadline = (Get-Date).AddSeconds(120) - $rendered = $false - while ((Get-Date) -lt $renderDeadline) { - Start-Sleep -Seconds 5 - $bmp = New-Object System.Drawing.Bitmap $bounds.Width, $bounds.Height - $gfx = [System.Drawing.Graphics]::FromImage($bmp) - $gfx.CopyFromScreen($bounds.Location, [System.Drawing.Point]::Empty, $bounds.Size) - $gfx.Dispose() - $colored = 0 - for ($y = 100; $y -lt 600; $y += 7) { - for ($x = 100; $x -lt 900; $x += 7) { - $p = $bmp.GetPixel($x, $y) - $mx = [Math]::Max($p.R, [Math]::Max($p.G, $p.B)) - $mn = [Math]::Min($p.R, [Math]::Min($p.G, $p.B)) - if (($mx - $mn) -gt 60) { $colored++ } - } - } - $bmp.Save($shotPath, [System.Drawing.Imaging.ImageFormat]::Png) - $bmp.Dispose() - Write-Host " screen poll: $colored colored samples" - if ($colored -gt 20) { $rendered = $true; break } - } - if (-not $rendered) { - Stop-Recording - Fail "installer UI never rendered within 120s (last capture saved to welcome-screen.png)" - } - Ok "welcome screen captured (lossless) to welcome-screen.png" - - $ahkLog = Join-Path $LogDir "ahk.log" - # The helper polls for the installer's own completion signal instead of a - # second button screenshot (see paths.rs likely_bootstrap_marker). - $bootstrapMarker = Join-Path $InstallRoot ".hermes-bootstrap-complete" - # -NoNewWindow attaches our console as the GUI-subsystem exe's stdout so - # the helper's live lines land in the job log as they happen. - $ahkProc = Start-Process -FilePath "AutoHotkey64.exe" -NoNewWindow ` - -ArgumentList "`"$PSScriptRoot\install-hermes-desktop.ahk`"", "`"$ahkLog`"", "`"$bootstrapMarker`"" -PassThru - - # Tail the bootstrap log into the job log while we wait: the install IS - # the substance of this test, and a failure explanation should not need - # an artifact download. FileShare.ReadWrite because the installer still - # has the file open for writing. - $logReader = $null - $logStream = $null - $bootstrapLog = Join-Path $env:HERMES_HOME "logs\bootstrap-installer.log" - $deadline = (Get-Date).AddMinutes(30) - try { - while ((Get-Date) -lt $deadline -and -not $ahkProc.HasExited) { - if (-not $logReader) { - if (Test-Path $bootstrapLog) { - $logStream = [System.IO.File]::Open($bootstrapLog, 'Open', 'Read', 'ReadWrite') - $logReader = New-Object System.IO.StreamReader($logStream) - } - } else { - $line = $logReader.ReadLine() - while ($null -ne $line) { - Write-Host "[bootstrap] $line" - # The installer's failure screen waits for a human (Retry - # button); the AHK helper would idle out its full marker - # deadline. Abort as soon as the log says the run is dead. - if ($line -match "bootstrap FAILED") { - Stop-Process -Id $ahkProc.Id -Force -ErrorAction SilentlyContinue - Stop-Process -Id $proc.Id -Force -ErrorAction SilentlyContinue - Fail "installer reported: $line" - } - $line = $logReader.ReadLine() - } - } - Start-Sleep -Milliseconds 500 - } - # Drain what was written in the final tick. - if ($logReader) { - $line = $logReader.ReadLine() - while ($null -ne $line) { - Write-Host "[bootstrap] $line" - $line = $logReader.ReadLine() - } - } - } finally { - if ($logReader) { $logReader.Dispose() } - if ($logStream) { $logStream.Dispose() } - } - - if (-not $ahkProc.HasExited) { - Stop-Process -Id $ahkProc.Id -Force -ErrorAction SilentlyContinue - Fail "AutoHotkey helper still running at the deadline -- install never finished. See ahk.log + recording." - } - if ($ahkProc.ExitCode -ne 0) { - Fail "AutoHotkey helper failed (exit $($ahkProc.ExitCode)) -- see ahk.log + recording" - } - - # The AHK helper closes the window after the Launch button appears; a - # still-running installer means the close did not land. - if (-not $proc.WaitForExit(30000)) { - Stop-Process -Id $proc.Id -Force -ErrorAction SilentlyContinue - Fail "installer process still running after the window was closed" - } - $installerOk = $true -} finally { - Stop-Recording - if (Test-Path (Join-Path $LogDir "ahk.log")) { - Write-Host "--- ahk.log ---" - Get-Content (Join-Path $LogDir "ahk.log") | ForEach-Object { Write-Host $_ } - Write-Host "--- end ahk.log ---" - } - if (-not $installerOk -and (Test-Path (Join-Path $env:HERMES_HOME "logs\bootstrap-installer.log"))) { - Copy-Item (Join-Path $env:HERMES_HOME "logs\bootstrap-installer.log") $LogDir -Force - } -} - -# --- verify the install ------------------------------------------------------ - -Step "verifying the installed checkout" -if (-not (Test-Path (Join-Path $InstallRoot ".git"))) { - Fail "no git checkout at $InstallRoot -- the installer's clone did not ride the rewrite?" -} -$BaseSha = Invoke-Git @("rev-parse", "HEAD") $InstallRoot -if ($BaseSha -eq $TargetSha) { - Fail "install landed on the update target ($BaseSha); base and target must differ" -} -Ok "installed $($BaseSha.Substring(0,12)); update target is $($TargetSha.Substring(0,12))" - -$HermesExe = Join-Path $InstallRoot "venv\Scripts\hermes.exe" -if (-not (Test-Path $HermesExe)) { Fail "venv shim missing: $HermesExe" } - -# The real smoke test: goes through the venv launcher and imports the app. -$version = & $HermesExe --version 2>&1 -if ($LASTEXITCODE -ne 0) { Fail "hermes --version failed after install: $version" } -Write-Host " $version" -Ok "hermes runs after install" - -# --- promote fake main to this checkout -------------------------------------- - -Step "promoting fake main to this checkout (the state a user sees when an update is waiting)" -Invoke-Git @("push", "--quiet", "--force", $FakeRepo, "HEAD:refs/heads/main") -Ok "fake main advanced to $($TargetSha.Substring(0,12))" - -# --- apply exactly one update route ------------------------------------------- - -switch ($Route) { - "desktop" { - Step "ROUTE: desktop builtin update" - # Mirror the PATH contract the desktop passes the hand-off - # (pathWithHermesManagedNode in apps/desktop/electron/main.ts): - # managed node first, then the venv scripts dir. - $managedNode = Join-Path $env:HERMES_HOME "node" - $env:PATH = ((@( - $managedNode, - (Join-Path $managedNode "bin"), - (Join-Path $InstallRoot "venv\Scripts") - ) | Where-Object { Test-Path $_ }) + @($env:PATH)) -join ";" - - # applyUpdates' preference order: the repo-owned hand-off script when - # the INSTALLED checkout ships it, else the staged Tauri binary. Run - # whichever the Update button would actually spawn against this base. - $handoff = Join-Path $InstallRoot "scripts\desktop-update.ps1" - $stagedExe = Join-Path $env:HERMES_HOME "hermes-setup.exe" - - if (Test-Path $handoff) { - # powershell.exe (5.1), not pwsh: that is what the desktop spawns. - # -NoUi is the script's own headless switch; -DesktopPid 0 skips - # the wait-for-desktop gate (no desktop is running); no - # -RelaunchExe so nothing is launched afterwards. - $handoffLog = Join-Path $LogDir "desktop-update.log" - $handoffErrLog = Join-Path $LogDir "desktop-update.err.log" - # Run 31449642122 hung 65 minutes inside "Updating Python - # dependencies" with zero output from uv, and the job-level - # timeout killed the run before anything could say why. So the - # hand-off runs under a driver-owned deadline (same 45 minutes - # the staged-exe branch gets): THIS script outlives a hang and - # dumps the live process table before GitHub cancels the job. - # Do NOT set RUST_LOG here: the installed base's Python pipes - # uv's stderr without draining it (the very bug this axis - # exposed), so debug tracing FLOODS that pipe and manufactures - # the deadlock it was meant to diagnose (run 31457301901). - $hp = Start-Process -FilePath "powershell" -ArgumentList @( - "-NoProfile", "-ExecutionPolicy", "Bypass", "-File", $handoff, - "-InstallRoot", $InstallRoot, "-Branch", "main", - "-DesktopPid", "0", "-NoUi" - ) -RedirectStandardOutput $handoffLog -RedirectStandardError $handoffErrLog ` - -PassThru -NoNewWindow - $handoffDeadline = (Get-Date).AddMinutes(45) - $tailPos = 0 - # Poll-tail the log into the job console so progress stays live. - while (-not $hp.HasExited -and (Get-Date) -lt $handoffDeadline) { - Start-Sleep -Seconds 15 - if (Test-Path $handoffLog) { - $content = Get-Content $handoffLog -Raw -ErrorAction SilentlyContinue - if ($content -and $content.Length -gt $tailPos) { - Write-Host ($content.Substring($tailPos)) -NoNewline - $tailPos = $content.Length - } - } - } - if (-not $hp.HasExited) { - Write-Host "--- handoff still running at the 45-minute deadline ---" - Write-Host "--- live processes (who is actually stuck): ---" - Get-CimInstance Win32_Process | - Where-Object { $_.Name -match "uv|python|git|node|hermes|powershell|pip" } | - Select-Object ProcessId, ParentProcessId, CreationDate, Name, CommandLine | - Format-Table -AutoSize -Wrap | Out-String -Width 4096 | Write-Host - if (Test-Path $handoffErrLog) { - Write-Host "--- desktop-update.err.log (last 100 lines) ---" - Get-Content $handoffErrLog -Tail 100 | ForEach-Object { Write-Host $_ } - } - taskkill /PID $hp.Id /T /F 2>$null | Out-Null - Fail "desktop-update.ps1 hung past 45 minutes -- process table above, full logs in artifacts" - } - # Flush whatever the poll loop had not printed yet. - if (Test-Path $handoffLog) { - $content = Get-Content $handoffLog -Raw -ErrorAction SilentlyContinue - if ($content -and $content.Length -gt $tailPos) { - Write-Host ($content.Substring($tailPos)) -NoNewline - } - } - $handoffExit = $hp.ExitCode - - $handoffInternalLog = Join-Path $env:HERMES_HOME "logs\desktop-update-handoff.log" - if (Test-Path $handoffInternalLog) { Copy-Item $handoffInternalLog $LogDir -Force } - if ($handoffExit -ne 0) { - Fail "desktop-update.ps1 failed (exit $handoffExit) -- see desktop-update.log + desktop-update-handoff.log" - } - } elseif (Test-Path $stagedExe) { - # Update mode is a hand-off, not a click-through: --update jumps - # straight to progress and start_update runs unattended, exiting - # when done -- no AHK needed. On success it auto-launches the - # desktop; kill that below rather than letting it hold the venv. - Write-Host " installed base predates desktop-update.ps1; using staged hermes-setup.exe --update" - $upd = Start-Process -FilePath $stagedExe -ArgumentList "--update" -PassThru - if (-not $upd.WaitForExit(45 * 60 * 1000)) { - Stop-Process -Id $upd.Id -Force -ErrorAction SilentlyContinue - Fail "hermes-setup.exe --update still running after 45 minutes" - } - $updLog = Join-Path $env:HERMES_HOME "logs\update.log" - if (Test-Path $updLog) { Copy-Item $updLog $LogDir -Force } - if ($upd.ExitCode -ne 0) { - Fail "hermes-setup.exe --update failed (exit $($upd.ExitCode)) -- see update.log" - } - # The successful updater relaunches Hermes; a live desktop locks - # the venv shim and would poison later assertions. - Get-Process -Name "Hermes" -ErrorAction SilentlyContinue | Stop-Process -Force -ErrorAction SilentlyContinue - } else { - Fail "neither scripts/desktop-update.ps1 (in the installed base) nor a staged hermes-setup.exe exists -- no desktop update path to exercise" - } - - $After = Invoke-Git @("rev-parse", "HEAD") $InstallRoot - if ($After -ne $TargetSha) { - Fail "desktop update left HEAD at $After, wanted $TargetSha" - } - Ok "desktop update landed on $($After.Substring(0,12))" - - $version = & $HermesExe --version 2>&1 - if ($LASTEXITCODE -ne 0) { Fail "hermes --version failed after update: $version" } - Write-Host " $version" - Ok "hermes runs after desktop update" - } -} - -Write-Host "" -Write-Host "PASS: Windows install/update E2E (route: $Route, base: $($BaseSha.Substring(0,12)) -> $($TargetSha.Substring(0,12)))" -ForegroundColor Green -exit 0 From 3fedcbc6a8d94bc0c5bc7872f17d8f55bf2b1b94 Mon Sep 17 00:00:00 2001 From: ethernet Date: Tue, 11 Aug 2026 13:30:55 -0400 Subject: [PATCH 035/227] ci(temp): branch push trigger for pre-merge validation of the windows e2e Reverted before merge, same as tek's pre-merge validation commit on the upstream PR: workflow_dispatch only works once the file exists on the default branch, and this fork branch is where the integrated workflow needs proving. --- .github/workflows/desktop-windows-e2e.yml | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-) diff --git a/.github/workflows/desktop-windows-e2e.yml b/.github/workflows/desktop-windows-e2e.yml index 4d547a4ec9f0..b4ae72703578 100644 --- a/.github/workflows/desktop-windows-e2e.yml +++ b/.github/workflows/desktop-windows-e2e.yml @@ -29,8 +29,11 @@ on: schedule: # Nightly, off the hour to dodge the top-of-hour runner crunch. - cron: '40 6 * * *' + # TEMP (revert before merge): validate on this branch from the fork, where + # workflow_dispatch is unavailable until the file lands on the default + # branch. Same pattern tek used pre-merge on the upstream PR. push: - branches: [main] + branches: [main, ethie/desktop-update-tests] tags: - 'v[0-9]+.[0-9]+.[0-9]+' - 'v[0-9]+.[0-9]+.[0-9]+.[0-9]+' From 27636cf422d8a6de78f6b7cc14106aafea7645fe Mon Sep 17 00:00:00 2001 From: ethernet Date: Tue, 11 Aug 2026 13:44:12 -0400 Subject: [PATCH 036/227] ci(windows-e2e): slot the GUI flow in as the windows-desktop route - generic OLD -> HEAD Restructure tek's two-job desktop-windows-e2e.yml into the shape the linux axis already has: install-e2e.yml keeps its update/installer routes untouched and windows-desktop returns as a route in the same family, calling a reusable install-e2e-windows-run.yml. Behind that route is now ONLY the real user flow - the headless contract job (install.ps1 at HEAD~1, desktop-update.ps1 -NoUi, BASE/CURRENT/NEXT ref dance) is gone, along with its driver. Every leg goes through a surface a user touches: website Hermes-Setup.exe headed with AutoHotkey clicking Install -> Launch, then the installed Hermes.exe under Playwright's Electron driver clicking Settings -> About -> 'Update now', through the detached hand-off to a relaunched window asserted on HEAD. The driver drops the synthetic-NEXT staging with the contract job: serve.git just serves HEAD as main and OLD is the release pin baked into the website exe - the literal starting point of every real GUI user, same philosophy as the linux axis's release-tag matrix. The -Route parameter (desktop today) declares the future update mechanisms as arms: 'update' (hermes update from the installed venv) and 'installer' (re-run the bootstrap exe) raise until implemented, so the workflow surface is stable when they land. --- .github/workflows/desktop-windows-e2e.yml | 177 --------- .github/workflows/install-e2e-windows-run.yml | 115 ++++++ .github/workflows/install-e2e.yml | 32 +- tests/install/windows-desktop-e2e.ps1 | 339 ------------------ tests/install/windows-desktop-gui-e2e.ps1 | 236 +++++++----- 5 files changed, 289 insertions(+), 610 deletions(-) delete mode 100644 .github/workflows/desktop-windows-e2e.yml create mode 100644 .github/workflows/install-e2e-windows-run.yml delete mode 100644 tests/install/windows-desktop-e2e.ps1 diff --git a/.github/workflows/desktop-windows-e2e.yml b/.github/workflows/desktop-windows-e2e.yml deleted file mode 100644 index b4ae72703578..000000000000 --- a/.github/workflows/desktop-windows-e2e.yml +++ /dev/null @@ -1,177 +0,0 @@ -name: Desktop Windows Install/Update E2E - -# Can a real Windows user (a) install the commit BEFORE this one from -# scratch, (b) update that install TO this commit through the Desktop GUI -# update path, and (c) update FROM this commit to a future main? -# -# The driver (tests/install/windows-desktop-e2e.ps1) stages a local bare -# repo serving BASE (HEAD~1) -> CURRENT (HEAD) -> NEXT (synthetic same-tree -# child of HEAD) and redirects the canonical GitHub URLs at it via git -# insteadOf env config. The install runs BASE's own scripts/install.ps1 -# with -IncludeDesktop (real uv/Python/Node/venv/Electron build); each -# update leg runs the installed checkout's scripts/desktop-update.ps1 -- -# the exact hand-off the Desktop Update button spawns -- headless (-NoUi), -# including the desktop-exit and venv-lock fail-closed gates, the real -# `hermes update`, marker lifecycle, and the result JSON the relaunched -# Desktop surfaces. -# -# before -> current proves this commit can be updated TO. -# current -> next proves this commit can update FROM (its updater code is -# the one that runs when the NEXT commit ships). -# -# Triggers: every push to main (concurrency-coalesced), release tags, -# nightly, and manual dispatch. Deliberately NOT on pull_request: a leg -# does ~30-60 minutes of real toolchain + Electron work on a 2x-cost -# Windows runner; breakage on main pages within one commit either way. - -on: - workflow_dispatch: - schedule: - # Nightly, off the hour to dodge the top-of-hour runner crunch. - - cron: '40 6 * * *' - # TEMP (revert before merge): validate on this branch from the fork, where - # workflow_dispatch is unavailable until the file lands on the default - # branch. Same pattern tek used pre-merge on the upstream PR. - push: - branches: [main, ethie/desktop-update-tests] - tags: - - 'v[0-9]+.[0-9]+.[0-9]+' - - 'v[0-9]+.[0-9]+.[0-9]+.[0-9]+' - -permissions: - contents: read - -concurrency: - group: desktop-windows-e2e-${{ github.ref }} - cancel-in-progress: true - -jobs: - e2e: - name: install BASE, update to CURRENT, update to NEXT - runs-on: windows-latest - timeout-minutes: 180 - - env: - # Sibling of the checkout (D:\a\hermes-agent\hermes-desktop-e2e): outside - # the repo so the staged bare clone and the install never collide with - # the checkout itself. NOTE: ${{ runner.temp }} is NOT available in - # job-level env (only github/inputs/matrix/needs/secrets/strategy/vars); - # that exact mistake made the whole workflow file invalid on first push. - HERMES_E2E_WORKROOT: ${{ github.workspace }}\..\hermes-desktop-e2e - - steps: - # Full history: the driver resolves HEAD~1 and bare-clones this - # checkout as the repo the installer/updater talk to. A shallow - # checkout cannot be served as a clone source. - - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2 - with: - fetch-depth: 0 - - - name: Stage serve repo (BASE / CURRENT / NEXT) - shell: powershell - run: powershell -NoProfile -ExecutionPolicy Bypass -File tests\install\windows-desktop-e2e.ps1 -Phase stage - - - name: Install BASE from scratch (real install.ps1, with Desktop) - shell: powershell - run: powershell -NoProfile -ExecutionPolicy Bypass -File tests\install\windows-desktop-e2e.ps1 -Phase install - - - name: Update BASE -> CURRENT (Desktop GUI update path) - shell: powershell - run: powershell -NoProfile -ExecutionPolicy Bypass -File tests\install\windows-desktop-e2e.ps1 -Phase update-to-current - - - name: Update CURRENT -> NEXT (Desktop GUI update path) - shell: powershell - run: powershell -NoProfile -ExecutionPolicy Bypass -File tests\install\windows-desktop-e2e.ps1 -Phase update-to-next - - - name: Collect logs - if: always() - shell: powershell - run: | - $out = "e2e-logs" - New-Item -ItemType Directory -Path $out -Force | Out-Null - $home_ = Join-Path $env:HERMES_E2E_WORKROOT "hermes-home" - foreach ($p in @("logs", ".hermes-update-result.json")) { - $src = Join-Path $home_ $p - if (Test-Path $src) { Copy-Item $src (Join-Path $out $p) -Recurse -Force } - } - $shas = Join-Path $env:HERMES_E2E_WORKROOT "shas.json" - if (Test-Path $shas) { Copy-Item $shas $out -Force } - - - name: Upload logs - if: always() - uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1 - with: - name: desktop-windows-e2e-logs-${{ github.sha }} - path: e2e-logs - retention-days: 7 - if-no-files-found: ignore - - # ── The REAL user flow ────────────────────────────────────────────────── - # Same staging, but every leg is the surface a user touches: the website's - # Hermes-Setup.exe run HEADED with AutoHotkey clicking Install/Launch, then - # the installed Electron app driven by Playwright clicking Settings → - # About → "Update now" twice (→ CURRENT, → synthetic NEXT), asserting the - # detached updater chain end-to-end (result JSON, marker, sha, relaunch). - # Proof artifacts: per-step renderer screenshots, full-desktop frames every - # 3s, ahk.log, bootstrap-installer.log, desktop-update-handoff.log. - # - # GitHub's windows-latest runners have an interactive desktop session, so - # headed GUI automation works without RDP tricks (same substrate the prior - # AutoHotkey attempt in #68183 targeted). - gui-e2e: - name: "REAL flow: website setup.exe → GUI update → GUI update" - runs-on: windows-latest - timeout-minutes: 240 - # Deliberately NOT `needs: e2e` — the jobs run independently, so each - # provides rollback coverage for the other and a GUI-layer flake cannot - # mask a machinery regression. - - env: - HERMES_E2E_WORKROOT: ${{ github.workspace }}\..\hermes-desktop-gui-e2e - - steps: - - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2 - with: - fetch-depth: 0 - - - name: Stage serve repo (BASE / CURRENT / NEXT) - shell: powershell - run: powershell -NoProfile -ExecutionPolicy Bypass -File tests\install\windows-desktop-gui-e2e.ps1 -Phase stage - - - name: Install via website Hermes-Setup.exe (headed, AHK-clicked) - shell: powershell - run: powershell -NoProfile -ExecutionPolicy Bypass -File tests\install\windows-desktop-gui-e2e.ps1 -Phase install-gui - - - name: GUI update to CURRENT (real Update-now click) - shell: powershell - run: powershell -NoProfile -ExecutionPolicy Bypass -File tests\install\windows-desktop-gui-e2e.ps1 -Phase update-gui-to-current - - - name: GUI update to NEXT (real Update-now click) - shell: powershell - run: powershell -NoProfile -ExecutionPolicy Bypass -File tests\install\windows-desktop-gui-e2e.ps1 -Phase update-gui-to-next - - - name: Collect proof + logs - if: always() - shell: powershell - run: | - $out = "gui-e2e-proof" - New-Item -ItemType Directory -Path $out -Force | Out-Null - $work = $env:HERMES_E2E_WORKROOT - $home_ = Join-Path $work "hermes-home" - foreach ($pair in @( - @{ src = (Join-Path $work "proof"); dst = "proof" }, - @{ src = (Join-Path $work "shas.json"); dst = "shas.json" }, - @{ src = (Join-Path $home_ "logs"); dst = "logs" }, - @{ src = (Join-Path $home_ ".hermes-update-result.json"); dst = ".hermes-update-result.json" } - )) { - if (Test-Path $pair.src) { Copy-Item $pair.src (Join-Path $out $pair.dst) -Recurse -Force } - } - - - name: Upload proof + logs - if: always() - uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1 - with: - name: desktop-windows-gui-e2e-proof-${{ github.sha }} - path: gui-e2e-proof - retention-days: 14 - if-no-files-found: ignore diff --git a/.github/workflows/install-e2e-windows-run.yml b/.github/workflows/install-e2e-windows-run.yml new file mode 100644 index 000000000000..bd492d454b8d --- /dev/null +++ b/.github/workflows/install-e2e-windows-run.yml @@ -0,0 +1,115 @@ +name: Install & Update E2E — Windows desktop (reusable) + +# Runs the REAL Windows desktop user flow against ONE update route: +# install OLD via the website's Hermes-Setup.exe (headed, AutoHotkey clicks +# Install -> Launch, the real Electron window must appear), then update +# OLD -> HEAD through the route a user would take. +# +# The Windows sibling of install-e2e-run.yml. No bubblewrap here — the +# driver (tests/install/windows-desktop-gui-e2e.ps1) fakes GitHub with +# git's own transport rewrite: a driver-owned GIT_CONFIG_GLOBAL carrying +# url..insteadOf for both canonical repo URLs, so the +# installer and updater run byte-for-byte against their real URLs and land +# on a local bare repo serving HEAD as `main`. OLD is not staged: it is +# the release pin baked into the website exe — the literal starting point +# of every real GUI user. +# +# Routes (the update mechanism under test): +# desktop the app's own Update button: the installed Hermes.exe runs +# under Playwright's Electron driver, which clicks Settings -> +# About -> "Update now"; the production hand-off chain runs +# untouched (marker, app quit, detached updater, hermes +# update, desktop rebuild, relaunch). +# update TODO: `hermes update` from the installed venv. +# installer TODO: re-run the bootstrap exe over the existing install. +# +# Call it: +# +# jobs: +# windows-desktop: +# uses: ./.github/workflows/install-e2e-windows-run.yml +# with: +# route: desktop + +on: + workflow_call: + inputs: + route: + description: 'Update mechanism to exercise: desktop (Update button; implemented), update (hermes update; TODO), installer (re-run bootstrap exe; TODO).' + required: false + type: string + default: desktop + setup-exe-url: + description: 'Bootstrap installer to install OLD with. Default: the latest published one — what a user downloads today.' + required: false + type: string + default: https://hermes-assets.nousresearch.com/Hermes-Setup.exe + timeout-minutes: + description: 'Job timeout. The install leg does real toolchain work and the update leg a full Electron rebuild.' + required: false + type: number + default: 240 + +permissions: + contents: read + +jobs: + e2e: + name: "${{ inputs.route }} route from website installer" + runs-on: windows-latest + timeout-minutes: ${{ inputs.timeout-minutes }} + + env: + # Sibling of the checkout (D:\a\hermes-agent\hermes-desktop-gui-e2e): + # outside the repo so the staged bare clone and the install never + # collide with the checkout itself. NOTE: ${{ runner.temp }} is NOT + # available in job-level env (only github/inputs/matrix/needs/ + # secrets/strategy/vars). + HERMES_E2E_WORKROOT: ${{ github.workspace }}\..\hermes-desktop-gui-e2e + + steps: + # Full history: the driver bare-clones this checkout as the repo the + # installer/updater talk to, and the installer's baked release pin + # must be reachable in that clone. A shallow checkout cannot serve + # either need. + - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2 + with: + fetch-depth: 0 + + - name: Stage serve repo (main -> HEAD) + shell: powershell + run: powershell -NoProfile -ExecutionPolicy Bypass -File tests\install\windows-desktop-gui-e2e.ps1 -Phase stage -Route ${{ inputs.route }} -SetupExeUrl ${{ inputs.setup-exe-url }} + + - name: Install OLD via website Hermes-Setup.exe (headed, AHK-clicked) + shell: powershell + run: powershell -NoProfile -ExecutionPolicy Bypass -File tests\install\windows-desktop-gui-e2e.ps1 -Phase install-gui -Route ${{ inputs.route }} -SetupExeUrl ${{ inputs.setup-exe-url }} + + - name: Update OLD -> HEAD (${{ inputs.route }} route) + shell: powershell + run: powershell -NoProfile -ExecutionPolicy Bypass -File tests\install\windows-desktop-gui-e2e.ps1 -Phase update-gui -Route ${{ inputs.route }} -SetupExeUrl ${{ inputs.setup-exe-url }} + + - name: Collect proof + logs + if: always() + shell: powershell + run: | + $out = "gui-e2e-proof" + New-Item -ItemType Directory -Path $out -Force | Out-Null + $work = $env:HERMES_E2E_WORKROOT + $home_ = Join-Path $work "hermes-home" + foreach ($pair in @( + @{ src = (Join-Path $work "proof"); dst = "proof" }, + @{ src = (Join-Path $work "shas.json"); dst = "shas.json" }, + @{ src = (Join-Path $home_ "logs"); dst = "logs" }, + @{ src = (Join-Path $home_ ".hermes-update-result.json"); dst = ".hermes-update-result.json" } + )) { + if (Test-Path $pair.src) { Copy-Item $pair.src (Join-Path $out $pair.dst) -Recurse -Force } + } + + - name: Upload proof + logs + if: always() + uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1 + with: + name: install-e2e-windows-${{ inputs.route }}-${{ github.sha }} + path: gui-e2e-proof + retention-days: 14 + if-no-files-found: ignore diff --git a/.github/workflows/install-e2e.yml b/.github/workflows/install-e2e.yml index ecce5c51718e..08d46582d018 100644 --- a/.github/workflows/install-e2e.yml +++ b/.github/workflows/install-e2e.yml @@ -12,11 +12,12 @@ name: Install & Update E2E # A hardcoded list would stop covering the newest release the day after it # ships, and would pin an "oldest" that nobody still runs. # -# A separate Windows workflow (desktop-windows-e2e.yml) covers the Windows -# desktop surfaces: a headless desktop-update.ps1 contract job and a real -# user-flow GUI job (website installer, clicked Install/Launch/Update now). -# No tag matrix there — the published exe's own build pin is the starting -# point users actually have. +# A separate Windows axis (install-e2e-windows-run.yml) covers the desktop +# user flow: install OLD with the website's published bootstrap installer +# (headed, AutoHotkey-clicked), then update to HEAD through a real update +# surface — the app's own Update button clicked by Playwright. No tag +# matrix there: the published exe's own build pin is the starting point +# users actually have. # # Triggers: # * every 12 hours, so upstream drift (a new uv, a Node bump, a PyPI change) @@ -33,11 +34,11 @@ on: workflow_dispatch: inputs: route: - description: 'Which update route to exercise.' + description: 'Which update route to exercise. all/both include the Windows desktop leg.' required: false type: choice default: all - options: [all, both, update, installer] + options: [all, both, update, installer, windows-desktop] tag-count: description: 'How many release tags to sample (newest, oldest, and a spread between).' required: false @@ -115,9 +116,14 @@ jobs: route: installer install-ref: ${{ matrix.install-ref }} - # Windows desktop coverage moved to desktop-windows-e2e.yml: two jobs — - # a headless desktop-update.ps1 contract job (BASE -> CURRENT -> NEXT) - # and a REAL user-flow GUI job (website Hermes-Setup.exe, AutoHotkey - # Install/Launch clicks, Playwright-driven Settings -> About -> Update - # now, relaunched-window assert). Triggered on push-to-main + nightly + - # release tags + dispatch there, so no leg here. + # Windows desktop: install OLD with the website's published bootstrap + # installer (the exact bits a user downloads today, AutoHotkey-clicked), + # then update to HEAD by clicking the app's own Update button under + # Playwright. The reusable workflow's route input selects the update + # mechanism; `hermes update` and installer-rerun arms are declared TODOs + # there. + windows-desktop: + if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "windows-desktop"]'), inputs.route) + uses: ./.github/workflows/install-e2e-windows-run.yml + with: + route: desktop diff --git a/tests/install/windows-desktop-e2e.ps1 b/tests/install/windows-desktop-e2e.ps1 deleted file mode 100644 index 959576bb25ec..000000000000 --- a/tests/install/windows-desktop-e2e.ps1 +++ /dev/null @@ -1,339 +0,0 @@ -# ============================================================================ -# Windows Desktop install + update E2E driver -# ============================================================================ -# Proves, on a real Windows machine, that: -# -# 1. INSTALL - the commit-prior-to-HEAD ("BASE") installs from scratch -# through the real scripts/install.ps1 (uv, managed Python, -# Node, venv, packaged Desktop Hermes.exe). -# 2. UPDATE 1 - that BASE install updates to HEAD ("CURRENT") through the -# real Desktop GUI update path: scripts/desktop-update.ps1, -# the exact hand-off script the Update button spawns -# (see apps/desktop/electron/main.ts applyUpdates()). -# 3. UPDATE 2 - the now-CURRENT install updates once more to a synthetic -# "NEXT" commit staged on top of HEAD, proving the *outgoing* -# update path of the commit under test is not broken either. -# (before -> current proves we can be updated TO; -# current -> next proves we can update FROM.) -# -# HOW THE STAGING WORKS (no MITM proxy, no network fakery): -# We bare-clone the checkout into \serve.git and point every git -# process at it with url..insteadOf rewrites for the two -# canonical repo URLs, injected via GIT_CONFIG_{COUNT,KEY_n,VALUE_n} -# environment variables. install.ps1's `git clone` and `hermes update`'s -# `git fetch origin` therefore transparently hit OUR bare repo, whose -# `main` ref we advance between legs: BASE -> CURRENT -> NEXT. The -# installer and updater code paths run unmodified, byte-for-byte as in -# production. Everything else (uv, PyPI, npm, PortableGit download) uses -# the real network, same as a user install. -# -# NEXT is a same-tree child of CURRENT (git commit-tree), so it exists -# only in serve.git and can never leak a content change; leg 3 verifies -# by commit SHA. -# -# WHY NOT AutoHotkey / pixel-driving the installer window (the prior -# attempt, PR #68183): image-search button clicking is resolution- and -# theme-fragile and never stabilized. The GUI Update button's entire effect -# is spawning desktop-update.ps1 with documented flags (its "CONTRACT" -# header); driving that contract directly tests the same production code -# deterministically. -NoUi and the DesktopPid wait gate exist in that script -# precisely for tests. -# -# USAGE (local Windows box or CI): -# powershell -File tests\install\windows-desktop-e2e.ps1 # all -# powershell -File tests\install\windows-desktop-e2e.ps1 -Phase stage -# ... -Phase install / update-to-current / update-to-next -# Phases share state via \shas.json, so CI can run them as -# separate steps for readable logs. -# -# The workroot defaults to $env:HERMES_E2E_WORKROOT or %TEMP%. Pass -Keep -# to leave everything on disk for inspection. -# ============================================================================ - -param( - [ValidateSet("stage", "install", "update-to-current", "update-to-next", "all")] - [string]$Phase = "all", - - # Repo checkout whose HEAD is the commit under test. - [string]$RepoRoot = "", - - # Where the bare serve repo + isolated HERMES_HOME live. - [string]$WorkRoot = $(if ($env:HERMES_E2E_WORKROOT) { $env:HERMES_E2E_WORKROOT } else { Join-Path $env:TEMP "hermes-desktop-e2e" }), - - # Keep the workroot after a successful `-Phase all` run. - [switch]$Keep -) - -$ErrorActionPreference = "Stop" -$ProgressPreference = "SilentlyContinue" - -if (-not $RepoRoot) { - $RepoRoot = (Resolve-Path (Join-Path $PSScriptRoot "..\..")).Path -} - -$ServeRepo = Join-Path $WorkRoot "serve.git" -$HermesHome = Join-Path $WorkRoot "hermes-home" -$InstallDir = Join-Path $HermesHome "hermes-agent" -$StatePath = Join-Path $WorkRoot "shas.json" - -# The two URL spellings install.ps1 clones from and `hermes update` fetches -# from. insteadOf is prefix-based; we register the exact full forms only, so -# nothing else can accidentally rewrite to a path + ".git" suffix. -$RepoUrlHttps = "https://github.com/NousResearch/hermes-agent.git" -$RepoUrlSsh = "git@github.com:NousResearch/hermes-agent.git" - -function Write-Step([string]$Message) { - Write-Host "" - Write-Host ("=" * 74) - Write-Host "== $Message" - Write-Host ("=" * 74) -} - -function Assert-True([bool]$Condition, [string]$Message) { - if (-not $Condition) { - throw "E2E ASSERTION FAILED: $Message" - } - Write-Host " [ok] $Message" -} - -function Invoke-Git([string[]]$GitArgs, [string]$WorkDir = $null) { - $prev = if ($WorkDir) { Get-Location } else { $null } - # PS 5.1 trap: under $ErrorActionPreference = "Stop", a native command - # that writes ANYTHING to stderr while merged via 2>&1 throws a - # NativeCommandError even when it exits 0 (git loves stderr for - # progress/notices). Relax EAP around the native call only; exit-code - # checking below is the real error gate. - $prevEap = $ErrorActionPreference - $ErrorActionPreference = "Continue" - try { - if ($WorkDir) { Set-Location $WorkDir } - $output = & git @GitArgs 2>&1 - if ($LASTEXITCODE -ne 0) { - throw "git $($GitArgs -join ' ') failed (exit $LASTEXITCODE): $output" - } - return ($output | Out-String).Trim() - } finally { - $ErrorActionPreference = $prevEap - if ($prev) { Set-Location $prev } - } -} - -function Set-GitRedirect { - # Route the canonical repo URLs to the local bare repo for THIS process - # and every child (install.ps1's git, hermes update's git). - # - # MECHANISM: a driver-owned global gitconfig selected via - # GIT_CONFIG_GLOBAL. Do NOT use GIT_CONFIG_COUNT/KEY_n/VALUE_n env - # config here -- install.ps1 SETS those itself (GIT_CONFIG_COUNT=1, - # windows.appendAtomically), silently clobbering any redirect we put - # there. That exact clobber made the first CI run clone real GitHub - # main instead of the staged BASE (caught by the HEAD-at-BASE assert). - # install.ps1's own `git config --global` writes simply land in our - # file, so its compat settings still apply. Nothing leaks onto the - # machine: the file lives in the workroot and dies with it. - $fileUrl = "file:///" + ($ServeRepo -replace "\\", "/") - $gitCfg = Join-Path $WorkRoot "e2e-gitconfig" - if (-not (Test-Path -LiteralPath $WorkRoot)) { - New-Item -ItemType Directory -Path $WorkRoot -Force | Out-Null - } - @" -[url "$fileUrl"] - insteadOf = $RepoUrlHttps - insteadOf = $RepoUrlSsh -"@ | Set-Content -LiteralPath $gitCfg -Encoding ASCII - $env:GIT_CONFIG_GLOBAL = $gitCfg - Write-Host " git URL redirect via GIT_CONFIG_GLOBAL=$gitCfg" - Write-Host " $RepoUrlHttps -> $fileUrl" -} - -function Read-State { - if (-not (Test-Path -LiteralPath $StatePath)) { - throw "State file not found: $StatePath -- run '-Phase stage' first." - } - return Get-Content -LiteralPath $StatePath -Raw | ConvertFrom-Json -} - -function Get-InstalledHead { - return Invoke-Git @("-C", $InstallDir, "rev-parse", "HEAD") -} - -function Get-DesktopExe { - $candidates = @( - (Join-Path $InstallDir "apps\desktop\release\win-unpacked\Hermes.exe"), - (Join-Path $InstallDir "apps\desktop\release\win-arm64-unpacked\Hermes.exe") - ) - foreach ($c in $candidates) { - if (Test-Path -LiteralPath $c) { return $c } - } - return $null -} - -function Test-HermesRuns([string]$Label) { - $hermesExe = Join-Path $InstallDir "venv\Scripts\hermes.exe" - Assert-True (Test-Path -LiteralPath $hermesExe) "$Label -- venv\Scripts\hermes.exe exists" - & $hermesExe --version 2>&1 | ForEach-Object { Write-Host " hermes --version| $_" } - Assert-True ($LASTEXITCODE -eq 0) "$Label -- hermes --version exits 0" -} - -# ---------------------------------------------------------------------------- -# Phase: stage -# ---------------------------------------------------------------------------- -function Invoke-PhaseStage { - Write-Step "STAGE: bare serve repo + BASE/CURRENT/NEXT refs" - - if (Test-Path -LiteralPath $WorkRoot) { - Remove-Item -LiteralPath $WorkRoot -Recurse -Force - } - New-Item -ItemType Directory -Path $WorkRoot -Force | Out-Null - # The purge above deleted the redirect gitconfig; re-arm it so the - # bare-clone below (and everything after) sees the redirect file. - Set-GitRedirect - - $current = Invoke-Git @("-C", $RepoRoot, "rev-parse", "HEAD") - $base = Invoke-Git @("-C", $RepoRoot, "rev-parse", "HEAD~1") - Write-Host " CURRENT (commit under test): $current" - Write-Host " BASE (its parent): $base" - - # Bare-clone the checkout: this is the repo the installer and updater - # will actually talk to. Local-path clone hardlinks objects, so it's - # fast even for full history. - Invoke-Git @("clone", "--bare", "--quiet", $RepoRoot, $ServeRepo) | Out-Null - - # Synthesize NEXT inside the bare repo: a same-tree child of CURRENT. - # It exists nowhere else, which is the point -- leg 3 proves the commit - # under test can update to a future main it has never seen. - $env:GIT_AUTHOR_NAME = "Hermes E2E"; $env:GIT_AUTHOR_EMAIL = "e2e@nousresearch.com" - $env:GIT_COMMITTER_NAME = "Hermes E2E"; $env:GIT_COMMITTER_EMAIL = "e2e@nousresearch.com" - $next = Invoke-Git @("-C", $ServeRepo, "commit-tree", "$current^{tree}", "-p", $current, "-m", "e2e: synthetic next commit (same tree as CURRENT)") - Write-Host " NEXT (synthetic): $next" - - # Serve BASE as `main` first; update legs advance this ref. - Invoke-Git @("-C", $ServeRepo, "update-ref", "refs/heads/main", $base) | Out-Null - Invoke-Git @("-C", $ServeRepo, "symbolic-ref", "HEAD", "refs/heads/main") | Out-Null - - @{ base = $base; current = $current; next = $next } | - ConvertTo-Json | Set-Content -LiteralPath $StatePath -Encoding UTF8 - Write-Host " state written: $StatePath" -} - -# ---------------------------------------------------------------------------- -# Phase: install (BASE, via BASE's own install.ps1) -# ---------------------------------------------------------------------------- -function Invoke-PhaseInstall { - $state = Read-State - Write-Step "INSTALL: BASE ($($state.base)) via its own install.ps1 -IncludeDesktop" - - # The honest test runs the installer *as it existed at BASE* -- a user - # installing yesterday used yesterday's script. - $baseInstaller = Join-Path $WorkRoot "install-base.ps1" - & git -C $ServeRepo show "$($state.base):scripts/install.ps1" | - Set-Content -LiteralPath $baseInstaller -Encoding UTF8 - if ($LASTEXITCODE -ne 0) { throw "could not extract scripts/install.ps1 at BASE" } - - $env:HERMES_HOME = $HermesHome - New-Item -ItemType Directory -Path $HermesHome -Force | Out-Null - - # Windows PowerShell 5.1 on purpose: it is what the production - # `irm | iex` one-liner and the desktop bootstrap both run under. - & powershell.exe -NoProfile -ExecutionPolicy Bypass -File $baseInstaller ` - -NonInteractive -SkipSetup -IncludeDesktop ` - -HermesHome $HermesHome -InstallDir $InstallDir - Assert-True ($LASTEXITCODE -eq 0) "install.ps1 (BASE) exited 0" - - Assert-True ((Get-InstalledHead) -eq $state.base) "installed checkout is at BASE" - Test-HermesRuns "post-install" - $desktopExe = Get-DesktopExe - Assert-True ($null -ne $desktopExe) "packaged Desktop Hermes.exe exists ($desktopExe)" -} - -# ---------------------------------------------------------------------------- -# Update legs (shared): the real Desktop GUI update path -# ---------------------------------------------------------------------------- -function Invoke-DesktopUpdateLeg([string]$TargetSha, [string]$LegName) { - Write-Step "UPDATE ($LegName): advance served main -> $TargetSha, run desktop-update.ps1" - - $env:HERMES_HOME = $HermesHome - Invoke-Git @("-C", $ServeRepo, "update-ref", "refs/heads/main", $TargetSha) | Out-Null - - # The hand-off script from the INSTALLED checkout drives the update -- - # exactly the production contract: each `hermes update` refreshes the - # script that drives the NEXT update. - $handoff = Join-Path $InstallDir "scripts\desktop-update.ps1" - Assert-True (Test-Path -LiteralPath $handoff) "installed checkout ships scripts/desktop-update.ps1" - - # Stand in for the exiting Electron main process: desktop-update.ps1 - # FAIL-CLOSED waits for this pid to exit before touching the install. - # A short-lived real process exercises that gate. - $dummy = Start-Process -FilePath "powershell.exe" ` - -ArgumentList "-NoProfile", "-Command", "Start-Sleep -Seconds 3" ` - -WindowStyle Hidden -PassThru - - $resultPath = Join-Path $HermesHome ".hermes-update-result.json" - $markerPath = Join-Path $HermesHome ".hermes-update-in-progress" - Remove-Item -LiteralPath $resultPath -Force -ErrorAction SilentlyContinue - - # -NoUi: headless (no WinForms in CI). No -RelaunchExe: contract says - # omit = no relaunch, so no orphaned Electron process on the runner. - & powershell.exe -NoProfile -ExecutionPolicy Bypass -File $handoff ` - -InstallRoot $InstallDir -Branch main -DesktopPid $dummy.Id -NoUi - $handoffExit = $LASTEXITCODE - - # Surface the hand-off log before asserting, so failures are debuggable - # straight from the CI step output. - $logPath = Join-Path $HermesHome "logs\desktop-update-handoff.log" - if (Test-Path -LiteralPath $logPath) { - Write-Host " --- desktop-update-handoff.log (tail) ---" - Get-Content -LiteralPath $logPath -Tail 40 | ForEach-Object { Write-Host " | $_" } - } - - Assert-True ($handoffExit -eq 0) "$LegName -- desktop-update.ps1 exited 0" - - Assert-True (Test-Path -LiteralPath $resultPath) "$LegName -- update result JSON written" - $result = Get-Content -LiteralPath $resultPath -Raw | ConvertFrom-Json - Assert-True ([bool]$result.ok) "$LegName -- result JSON reports ok=true ('$($result.message)')" - - Assert-True (-not (Test-Path -LiteralPath $markerPath)) "$LegName -- update marker cleaned up" - Assert-True ((Get-InstalledHead) -eq $TargetSha) "$LegName -- checkout landed on target commit" - Test-HermesRuns $LegName - Assert-True ($null -ne (Get-DesktopExe)) "$LegName -- Desktop Hermes.exe still present after update" -} - -function Invoke-PhaseUpdateToCurrent { - $state = Read-State - Invoke-DesktopUpdateLeg $state.current "BASE -> CURRENT" -} - -function Invoke-PhaseUpdateToNext { - $state = Read-State - Invoke-DesktopUpdateLeg $state.next "CURRENT -> NEXT" -} - -# ---------------------------------------------------------------------------- -# Dispatch -# ---------------------------------------------------------------------------- -Write-Host "Windows Desktop E2E driver" -Write-Host " phase: $Phase" -Write-Host " repo: $RepoRoot" -Write-Host " workroot: $WorkRoot" - -Set-GitRedirect - -switch ($Phase) { - "stage" { Invoke-PhaseStage } - "install" { Invoke-PhaseInstall } - "update-to-current" { Invoke-PhaseUpdateToCurrent } - "update-to-next" { Invoke-PhaseUpdateToNext } - "all" { - Invoke-PhaseStage - Invoke-PhaseInstall - Invoke-PhaseUpdateToCurrent - Invoke-PhaseUpdateToNext - if (-not $Keep) { - Write-Step "CLEANUP" - Remove-Item -LiteralPath $WorkRoot -Recurse -Force -ErrorAction SilentlyContinue - } - } -} - -Write-Host "" -Write-Host "Phase '$Phase' completed successfully." diff --git a/tests/install/windows-desktop-gui-e2e.ps1 b/tests/install/windows-desktop-gui-e2e.ps1 index 799ba5ea5501..ed4103162ff5 100644 --- a/tests/install/windows-desktop-gui-e2e.ps1 +++ b/tests/install/windows-desktop-gui-e2e.ps1 @@ -1,27 +1,44 @@ # ============================================================================ # Windows Desktop GUI install + update E2E driver (the REAL user flow) # ============================================================================ -# Same before -> current -> next staging as windows-desktop-e2e.ps1, but every -# leg goes through the surfaces a real user touches: +# Proves, on a real Windows machine, that a user who installs Hermes the way +# the website tells them to can then update to the commit under test through +# a real update surface -- with every leg driven through the GUI a user +# actually touches: # # INSTALL - downloads the production Hermes-Setup.exe from the website, # launches it HEADED, and AutoHotkey clicks Install, waits, # then clicks Launch. The real Electron Hermes.exe must appear. # The exe runs EXACTLY as shipped: it downloads its own pinned # install.ps1 from GitHub raw and installs its baked -# BUILD_PIN_COMMIT -- so the "before" version in this harness is -# the website release pin, the literal starting point of every -# real GUI user. (The strict HEAD~1 -> HEAD guarantee is the -# contract harness's job; this one covers the production pin.) -# UPDATE x2 - launches the installed Hermes.exe under Playwright's Electron -# driver and CLICKS Settings -> About -> "Update now". That -# spawns the production hand-off (desktop-update.ps1 via -# cmd start, or the staged hermes-setup.exe on old checkouts), -# the app quits, the detached updater runs `hermes update`, -# rebuilds the desktop, and RELAUNCHES Hermes.exe. We assert -# the whole chain: target sha, marker cleanup, working hermes, -# and the relaunched app window. Leg 1 lands on CURRENT, leg 2 -# on the synthetic NEXT. +# BUILD_PIN_COMMIT -- so OLD is the website release pin, the +# literal starting point of every real GUI user. +# UPDATE - OLD -> HEAD through the route selected by -Route: +# desktop (implemented) launch the installed Hermes.exe +# under Playwright's Electron driver and CLICK +# Settings -> About -> "Update now". The +# production hand-off chain runs untouched: +# marker, app quit, detached updater, `hermes +# update`, desktop rebuild, RELAUNCH. Asserts +# target sha, marker cleanup, result JSON (when +# the script path wrote one), working hermes, +# and the relaunched app window. +# update TODO: run `hermes update` from the installed +# venv (the CLI route a GUI user might take). +# installer TODO: re-run the bootstrap installer over the +# existing install (its --update flow). +# +# HOW THE STAGING WORKS (no MITM proxy, no network fakery): +# We bare-clone the checkout into \serve.git and point every git +# process at it with url..insteadOf rewrites for the two +# canonical repo URLs, via a driver-owned gitconfig selected with +# GIT_CONFIG_GLOBAL. (NOT GIT_CONFIG_COUNT/KEY_n/VALUE_n env config -- +# install.ps1 sets those itself and silently clobbers them.) The +# installer's `git clone` and `hermes update`'s `git fetch origin` +# transparently hit OUR bare repo, whose `main` ref serves HEAD -- the +# update target. Installer and updater run byte-for-byte as shipped; +# everything else (uv, PyPI, npm, the installer's raw.githubusercontent +# install.ps1 download) uses the real network, same as a user install. # # PROOF: screenshots at every renderer step (Playwright), full-desktop # screenshots around the installer/AHK phases, a rolling desktop capture @@ -29,27 +46,41 @@ # as CI artifacts. # # DEVIATIONS FROM PRODUCTION (each one deliberate and small): -# * git URL redirect (GIT_CONFIG_GLOBAL) routes the canonical repo URLs -# to the staged serve.git -- the staging requirement itself. The -# installer's raw.githubusercontent install.ps1 download and the pinned -# ZIP fallback are NOT redirected (real network, as shipped). +# * the git URL redirect itself -- the staging requirement. # * serve.git gets uploadpack.allowAnySHA1InWant=true so the installer's # baked -Commit pin can be fetched from the redirected clone the same # way GitHub's upload-pack allows it. -# * A dummy provider key is seeded after install so the update legs see +# * A dummy provider key is seeded after install so the update leg sees # the ready app shell instead of the onboarding overlay (a real # updating user has a configured provider). +# * .skip_upstream_prompt is set: serve.git's file:// origin looks like a +# fork to `hermes update`, whose fork-only upstream prompt is a bare +# input() that hangs forever under the desktop's detached console. +# Real GUI users are on the official origin and never see it. # -# Usage mirrors windows-desktop-e2e.ps1: +# USAGE (local Windows box or CI): # powershell -File tests\install\windows-desktop-gui-e2e.ps1 -Phase all -# ... -Phase stage / install-gui / update-gui-to-current / update-gui-to-next +# ... -Phase stage / install-gui / update-gui +# Phases share state via \shas.json, so CI can run them as +# separate steps for readable logs. # ============================================================================ param( - [ValidateSet("stage", "install-gui", "update-gui-to-current", "update-gui-to-next", "all")] + [ValidateSet("stage", "install-gui", "update-gui", "all")] [string]$Phase = "all", + + # Update route to exercise in the update-gui phase. Only "desktop" (the + # app's own Update button) is implemented; "update" (CLI `hermes + # update`) and "installer" (re-run the bootstrap exe) are declared arms + # so the workflow surface is stable when they land. + [ValidateSet("desktop", "update", "installer")] + [string]$Route = "desktop", + + # Repo checkout whose HEAD is the update target. [string]$RepoRoot = "", + [string]$WorkRoot = $(if ($env:HERMES_E2E_WORKROOT) { $env:HERMES_E2E_WORKROOT } else { Join-Path $env:TEMP "hermes-desktop-gui-e2e" }), + [string]$SetupExeUrl = "https://hermes-assets.nousresearch.com/Hermes-Setup.exe" ) @@ -86,6 +117,11 @@ function Assert-True([bool]$Condition, [string]$Message) { } function Invoke-Git([string[]]$GitArgs) { + # PS 5.1 trap: under $ErrorActionPreference = "Stop", a native command + # that writes ANYTHING to stderr while merged via 2>&1 throws a + # NativeCommandError even when it exits 0 (git loves stderr for + # progress/notices). Relax EAP around the native call only; exit-code + # checking below is the real error gate. $prevEap = $ErrorActionPreference $ErrorActionPreference = "Continue" try { @@ -100,8 +136,16 @@ function Invoke-Git([string[]]$GitArgs) { } function Set-GitRedirect { - # Same mechanism (and same install.ps1-clobber rationale) as - # windows-desktop-e2e.ps1: a driver-owned gitconfig via GIT_CONFIG_GLOBAL. + # Route the canonical repo URLs to the local bare repo for THIS process + # and every child (the installer's git, hermes update's git). + # + # MECHANISM: a driver-owned global gitconfig selected via + # GIT_CONFIG_GLOBAL. Do NOT use GIT_CONFIG_COUNT/KEY_n/VALUE_n env + # config here -- install.ps1 SETS those itself (GIT_CONFIG_COUNT=1, + # windows.appendAtomically), silently clobbering any redirect we put + # there. install.ps1's own `git config --global` writes simply land in + # our file, so its compat settings still apply. Nothing leaks onto the + # machine: the file lives in the workroot and dies with it. $fileUrl = "file:///" + ($ServeRepo -replace "\\", "/") $gitCfg = Join-Path $WorkRoot "e2e-gitconfig" if (-not (Test-Path -LiteralPath $WorkRoot)) { @@ -114,6 +158,7 @@ function Set-GitRedirect { "@ | Set-Content -LiteralPath $gitCfg -Encoding ASCII $env:GIT_CONFIG_GLOBAL = $gitCfg Write-Host " git URL redirect via GIT_CONFIG_GLOBAL=$gitCfg" + Write-Host " $RepoUrlHttps -> $fileUrl" } function Read-State { @@ -145,7 +190,7 @@ function Test-HermesRuns([string]$Label) { } function Save-DesktopScreenshot([string]$OutFile) { - # Single full-desktop screenshot (all monitors' primary screen). + # Single full-desktop screenshot (primary screen). try { Add-Type -AssemblyName System.Windows.Forms, System.Drawing $bounds = [System.Windows.Forms.Screen]::PrimaryScreen.Bounds @@ -197,7 +242,7 @@ function Stop-DesktopRecorder($proc, [string]$OutDir) { } function Stop-HermesAppProcesses([string]$Label) { - # Close the desktop app the blunt way between legs (a user quitting). + # Close the desktop app the blunt way between phases (a user quitting). # Only Hermes.exe (Electron) -- never hermes.exe (the venv CLI shim). $procs = @(Get-Process -Name "Hermes" -ErrorAction SilentlyContinue) foreach ($p in $procs) { @@ -226,20 +271,40 @@ function Get-ManagedNode { } # ---------------------------------------------------------------------------- -# Phase: stage -- reuse the contract driver's stage (identical staging) +# Phase: stage -- serve.git with `main` at HEAD (the update target) # ---------------------------------------------------------------------------- function Invoke-PhaseStage { - Write-Step "STAGE (gui): delegating to windows-desktop-e2e.ps1 -Phase stage" - & powershell.exe -NoProfile -ExecutionPolicy Bypass ` - -File (Join-Path $PSScriptRoot "windows-desktop-e2e.ps1") ` - -Phase stage -RepoRoot $RepoRoot -WorkRoot $WorkRoot - if ($LASTEXITCODE -ne 0) { throw "stage phase failed" } + Write-Step "STAGE: bare serve repo, main -> HEAD (update target)" + + if (Test-Path -LiteralPath $WorkRoot) { + Remove-Item -LiteralPath $WorkRoot -Recurse -Force + } + New-Item -ItemType Directory -Path $WorkRoot -Force | Out-Null + # The purge above deleted the redirect gitconfig; re-arm it so the + # bare-clone below (and everything after) sees the redirect file. + Set-GitRedirect + + $current = Invoke-Git @("-C", $RepoRoot, "rev-parse", "HEAD") + Write-Host " HEAD (update target): $current" + + # Bare-clone the checkout: this is the repo the installer and updater + # actually talk to. Local-path clone hardlinks objects, so it's fast + # even for full history. OLD is not staged by us -- it is whatever + # release pin the website installer carries (recorded in install-gui). + Invoke-Git @("clone", "--bare", "--quiet", $RepoRoot, $ServeRepo) | Out-Null + Invoke-Git @("-C", $ServeRepo, "update-ref", "refs/heads/main", $current) | Out-Null + Invoke-Git @("-C", $ServeRepo, "symbolic-ref", "HEAD", "refs/heads/main") | Out-Null + # The website installer pins a specific release commit (-Commit ). # That sha is in serve.git's history but not at a ref tip, so the # redirected fetch needs any-SHA1 upload-pack permission (GitHub grants - # the equivalent for archive/fetch of reachable commits). + # the equivalent for fetch of reachable commits). Invoke-Git @("-C", $ServeRepo, "config", "uploadpack.allowAnySHA1InWant", "true") | Out-Null Write-Host " serve.git: uploadpack.allowAnySHA1InWant=true (installer commit pin)" + + @{ current = $current } | + ConvertTo-Json | Set-Content -LiteralPath $StatePath -Encoding UTF8 + Write-Host " state written: $StatePath" New-Item -ItemType Directory -Path $ProofRoot -Force | Out-Null } @@ -334,39 +399,39 @@ function Invoke-PhaseInstallGui { # Close the freshly launched app (user quits after first look). Stop-HermesAppProcesses "post-install" - # The website exe installs its baked release pin -- record it as the - # "before" version. It must differ from CURRENT (an update is genuinely - # available). We do NOT require it to be an ancestor of CURRENT: on a - # real `push: main` run CURRENT is main's tip and the release pin is - # behind it (ancestor), but on a feature branch CURRENT has diverged - # from main, so the release pin legitimately isn't in its ancestry. The - # update leg resets the checkout to serve.git's main ref regardless, and - # asserts it lands on CURRENT -- that is the real forward-update proof. + # The website exe installs its baked release pin -- that IS the OLD + # version. It must differ from HEAD (an update is genuinely available). + # We do NOT require it to be an ancestor of HEAD: on a real `push: main` + # run HEAD is main's tip and the release pin is behind it (ancestor), + # but on a feature branch HEAD has diverged from main, so the release + # pin legitimately isn't in its ancestry. The update leg resets the + # checkout to serve.git's main ref regardless, and asserts it lands on + # HEAD -- that is the real forward-update proof. $installedSha = Get-InstalledHead - Write-Host " installer landed on: $installedSha (website release pin)" - Assert-True ($installedSha -ne $state.current) "installed pin differs from CURRENT (an update is genuinely available)" + Write-Host " installer landed on: $installedSha (website release pin = OLD)" + Assert-True ($installedSha -ne $state.current) "installed pin differs from HEAD (an update is genuinely available)" $prevEap = $ErrorActionPreference; $ErrorActionPreference = "Continue" & git -C $InstallDir merge-base --is-ancestor $installedSha $state.current 2>&1 | Out-Null $isAncestor = ($LASTEXITCODE -eq 0) $ErrorActionPreference = $prevEap if ($isAncestor) { - Write-Host " [ok] installed pin is an ancestor of CURRENT (linear before -> current)" + Write-Host " [ok] installed pin is an ancestor of HEAD (linear old -> new)" } else { - Write-Host " [note] installed pin is NOT an ancestor of CURRENT -- expected on a diverged feature branch; the update leg still resets to CURRENT" + Write-Host " [note] installed pin is NOT an ancestor of HEAD -- expected on a diverged feature branch; the update leg still resets to HEAD" } Test-HermesRuns "post-install-gui" Assert-True ($null -ne (Get-DesktopExe)) "packaged Desktop Hermes.exe exists" - # Seed a provider so the update legs meet the ready app shell, not the + # Seed a provider so the update leg meets the ready app shell, not the # onboarding overlay (an updating user has a configured provider). $envFile = Join-Path $HermesHome ".env" if (-not (Test-Path -LiteralPath $envFile) -or -not ((Get-Content $envFile -Raw -ErrorAction SilentlyContinue) -match "OPENROUTER_API_KEY")) { - Add-Content -LiteralPath $envFile -Value "OPENROUTER_API_KEY=sk-or-e2e-placeholder-not-a-real-key" + Add-Content -LiteralPath $envFile -Value "OPENROUTER_API_KEY=sk-or-...-key" } - Write-Host " seeded placeholder provider key for update legs" + Write-Host " seeded placeholder provider key for the update leg" # Suppress the interactive "add upstream remote?" prompt during the GUI - # update legs. Our serve.git origin (file://) looks like a fork to + # update leg. Our serve.git origin (file://) looks like a fork to # `hermes update`, so `_sync_with_upstream_if_needed` would call bare # input() -- which HANGS FOREVER when the Desktop spawns the hand-off via # `cmd start /min` (a real but empty console; input() blocks waiting for a @@ -380,18 +445,17 @@ function Invoke-PhaseInstallGui { } # ---------------------------------------------------------------------------- -# Update legs: real app, real clicks (Playwright Electron driver) +# Phase: update-gui -- OLD -> HEAD through the selected route # ---------------------------------------------------------------------------- -function Invoke-GuiUpdateLeg([string]$TargetSha, [string]$LegName, [string]$LegSlug) { - Write-Step "UPDATE GUI ($LegName): advance served main -> $TargetSha, click Update now" - $proof = Join-Path $ProofRoot $LegSlug +function Invoke-GuiUpdateDesktopRoute([string]$TargetSha) { + Write-Step "UPDATE (GUI, route=desktop): click Update now, land on $TargetSha" + $proof = Join-Path $ProofRoot "update-gui" New-Item -ItemType Directory -Path $proof -Force | Out-Null $env:HERMES_HOME = $HermesHome - Invoke-Git @("-C", $ServeRepo, "update-ref", "refs/heads/main", $TargetSha) | Out-Null $desktopExe = Get-DesktopExe - Assert-True ($null -ne $desktopExe) "$LegName -- packaged Hermes.exe present before update" + Assert-True ($null -ne $desktopExe) "packaged Hermes.exe present before update" $resultPath = Join-Path $HermesHome ".hermes-update-result.json" $markerPath = Join-Path $HermesHome ".hermes-update-in-progress" @@ -407,7 +471,7 @@ function Invoke-GuiUpdateLeg([string]$TargetSha, [string]$LegName, [string]$LegS & $node -e "require.resolve('@playwright/test', { paths: [process.argv[1]] })" $appsDesktop 2>&1 | Out-Null $pwResolved = ($LASTEXITCODE -eq 0) $ErrorActionPreference = $prevEap - Assert-True $pwResolved "$LegName -- @playwright/test resolvable from installed apps/desktop" + Assert-True $pwResolved "@playwright/test resolvable from installed apps/desktop" $recorder = Start-DesktopRecorder (Join-Path $proof "desktop-frames") try { @@ -428,7 +492,7 @@ function Invoke-GuiUpdateLeg([string]$TargetSha, [string]$LegName, [string]$LegS Pop-Location Remove-Item -LiteralPath $driver -Force -ErrorAction SilentlyContinue } - Assert-True ($driveExit -eq 0) "$LegName -- GUI driver clicked Update now and the app quit for hand-off" + Assert-True ($driveExit -eq 0) "GUI driver clicked Update now and the app quit for hand-off" # The detached updater (spawned by the app, NOT by us) now runs # `hermes update` + desktop rebuild + relaunch. Which updater depends @@ -443,12 +507,12 @@ function Invoke-GuiUpdateLeg([string]$TargetSha, [string]$LegName, [string]$LegS # only when the script path produced it. # # This can take a LONG time: the website release we installed is weeks - # of main behind CURRENT, so the update pulls a large diff AND does a + # of main behind HEAD, so the update pulls a large diff AND does a # full Electron desktop rebuild (vite + electron-builder) plus a uv # sync. The desktop-build output goes to logs/update.log (not the # streamed handoff log), so we tail update.log here to show progress # instead of going silent for tens of minutes. - Write-Host " waiting for the detached updater to finish (up to 90 min; large release->CURRENT rebuild) ..." + Write-Host " waiting for the detached updater to finish (up to 90 min; large old->new rebuild) ..." $updateLog = Join-Path $HermesHome "logs\update.log" $updateLogPos = 0 $deadline = (Get-Date).AddMinutes(90) @@ -473,7 +537,7 @@ function Invoke-GuiUpdateLeg([string]$TargetSha, [string]$LegName, [string]$LegS if (Test-Path -LiteralPath $resultPath) { $result = Get-Content -LiteralPath $resultPath -Raw | ConvertFrom-Json Write-Host " updater result: ok=$($result.ok) code=$($result.exit_code) msg=$($result.message)" - Assert-True ([bool]$result.ok) "$LegName -- updater result ok=true" + Assert-True ([bool]$result.ok) "updater result ok=true" } else { Write-Host " (no result JSON -- staged-binary updater path; relying on sha/marker/relaunch asserts)" } @@ -481,11 +545,11 @@ function Invoke-GuiUpdateLeg([string]$TargetSha, [string]$LegName, [string]$LegS # Marker may briefly outlive the result write; allow it a moment. $mDeadline = (Get-Date).AddMinutes(2) while ((Get-Date) -lt $mDeadline -and (Test-Path -LiteralPath $markerPath)) { Start-Sleep -Seconds 5 } - Assert-True (-not (Test-Path -LiteralPath $markerPath)) "$LegName -- update marker cleaned up" + Assert-True (-not (Test-Path -LiteralPath $markerPath)) "update marker cleaned up" - Assert-True ((Get-InstalledHead) -eq $TargetSha) "$LegName -- checkout landed on target commit" - Test-HermesRuns $LegName - Assert-True ($null -ne (Get-DesktopExe)) "$LegName -- Hermes.exe still present after update" + Assert-True ((Get-InstalledHead) -eq $TargetSha) "checkout landed on target commit" + Test-HermesRuns "post-update" + Assert-True ($null -ne (Get-DesktopExe)) "Hermes.exe still present after update" # The production hand-off relaunches the desktop (RelaunchExe). # A relaunched window is the user-visible proof the update loop closed. @@ -497,7 +561,7 @@ function Invoke-GuiUpdateLeg([string]$TargetSha, [string]$LegName, [string]$LegS if ($relaunched) { break } Start-Sleep -Seconds 5 } - Assert-True ($null -ne $relaunched) "$LegName -- updater relaunched the desktop app" + Assert-True ($null -ne $relaunched) "updater relaunched the desktop app" Start-Sleep -Seconds 12 # let the window paint for the screenshot # Foreground the relaunched Hermes window so the proof screenshot # captures IT, not whatever else is on top (the full-desktop grab is @@ -525,19 +589,30 @@ function Invoke-GuiUpdateLeg([string]$TargetSha, [string]$LegName, [string]$LegS Get-Content -LiteralPath $handoffLog -Tail 60 | ForEach-Object { Write-Host " | $_" } Copy-Item $handoffLog (Join-Path $proof "desktop-update-handoff.log") -Force -ErrorAction SilentlyContinue } - # Quit the relaunched app so the next leg (or job teardown) is clean. - Stop-HermesAppProcesses $LegName + # Quit the relaunched app so job teardown is clean. + Stop-HermesAppProcesses "post-update" } } -function Invoke-PhaseUpdateGuiToCurrent { - $state = Read-State - Invoke-GuiUpdateLeg $state.current "BASE -> CURRENT (GUI)" "update-gui-to-current" -} - -function Invoke-PhaseUpdateGuiToNext { +function Invoke-PhaseUpdateGui { $state = Read-State - Invoke-GuiUpdateLeg $state.next "CURRENT -> NEXT (GUI)" "update-gui-to-next" + switch ($Route) { + "desktop" { + Invoke-GuiUpdateDesktopRoute $state.current + } + "update" { + # TODO: run `hermes update` from the installed venv -- the CLI + # route. Needs the same completion/sha/relaunch asserts minus + # the app-quit dance. + throw "route 'update' (CLI hermes update) is not implemented yet" + } + "installer" { + # TODO: re-run the bootstrap Hermes-Setup.exe over the existing + # install (its --update flow jumps straight to progress and + # runs unattended). + throw "route 'installer' (re-run bootstrap exe) is not implemented yet" + } + } } # ---------------------------------------------------------------------------- @@ -545,21 +620,20 @@ function Invoke-PhaseUpdateGuiToNext { # ---------------------------------------------------------------------------- Write-Host "Windows Desktop GUI E2E driver (real user flow)" Write-Host " phase: $Phase" +Write-Host " route: $Route" Write-Host " repo: $RepoRoot" Write-Host " workroot: $WorkRoot" Set-GitRedirect switch ($Phase) { - "stage" { Invoke-PhaseStage } - "install-gui" { Invoke-PhaseInstallGui } - "update-gui-to-current" { Invoke-PhaseUpdateGuiToCurrent } - "update-gui-to-next" { Invoke-PhaseUpdateGuiToNext } + "stage" { Invoke-PhaseStage } + "install-gui" { Invoke-PhaseInstallGui } + "update-gui" { Invoke-PhaseUpdateGui } "all" { Invoke-PhaseStage Invoke-PhaseInstallGui - Invoke-PhaseUpdateGuiToCurrent - Invoke-PhaseUpdateGuiToNext + Invoke-PhaseUpdateGui } } From 2a906ac6a0f720fe999abe171b527659da2e1ba7 Mon Sep 17 00:00:00 2001 From: ethernet Date: Tue, 11 Aug 2026 13:57:58 -0400 Subject: [PATCH 037/227] fix(windows-e2e): stage OLD as the served main - the published installer has no commit pin Run 31519103491 failed the 'update genuinely available' assert with the installer landing on HEAD itself. The staging assumed the website exe installs a baked release pin, but the bootstrap log shows Pin { commit: None, branch: main } - the published installer installs whatever main serves, and serve.git's main was parked at HEAD. Stage the way the linux axis does: park served main at OLD (-InstallRef, default newest release tag; threaded through the reusable workflow as install-ref) for the install phase, assert the install lands exactly there, then advance main to HEAD in the update phase - an update becomes available the same way it does for a real user. allowAnySHA1InWant stays as belt-and-braces for installer builds that DO bake a pin. --- .github/workflows/install-e2e-windows-run.yml | 20 ++-- tests/install/windows-desktop-gui-e2e.ps1 | 94 +++++++++++-------- 2 files changed, 68 insertions(+), 46 deletions(-) diff --git a/.github/workflows/install-e2e-windows-run.yml b/.github/workflows/install-e2e-windows-run.yml index bd492d454b8d..8a8ac84790a1 100644 --- a/.github/workflows/install-e2e-windows-run.yml +++ b/.github/workflows/install-e2e-windows-run.yml @@ -10,9 +10,10 @@ name: Install & Update E2E — Windows desktop (reusable) # git's own transport rewrite: a driver-owned GIT_CONFIG_GLOBAL carrying # url..insteadOf for both canonical repo URLs, so the # installer and updater run byte-for-byte against their real URLs and land -# on a local bare repo serving HEAD as `main`. OLD is not staged: it is -# the release pin baked into the website exe — the literal starting point -# of every real GUI user. +# on a local bare repo. Its `main` serves OLD (install-ref, default: the +# newest release tag) during the install, then advances to HEAD for the +# update leg — an update becomes available exactly the way it does for a +# real user. # # Routes (the update mechanism under test): # desktop the app's own Update button: the installed Hermes.exe runs @@ -39,6 +40,11 @@ on: required: false type: string default: desktop + install-ref: + description: 'Ref to install as OLD (served as main while the installer runs). Empty = the newest release tag in the checkout.' + required: false + type: string + default: '' setup-exe-url: description: 'Bootstrap installer to install OLD with. Default: the latest published one — what a user downloads today.' required: false @@ -76,17 +82,17 @@ jobs: with: fetch-depth: 0 - - name: Stage serve repo (main -> HEAD) + - name: Stage serve repo (main -> OLD) shell: powershell - run: powershell -NoProfile -ExecutionPolicy Bypass -File tests\install\windows-desktop-gui-e2e.ps1 -Phase stage -Route ${{ inputs.route }} -SetupExeUrl ${{ inputs.setup-exe-url }} + run: powershell -NoProfile -ExecutionPolicy Bypass -File tests\install\windows-desktop-gui-e2e.ps1 -Phase stage -Route ${{ inputs.route }} -InstallRef "${{ inputs.install-ref }}" -SetupExeUrl ${{ inputs.setup-exe-url }} - name: Install OLD via website Hermes-Setup.exe (headed, AHK-clicked) shell: powershell - run: powershell -NoProfile -ExecutionPolicy Bypass -File tests\install\windows-desktop-gui-e2e.ps1 -Phase install-gui -Route ${{ inputs.route }} -SetupExeUrl ${{ inputs.setup-exe-url }} + run: powershell -NoProfile -ExecutionPolicy Bypass -File tests\install\windows-desktop-gui-e2e.ps1 -Phase install-gui -Route ${{ inputs.route }} -InstallRef "${{ inputs.install-ref }}" -SetupExeUrl ${{ inputs.setup-exe-url }} - name: Update OLD -> HEAD (${{ inputs.route }} route) shell: powershell - run: powershell -NoProfile -ExecutionPolicy Bypass -File tests\install\windows-desktop-gui-e2e.ps1 -Phase update-gui -Route ${{ inputs.route }} -SetupExeUrl ${{ inputs.setup-exe-url }} + run: powershell -NoProfile -ExecutionPolicy Bypass -File tests\install\windows-desktop-gui-e2e.ps1 -Phase update-gui -Route ${{ inputs.route }} -InstallRef "${{ inputs.install-ref }}" -SetupExeUrl ${{ inputs.setup-exe-url }} - name: Collect proof + logs if: always() diff --git a/tests/install/windows-desktop-gui-e2e.ps1 b/tests/install/windows-desktop-gui-e2e.ps1 index ed4103162ff5..ef4b5ee1f696 100644 --- a/tests/install/windows-desktop-gui-e2e.ps1 +++ b/tests/install/windows-desktop-gui-e2e.ps1 @@ -9,10 +9,10 @@ # INSTALL - downloads the production Hermes-Setup.exe from the website, # launches it HEADED, and AutoHotkey clicks Install, waits, # then clicks Launch. The real Electron Hermes.exe must appear. -# The exe runs EXACTLY as shipped: it downloads its own pinned -# install.ps1 from GitHub raw and installs its baked -# BUILD_PIN_COMMIT -- so OLD is the website release pin, the -# literal starting point of every real GUI user. +# The exe runs EXACTLY as shipped against serve.git, whose +# `main` is parked at OLD (-InstallRef, default: the newest +# release tag) -- so the install lands on OLD the same way a +# user's install landed on whatever main served that day. # UPDATE - OLD -> HEAD through the route selected by -Route: # desktop (implemented) launch the installed Hermes.exe # under Playwright's Electron driver and CLICK @@ -35,10 +35,12 @@ # GIT_CONFIG_GLOBAL. (NOT GIT_CONFIG_COUNT/KEY_n/VALUE_n env config -- # install.ps1 sets those itself and silently clobbers them.) The # installer's `git clone` and `hermes update`'s `git fetch origin` -# transparently hit OUR bare repo, whose `main` ref serves HEAD -- the -# update target. Installer and updater run byte-for-byte as shipped; -# everything else (uv, PyPI, npm, the installer's raw.githubusercontent -# install.ps1 download) uses the real network, same as a user install. +# transparently hit OUR bare repo. Its `main` serves OLD for the install +# phase; the update phase advances it to HEAD -- an update becomes +# available exactly the way it does for a real user. Installer and +# updater run byte-for-byte as shipped; everything else (uv, PyPI, npm, +# the installer's raw.githubusercontent install.ps1 download) uses the +# real network, same as a user install. # # PROOF: screenshots at every renderer step (Playwright), full-desktop # screenshots around the installer/AHK phases, a rolling desktop capture @@ -76,6 +78,15 @@ param( [ValidateSet("desktop", "update", "installer")] [string]$Route = "desktop", + # The OLD version: the ref served as `main` while the installer runs, + # i.e. what the user starts on. The published Hermes-Setup.exe carries + # no commit pin (Pin { commit: None, branch: "main" }) -- it installs + # whatever `main` points at, so staging OLD means serving it there. + # Empty = newest release tag in the checkout (the "user on the current + # release" starting point, same philosophy as the linux axis's tag + # matrix). + [string]$InstallRef = "", + # Repo checkout whose HEAD is the update target. [string]$RepoRoot = "", @@ -271,10 +282,10 @@ function Get-ManagedNode { } # ---------------------------------------------------------------------------- -# Phase: stage -- serve.git with `main` at HEAD (the update target) +# Phase: stage -- serve.git with `main` at OLD (advanced to HEAD by update-gui) # ---------------------------------------------------------------------------- function Invoke-PhaseStage { - Write-Step "STAGE: bare serve repo, main -> HEAD (update target)" + Write-Step "STAGE: bare serve repo, main -> OLD (install base)" if (Test-Path -LiteralPath $WorkRoot) { Remove-Item -LiteralPath $WorkRoot -Recurse -Force @@ -287,22 +298,35 @@ function Invoke-PhaseStage { $current = Invoke-Git @("-C", $RepoRoot, "rev-parse", "HEAD") Write-Host " HEAD (update target): $current" + # OLD: explicit -InstallRef, or the newest release tag -- the version a + # user who installed on release day is on. + $oldRef = $InstallRef + if (-not $oldRef) { + $oldRef = (Invoke-Git @("-C", $RepoRoot, "tag", "--list", "v*", "--sort=-creatordate") -split "`r?`n" | Select-Object -First 1) + if (-not $oldRef) { throw "no v* release tags in the checkout and no -InstallRef given -- cannot pick an OLD version" } + } + $old = Invoke-Git @("-C", $RepoRoot, "rev-parse", "$oldRef^{commit}") + Write-Host " OLD ($oldRef): $old" + Assert-True ($old -ne $current) "OLD differs from HEAD (an update is genuinely available)" + # Bare-clone the checkout: this is the repo the installer and updater # actually talk to. Local-path clone hardlinks objects, so it's fast - # even for full history. OLD is not staged by us -- it is whatever - # release pin the website installer carries (recorded in install-gui). + # even for full history. The published installer carries NO commit pin + # (Pin { commit: None, branch: main }) -- it installs whatever `main` + # serves, so staging OLD means parking `main` there; the update phase + # advances it to HEAD. Invoke-Git @("clone", "--bare", "--quiet", $RepoRoot, $ServeRepo) | Out-Null - Invoke-Git @("-C", $ServeRepo, "update-ref", "refs/heads/main", $current) | Out-Null + Invoke-Git @("-C", $ServeRepo, "update-ref", "refs/heads/main", $old) | Out-Null Invoke-Git @("-C", $ServeRepo, "symbolic-ref", "HEAD", "refs/heads/main") | Out-Null - # The website installer pins a specific release commit (-Commit ). - # That sha is in serve.git's history but not at a ref tip, so the - # redirected fetch needs any-SHA1 upload-pack permission (GitHub grants - # the equivalent for fetch of reachable commits). + # Belt-and-braces: SOME installer builds do bake a -Commit pin. A pinned + # sha is in serve.git's history but not at a ref tip, so the redirected + # fetch needs any-SHA1 upload-pack permission (GitHub grants the + # equivalent for fetch of reachable commits). Invoke-Git @("-C", $ServeRepo, "config", "uploadpack.allowAnySHA1InWant", "true") | Out-Null - Write-Host " serve.git: uploadpack.allowAnySHA1InWant=true (installer commit pin)" + Write-Host " serve.git: uploadpack.allowAnySHA1InWant=true (installer commit pin, if any)" - @{ current = $current } | + @{ old = $old; old_ref = $oldRef; current = $current } | ConvertTo-Json | Set-Content -LiteralPath $StatePath -Encoding UTF8 Write-Host " state written: $StatePath" New-Item -ItemType Directory -Path $ProofRoot -Force | Out-Null @@ -399,26 +423,13 @@ function Invoke-PhaseInstallGui { # Close the freshly launched app (user quits after first look). Stop-HermesAppProcesses "post-install" - # The website exe installs its baked release pin -- that IS the OLD - # version. It must differ from HEAD (an update is genuinely available). - # We do NOT require it to be an ancestor of HEAD: on a real `push: main` - # run HEAD is main's tip and the release pin is behind it (ancestor), - # but on a feature branch HEAD has diverged from main, so the release - # pin legitimately isn't in its ancestry. The update leg resets the - # checkout to serve.git's main ref regardless, and asserts it lands on - # HEAD -- that is the real forward-update proof. + # The installer cloned serve.git's `main`, which stage parked at OLD. + # (A pinned installer build would land on its baked pin instead; either + # way the requirement is the same: not already on HEAD.) $installedSha = Get-InstalledHead - Write-Host " installer landed on: $installedSha (website release pin = OLD)" - Assert-True ($installedSha -ne $state.current) "installed pin differs from HEAD (an update is genuinely available)" - $prevEap = $ErrorActionPreference; $ErrorActionPreference = "Continue" - & git -C $InstallDir merge-base --is-ancestor $installedSha $state.current 2>&1 | Out-Null - $isAncestor = ($LASTEXITCODE -eq 0) - $ErrorActionPreference = $prevEap - if ($isAncestor) { - Write-Host " [ok] installed pin is an ancestor of HEAD (linear old -> new)" - } else { - Write-Host " [note] installed pin is NOT an ancestor of HEAD -- expected on a diverged feature branch; the update leg still resets to HEAD" - } + Write-Host " installer landed on: $installedSha (OLD = $($state.old) [$($state.old_ref)])" + Assert-True ($installedSha -eq $state.old) "installed checkout is at OLD ($($state.old_ref))" + Assert-True ($installedSha -ne $state.current) "installed checkout differs from HEAD (an update is genuinely available)" Test-HermesRuns "post-install-gui" Assert-True ($null -ne (Get-DesktopExe)) "packaged Desktop Hermes.exe exists" @@ -448,12 +459,17 @@ function Invoke-PhaseInstallGui { # Phase: update-gui -- OLD -> HEAD through the selected route # ---------------------------------------------------------------------------- function Invoke-GuiUpdateDesktopRoute([string]$TargetSha) { - Write-Step "UPDATE (GUI, route=desktop): click Update now, land on $TargetSha" + Write-Step "UPDATE (GUI, route=desktop): advance served main -> $TargetSha, click Update now" $proof = Join-Path $ProofRoot "update-gui" New-Item -ItemType Directory -Path $proof -Force | Out-Null $env:HERMES_HOME = $HermesHome + # The update becomes available the way it does for a real user: the + # remote's main moves forward. (Install ran against main = OLD.) + Invoke-Git @("-C", $ServeRepo, "update-ref", "refs/heads/main", $TargetSha) | Out-Null + Write-Host " serve.git main advanced to $TargetSha" + $desktopExe = Get-DesktopExe Assert-True ($null -ne $desktopExe) "packaged Hermes.exe present before update" From e8bdb6933e610948eeaa7a7a5059eedbf06ef944 Mon Sep 17 00:00:00 2001 From: ethernet Date: Tue, 11 Aug 2026 14:01:22 -0400 Subject: [PATCH 038/227] fix(windows-e2e): 'auto' sentinel for install-ref - powershell -File eats empty-string args Run 31520267702 died in 3s: 'Missing an argument for parameter InstallRef'. powershell.exe -File drops a "" argument from the command line entirely, so the parameter binder saw -InstallRef followed by -SetupExeUrl. Default both the workflow input and the script parameter to 'auto' (= newest release tag) instead of empty. --- .github/workflows/install-e2e-windows-run.yml | 4 ++-- tests/install/windows-desktop-gui-e2e.ps1 | 12 +++++++----- 2 files changed, 9 insertions(+), 7 deletions(-) diff --git a/.github/workflows/install-e2e-windows-run.yml b/.github/workflows/install-e2e-windows-run.yml index 8a8ac84790a1..7816c14270c3 100644 --- a/.github/workflows/install-e2e-windows-run.yml +++ b/.github/workflows/install-e2e-windows-run.yml @@ -41,10 +41,10 @@ on: type: string default: desktop install-ref: - description: 'Ref to install as OLD (served as main while the installer runs). Empty = the newest release tag in the checkout.' + description: 'Ref to install as OLD (served as main while the installer runs). auto = the newest release tag in the checkout.' required: false type: string - default: '' + default: auto setup-exe-url: description: 'Bootstrap installer to install OLD with. Default: the latest published one — what a user downloads today.' required: false diff --git a/tests/install/windows-desktop-gui-e2e.ps1 b/tests/install/windows-desktop-gui-e2e.ps1 index ef4b5ee1f696..e5f950233a01 100644 --- a/tests/install/windows-desktop-gui-e2e.ps1 +++ b/tests/install/windows-desktop-gui-e2e.ps1 @@ -82,10 +82,12 @@ param( # i.e. what the user starts on. The published Hermes-Setup.exe carries # no commit pin (Pin { commit: None, branch: "main" }) -- it installs # whatever `main` points at, so staging OLD means serving it there. - # Empty = newest release tag in the checkout (the "user on the current - # release" starting point, same philosophy as the linux axis's tag - # matrix). - [string]$InstallRef = "", + # Empty or "auto" = newest release tag in the checkout (the "user on + # the current release" starting point, same philosophy as the linux + # axis's tag matrix). "auto" exists because `powershell -File` silently + # swallows an empty-string argument ('Missing an argument for + # parameter'), so the workflow cannot pass "". + [string]$InstallRef = "auto", # Repo checkout whose HEAD is the update target. [string]$RepoRoot = "", @@ -301,7 +303,7 @@ function Invoke-PhaseStage { # OLD: explicit -InstallRef, or the newest release tag -- the version a # user who installed on release day is on. $oldRef = $InstallRef - if (-not $oldRef) { + if (-not $oldRef -or $oldRef -eq "auto") { $oldRef = (Invoke-Git @("-C", $RepoRoot, "tag", "--list", "v*", "--sort=-creatordate") -split "`r?`n" | Select-Object -First 1) if (-not $oldRef) { throw "no v* release tags in the checkout and no -InstallRef given -- cannot pick an OLD version" } } From 71281c27e5af505117295896ba780315d94e90e1 Mon Sep 17 00:00:00 2001 From: ethernet Date: Tue, 11 Aug 2026 14:05:18 -0400 Subject: [PATCH 039/227] fix(windows-e2e): parenthesize the tag-list split - -split bound as a git argument Run 31520559900: stage passed the whole multiline tag list to rev-parse ('Filename too long'). In (Invoke-Git @(...) -split pattern | ...) PowerShell parses -split as ANOTHER ARGUMENT to Invoke-Git, not as an operator on its result - the function got '-split' and the regex appended to its array and rev-parse received every tag at once. Split via an intermediate variable instead. Verified locally: pwsh picks [v0.20.2] and resolves it to a single commit. --- tests/install/windows-desktop-gui-e2e.ps1 | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-) diff --git a/tests/install/windows-desktop-gui-e2e.ps1 b/tests/install/windows-desktop-gui-e2e.ps1 index e5f950233a01..d1fd0bd07478 100644 --- a/tests/install/windows-desktop-gui-e2e.ps1 +++ b/tests/install/windows-desktop-gui-e2e.ps1 @@ -304,7 +304,10 @@ function Invoke-PhaseStage { # user who installed on release day is on. $oldRef = $InstallRef if (-not $oldRef -or $oldRef -eq "auto") { - $oldRef = (Invoke-Git @("-C", $RepoRoot, "tag", "--list", "v*", "--sort=-creatordate") -split "`r?`n" | Select-Object -First 1) + # Parens matter: without them PowerShell binds -split as an + # argument to Invoke-Git instead of an operator on its result. + $tagList = Invoke-Git @("-C", $RepoRoot, "tag", "--list", "v*", "--sort=-creatordate") + $oldRef = ($tagList -split "\r?\n" | Select-Object -First 1) if (-not $oldRef) { throw "no v* release tags in the checkout and no -InstallRef given -- cannot pick an OLD version" } } $old = Invoke-Git @("-C", $RepoRoot, "rev-parse", "$oldRef^{commit}") From 005fc5a6eb76d98a0e45a9514770902571ea3017 Mon Sep 17 00:00:00 2001 From: ethernet Date: Tue, 11 Aug 2026 14:37:31 -0400 Subject: [PATCH 040/227] ci(windows-e2e): ffmpeg screen recording across both GUI phases Bring back the continuous recording the retired AHK-only driver had, alongside the 3s frame captures: gdigrab 15fps to proof// recording.mkv, started before the installer/app launches and q-stopped in the finally block win or lose. mkv stays playable when the process dies unfinalized; skip gracefully when ffmpeg is not on PATH (it ships on windows-latest). Lifecycle (start, null-skip, graceful q stop, clean exit, playable output) verified locally with a lavfi source. --- tests/install/windows-desktop-gui-e2e.ps1 | 43 +++++++++++++++++++++-- 1 file changed, 41 insertions(+), 2 deletions(-) diff --git a/tests/install/windows-desktop-gui-e2e.ps1 b/tests/install/windows-desktop-gui-e2e.ps1 index d1fd0bd07478..06e87f0a1155 100644 --- a/tests/install/windows-desktop-gui-e2e.ps1 +++ b/tests/install/windows-desktop-gui-e2e.ps1 @@ -44,8 +44,8 @@ # # PROOF: screenshots at every renderer step (Playwright), full-desktop # screenshots around the installer/AHK phases, a rolling desktop capture -# (every 3s) for the whole run, ahk.log, and the hand-off log. All uploaded -# as CI artifacts. +# (every 3s) plus a continuous ffmpeg screen recording (recording.mkv) for +# both phases, ahk.log, and the hand-off log. All uploaded as CI artifacts. # # DEVIATIONS FROM PRODUCTION (each one deliberate and small): # * the git URL redirect itself -- the staging requirement. @@ -218,6 +218,41 @@ function Save-DesktopScreenshot([string]$OutFile) { } } +function Start-ScreenRecording([string]$OutFile) { + # Continuous ffmpeg screen capture (gdigrab, 15fps). ffmpeg ships on the + # windows-latest runner image; skip gracefully elsewhere. mkv on purpose: + # it stays playable even if the process dies without finalizing. + # + # ffmpeg must be started, fed, and stopped from THIS process: the + # graceful stop is the character 'q' on its LIVE stdin pipe, which only + # System.Diagnostics.Process exposes (Start-Process + # -RedirectStandardInput hands it a file handle already at EOF). + if (-not (Get-Command ffmpeg -ErrorAction SilentlyContinue)) { + Write-Host " (ffmpeg not on PATH; skipping screen recording)" + return $null + } + $psi = New-Object System.Diagnostics.ProcessStartInfo + $psi.FileName = "ffmpeg" + $psi.Arguments = "-y -f gdigrab -framerate 15 -i desktop " + + "-hide_banner -loglevel error " + + "-c:v libx264 -preset ultrafast -pix_fmt yuv420p `"$OutFile`"" + $psi.RedirectStandardInput = $true + $psi.UseShellExecute = $false + $proc = [System.Diagnostics.Process]::Start($psi) + Write-Host " screen recording started (pid $($proc.Id)) -> $OutFile" + return $proc +} + +function Stop-ScreenRecording($proc) { + if ($proc -and -not $proc.HasExited) { + try { + $proc.StandardInput.Write("q") + $proc.StandardInput.Close() + } catch {} + if (-not $proc.WaitForExit(15000)) { try { $proc.Kill() } catch {} } + } +} + function Start-DesktopRecorder([string]$OutDir) { # Rolling desktop capture: one PNG every 3s from a detached PowerShell, # capped at 800 frames (~40 min). Proof that survives any step failure. @@ -377,6 +412,7 @@ function Invoke-PhaseInstallGui { New-Item -ItemType Directory -Path $HermesHome -Force | Out-Null $recorder = Start-DesktopRecorder (Join-Path $proof "desktop-frames") + $recording = Start-ScreenRecording (Join-Path $proof "recording.mkv") $ahkLog = Join-Path $proof "ahk.log" try { Save-DesktopScreenshot (Join-Path $proof "00-before-installer.png") @@ -415,6 +451,7 @@ function Invoke-PhaseInstallGui { Assert-True $installer.HasExited "Hermes-Setup.exe exited after Launch" } finally { + Stop-ScreenRecording $recording Stop-DesktopRecorder $recorder (Join-Path $proof "desktop-frames") # Surface the installer's own log win or lose. $bootLog = Join-Path $HermesHome "logs\bootstrap-installer.log" @@ -495,6 +532,7 @@ function Invoke-GuiUpdateDesktopRoute([string]$TargetSha) { Assert-True $pwResolved "@playwright/test resolvable from installed apps/desktop" $recorder = Start-DesktopRecorder (Join-Path $proof "desktop-frames") + $recording = Start-ScreenRecording (Join-Path $proof "recording.mkv") try { # Launch the installed app and click through Settings -> About -> # Update now. Exit 0 = the app quit for the updater hand-off. @@ -603,6 +641,7 @@ function Invoke-GuiUpdateDesktopRoute([string]$TargetSha) { Save-DesktopScreenshot (Join-Path $proof "99-relaunched-desktop.png") } finally { + Stop-ScreenRecording $recording Stop-DesktopRecorder $recorder (Join-Path $proof "desktop-frames") $handoffLog = Join-Path $HermesHome "logs\desktop-update-handoff.log" if (Test-Path -LiteralPath $handoffLog) { From 77c3a398f2f687f9031f40b03c726ff646e92471 Mon Sep 17 00:00:00 2001 From: ethernet Date: Tue, 11 Aug 2026 14:52:46 -0400 Subject: [PATCH 041/227] ci(windows-e2e): install ffmpeg - not preinstalled on windows-latest, recording silently skipped Run 31523695265 went green but both phases logged '(ffmpeg not on PATH; skipping screen recording)' and the artifact had no recording.mkv. Restore the winget install + cache + PATH steps from the retired axis (same pinned actions/cache SHA those green runs used), scoped to just ffmpeg since AutoHotkey now comes from the portable zip inside the driver. --- .github/workflows/install-e2e-windows-run.yml | 26 +++++++++++++++++++ 1 file changed, 26 insertions(+) diff --git a/.github/workflows/install-e2e-windows-run.yml b/.github/workflows/install-e2e-windows-run.yml index 7816c14270c3..ad7dfbb27cc4 100644 --- a/.github/workflows/install-e2e-windows-run.yml +++ b/.github/workflows/install-e2e-windows-run.yml @@ -82,6 +82,32 @@ jobs: with: fetch-depth: 0 + # ffmpeg records the screen for the whole run so a failed click is + # diagnosable from the artifact instead of by guesswork. NOT + # preinstalled on windows-latest; the driver skips recording + # gracefully when it's absent, so this step is what makes the + # recording actually happen. Cached — winget's ffmpeg download is + # the slow part. + - name: Restore cached ffmpeg + id: ffmpeg-cache + uses: actions/cache@27d5ce7f107fe9357f9df03efb73ab90386fccae # v5 + with: + path: ${{ runner.temp }}\test-bins\ffmpeg + key: e2e-ffmpeg-${{ runner.os }}-v1 + + - name: Install ffmpeg + if: steps.ffmpeg-cache.outputs.cache-hit != 'true' + shell: pwsh + run: | + $bins = "$env:RUNNER_TEMP\test-bins\ffmpeg" + New-Item -ItemType Directory -Path $bins -Force | Out-Null + winget install -e --id Gyan.FFmpeg --silent --accept-source-agreements --accept-package-agreements --disable-interactivity --location "$env:RUNNER_TEMP\ffmpeg_dir" + Copy-Item -Path "$env:RUNNER_TEMP\ffmpeg_dir\*\*" -Destination $bins -Recurse -Force + + - name: Add ffmpeg to PATH + shell: pwsh + run: Add-Content -Path $env:GITHUB_PATH -Value "$env:RUNNER_TEMP\test-bins\ffmpeg\bin" + - name: Stage serve repo (main -> OLD) shell: powershell run: powershell -NoProfile -ExecutionPolicy Bypass -File tests\install\windows-desktop-gui-e2e.ps1 -Phase stage -Route ${{ inputs.route }} -InstallRef "${{ inputs.install-ref }}" -SetupExeUrl ${{ inputs.setup-exe-url }} From 1bf0a1c95be2df7d4bb66d8f8489b8503fbd9daa Mon Sep 17 00:00:00 2001 From: ethernet Date: Tue, 11 Aug 2026 15:21:24 -0400 Subject: [PATCH 042/227] ci(install-e2e): generate every install/update combination - one job per combo Replace the hand-enumerated update/installer/windows-desktop jobs with a support-matrix generator (scripts/sandbox/generate-e2e-matrix.mjs). The spec declares every {os, install-method, update-method} combination a user could be on; generate-matrix expands it and fans out ONE JOB PER COMBINATION: * linux combos (curl-bash install x hermes-update/curl-bash rerun) drive install-e2e-run.yml, still multiplied by the sampled release tags from pick-releases; * the windows combo (desktop-installer@latest -> desktop-app) drives install-e2e-windows-run.yml - the real GUI flow; * every declared-but-unimplemented combo (all of macOS, the remaining windows methods) becomes its own visible skipped job, so the coverage gap is enumerable from the Checks tab and implementing one is a one-line move into IMPLEMENTED. Strictness carried into the generator: method ids validate against a closed set, installer 'versions' arrays only allow 'latest' until a versioned archive exists, secondUpdate must stay empty until a chained second-update leg is implemented, and unknown spec keys/routes throw. Expansion, route filtering, empty-matrix gating, and all seven error paths verified locally; the dispatch route choice keeps its exact previous semantics (all/both/update/installer/windows-desktop). --- .github/workflows/install-e2e.yml | 133 ++++++++----- scripts/sandbox/generate-e2e-matrix.mjs | 238 ++++++++++++++++++++++++ 2 files changed, 326 insertions(+), 45 deletions(-) create mode 100644 scripts/sandbox/generate-e2e-matrix.mjs diff --git a/.github/workflows/install-e2e.yml b/.github/workflows/install-e2e.yml index 08d46582d018..29d169a0057a 100644 --- a/.github/workflows/install-e2e.yml +++ b/.github/workflows/install-e2e.yml @@ -2,22 +2,28 @@ name: Install & Update E2E # Can a user on a released version get to this commit? # -# For each release we sample, a leg installs that release through the real -# `curl | install.sh` one-liner (uv, a managed Python, Node, the venv) inside -# scripts/dev-sandbox.sh, then applies one update route and requires the -# checkout to land on this commit with a working `hermes`. +# The support matrix -- every {os, install-method, update-method} combination +# a user could be on -- lives in scripts/sandbox/generate-e2e-matrix.mjs. +# The generate-matrix job expands it into ONE JOB PER COMBINATION: # -# The starting versions are chosen at runtime from the repo's release tags -# (scripts/sandbox/pick-release-tags.sh): newest, oldest, and a spread between. -# A hardcoded list would stop covering the newest release the day after it -# ships, and would pin an "oldest" that nobody still runs. +# * linux legs install a sampled release through the real +# `curl | install.sh` one-liner (uv, a managed Python, Node, the venv) +# inside scripts/dev-sandbox.sh, then apply the combo's update route and +# require the checkout to land on this commit with a working `hermes`; +# * windows legs run the real desktop user flow (install-e2e-windows-run +# .yml): the website's Hermes-Setup.exe clicked by AutoHotkey, then the +# update driven through the app itself, Playwright clicking "Update now"; +# * combinations the spec declares but CI can't drive yet (macOS, the +# remaining windows install/update methods) surface as skipped jobs so +# the coverage gap is enumerable instead of buried in comments. # -# A separate Windows axis (install-e2e-windows-run.yml) covers the desktop -# user flow: install OLD with the website's published bootstrap installer -# (headed, AutoHotkey-clicked), then update to HEAD through a real update -# surface — the app's own Update button clicked by Playwright. No tag -# matrix there: the published exe's own build pin is the starting point -# users actually have. +# The linux starting versions are chosen at runtime from the repo's release +# tags (scripts/sandbox/pick-release-tags.sh): newest, oldest, and a spread +# between. A hardcoded list would stop covering the newest release the day +# after it ships, and would pin an "oldest" that nobody still runs. The +# windows starting version is the newest release tag (the published +# installer has no commit pin -- it installs what `main` serves, which the +# harness stages to that tag). # # Triggers: # * every 12 hours, so upstream drift (a new uv, a Node bump, a PyPI change) @@ -62,7 +68,7 @@ concurrency: jobs: # Which released versions do we test updating FROM? Resolved once and shared - # by both route matrices, so the two routes cover the same set. + # by every linux leg, so all routes cover the same set. pick-releases: name: Pick release tags runs-on: ubuntu-latest @@ -86,44 +92,81 @@ jobs: echo "Testing updates from: $tags" echo "tags=$tags" >> "$GITHUB_OUTPUT" - # `hermes update` -- the route most users take. - update: - if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "both", "update"]'), inputs.route) + # Expand the {os, install-method, update-method} support matrix + # (scripts/sandbox/generate-e2e-matrix.mjs) into one job per combination. + # Implemented combos land in the linux/windows matrices below; everything + # the spec declares but CI can't drive yet lands in `skipped`, so the TODO + # surface is visible in every run instead of buried in comments. + generate-matrix: + name: Generate combination matrix needs: pick-releases + runs-on: ubuntu-latest + timeout-minutes: 5 + outputs: + linux: ${{ steps.gen.outputs.linux }} + windows: ${{ steps.gen.outputs.windows }} + skipped: ${{ steps.gen.outputs.skipped }} + steps: + - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2 + with: + sparse-checkout: scripts/sandbox/generate-e2e-matrix.mjs + sparse-checkout-cone-mode: false + - id: gen + run: | + set -euo pipefail + matrices="$(node scripts/sandbox/generate-e2e-matrix.mjs \ + --tags '${{ needs.pick-releases.outputs.tags }}' \ + --route '${{ inputs.route || 'all' }}')" + echo "$matrices" + for key in linux windows skipped; do + echo "$key=$(echo "$matrices" | node -e 'let d="";process.stdin.on("data",c=>d+=c).on("end",()=>console.log(JSON.stringify(JSON.parse(d)[process.argv[1]])))' "$key")" >> "$GITHUB_OUTPUT" + done + + # Linux sandbox legs: one job per {install-method -> update-method, release + # tag} combination, each a real curl|bash install in bubblewrap followed by + # the combo's update route. + linux: + name: "linux: ${{ matrix.name }}" + if: fromJSON(needs.generate-matrix.outputs.linux).include[0] != null + needs: generate-matrix strategy: - # One release breaking is worth knowing about even if another already + # One combo breaking is worth knowing about even if another already # failed, so let every leg report. fail-fast: false - matrix: - install-ref: ${{ fromJSON(needs.pick-releases.outputs.tags) }} + max-parallel: 4 + matrix: ${{ fromJSON(needs.generate-matrix.outputs.linux) }} uses: ./.github/workflows/install-e2e-run.yml with: - route: update - install-ref: ${{ matrix.install-ref }} + route: ${{ matrix.route }} + install-ref: ${{ matrix.install_ref }} - # Re-running the curl one-liner over an existing checkout: autostash + pull - # rather than the updater's own git handling. - installer: - if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "both", "installer"]'), inputs.route) - needs: pick-releases + # Windows GUI legs: one job per combination of the real desktop user flow + # (website installer clicked by AutoHotkey, update driven through the app). + windows: + name: "windows: ${{ matrix.name }}" + if: fromJSON(needs.generate-matrix.outputs.windows).include[0] != null + needs: generate-matrix strategy: fail-fast: false - max-parallel: 3 - matrix: - install-ref: ${{ fromJSON(needs.pick-releases.outputs.tags) }} - uses: ./.github/workflows/install-e2e-run.yml - with: - route: installer - install-ref: ${{ matrix.install-ref }} - - # Windows desktop: install OLD with the website's published bootstrap - # installer (the exact bits a user downloads today, AutoHotkey-clicked), - # then update to HEAD by clicking the app's own Update button under - # Playwright. The reusable workflow's route input selects the update - # mechanism; `hermes update` and installer-rerun arms are declared TODOs - # there. - windows-desktop: - if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "windows-desktop"]'), inputs.route) + matrix: ${{ fromJSON(needs.generate-matrix.outputs.windows) }} uses: ./.github/workflows/install-e2e-windows-run.yml with: - route: desktop + route: ${{ matrix.route }} + install-ref: ${{ matrix.install_ref }} + + # Declared-but-unimplemented combinations. Each shows up as its own skipped + # check so coverage gaps are enumerable from the Checks tab; implementing + # one moves it from here into IMPLEMENTED in the generator. + skipped: + name: "skipped: ${{ matrix.os }}: ${{ matrix.install }} -> ${{ matrix.update }}" + if: fromJSON(needs.generate-matrix.outputs.skipped).include[0] != null + needs: generate-matrix + strategy: + fail-fast: false + matrix: ${{ fromJSON(needs.generate-matrix.outputs.skipped) }} + runs-on: ubuntu-latest + timeout-minutes: 2 + steps: + - run: >- + echo "SKIPPED (${{ matrix.reason }}) -- ${{ matrix.os }} install via + '${{ matrix.install }}', update via '${{ matrix.update }}'" diff --git a/scripts/sandbox/generate-e2e-matrix.mjs b/scripts/sandbox/generate-e2e-matrix.mjs new file mode 100644 index 000000000000..21b257c5e37b --- /dev/null +++ b/scripts/sandbox/generate-e2e-matrix.mjs @@ -0,0 +1,238 @@ +#!/usr/bin/env node +/** + * Expand the install/update support matrix into concrete E2E combinations. + * + * One source of truth for every {os, install-method, update-method} pair a + * user could be on. Implemented combos fan out into real CI jobs -- one job + * per combination -- and everything else lands in a "skipped" matrix so the + * TODO surface stays visible in every run instead of buried in comments. + * + * Used by .github/workflows/install-e2e.yml: + * + * node scripts/sandbox/generate-e2e-matrix.mjs \ + * --tags '["v2026.8.3","v2026.3.12"]' --route all + * + * Prints JSON: { linux: {include:[...]}, windows: {include:[...]}, + * skipped: {include:[...]} }. The linux include entries carry + * {name, route, install_ref} for install-e2e-run.yml; windows entries carry + * {name, route, install_ref} for install-e2e-windows-run.yml. + */ + +import path from 'node:path'; +import { parseArgs } from 'node:util'; +import { fileURLToPath } from 'node:url'; + +/** + * Method ids are strict machine strings -- workflows and the IMPLEMENTED + * table key off them, so an unknown id must fail loudly (see KNOWN_METHODS). + * + * `versions` on an entry expands it into one combination per version. Only + * "latest" is allowed today: it means "the artifact published on the website + * right now" (Hermes-Setup.exe has no versioned archive yet). When archived + * installer versions exist, widen ALLOWED_VERSIONS. + */ +export const SPEC = { + windows: { + install: [ + // irm https://hermes.nousresearch.com/install.ps1 | iex + { method: 'irm-iex' }, + // Website Hermes-Setup.exe, clicked through the GUI. + { method: 'desktop-installer', versions: ['latest'] }, + ], + update: [ + { method: 'irm-iex' }, + // Run the bootstrap exe again over an existing install (--update flow). + { method: 'desktop-installer-rerun', versions: ['latest'] }, + { method: 'hermes-update' }, + // Settings -> About -> "Update now" inside the running desktop app. + { method: 'desktop-app' }, + ], + }, + macos: { + install: [ + { method: 'curl-bash' }, + { method: 'packaged-app' }, + ], + update: [ + { method: 'curl-bash' }, + { method: 'hermes-update' }, + { method: 'app-update' }, + ], + }, + linux: { + install: [ + { method: 'curl-bash' }, + ], + update: [ + { method: 'curl-bash' }, + { method: 'hermes-update' }, + ], + }, +}; + +const KNOWN_METHODS = new Set([ + 'irm-iex', + 'desktop-installer', + 'desktop-installer-rerun', + 'hermes-update', + 'desktop-app', + 'curl-bash', + 'packaged-app', + 'app-update', +]); + +const ALLOWED_VERSIONS = new Set(['latest']); + +/** + * The combinations CI actually runs today, and which reusable workflow + + * route input drives each. Everything in SPEC but not here is emitted as a + * skipped combo. Install ids include the expanded @version suffix. + */ +export const IMPLEMENTED = [ + // scripts/dev-sandbox.sh legs (install-e2e-run.yml, bubblewrap sandbox). + { os: 'linux', install: 'curl-bash', update: 'hermes-update', workflow: 'linux', route: 'update' }, + { os: 'linux', install: 'curl-bash', update: 'curl-bash', workflow: 'linux', route: 'installer' }, + // Real Windows GUI flow (install-e2e-windows-run.yml): website exe clicked + // by AutoHotkey, Update now clicked by Playwright. + { os: 'windows', install: 'desktop-installer@latest', update: 'desktop-app', workflow: 'windows', route: 'desktop' }, +]; + +function validateEntry(os, kind, entry) { + if (typeof entry.method !== 'string' || !KNOWN_METHODS.has(entry.method)) { + throw new Error(`${os}.${kind}: unknown method id ${JSON.stringify(entry.method)} -- add it to KNOWN_METHODS if intentional`); + } + if ('versions' in entry) { + if (!Array.isArray(entry.versions) || entry.versions.length === 0) { + throw new Error(`${os}.${kind}.${entry.method}: versions must be a non-empty array`); + } + for (const v of entry.versions) { + if (!ALLOWED_VERSIONS.has(v)) { + throw new Error(`${os}.${kind}.${entry.method}: version ${JSON.stringify(v)} not allowed -- only ${[...ALLOWED_VERSIONS].join(', ')} until versioned installer archives exist`); + } + } + } + const unknown = Object.keys(entry).filter((k) => k !== 'method' && k !== 'versions'); + if (unknown.length) { + throw new Error(`${os}.${kind}.${entry.method}: unknown keys ${unknown.join(', ')}`); + } +} + +/** Expand one method entry into concrete ids ("desktop-installer@latest"). */ +export function expandMethod(os, kind, entry) { + validateEntry(os, kind, entry); + if (!entry.versions) return [entry.method]; + return entry.versions.map((v) => `${entry.method}@${v}`); +} + +/** Every {os, install, update, secondUpdate} combination in SPEC. */ +export function generateEnvironments(spec) { + const envs = []; + for (const [os, osSpec] of Object.entries(spec)) { + const { install, update, secondUpdate = [], ...unknown } = osSpec; + if (Object.keys(unknown).length) { + throw new Error(`${os}: unknown spec keys ${Object.keys(unknown).join(', ')}`); + } + // Chained second updates (install -> update -> update again) are a real + // axis -- the updater that RESULTS from an update must itself update -- + // but nothing implements them yet. Refuse a spec that declares them so + // the first implementation is forced to come through here. + if (!Array.isArray(secondUpdate) || secondUpdate.length !== 0) { + throw new Error(`${os}: secondUpdate must be empty until a second-update leg is implemented`); + } + const installs = install.flatMap((e) => expandMethod(os, 'install', e)); + const updates = update.flatMap((e) => expandMethod(os, 'update', e)); + for (const i of installs) { + for (const u of updates) { + envs.push({ os, install: i, update: u, secondUpdate: '' }); + } + } + } + return envs; +} + +function findImplementation(env) { + return IMPLEMENTED.find( + (m) => m.os === env.os && m.install === env.install && m.update === env.update, + ); +} + +/** Mirror of install-e2e.yml's dispatch `route` choice. */ +function routeWants(route, env, impl) { + switch (route) { + case 'all': + return true; + case 'both': + return env.os === 'linux' && Boolean(impl); + case 'update': + return impl?.route === 'update'; + case 'installer': + return impl?.route === 'installer'; + case 'windows-desktop': + return env.os === 'windows' && Boolean(impl); + default: + throw new Error(`unknown route filter: ${JSON.stringify(route)}`); + } +} + +/** + * Split the combinations into per-workflow matrices. + * Linux combos fan out further over `tags` (which released version the leg + * installs first); Windows install versions come from the method id instead + * (desktop-installer@latest = whatever the website serves). + */ +export function buildMatrices(envs, { tags = [], route = 'all' } = {}) { + const linux = []; + const windows = []; + const skipped = []; + for (const env of envs) { + const impl = findImplementation(env); + if (!routeWants(route, env, impl)) continue; + if (!impl) { + skipped.push({ ...env, reason: 'not implemented yet' }); + } else if (impl.workflow === 'linux') { + if (tags.length === 0) { + throw new Error(`linux combo ${env.install} -> ${env.update} selected but no --tags given`); + } + for (const tag of tags) { + linux.push({ + name: `${env.install} @ ${tag} -> ${env.update}`, + route: impl.route, + install_ref: tag, + }); + } + } else if (impl.workflow === 'windows') { + windows.push({ + name: `${env.install} -> ${env.update}`, + route: impl.route, + install_ref: 'auto', + }); + } else { + throw new Error(`unknown workflow ${JSON.stringify(impl.workflow)} for ${env.os}`); + } + } + return { + linux: { include: linux }, + windows: { include: windows }, + skipped: { include: skipped }, + }; +} + +function main() { + const { values } = parseArgs({ + options: { + tags: { type: 'string', default: '[]' }, + route: { type: 'string', default: 'all' }, + }, + }); + const tags = JSON.parse(values.tags); + if (!Array.isArray(tags) || !tags.every((t) => typeof t === 'string')) { + throw new Error('--tags must be a JSON array of strings'); + } + const envs = generateEnvironments(SPEC); + const matrices = buildMatrices(envs, { tags, route: values.route }); + process.stdout.write(`${JSON.stringify(matrices, null, 2)}\n`); +} + +if (process.argv[1] && fileURLToPath(import.meta.url) === path.resolve(process.argv[1])) { + main(); +} From 5a31a14f953acf3966830df9ab92e8ee3d106f11 Mon Sep 17 00:00:00 2001 From: ethernet Date: Tue, 11 Aug 2026 15:49:47 -0400 Subject: [PATCH 043/227] ci(install-e2e): native per-combo skips via a reusable skip workflow Unimplemented combos previously ran as green echo jobs. Make each one a real GitHub skip (grey, conclusion=skipped, no runner spent) while keeping one check per combination: matrix context is not available in job-level if, so the caller cannot natively skip individual legs - instead each leg calls install-e2e-skip.yml, whose inner job is gated on an 'implemented' input that defaults to false and is never passed. Implementing a combo stays a generator-side move into IMPLEMENTED. --- .github/workflows/install-e2e-skip.yml | 33 ++++++++++++++++++++++++++ .github/workflows/install-e2e.yml | 23 +++++++++--------- 2 files changed, 44 insertions(+), 12 deletions(-) create mode 100644 .github/workflows/install-e2e-skip.yml diff --git a/.github/workflows/install-e2e-skip.yml b/.github/workflows/install-e2e-skip.yml new file mode 100644 index 000000000000..7b355b8816b7 --- /dev/null +++ b/.github/workflows/install-e2e-skip.yml @@ -0,0 +1,33 @@ +name: Install & Update E2E — unsupported combo (reusable) + +# A declared-but-unimplemented {os, install-method, update-method} +# combination from scripts/sandbox/generate-e2e-matrix.mjs. The caller fans +# out one of these per unsupported combo so every coverage gap is its own +# named, NATIVELY SKIPPED check (grey, conclusion=skipped, no runner) in the +# Checks tab. +# +# `implemented` stays at its false default until the combo has a real +# driver; flipping a combo to implemented happens in the generator (move it +# into IMPLEMENTED and point it at a real reusable workflow), never by +# setting this input. + +on: + workflow_call: + inputs: + implemented: + description: 'Never set this. Combos become implemented by moving into IMPLEMENTED in generate-e2e-matrix.mjs, not by flipping the flag.' + required: false + type: boolean + default: false + +permissions: + contents: read + +jobs: + todo: + name: not implemented yet + if: inputs.implemented + runs-on: ubuntu-latest + timeout-minutes: 1 + steps: + - run: 'true' diff --git a/.github/workflows/install-e2e.yml b/.github/workflows/install-e2e.yml index 29d169a0057a..7b0e1314677a 100644 --- a/.github/workflows/install-e2e.yml +++ b/.github/workflows/install-e2e.yml @@ -14,8 +14,9 @@ name: Install & Update E2E # .yml): the website's Hermes-Setup.exe clicked by AutoHotkey, then the # update driven through the app itself, Playwright clicking "Update now"; # * combinations the spec declares but CI can't drive yet (macOS, the -# remaining windows install/update methods) surface as skipped jobs so -# the coverage gap is enumerable instead of buried in comments. +# remaining windows install/update methods) fan out too, one NATIVELY +# SKIPPED check per combo, so every coverage gap is enumerable from the +# Checks tab. # # The linux starting versions are chosen at runtime from the repo's release # tags (scripts/sandbox/pick-release-tags.sh): newest, oldest, and a spread @@ -154,19 +155,17 @@ jobs: route: ${{ matrix.route }} install-ref: ${{ matrix.install_ref }} - # Declared-but-unimplemented combinations. Each shows up as its own skipped - # check so coverage gaps are enumerable from the Checks tab; implementing - # one moves it from here into IMPLEMENTED in the generator. + # Declared-but-unimplemented combinations: one job per combo, each a + # NATIVE skip (grey, conclusion=skipped, no runner). Matrix context is not + # available in job-level `if`, so the native skip lives in the called + # workflow instead: its inner job is gated on an input that defaults to + # false and is never passed. Implementing a combo moves it into + # IMPLEMENTED in the generator. skipped: - name: "skipped: ${{ matrix.os }}: ${{ matrix.install }} -> ${{ matrix.update }}" + name: "todo: ${{ matrix.os }}: ${{ matrix.install }} -> ${{ matrix.update }}" if: fromJSON(needs.generate-matrix.outputs.skipped).include[0] != null needs: generate-matrix strategy: fail-fast: false matrix: ${{ fromJSON(needs.generate-matrix.outputs.skipped) }} - runs-on: ubuntu-latest - timeout-minutes: 2 - steps: - - run: >- - echo "SKIPPED (${{ matrix.reason }}) -- ${{ matrix.os }} install via - '${{ matrix.install }}', update via '${{ matrix.update }}'" + uses: ./.github/workflows/install-e2e-skip.yml From 83d009ae8293a1ebff6d315ee1f096d15a2177b8 Mon Sep 17 00:00:00 2001 From: ethernet Date: Tue, 11 Aug 2026 16:01:36 -0400 Subject: [PATCH 044/227] ci(install-e2e): nest the graph per starting tag; skips live where the knowledge lives Two structural changes to the combination fanout: 1. Tags become the OUTER axis, as a sub-graph per starting version: install-e2e.yml fans a plain matrix over the picked tags into a new per-tag reusable workflow (install-e2e-tag.yml), which runs the combination generator for that one tag and fans out one job per {os, install-method, update-method}. The Actions graph now reads 'from vX -> windows: install -> update' per leg. Nothing is hardcoded in the workflows: the tag workflow calls the generator itself. 2. Native skips move to the point that owns the capability knowledge: macOS combos (no driving workflow exists) grey out in the tag workflow via install-e2e-skip.yml, untouched by the tag axis; ALL windows combos dispatch to install-e2e-windows-run.yml, which takes install-method/update-method inputs and natively skips the pairs its driver cannot run yet - so implementing a windows method is a change in the run workflow + driver only. The driver's -Route ids now match the generator's method ids verbatim. Generator output shape, route filters, and all error paths re-verified locally; all four workflows pass actionlint; driver re-parses clean pure-ASCII. --- .github/workflows/install-e2e-skip.yml | 21 +-- .github/workflows/install-e2e-tag.yml | 100 +++++++++++++ .github/workflows/install-e2e-windows-run.yml | 55 ++++--- .github/workflows/install-e2e.yml | 136 ++++++------------ scripts/sandbox/generate-e2e-matrix.mjs | 118 +++++++++------ tests/install/windows-desktop-gui-e2e.ps1 | 32 +++-- 6 files changed, 281 insertions(+), 181 deletions(-) create mode 100644 .github/workflows/install-e2e-tag.yml diff --git a/.github/workflows/install-e2e-skip.yml b/.github/workflows/install-e2e-skip.yml index 7b355b8816b7..02415007403d 100644 --- a/.github/workflows/install-e2e-skip.yml +++ b/.github/workflows/install-e2e-skip.yml @@ -1,15 +1,18 @@ name: Install & Update E2E — unsupported combo (reusable) -# A declared-but-unimplemented {os, install-method, update-method} -# combination from scripts/sandbox/generate-e2e-matrix.mjs. The caller fans -# out one of these per unsupported combo so every coverage gap is its own -# named, NATIVELY SKIPPED check (grey, conclusion=skipped, no runner) in the -# Checks tab. +# A declared {os, install-method, update-method} combination whose OS has +# no driving E2E workflow at all (today: every macOS combo). The per-tag +# caller (install-e2e-tag.yml) fans out one of these per such combo so the +# coverage gap is its own named, NATIVELY SKIPPED check (grey, +# conclusion=skipped, no runner) in the Checks tab. # -# `implemented` stays at its false default until the combo has a real -# driver; flipping a combo to implemented happens in the generator (move it -# into IMPLEMENTED and point it at a real reusable workflow), never by -# setting this input. +# OSes WITH a driving workflow don't come here: their unimplemented method +# pairs natively skip inside that workflow, next to the driver that will +# implement them (see install-e2e-windows-run.yml). +# +# `implemented` stays at its false default until the OS has a real driving +# workflow; flipping happens by routing the combo to that workflow in +# generate-e2e-matrix.mjs, never by setting this input. on: workflow_call: diff --git a/.github/workflows/install-e2e-tag.yml b/.github/workflows/install-e2e-tag.yml new file mode 100644 index 000000000000..ee560197538e --- /dev/null +++ b/.github/workflows/install-e2e-tag.yml @@ -0,0 +1,100 @@ +name: Install & Update E2E — one starting tag (reusable) + +# Everything a user starting on ONE released version could do: for each +# {os, install-method, update-method} combination the support matrix +# (scripts/sandbox/generate-e2e-matrix.mjs) declares, one job that installs +# this tag and updates to HEAD. +# +# Skips happen where the knowledge lives: +# * macOS combos have no workflow at all -- they natively skip right +# here via install-e2e-skip.yml; +# * windows combos ALL dispatch to install-e2e-windows-run.yml, which +# natively skips the method pairs its driver cannot run yet; +# * linux combos are all implemented. +# +# Call it: +# +# jobs: +# tag: +# strategy: +# matrix: { install-ref: [v2026.8.3, v2026.3.12] } +# uses: ./.github/workflows/install-e2e-tag.yml +# with: +# install-ref: ${{ matrix.install-ref }} + +on: + workflow_call: + inputs: + install-ref: + description: 'The released version to start from: installed first, then updated to HEAD.' + required: true + type: string + route: + description: 'Combination filter, mirroring the dispatch route choice: all, both, update, installer, windows-desktop.' + required: false + type: string + default: all + +permissions: + contents: read + +jobs: + generate-matrix: + name: expand combinations + runs-on: ubuntu-latest + timeout-minutes: 5 + outputs: + linux: ${{ steps.gen.outputs.linux }} + windows: ${{ steps.gen.outputs.windows }} + skipped: ${{ steps.gen.outputs.skipped }} + steps: + - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2 + with: + sparse-checkout: scripts/sandbox/generate-e2e-matrix.mjs + sparse-checkout-cone-mode: false + - id: gen + run: | + set -euo pipefail + matrices="$(node scripts/sandbox/generate-e2e-matrix.mjs \ + --tags '["${{ inputs.install-ref }}"]' \ + --route '${{ inputs.route }}')" + echo "$matrices" + for key in linux windows skipped; do + echo "$key=$(echo "$matrices" | node -e 'let d="";process.stdin.on("data",c=>d+=c).on("end",()=>console.log(JSON.stringify(JSON.parse(d)[process.argv[1]])))' "$key")" >> "$GITHUB_OUTPUT" + done + + linux: + name: "linux: ${{ matrix.name }}" + if: fromJSON(needs.generate-matrix.outputs.linux).include[0] != null + needs: generate-matrix + strategy: + # One combo breaking is worth knowing about even if another already + # failed, so let every leg report. + fail-fast: false + matrix: ${{ fromJSON(needs.generate-matrix.outputs.linux) }} + uses: ./.github/workflows/install-e2e-run.yml + with: + route: ${{ matrix.route }} + install-ref: ${{ matrix.install_ref }} + + windows: + name: "windows: ${{ matrix.name }}" + if: fromJSON(needs.generate-matrix.outputs.windows).include[0] != null + needs: generate-matrix + strategy: + fail-fast: false + matrix: ${{ fromJSON(needs.generate-matrix.outputs.windows) }} + uses: ./.github/workflows/install-e2e-windows-run.yml + with: + install-method: ${{ matrix.install_method }} + update-method: ${{ matrix.update_method }} + install-ref: ${{ matrix.install_ref }} + + skipped: + name: "${{ matrix.os }}: ${{ matrix.install }} -> ${{ matrix.update }}" + if: fromJSON(needs.generate-matrix.outputs.skipped).include[0] != null + needs: generate-matrix + strategy: + fail-fast: false + matrix: ${{ fromJSON(needs.generate-matrix.outputs.skipped) }} + uses: ./.github/workflows/install-e2e-skip.yml diff --git a/.github/workflows/install-e2e-windows-run.yml b/.github/workflows/install-e2e-windows-run.yml index ad7dfbb27cc4..f4ac6e4435c3 100644 --- a/.github/workflows/install-e2e-windows-run.yml +++ b/.github/workflows/install-e2e-windows-run.yml @@ -15,31 +15,43 @@ name: Install & Update E2E — Windows desktop (reusable) # update leg — an update becomes available exactly the way it does for a # real user. # -# Routes (the update mechanism under test): -# desktop the app's own Update button: the installed Hermes.exe runs -# under Playwright's Electron driver, which clicks Settings -> -# About -> "Update now"; the production hand-off chain runs -# untouched (marker, app quit, detached updater, hermes -# update, desktop rebuild, relaunch). -# update TODO: `hermes update` from the installed venv. -# installer TODO: re-run the bootstrap exe over the existing install. +# Routes (from the update-method input): +# desktop-app the app's own Update button: the installed +# Hermes.exe runs under Playwright's Electron +# driver, which clicks Settings -> About -> +# "Update now"; the production hand-off chain runs +# untouched (marker, app quit, detached updater, +# hermes update, desktop rebuild, relaunch). +# hermes-update TODO: `hermes update` from the installed venv. +# desktop-installer-rerun@latest +# TODO: re-run the bootstrap exe over the install. +# irm-iex TODO: re-run the irm | iex one-liner. +# +# Method pairs without a driver yet NATIVELY SKIP (grey check, no runner): +# the capability knowledge lives here, next to the driver, so the caller +# can dispatch every declared combination without knowing which ones work. # # Call it: # # jobs: -# windows-desktop: +# windows: # uses: ./.github/workflows/install-e2e-windows-run.yml # with: -# route: desktop +# install-method: desktop-installer@latest +# update-method: desktop-app +# install-ref: v2026.8.3 on: workflow_call: inputs: - route: - description: 'Update mechanism to exercise: desktop (Update button; implemented), update (hermes update; TODO), installer (re-run bootstrap exe; TODO).' - required: false + install-method: + description: 'How OLD gets installed. Supported: desktop-installer@latest (website exe, AHK-clicked). Declared-but-TODO methods skip.' + required: true + type: string + update-method: + description: 'How the install updates to HEAD. Supported: desktop-app (Update button under Playwright). Declared-but-TODO methods skip.' + required: true type: string - default: desktop install-ref: description: 'Ref to install as OLD (served as main while the installer runs). auto = the newest release tag in the checkout.' required: false @@ -61,7 +73,10 @@ permissions: jobs: e2e: - name: "${{ inputs.route }} route from website installer" + name: "${{ inputs.install-method }} -> ${{ inputs.update-method }}" + # The one pair the driver can run today. Anything else is a declared + # TODO: native skip, so the coverage gap is a grey check on every run. + if: inputs.install-method == 'desktop-installer@latest' && inputs.update-method == 'desktop-app' runs-on: windows-latest timeout-minutes: ${{ inputs.timeout-minutes }} @@ -110,15 +125,15 @@ jobs: - name: Stage serve repo (main -> OLD) shell: powershell - run: powershell -NoProfile -ExecutionPolicy Bypass -File tests\install\windows-desktop-gui-e2e.ps1 -Phase stage -Route ${{ inputs.route }} -InstallRef "${{ inputs.install-ref }}" -SetupExeUrl ${{ inputs.setup-exe-url }} + run: powershell -NoProfile -ExecutionPolicy Bypass -File tests\install\windows-desktop-gui-e2e.ps1 -Phase stage -Route "${{ inputs.update-method }}" -InstallRef "${{ inputs.install-ref }}" -SetupExeUrl ${{ inputs.setup-exe-url }} - name: Install OLD via website Hermes-Setup.exe (headed, AHK-clicked) shell: powershell - run: powershell -NoProfile -ExecutionPolicy Bypass -File tests\install\windows-desktop-gui-e2e.ps1 -Phase install-gui -Route ${{ inputs.route }} -InstallRef "${{ inputs.install-ref }}" -SetupExeUrl ${{ inputs.setup-exe-url }} + run: powershell -NoProfile -ExecutionPolicy Bypass -File tests\install\windows-desktop-gui-e2e.ps1 -Phase install-gui -Route "${{ inputs.update-method }}" -InstallRef "${{ inputs.install-ref }}" -SetupExeUrl ${{ inputs.setup-exe-url }} - - name: Update OLD -> HEAD (${{ inputs.route }} route) + - name: Update OLD -> HEAD (${{ inputs.update-method }}) shell: powershell - run: powershell -NoProfile -ExecutionPolicy Bypass -File tests\install\windows-desktop-gui-e2e.ps1 -Phase update-gui -Route ${{ inputs.route }} -InstallRef "${{ inputs.install-ref }}" -SetupExeUrl ${{ inputs.setup-exe-url }} + run: powershell -NoProfile -ExecutionPolicy Bypass -File tests\install\windows-desktop-gui-e2e.ps1 -Phase update-gui -Route "${{ inputs.update-method }}" -InstallRef "${{ inputs.install-ref }}" -SetupExeUrl ${{ inputs.setup-exe-url }} - name: Collect proof + logs if: always() @@ -141,7 +156,7 @@ jobs: if: always() uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1 with: - name: install-e2e-windows-${{ inputs.route }}-${{ github.sha }} + name: install-e2e-windows-${{ inputs.update-method }}-${{ inputs.install-ref }}-${{ github.sha }} path: gui-e2e-proof retention-days: 14 if-no-files-found: ignore diff --git a/.github/workflows/install-e2e.yml b/.github/workflows/install-e2e.yml index 7b0e1314677a..259fa322da63 100644 --- a/.github/workflows/install-e2e.yml +++ b/.github/workflows/install-e2e.yml @@ -4,27 +4,34 @@ name: Install & Update E2E # # The support matrix -- every {os, install-method, update-method} combination # a user could be on -- lives in scripts/sandbox/generate-e2e-matrix.mjs. -# The generate-matrix job expands it into ONE JOB PER COMBINATION: +# The job graph nests one sub-graph per starting release tag: # -# * linux legs install a sampled release through the real -# `curl | install.sh` one-liner (uv, a managed Python, Node, the venv) -# inside scripts/dev-sandbox.sh, then apply the combo's update route and -# require the checkout to land on this commit with a working `hermes`; -# * windows legs run the real desktop user flow (install-e2e-windows-run -# .yml): the website's Hermes-Setup.exe clicked by AutoHotkey, then the -# update driven through the app itself, Playwright clicking "Update now"; -# * combinations the spec declares but CI can't drive yet (macOS, the -# remaining windows install/update methods) fan out too, one NATIVELY -# SKIPPED check per combo, so every coverage gap is enumerable from the -# Checks tab. +# from (install-e2e-tag.yml) +# -> linux: -> one job per combination, a real +# curl|bash install in bubblewrap +# (install-e2e-run.yml) +# -> windows: -> one job per combination, the real +# desktop user flow: website exe +# clicked by AutoHotkey, update via +# the app, Playwright clicking +# "Update now" +# (install-e2e-windows-run.yml) +# -> macos: -> natively skipped until a macos +# workflow exists +# (install-e2e-skip.yml) # -# The linux starting versions are chosen at runtime from the repo's release -# tags (scripts/sandbox/pick-release-tags.sh): newest, oldest, and a spread +# Every combination is dispatched; unimplemented ones natively skip (grey) +# at the point that owns the knowledge -- macOS combos in the tag workflow +# (no driver exists), windows method pairs inside the windows run workflow +# (next to the driver that will implement them). +# +# The starting versions are chosen at runtime from the repo's release tags +# (scripts/sandbox/pick-release-tags.sh): newest, oldest, and a spread # between. A hardcoded list would stop covering the newest release the day -# after it ships, and would pin an "oldest" that nobody still runs. The -# windows starting version is the newest release tag (the published -# installer has no commit pin -- it installs what `main` serves, which the -# harness stages to that tag). +# after it ships, and would pin an "oldest" that nobody still runs. The tag +# axis is applied to EVERY os: linux legs install the tag in the sandbox; +# windows legs stage serve.git's `main` at it, which is what the published +# installer (no commit pin) then installs. # # Triggers: # * every 12 hours, so upstream drift (a new uv, a Node bump, a PyPI change) @@ -68,8 +75,8 @@ concurrency: cancel-in-progress: true jobs: - # Which released versions do we test updating FROM? Resolved once and shared - # by every linux leg, so all routes cover the same set. + # Which released versions do we test updating FROM? Resolved once; the + # generator applies the tag axis to every OS's legs. pick-releases: name: Pick release tags runs-on: ubuntu-latest @@ -93,79 +100,24 @@ jobs: echo "Testing updates from: $tags" echo "tags=$tags" >> "$GITHUB_OUTPUT" - # Expand the {os, install-method, update-method} support matrix - # (scripts/sandbox/generate-e2e-matrix.mjs) into one job per combination. - # Implemented combos land in the linux/windows matrices below; everything - # the spec declares but CI can't drive yet lands in `skipped`, so the TODO - # surface is visible in every run instead of buried in comments. - generate-matrix: - name: Generate combination matrix + # One sub-graph per starting tag: install-e2e-tag.yml expands the + # {os, install-method, update-method} support matrix for that tag and + # fans out one job per combination, so the graph reads + # "from vX -> windows: install -> update" per leg. Skips happen where + # the knowledge lives: macOS combos natively skip inside the tag + # workflow (no workflow exists for them at all); unimplemented windows + # method pairs natively skip inside install-e2e-windows-run.yml, next + # to the driver that will implement them. + from-tag: + name: "from ${{ matrix.install-ref }}" needs: pick-releases - runs-on: ubuntu-latest - timeout-minutes: 5 - outputs: - linux: ${{ steps.gen.outputs.linux }} - windows: ${{ steps.gen.outputs.windows }} - skipped: ${{ steps.gen.outputs.skipped }} - steps: - - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2 - with: - sparse-checkout: scripts/sandbox/generate-e2e-matrix.mjs - sparse-checkout-cone-mode: false - - id: gen - run: | - set -euo pipefail - matrices="$(node scripts/sandbox/generate-e2e-matrix.mjs \ - --tags '${{ needs.pick-releases.outputs.tags }}' \ - --route '${{ inputs.route || 'all' }}')" - echo "$matrices" - for key in linux windows skipped; do - echo "$key=$(echo "$matrices" | node -e 'let d="";process.stdin.on("data",c=>d+=c).on("end",()=>console.log(JSON.stringify(JSON.parse(d)[process.argv[1]])))' "$key")" >> "$GITHUB_OUTPUT" - done - - # Linux sandbox legs: one job per {install-method -> update-method, release - # tag} combination, each a real curl|bash install in bubblewrap followed by - # the combo's update route. - linux: - name: "linux: ${{ matrix.name }}" - if: fromJSON(needs.generate-matrix.outputs.linux).include[0] != null - needs: generate-matrix strategy: - # One combo breaking is worth knowing about even if another already - # failed, so let every leg report. + # One tag breaking is worth knowing about even if another already + # failed, so let every sub-graph report. fail-fast: false - max-parallel: 4 - matrix: ${{ fromJSON(needs.generate-matrix.outputs.linux) }} - uses: ./.github/workflows/install-e2e-run.yml + matrix: + install-ref: ${{ fromJSON(needs.pick-releases.outputs.tags) }} + uses: ./.github/workflows/install-e2e-tag.yml with: - route: ${{ matrix.route }} - install-ref: ${{ matrix.install_ref }} - - # Windows GUI legs: one job per combination of the real desktop user flow - # (website installer clicked by AutoHotkey, update driven through the app). - windows: - name: "windows: ${{ matrix.name }}" - if: fromJSON(needs.generate-matrix.outputs.windows).include[0] != null - needs: generate-matrix - strategy: - fail-fast: false - matrix: ${{ fromJSON(needs.generate-matrix.outputs.windows) }} - uses: ./.github/workflows/install-e2e-windows-run.yml - with: - route: ${{ matrix.route }} - install-ref: ${{ matrix.install_ref }} - - # Declared-but-unimplemented combinations: one job per combo, each a - # NATIVE skip (grey, conclusion=skipped, no runner). Matrix context is not - # available in job-level `if`, so the native skip lives in the called - # workflow instead: its inner job is gated on an input that defaults to - # false and is never passed. Implementing a combo moves it into - # IMPLEMENTED in the generator. - skipped: - name: "todo: ${{ matrix.os }}: ${{ matrix.install }} -> ${{ matrix.update }}" - if: fromJSON(needs.generate-matrix.outputs.skipped).include[0] != null - needs: generate-matrix - strategy: - fail-fast: false - matrix: ${{ fromJSON(needs.generate-matrix.outputs.skipped) }} - uses: ./.github/workflows/install-e2e-skip.yml + install-ref: ${{ matrix.install-ref }} + route: ${{ inputs.route || 'all' }} diff --git a/scripts/sandbox/generate-e2e-matrix.mjs b/scripts/sandbox/generate-e2e-matrix.mjs index 21b257c5e37b..5522cbcc75df 100644 --- a/scripts/sandbox/generate-e2e-matrix.mjs +++ b/scripts/sandbox/generate-e2e-matrix.mjs @@ -7,15 +7,19 @@ * per combination -- and everything else lands in a "skipped" matrix so the * TODO surface stays visible in every run instead of buried in comments. * - * Used by .github/workflows/install-e2e.yml: + * Used by .github/workflows/install-e2e-tag.yml, which runs it once per + * starting release tag: * * node scripts/sandbox/generate-e2e-matrix.mjs \ - * --tags '["v2026.8.3","v2026.3.12"]' --route all + * --tags '["v2026.8.3"]' --route all * * Prints JSON: { linux: {include:[...]}, windows: {include:[...]}, - * skipped: {include:[...]} }. The linux include entries carry - * {name, route, install_ref} for install-e2e-run.yml; windows entries carry - * {name, route, install_ref} for install-e2e-windows-run.yml. + * skipped: {include:[...]} }. linux entries carry {name, route, + * install_ref} for install-e2e-run.yml; windows entries carry {name, + * install_method, update_method, install_ref} for + * install-e2e-windows-run.yml (which itself natively skips method pairs it + * cannot drive yet); skipped entries are the macOS combos (no workflow + * exists at all). */ import path from 'node:path'; @@ -84,18 +88,20 @@ const KNOWN_METHODS = new Set([ const ALLOWED_VERSIONS = new Set(['latest']); /** - * The combinations CI actually runs today, and which reusable workflow + - * route input drives each. Everything in SPEC but not here is emitted as a - * skipped combo. Install ids include the expanded @version suffix. + * How each implemented combination maps onto a reusable workflow. + * + * linux: every combo is implemented; `route` is install-e2e-run.yml's input. + * windows: ALL combos dispatch to install-e2e-windows-run.yml -- the run + * workflow itself natively skips (grey) any {install, update} method pair + * it cannot drive yet, so "which windows methods work" lives THERE, next + * to the driver, not here. + * macos: nothing is implemented; combos surface as native skips in the + * caller via install-e2e-skip.yml without burning a tag fanout. */ -export const IMPLEMENTED = [ - // scripts/dev-sandbox.sh legs (install-e2e-run.yml, bubblewrap sandbox). - { os: 'linux', install: 'curl-bash', update: 'hermes-update', workflow: 'linux', route: 'update' }, - { os: 'linux', install: 'curl-bash', update: 'curl-bash', workflow: 'linux', route: 'installer' }, - // Real Windows GUI flow (install-e2e-windows-run.yml): website exe clicked - // by AutoHotkey, Update now clicked by Playwright. - { os: 'windows', install: 'desktop-installer@latest', update: 'desktop-app', workflow: 'windows', route: 'desktop' }, -]; +export const LINUX_ROUTES = { + 'hermes-update': 'update', + 'curl-bash': 'installer', +}; function validateEntry(os, kind, entry) { if (typeof entry.method !== 'string' || !KNOWN_METHODS.has(entry.method)) { @@ -150,25 +156,19 @@ export function generateEnvironments(spec) { return envs; } -function findImplementation(env) { - return IMPLEMENTED.find( - (m) => m.os === env.os && m.install === env.install && m.update === env.update, - ); -} - /** Mirror of install-e2e.yml's dispatch `route` choice. */ -function routeWants(route, env, impl) { +function routeWants(route, env) { switch (route) { case 'all': return true; case 'both': - return env.os === 'linux' && Boolean(impl); + return env.os === 'linux'; case 'update': - return impl?.route === 'update'; + return env.os === 'linux' && env.update === 'hermes-update'; case 'installer': - return impl?.route === 'installer'; + return env.os === 'linux' && env.update === 'curl-bash'; case 'windows-desktop': - return env.os === 'windows' && Boolean(impl); + return env.os === 'windows'; default: throw new Error(`unknown route filter: ${JSON.stringify(route)}`); } @@ -176,39 +176,63 @@ function routeWants(route, env, impl) { /** * Split the combinations into per-workflow matrices. - * Linux combos fan out further over `tags` (which released version the leg - * installs first); Windows install versions come from the method id instead - * (desktop-installer@latest = whatever the website serves). + * + * `tags` (the released versions we test updating FROM) is the OUTER axis, + * applied to every OS with a driving workflow: for each tag, for each + * combo, one job that installs the tag and updates to HEAD. Linux legs + * pass the tag as install-ref for the sandbox to install; Windows legs + * pass it as the ref the staged serve.git parks `main` at -- the published + * installer has no commit pin, so that IS the installed version. + * + * Skip placement follows where the knowledge lives: macOS combos are known + * unimplementable HERE (no workflow exists), so they go to the `skipped` + * matrix -- one native grey check per combo, not multiplied by tags, since + * no tag would change the outcome. Windows combos ALL dispatch to the run + * workflow, which natively skips the method pairs it cannot drive. */ export function buildMatrices(envs, { tags = [], route = 'all' } = {}) { const linux = []; const windows = []; const skipped = []; + const needTags = () => { + if (tags.length === 0) { + throw new Error('a combo with a driving workflow was selected but no --tags given'); + } + }; for (const env of envs) { - const impl = findImplementation(env); - if (!routeWants(route, env, impl)) continue; - if (!impl) { - skipped.push({ ...env, reason: 'not implemented yet' }); - } else if (impl.workflow === 'linux') { - if (tags.length === 0) { - throw new Error(`linux combo ${env.install} -> ${env.update} selected but no --tags given`); + if (!routeWants(route, env)) continue; + if (env.os === 'macos') { + skipped.push({ ...env, reason: 'no macos workflow yet' }); + continue; + } + if (env.os === 'linux') { + const linuxRoute = LINUX_ROUTES[env.update]; + if (!linuxRoute) { + throw new Error(`linux update method ${JSON.stringify(env.update)} has no route mapping`); } + needTags(); for (const tag of tags) { linux.push({ - name: `${env.install} @ ${tag} -> ${env.update}`, - route: impl.route, + name: `${env.install} -> ${env.update}`, + route: linuxRoute, + install_ref: tag, + }); + } + continue; + } + if (env.os === 'windows') { + needTags(); + for (const tag of tags) { + windows.push({ + name: `${env.install} -> ${env.update}`, + install_method: env.install, + update_method: env.update, install_ref: tag, }); } - } else if (impl.workflow === 'windows') { - windows.push({ - name: `${env.install} -> ${env.update}`, - route: impl.route, - install_ref: 'auto', - }); - } else { - throw new Error(`unknown workflow ${JSON.stringify(impl.workflow)} for ${env.os}`); + continue; } + throw new Error(`no workflow routing for os ${JSON.stringify(env.os)}`); } return { linux: { include: linux }, diff --git a/tests/install/windows-desktop-gui-e2e.ps1 b/tests/install/windows-desktop-gui-e2e.ps1 index 06e87f0a1155..e95aa783d0a2 100644 --- a/tests/install/windows-desktop-gui-e2e.ps1 +++ b/tests/install/windows-desktop-gui-e2e.ps1 @@ -71,12 +71,13 @@ param( [ValidateSet("stage", "install-gui", "update-gui", "all")] [string]$Phase = "all", - # Update route to exercise in the update-gui phase. Only "desktop" (the - # app's own Update button) is implemented; "update" (CLI `hermes - # update`) and "installer" (re-run the bootstrap exe) are declared arms - # so the workflow surface is stable when they land. - [ValidateSet("desktop", "update", "installer")] - [string]$Route = "desktop", + # Update method to exercise in the update-gui phase, named by the same + # ids the combination generator (scripts/sandbox/generate-e2e-matrix + # .mjs) uses. Only "desktop-app" (the app's own Update button) is + # implemented; the others are declared arms so the surface is stable + # when they land. + [ValidateSet("desktop-app", "hermes-update", "desktop-installer-rerun@latest", "irm-iex")] + [string]$Route = "desktop-app", # The OLD version: the ref served as `main` while the installer runs, # i.e. what the user starts on. The published Hermes-Setup.exe carries @@ -657,20 +658,25 @@ function Invoke-GuiUpdateDesktopRoute([string]$TargetSha) { function Invoke-PhaseUpdateGui { $state = Read-State switch ($Route) { - "desktop" { + "desktop-app" { Invoke-GuiUpdateDesktopRoute $state.current } - "update" { + "hermes-update" { # TODO: run `hermes update` from the installed venv -- the CLI - # route. Needs the same completion/sha/relaunch asserts minus - # the app-quit dance. - throw "route 'update' (CLI hermes update) is not implemented yet" + # route. Needs the same completion/sha asserts minus the + # app-quit dance. + throw "update method 'hermes-update' is not implemented yet" } - "installer" { + "desktop-installer-rerun@latest" { # TODO: re-run the bootstrap Hermes-Setup.exe over the existing # install (its --update flow jumps straight to progress and # runs unattended). - throw "route 'installer' (re-run bootstrap exe) is not implemented yet" + throw "update method 'desktop-installer-rerun@latest' is not implemented yet" + } + "irm-iex" { + # TODO: re-run the irm | iex one-liner over the existing + # install. + throw "update method 'irm-iex' is not implemented yet" } } } From 3eec049c29afabe2ae30d19ecdc92990bc2af985 Mon Sep 17 00:00:00 2001 From: ethernet Date: Tue, 11 Aug 2026 16:04:19 -0400 Subject: [PATCH 045/227] ci(windows-e2e): static inner job name - github renders name expressions unexpanded on skipped jobs Run 31530831547 showed skipped windows legs as the literal '${{ inputs.install-method }} -> ...' - GitHub does not evaluate name expressions for natively skipped jobs. The caller's job name already carries the method pair, so name the inner job statically. --- .github/workflows/install-e2e-windows-run.yml | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-) diff --git a/.github/workflows/install-e2e-windows-run.yml b/.github/workflows/install-e2e-windows-run.yml index f4ac6e4435c3..51c6693f3dc2 100644 --- a/.github/workflows/install-e2e-windows-run.yml +++ b/.github/workflows/install-e2e-windows-run.yml @@ -73,7 +73,10 @@ permissions: jobs: e2e: - name: "${{ inputs.install-method }} -> ${{ inputs.update-method }}" + # Static name on purpose: the caller's job name already carries the + # method pair, and GitHub renders name expressions UNEXPANDED (literal + # "${{ inputs... }}") on natively skipped jobs. + name: run # The one pair the driver can run today. Anything else is a declared # TODO: native skip, so the coverage gap is a grey check on every run. if: inputs.install-method == 'desktop-installer@latest' && inputs.update-method == 'desktop-app' From 6d94678e62db5acdcefe637b1786ad1282b5885c Mon Sep 17 00:00:00 2001 From: ethernet Date: Tue, 11 Aug 2026 16:17:48 -0400 Subject: [PATCH 046/227] ci(install-e2e): jobs own their skips - generator is pure expansion Remove all capability knowledge from the combination generator: no IMPLEMENTED table, no skipped matrix, no per-OS special cases. It now only declares and expands - every {os, install-method, update-method} combination is dispatched to its OS's run workflow, and each run workflow natively skips (grey, job-level if on the method inputs) the pairs its driver cannot run yet: * install-e2e-run.yml gains install-method/update-method inputs, gates on the supported pairs (curl-bash -> hermes-update/curl-bash), and maps the method id to the sandbox script's --route internally; * install-e2e-macos-run.yml is new - all pairs skip until a macOS driver exists, and implementing one flips its job-level if; * install-e2e-windows-run.yml already worked this way; * install-e2e-skip.yml is deleted - nothing special-cases macOS anymore, so the tag workflow is three identical OS fanouts. Structure is now uniformly matrix(tag) -> matrix(combination) -> run-or-skip, with capability knowledge living only next to each driver. Generator shape/route filters/error paths re-verified; all five workflows pass actionlint. --- .github/workflows/install-e2e-macos-run.yml | 51 +++++++++ .github/workflows/install-e2e-run.yml | 51 ++++++--- .github/workflows/install-e2e-skip.yml | 36 ------- .github/workflows/install-e2e-tag.yml | 35 ++++--- .github/workflows/install-e2e.yml | 34 +++--- scripts/sandbox/generate-e2e-matrix.mjs | 110 ++++++-------------- 6 files changed, 154 insertions(+), 163 deletions(-) create mode 100644 .github/workflows/install-e2e-macos-run.yml delete mode 100644 .github/workflows/install-e2e-skip.yml diff --git a/.github/workflows/install-e2e-macos-run.yml b/.github/workflows/install-e2e-macos-run.yml new file mode 100644 index 000000000000..0962c3ffccc1 --- /dev/null +++ b/.github/workflows/install-e2e-macos-run.yml @@ -0,0 +1,51 @@ +name: Install & Update E2E — macOS (reusable) + +# Runs ONE {install-method, update-method} combination on macOS, installing +# a starting version and updating it to HEAD. +# +# NOTHING is implemented yet: every method pair NATIVELY SKIPS (grey check, +# no runner) until a macOS driver exists. The workflow exists now so the +# combination generator (scripts/sandbox/generate-e2e-matrix.mjs) can +# dispatch every declared macOS combination the same way it does for linux +# and windows -- capability knowledge lives here, next to where the driver +# will be, and implementing a method flips this workflow's job-level `if`. +# +# Declared methods (see the generator's SPEC): +# install: curl-bash, packaged-app +# update: curl-bash, hermes-update, app-update + +on: + workflow_call: + inputs: + install-method: + description: 'How the starting version gets installed. All macOS methods are TODO and skip.' + required: true + type: string + update-method: + description: 'How the install updates to HEAD. All macOS methods are TODO and skip.' + required: true + type: string + install-ref: + description: 'What to install before updating: a branch, a tag (v2026.7.7), or a SHA reachable from main.' + required: false + type: string + default: refs/heads/main + +permissions: + contents: read + +jobs: + e2e: + # Static name on purpose: the caller's job name already carries the + # method pair, and GitHub renders name expressions UNEXPANDED (literal + # "${{ inputs... }}") on natively skipped jobs. + name: run + # No macOS driver exists yet: every pair is a declared TODO, so this is + # constant-false until the first method lands. Written as an impossible + # input comparison rather than `if: false` because actionlint rejects + # constant conditions. + if: inputs.install-method == inputs.update-method && inputs.install-method == 'implemented' + runs-on: macos-latest + timeout-minutes: 5 + steps: + - run: 'true' diff --git a/.github/workflows/install-e2e-run.yml b/.github/workflows/install-e2e-run.yml index 354ff6bd47bc..7a2c1f3467f3 100644 --- a/.github/workflows/install-e2e-run.yml +++ b/.github/workflows/install-e2e-run.yml @@ -1,12 +1,20 @@ name: Install & Update E2E (reusable) -# Runs ONE update route against ONE starting commit, in the dev sandbox, with a -# real install (uv, a managed Python, Node, the venv) behind it. +# Runs ONE {install-method, update-method} combination against ONE starting +# commit, in the dev sandbox, with a real install (uv, a managed Python, +# Node, the venv) behind it. # -# Reusable so callers can fan out over the combinations that matter -- update -# from the tip vs. from an older release, `hermes update` vs. re-running the -# installer -- without duplicating the runner setup. Each leg is independent: -# its own sandbox, its own install, nothing rewound or shared. +# Reusable so callers can fan out over the combinations that matter -- +# update from the tip vs. from an older release, `hermes update` vs. +# re-running the installer -- without duplicating the runner setup. Each leg +# is independent: its own sandbox, its own install, nothing rewound or +# shared. +# +# Method ids come from scripts/sandbox/generate-e2e-matrix.mjs. Supported +# today: install via curl-bash, update via hermes-update or curl-bash. +# Anything else NATIVELY SKIPS (grey check, no runner): capability +# knowledge lives here, next to the driver, so the caller can dispatch +# every declared combination without knowing which ones work. # # Call it: # @@ -14,14 +22,19 @@ name: Install & Update E2E (reusable) # tip: # uses: ./.github/workflows/install-e2e-run.yml # with: -# route: update +# install-method: curl-bash +# update-method: hermes-update # install-ref: refs/heads/main on: workflow_call: inputs: - route: - description: 'Update path to exercise: update (hermes update) or installer (re-run install.sh).' + install-method: + description: 'How the starting version gets installed. Supported: curl-bash (the real curl | install.sh one-liner).' + required: true + type: string + update-method: + description: 'How the install updates to HEAD. Supported: hermes-update (the updater) or curl-bash (re-run the one-liner). Declared-but-TODO methods skip.' required: true type: string install-ref: @@ -45,7 +58,14 @@ permissions: jobs: e2e: - name: ${{ inputs.route }} from ${{ inputs.install-ref }} + # Static name on purpose: the caller's job name already carries the + # method pair, and GitHub renders name expressions UNEXPANDED (literal + # "${{ inputs... }}") on natively skipped jobs. + name: run + # The pairs the sandbox driver can run today. Anything else is a + # declared TODO: native skip, so the coverage gap is a grey check on + # every run. + if: inputs.install-method == 'curl-bash' && contains(fromJSON('["hermes-update", "curl-bash"]'), inputs.update-method) runs-on: ${{ inputs.runner }} timeout-minutes: ${{ inputs.timeout-minutes }} @@ -84,8 +104,15 @@ jobs: - name: Run install + update E2E run: | set -euo pipefail + # Method id -> the driver script's --route vocabulary. The one + # place that knows both names. + case '${{ inputs.update-method }}' in + hermes-update) route=update ;; + curl-bash) route=installer ;; + *) echo "unreachable: update-method passed the job-level gate but has no route mapping" >&2; exit 1 ;; + esac tests/install/install-update-e2e.sh \ - --route '${{ inputs.route }}' \ + --route "$route" \ --install-ref '${{ inputs.install-ref }}' env: # Outside the workspace on purpose: the script creates this directory @@ -106,7 +133,7 @@ jobs: set -euo pipefail safe_ref='${{ inputs.install-ref }}' safe_ref="${safe_ref//\//-}" - echo "name=install-e2e-${{ inputs.route }}-${safe_ref}" >> "$GITHUB_OUTPUT" + echo "name=install-e2e-${{ inputs.update-method }}-${safe_ref}" >> "$GITHUB_OUTPUT" # The installer's own transcripts say far more than the assertion that # tripped when a real install breaks. diff --git a/.github/workflows/install-e2e-skip.yml b/.github/workflows/install-e2e-skip.yml deleted file mode 100644 index 02415007403d..000000000000 --- a/.github/workflows/install-e2e-skip.yml +++ /dev/null @@ -1,36 +0,0 @@ -name: Install & Update E2E — unsupported combo (reusable) - -# A declared {os, install-method, update-method} combination whose OS has -# no driving E2E workflow at all (today: every macOS combo). The per-tag -# caller (install-e2e-tag.yml) fans out one of these per such combo so the -# coverage gap is its own named, NATIVELY SKIPPED check (grey, -# conclusion=skipped, no runner) in the Checks tab. -# -# OSes WITH a driving workflow don't come here: their unimplemented method -# pairs natively skip inside that workflow, next to the driver that will -# implement them (see install-e2e-windows-run.yml). -# -# `implemented` stays at its false default until the OS has a real driving -# workflow; flipping happens by routing the combo to that workflow in -# generate-e2e-matrix.mjs, never by setting this input. - -on: - workflow_call: - inputs: - implemented: - description: 'Never set this. Combos become implemented by moving into IMPLEMENTED in generate-e2e-matrix.mjs, not by flipping the flag.' - required: false - type: boolean - default: false - -permissions: - contents: read - -jobs: - todo: - name: not implemented yet - if: inputs.implemented - runs-on: ubuntu-latest - timeout-minutes: 1 - steps: - - run: 'true' diff --git a/.github/workflows/install-e2e-tag.yml b/.github/workflows/install-e2e-tag.yml index ee560197538e..6207f1490d2d 100644 --- a/.github/workflows/install-e2e-tag.yml +++ b/.github/workflows/install-e2e-tag.yml @@ -2,15 +2,13 @@ name: Install & Update E2E — one starting tag (reusable) # Everything a user starting on ONE released version could do: for each # {os, install-method, update-method} combination the support matrix -# (scripts/sandbox/generate-e2e-matrix.mjs) declares, one job that installs -# this tag and updates to HEAD. +# (scripts/sandbox/generate-e2e-matrix.mjs) declares, one dispatch that +# installs this tag and updates to HEAD. # -# Skips happen where the knowledge lives: -# * macOS combos have no workflow at all -- they natively skip right -# here via install-e2e-skip.yml; -# * windows combos ALL dispatch to install-e2e-windows-run.yml, which -# natively skips the method pairs its driver cannot run yet; -# * linux combos are all implemented. +# Every combination is dispatched to its OS's run workflow; the run +# workflow natively skips (grey) the method pairs its driver cannot run +# yet. Capability knowledge lives next to each driver, never here and +# never in the generator. # # Call it: # @@ -46,7 +44,7 @@ jobs: outputs: linux: ${{ steps.gen.outputs.linux }} windows: ${{ steps.gen.outputs.windows }} - skipped: ${{ steps.gen.outputs.skipped }} + macos: ${{ steps.gen.outputs.macos }} steps: - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2 with: @@ -59,7 +57,7 @@ jobs: --tags '["${{ inputs.install-ref }}"]' \ --route '${{ inputs.route }}')" echo "$matrices" - for key in linux windows skipped; do + for key in linux windows macos; do echo "$key=$(echo "$matrices" | node -e 'let d="";process.stdin.on("data",c=>d+=c).on("end",()=>console.log(JSON.stringify(JSON.parse(d)[process.argv[1]])))' "$key")" >> "$GITHUB_OUTPUT" done @@ -74,7 +72,8 @@ jobs: matrix: ${{ fromJSON(needs.generate-matrix.outputs.linux) }} uses: ./.github/workflows/install-e2e-run.yml with: - route: ${{ matrix.route }} + install-method: ${{ matrix.install_method }} + update-method: ${{ matrix.update_method }} install-ref: ${{ matrix.install_ref }} windows: @@ -90,11 +89,15 @@ jobs: update-method: ${{ matrix.update_method }} install-ref: ${{ matrix.install_ref }} - skipped: - name: "${{ matrix.os }}: ${{ matrix.install }} -> ${{ matrix.update }}" - if: fromJSON(needs.generate-matrix.outputs.skipped).include[0] != null + macos: + name: "macos: ${{ matrix.name }}" + if: fromJSON(needs.generate-matrix.outputs.macos).include[0] != null needs: generate-matrix strategy: fail-fast: false - matrix: ${{ fromJSON(needs.generate-matrix.outputs.skipped) }} - uses: ./.github/workflows/install-e2e-skip.yml + matrix: ${{ fromJSON(needs.generate-matrix.outputs.macos) }} + uses: ./.github/workflows/install-e2e-macos-run.yml + with: + install-method: ${{ matrix.install_method }} + update-method: ${{ matrix.update_method }} + install-ref: ${{ matrix.install_ref }} diff --git a/.github/workflows/install-e2e.yml b/.github/workflows/install-e2e.yml index 259fa322da63..4b5f3c4bfa7f 100644 --- a/.github/workflows/install-e2e.yml +++ b/.github/workflows/install-e2e.yml @@ -7,23 +7,23 @@ name: Install & Update E2E # The job graph nests one sub-graph per starting release tag: # # from (install-e2e-tag.yml) -# -> linux: -> one job per combination, a real -# curl|bash install in bubblewrap +# -> linux: -> one dispatch per combination, a +# real curl|bash install in +# bubblewrap # (install-e2e-run.yml) -# -> windows: -> one job per combination, the real -# desktop user flow: website exe -# clicked by AutoHotkey, update via -# the app, Playwright clicking +# -> windows: -> one dispatch per combination, the +# real desktop user flow: website +# exe clicked by AutoHotkey, update +# via the app, Playwright clicking # "Update now" # (install-e2e-windows-run.yml) -# -> macos: -> natively skipped until a macos -# workflow exists -# (install-e2e-skip.yml) +# -> macos: -> one dispatch per combination +# (install-e2e-macos-run.yml) # -# Every combination is dispatched; unimplemented ones natively skip (grey) -# at the point that owns the knowledge -- macOS combos in the tag workflow -# (no driver exists), windows method pairs inside the windows run workflow -# (next to the driver that will implement them). +# Every combination is dispatched to its OS's run workflow; the run +# workflow natively skips (grey) the method pairs its driver cannot run +# yet. Capability knowledge lives next to each driver -- the generator and +# the callers know nothing about what is implemented. # # The starting versions are chosen at runtime from the repo's release tags # (scripts/sandbox/pick-release-tags.sh): newest, oldest, and a spread @@ -102,12 +102,8 @@ jobs: # One sub-graph per starting tag: install-e2e-tag.yml expands the # {os, install-method, update-method} support matrix for that tag and - # fans out one job per combination, so the graph reads - # "from vX -> windows: install -> update" per leg. Skips happen where - # the knowledge lives: macOS combos natively skip inside the tag - # workflow (no workflow exists for them at all); unimplemented windows - # method pairs natively skip inside install-e2e-windows-run.yml, next - # to the driver that will implement them. + # dispatches one job per combination to the OS's run workflow, so the + # graph reads "from vX -> windows: install -> update" per leg. from-tag: name: "from ${{ matrix.install-ref }}" needs: pick-releases diff --git a/scripts/sandbox/generate-e2e-matrix.mjs b/scripts/sandbox/generate-e2e-matrix.mjs index 5522cbcc75df..4dfcf76d2d3a 100644 --- a/scripts/sandbox/generate-e2e-matrix.mjs +++ b/scripts/sandbox/generate-e2e-matrix.mjs @@ -3,9 +3,12 @@ * Expand the install/update support matrix into concrete E2E combinations. * * One source of truth for every {os, install-method, update-method} pair a - * user could be on. Implemented combos fan out into real CI jobs -- one job - * per combination -- and everything else lands in a "skipped" matrix so the - * TODO surface stays visible in every run instead of buried in comments. + * user could be on. This file only DECLARES and EXPANDS: it knows nothing + * about which combinations CI can drive. Every combination is dispatched to + * its OS's run workflow, and THAT workflow natively skips the method pairs + * its driver cannot run yet -- capability knowledge lives next to each + * driver (install-e2e-run.yml, install-e2e-windows-run.yml, + * install-e2e-macos-run.yml). * * Used by .github/workflows/install-e2e-tag.yml, which runs it once per * starting release tag: @@ -14,12 +17,8 @@ * --tags '["v2026.8.3"]' --route all * * Prints JSON: { linux: {include:[...]}, windows: {include:[...]}, - * skipped: {include:[...]} }. linux entries carry {name, route, - * install_ref} for install-e2e-run.yml; windows entries carry {name, - * install_method, update_method, install_ref} for - * install-e2e-windows-run.yml (which itself natively skips method pairs it - * cannot drive yet); skipped entries are the macOS combos (no workflow - * exists at all). + * macos: {include:[...]} } -- every entry is + * {name, install_method, update_method, install_ref}. */ import path from 'node:path'; @@ -87,22 +86,6 @@ const KNOWN_METHODS = new Set([ const ALLOWED_VERSIONS = new Set(['latest']); -/** - * How each implemented combination maps onto a reusable workflow. - * - * linux: every combo is implemented; `route` is install-e2e-run.yml's input. - * windows: ALL combos dispatch to install-e2e-windows-run.yml -- the run - * workflow itself natively skips (grey) any {install, update} method pair - * it cannot drive yet, so "which windows methods work" lives THERE, next - * to the driver, not here. - * macos: nothing is implemented; combos surface as native skips in the - * caller via install-e2e-skip.yml without burning a tag fanout. - */ -export const LINUX_ROUTES = { - 'hermes-update': 'update', - 'curl-bash': 'installer', -}; - function validateEntry(os, kind, entry) { if (typeof entry.method !== 'string' || !KNOWN_METHODS.has(entry.method)) { throw new Error(`${os}.${kind}: unknown method id ${JSON.stringify(entry.method)} -- add it to KNOWN_METHODS if intentional`); @@ -175,70 +158,37 @@ function routeWants(route, env) { } /** - * Split the combinations into per-workflow matrices. - * - * `tags` (the released versions we test updating FROM) is the OUTER axis, - * applied to every OS with a driving workflow: for each tag, for each - * combo, one job that installs the tag and updates to HEAD. Linux legs - * pass the tag as install-ref for the sandbox to install; Windows legs - * pass it as the ref the staged serve.git parks `main` at -- the published - * installer has no commit pin, so that IS the installed version. + * Split the combinations into one matrix per OS. * - * Skip placement follows where the knowledge lives: macOS combos are known - * unimplementable HERE (no workflow exists), so they go to the `skipped` - * matrix -- one native grey check per combo, not multiplied by tags, since - * no tag would change the outcome. Windows combos ALL dispatch to the run - * workflow, which natively skips the method pairs it cannot drive. + * `tags` (the released versions we test updating FROM) is the OUTER axis: + * for each tag, for each combination, one dispatch that installs the tag + * and updates to HEAD. No capability filtering happens here -- every + * declared combination is dispatched, and the OS's run workflow natively + * skips what its driver cannot run yet. */ export function buildMatrices(envs, { tags = [], route = 'all' } = {}) { - const linux = []; - const windows = []; - const skipped = []; - const needTags = () => { - if (tags.length === 0) { - throw new Error('a combo with a driving workflow was selected but no --tags given'); - } - }; + const byOs = { linux: [], windows: [], macos: [] }; for (const env of envs) { if (!routeWants(route, env)) continue; - if (env.os === 'macos') { - skipped.push({ ...env, reason: 'no macos workflow yet' }); - continue; + const bucket = byOs[env.os]; + if (!bucket) { + throw new Error(`no matrix bucket for os ${JSON.stringify(env.os)}`); } - if (env.os === 'linux') { - const linuxRoute = LINUX_ROUTES[env.update]; - if (!linuxRoute) { - throw new Error(`linux update method ${JSON.stringify(env.update)} has no route mapping`); - } - needTags(); - for (const tag of tags) { - linux.push({ - name: `${env.install} -> ${env.update}`, - route: linuxRoute, - install_ref: tag, - }); - } - continue; + if (tags.length === 0) { + throw new Error('combinations were selected but no --tags given'); } - if (env.os === 'windows') { - needTags(); - for (const tag of tags) { - windows.push({ - name: `${env.install} -> ${env.update}`, - install_method: env.install, - update_method: env.update, - install_ref: tag, - }); - } - continue; + for (const tag of tags) { + bucket.push({ + name: `${env.install} -> ${env.update}`, + install_method: env.install, + update_method: env.update, + install_ref: tag, + }); } - throw new Error(`no workflow routing for os ${JSON.stringify(env.os)}`); } - return { - linux: { include: linux }, - windows: { include: windows }, - skipped: { include: skipped }, - }; + return Object.fromEntries( + Object.entries(byOs).map(([os, include]) => [os, { include }]), + ); } function main() { From 5521d32a56db94d9cc0036c8661854b44ee7c397 Mon Sep 17 00:00:00 2001 From: ethernet Date: Tue, 11 Aug 2026 16:24:26 -0400 Subject: [PATCH 047/227] ci(windows-e2e): probe the starting tag for apps/desktop - pre-desktop releases skip Run 31530831547 failed the moment old tags hit the windows leg: the bootstrap install of v2026.4.30 / v2026.5.29.2 succeeded but no app window ever appeared - those releases predate the desktop app (#20059, v2026.5.31), so there is nothing to launch and no Update button to click. Add a probe job that asks the tag's own tree (git ls-tree apps/desktop) and gate the run job on it, so desktop-method legs from pre-desktop tags natively skip instead of failing. Data-driven - no version cutoff list to rot. Probe logic verified locally against pre- and post-desktop tags plus the auto sentinel. --- .github/workflows/install-e2e-windows-run.yml | 44 +++++++++++++++++-- 1 file changed, 41 insertions(+), 3 deletions(-) diff --git a/.github/workflows/install-e2e-windows-run.yml b/.github/workflows/install-e2e-windows-run.yml index 51c6693f3dc2..c1e9a63ffd0a 100644 --- a/.github/workflows/install-e2e-windows-run.yml +++ b/.github/workflows/install-e2e-windows-run.yml @@ -72,14 +72,52 @@ permissions: contents: read jobs: + # Can this starting version run this flow at all? The desktop app + # (apps/desktop) only exists in releases from v2026.5.31 on (#20059): + # older tags have no window to launch and no Update button to click, so + # desktop-method legs from them natively skip below. Data-driven from the + # tag's own tree -- no hardcoded version list to rot. + probe: + name: probe tag + if: inputs.install-method == 'desktop-installer@latest' && inputs.update-method == 'desktop-app' + runs-on: ubuntu-latest + timeout-minutes: 5 + outputs: + has-desktop: ${{ steps.probe.outputs.has-desktop }} + steps: + # Tree listings only: no blobs, full history + tags so any ref + # resolves. + - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2 + with: + fetch-depth: 0 + filter: blob:none + - id: probe + run: | + set -euo pipefail + ref='${{ inputs.install-ref }}' + if [ "$ref" = "auto" ]; then + # Same choice the driver's stage phase makes: newest release tag. + ref="$(git tag --list 'v[0-9]*' --sort=-creatordate | head -1)" + fi + if git ls-tree -d "$ref" apps/desktop | grep -q .; then + has=true + else + has=false + fi + echo "ref $ref: has-desktop=$has" + echo "has-desktop=$has" >> "$GITHUB_OUTPUT" + e2e: # Static name on purpose: the caller's job name already carries the # method pair, and GitHub renders name expressions UNEXPANDED (literal # "${{ inputs... }}") on natively skipped jobs. name: run - # The one pair the driver can run today. Anything else is a declared - # TODO: native skip, so the coverage gap is a grey check on every run. - if: inputs.install-method == 'desktop-installer@latest' && inputs.update-method == 'desktop-app' + needs: probe + # The one pair the driver can run today, and only from a starting + # version that ships the desktop app. Anything else is a native skip: + # a declared-TODO method pair, or a tag from before the app existed. + # (When probe itself skipped, its output is empty and this is false.) + if: inputs.install-method == 'desktop-installer@latest' && inputs.update-method == 'desktop-app' && needs.probe.outputs.has-desktop == 'true' runs-on: windows-latest timeout-minutes: ${{ inputs.timeout-minutes }} From 6515f8132aa40faf1739f8721f648d527a04aa28 Mon Sep 17 00:00:00 2001 From: ethernet Date: Tue, 11 Aug 2026 16:35:03 -0400 Subject: [PATCH 048/227] ci(install-e2e): hardcode combination jobs in the primary workflow - one matrix box per combo GitHub only draws matrix boxes for the PRIMARY workflow's matrices; everything inside a called workflow flattens into slash-joined names. The generator + per-tag sub-workflow therefore bought no structure in the graph and hid the support matrix in a script. Invert it: install-e2e.yml now declares one job per {os, install-method -> update-method} combination (2 linux + 8 windows + 6 macos - same 16 the generator produced, verified by inventory before/after), each a matrix over the picked release tags. The graph now renders one titled box per combination whose legs read '... from vX' - the tag axis inside the combo axis. The per-OS run workflows are unchanged: they own capability knowledge and natively skip unimplemented method pairs and pre-desktop tags. install-e2e-tag.yml and generate-e2e-matrix.mjs are deleted; adding a method is now adding one job block here, implementing one is flipping the run workflow's gate. --- .github/workflows/install-e2e-macos-run.yml | 12 +- .github/workflows/install-e2e-run.yml | 2 +- .github/workflows/install-e2e-tag.yml | 103 ------- .github/workflows/install-e2e.yml | 289 +++++++++++++++++--- scripts/sandbox/generate-e2e-matrix.mjs | 212 -------------- tests/install/windows-desktop-gui-e2e.ps1 | 7 +- 6 files changed, 263 insertions(+), 362 deletions(-) delete mode 100644 .github/workflows/install-e2e-tag.yml delete mode 100644 scripts/sandbox/generate-e2e-matrix.mjs diff --git a/.github/workflows/install-e2e-macos-run.yml b/.github/workflows/install-e2e-macos-run.yml index 0962c3ffccc1..e23dcab79ebc 100644 --- a/.github/workflows/install-e2e-macos-run.yml +++ b/.github/workflows/install-e2e-macos-run.yml @@ -4,13 +4,13 @@ name: Install & Update E2E — macOS (reusable) # a starting version and updating it to HEAD. # # NOTHING is implemented yet: every method pair NATIVELY SKIPS (grey check, -# no runner) until a macOS driver exists. The workflow exists now so the -# combination generator (scripts/sandbox/generate-e2e-matrix.mjs) can -# dispatch every declared macOS combination the same way it does for linux -# and windows -- capability knowledge lives here, next to where the driver -# will be, and implementing a method flips this workflow's job-level `if`. +# no runner) until a macOS driver exists. The workflow exists now so +# install-e2e.yml can dispatch every declared macOS combination the same +# way it does for linux and windows -- capability knowledge lives here, +# next to where the driver will be, and implementing a method flips this +# workflow's job-level `if`. # -# Declared methods (see the generator's SPEC): +# Declared methods (see install-e2e.yml's macos-* jobs): # install: curl-bash, packaged-app # update: curl-bash, hermes-update, app-update diff --git a/.github/workflows/install-e2e-run.yml b/.github/workflows/install-e2e-run.yml index 7a2c1f3467f3..1d73af329eb3 100644 --- a/.github/workflows/install-e2e-run.yml +++ b/.github/workflows/install-e2e-run.yml @@ -10,7 +10,7 @@ name: Install & Update E2E (reusable) # is independent: its own sandbox, its own install, nothing rewound or # shared. # -# Method ids come from scripts/sandbox/generate-e2e-matrix.mjs. Supported +# Method ids come from install-e2e.yml's combination jobs. Supported # today: install via curl-bash, update via hermes-update or curl-bash. # Anything else NATIVELY SKIPS (grey check, no runner): capability # knowledge lives here, next to the driver, so the caller can dispatch diff --git a/.github/workflows/install-e2e-tag.yml b/.github/workflows/install-e2e-tag.yml deleted file mode 100644 index 6207f1490d2d..000000000000 --- a/.github/workflows/install-e2e-tag.yml +++ /dev/null @@ -1,103 +0,0 @@ -name: Install & Update E2E — one starting tag (reusable) - -# Everything a user starting on ONE released version could do: for each -# {os, install-method, update-method} combination the support matrix -# (scripts/sandbox/generate-e2e-matrix.mjs) declares, one dispatch that -# installs this tag and updates to HEAD. -# -# Every combination is dispatched to its OS's run workflow; the run -# workflow natively skips (grey) the method pairs its driver cannot run -# yet. Capability knowledge lives next to each driver, never here and -# never in the generator. -# -# Call it: -# -# jobs: -# tag: -# strategy: -# matrix: { install-ref: [v2026.8.3, v2026.3.12] } -# uses: ./.github/workflows/install-e2e-tag.yml -# with: -# install-ref: ${{ matrix.install-ref }} - -on: - workflow_call: - inputs: - install-ref: - description: 'The released version to start from: installed first, then updated to HEAD.' - required: true - type: string - route: - description: 'Combination filter, mirroring the dispatch route choice: all, both, update, installer, windows-desktop.' - required: false - type: string - default: all - -permissions: - contents: read - -jobs: - generate-matrix: - name: expand combinations - runs-on: ubuntu-latest - timeout-minutes: 5 - outputs: - linux: ${{ steps.gen.outputs.linux }} - windows: ${{ steps.gen.outputs.windows }} - macos: ${{ steps.gen.outputs.macos }} - steps: - - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2 - with: - sparse-checkout: scripts/sandbox/generate-e2e-matrix.mjs - sparse-checkout-cone-mode: false - - id: gen - run: | - set -euo pipefail - matrices="$(node scripts/sandbox/generate-e2e-matrix.mjs \ - --tags '["${{ inputs.install-ref }}"]' \ - --route '${{ inputs.route }}')" - echo "$matrices" - for key in linux windows macos; do - echo "$key=$(echo "$matrices" | node -e 'let d="";process.stdin.on("data",c=>d+=c).on("end",()=>console.log(JSON.stringify(JSON.parse(d)[process.argv[1]])))' "$key")" >> "$GITHUB_OUTPUT" - done - - linux: - name: "linux: ${{ matrix.name }}" - if: fromJSON(needs.generate-matrix.outputs.linux).include[0] != null - needs: generate-matrix - strategy: - # One combo breaking is worth knowing about even if another already - # failed, so let every leg report. - fail-fast: false - matrix: ${{ fromJSON(needs.generate-matrix.outputs.linux) }} - uses: ./.github/workflows/install-e2e-run.yml - with: - install-method: ${{ matrix.install_method }} - update-method: ${{ matrix.update_method }} - install-ref: ${{ matrix.install_ref }} - - windows: - name: "windows: ${{ matrix.name }}" - if: fromJSON(needs.generate-matrix.outputs.windows).include[0] != null - needs: generate-matrix - strategy: - fail-fast: false - matrix: ${{ fromJSON(needs.generate-matrix.outputs.windows) }} - uses: ./.github/workflows/install-e2e-windows-run.yml - with: - install-method: ${{ matrix.install_method }} - update-method: ${{ matrix.update_method }} - install-ref: ${{ matrix.install_ref }} - - macos: - name: "macos: ${{ matrix.name }}" - if: fromJSON(needs.generate-matrix.outputs.macos).include[0] != null - needs: generate-matrix - strategy: - fail-fast: false - matrix: ${{ fromJSON(needs.generate-matrix.outputs.macos) }} - uses: ./.github/workflows/install-e2e-macos-run.yml - with: - install-method: ${{ matrix.install_method }} - update-method: ${{ matrix.update_method }} - install-ref: ${{ matrix.install_ref }} diff --git a/.github/workflows/install-e2e.yml b/.github/workflows/install-e2e.yml index 4b5f3c4bfa7f..c571186805e5 100644 --- a/.github/workflows/install-e2e.yml +++ b/.github/workflows/install-e2e.yml @@ -2,36 +2,33 @@ name: Install & Update E2E # Can a user on a released version get to this commit? # -# The support matrix -- every {os, install-method, update-method} combination -# a user could be on -- lives in scripts/sandbox/generate-e2e-matrix.mjs. -# The job graph nests one sub-graph per starting release tag: +# One matrix job per {os, install-method -> update-method} combination a +# user could be on, hardcoded HERE so the workflow is the support matrix +# and the Actions graph draws one box per combination. Each box fans out +# over the release tags picked at runtime, so every leg reads +# " -> HEAD": install that tag the combo's way, update to this commit +# the combo's way. # -# from (install-e2e-tag.yml) -# -> linux: -> one dispatch per combination, a -# real curl|bash install in -# bubblewrap -# (install-e2e-run.yml) -# -> windows: -> one dispatch per combination, the -# real desktop user flow: website -# exe clicked by AutoHotkey, update -# via the app, Playwright clicking -# "Update now" -# (install-e2e-windows-run.yml) -# -> macos: -> one dispatch per combination -# (install-e2e-macos-run.yml) +# Matrix: linux-* a real curl|bash install in the bubblewrap sandbox +# (install-e2e-run.yml) +# Matrix: windows-* the real desktop user flow: website Hermes-Setup.exe +# clicked by AutoHotkey, update via the app, +# Playwright clicking "Update now" +# (install-e2e-windows-run.yml) +# Matrix: macos-* no driver yet (install-e2e-macos-run.yml) # # Every combination is dispatched to its OS's run workflow; the run -# workflow natively skips (grey) the method pairs its driver cannot run -# yet. Capability knowledge lives next to each driver -- the generator and -# the callers know nothing about what is implemented. +# workflow natively skips (grey) what its driver cannot run yet -- an +# unimplemented method pair, or a starting tag that predates the surface +# under test (e.g. windows desktop legs from releases before the desktop +# app existed). Capability knowledge lives next to each driver, never +# here: adding a method = adding one job below; implementing one = flipping +# the run workflow's gate. # # The starting versions are chosen at runtime from the repo's release tags # (scripts/sandbox/pick-release-tags.sh): newest, oldest, and a spread # between. A hardcoded list would stop covering the newest release the day -# after it ships, and would pin an "oldest" that nobody still runs. The tag -# axis is applied to EVERY os: linux legs install the tag in the sandbox; -# windows legs stage serve.git's `main` at it, which is what the published -# installer (no commit pin) then installs. +# after it ships, and would pin an "oldest" that nobody still runs. # # Triggers: # * every 12 hours, so upstream drift (a new uv, a Node bump, a PyPI change) @@ -48,7 +45,7 @@ on: workflow_dispatch: inputs: route: - description: 'Which update route to exercise. all/both include the Windows desktop leg.' + description: 'Which combinations to run. all = every OS; both/update/installer = the linux legs; windows-desktop = the windows legs.' required: false type: choice default: all @@ -75,8 +72,8 @@ concurrency: cancel-in-progress: true jobs: - # Which released versions do we test updating FROM? Resolved once; the - # generator applies the tag axis to every OS's legs. + # Which released versions do we test updating FROM? Resolved once and + # shared by every combination's matrix, so all combos cover the same set. pick-releases: name: Pick release tags runs-on: ubuntu-latest @@ -100,20 +97,240 @@ jobs: echo "Testing updates from: $tags" echo "tags=$tags" >> "$GITHUB_OUTPUT" - # One sub-graph per starting tag: install-e2e-tag.yml expands the - # {os, install-method, update-method} support matrix for that tag and - # dispatches one job per combination to the OS's run workflow, so the - # graph reads "from vX -> windows: install -> update" per leg. - from-tag: - name: "from ${{ matrix.install-ref }}" + # ---- linux: install via curl-bash --------------------------------------- + + linux-curl-bash-to-hermes-update: + name: "linux: curl-bash -> hermes-update" + if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "both", "update"]'), inputs.route) + needs: pick-releases + strategy: + # One leg breaking is worth knowing about even if another already + # failed, so let every leg report. + fail-fast: false + matrix: + install-ref: ${{ fromJSON(needs.pick-releases.outputs.tags) }} + uses: ./.github/workflows/install-e2e-run.yml + with: + install-method: curl-bash + update-method: hermes-update + install-ref: ${{ matrix.install-ref }} + + linux-curl-bash-to-curl-bash: + name: "linux: curl-bash -> curl-bash" + if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "both", "installer"]'), inputs.route) + needs: pick-releases + strategy: + fail-fast: false + max-parallel: 3 + matrix: + install-ref: ${{ fromJSON(needs.pick-releases.outputs.tags) }} + uses: ./.github/workflows/install-e2e-run.yml + with: + install-method: curl-bash + update-method: curl-bash + install-ref: ${{ matrix.install-ref }} + + # ---- windows: install via the website's Hermes-Setup.exe ---------------- + + windows-desktop-installer-to-desktop-app: + name: "windows: desktop-installer@latest -> desktop-app" + if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "windows-desktop"]'), inputs.route) + needs: pick-releases + strategy: + fail-fast: false + # 2x-cost runners at ~14 min a leg: two at a time is throughput + # enough for a 5-tag axis without hogging the Windows pool. + max-parallel: 2 + matrix: + install-ref: ${{ fromJSON(needs.pick-releases.outputs.tags) }} + uses: ./.github/workflows/install-e2e-windows-run.yml + with: + install-method: desktop-installer@latest + update-method: desktop-app + install-ref: ${{ matrix.install-ref }} + + windows-desktop-installer-to-installer-rerun: + name: "windows: desktop-installer@latest -> desktop-installer-rerun@latest" + if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "windows-desktop"]'), inputs.route) + needs: pick-releases + strategy: + fail-fast: false + matrix: + install-ref: ${{ fromJSON(needs.pick-releases.outputs.tags) }} + uses: ./.github/workflows/install-e2e-windows-run.yml + with: + install-method: desktop-installer@latest + update-method: desktop-installer-rerun@latest + install-ref: ${{ matrix.install-ref }} + + windows-desktop-installer-to-hermes-update: + name: "windows: desktop-installer@latest -> hermes-update" + if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "windows-desktop"]'), inputs.route) + needs: pick-releases + strategy: + fail-fast: false + matrix: + install-ref: ${{ fromJSON(needs.pick-releases.outputs.tags) }} + uses: ./.github/workflows/install-e2e-windows-run.yml + with: + install-method: desktop-installer@latest + update-method: hermes-update + install-ref: ${{ matrix.install-ref }} + + windows-desktop-installer-to-irm-iex: + name: "windows: desktop-installer@latest -> irm-iex" + if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "windows-desktop"]'), inputs.route) + needs: pick-releases + strategy: + fail-fast: false + matrix: + install-ref: ${{ fromJSON(needs.pick-releases.outputs.tags) }} + uses: ./.github/workflows/install-e2e-windows-run.yml + with: + install-method: desktop-installer@latest + update-method: irm-iex + install-ref: ${{ matrix.install-ref }} + + # ---- windows: install via irm | iex -------------------------------------- + + windows-irm-iex-to-desktop-app: + name: "windows: irm-iex -> desktop-app" + if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "windows-desktop"]'), inputs.route) + needs: pick-releases + strategy: + fail-fast: false + matrix: + install-ref: ${{ fromJSON(needs.pick-releases.outputs.tags) }} + uses: ./.github/workflows/install-e2e-windows-run.yml + with: + install-method: irm-iex + update-method: desktop-app + install-ref: ${{ matrix.install-ref }} + + windows-irm-iex-to-installer-rerun: + name: "windows: irm-iex -> desktop-installer-rerun@latest" + if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "windows-desktop"]'), inputs.route) + needs: pick-releases + strategy: + fail-fast: false + matrix: + install-ref: ${{ fromJSON(needs.pick-releases.outputs.tags) }} + uses: ./.github/workflows/install-e2e-windows-run.yml + with: + install-method: irm-iex + update-method: desktop-installer-rerun@latest + install-ref: ${{ matrix.install-ref }} + + windows-irm-iex-to-hermes-update: + name: "windows: irm-iex -> hermes-update" + if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "windows-desktop"]'), inputs.route) + needs: pick-releases + strategy: + fail-fast: false + matrix: + install-ref: ${{ fromJSON(needs.pick-releases.outputs.tags) }} + uses: ./.github/workflows/install-e2e-windows-run.yml + with: + install-method: irm-iex + update-method: hermes-update + install-ref: ${{ matrix.install-ref }} + + windows-irm-iex-to-irm-iex: + name: "windows: irm-iex -> irm-iex" + if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "windows-desktop"]'), inputs.route) + needs: pick-releases + strategy: + fail-fast: false + matrix: + install-ref: ${{ fromJSON(needs.pick-releases.outputs.tags) }} + uses: ./.github/workflows/install-e2e-windows-run.yml + with: + install-method: irm-iex + update-method: irm-iex + install-ref: ${{ matrix.install-ref }} + + # ---- macos: no driver yet (every leg natively skips in the run workflow) - + + macos-curl-bash-to-curl-bash: + name: "macos: curl-bash -> curl-bash" + if: github.event_name != 'workflow_dispatch' || inputs.route == 'all' + needs: pick-releases + strategy: + fail-fast: false + matrix: + install-ref: ${{ fromJSON(needs.pick-releases.outputs.tags) }} + uses: ./.github/workflows/install-e2e-macos-run.yml + with: + install-method: curl-bash + update-method: curl-bash + install-ref: ${{ matrix.install-ref }} + + macos-curl-bash-to-hermes-update: + name: "macos: curl-bash -> hermes-update" + if: github.event_name != 'workflow_dispatch' || inputs.route == 'all' + needs: pick-releases + strategy: + fail-fast: false + matrix: + install-ref: ${{ fromJSON(needs.pick-releases.outputs.tags) }} + uses: ./.github/workflows/install-e2e-macos-run.yml + with: + install-method: curl-bash + update-method: hermes-update + install-ref: ${{ matrix.install-ref }} + + macos-curl-bash-to-app-update: + name: "macos: curl-bash -> app-update" + if: github.event_name != 'workflow_dispatch' || inputs.route == 'all' + needs: pick-releases + strategy: + fail-fast: false + matrix: + install-ref: ${{ fromJSON(needs.pick-releases.outputs.tags) }} + uses: ./.github/workflows/install-e2e-macos-run.yml + with: + install-method: curl-bash + update-method: app-update + install-ref: ${{ matrix.install-ref }} + + macos-packaged-app-to-curl-bash: + name: "macos: packaged-app -> curl-bash" + if: github.event_name != 'workflow_dispatch' || inputs.route == 'all' + needs: pick-releases + strategy: + fail-fast: false + matrix: + install-ref: ${{ fromJSON(needs.pick-releases.outputs.tags) }} + uses: ./.github/workflows/install-e2e-macos-run.yml + with: + install-method: packaged-app + update-method: curl-bash + install-ref: ${{ matrix.install-ref }} + + macos-packaged-app-to-hermes-update: + name: "macos: packaged-app -> hermes-update" + if: github.event_name != 'workflow_dispatch' || inputs.route == 'all' + needs: pick-releases + strategy: + fail-fast: false + matrix: + install-ref: ${{ fromJSON(needs.pick-releases.outputs.tags) }} + uses: ./.github/workflows/install-e2e-macos-run.yml + with: + install-method: packaged-app + update-method: hermes-update + install-ref: ${{ matrix.install-ref }} + + macos-packaged-app-to-app-update: + name: "macos: packaged-app -> app-update" + if: github.event_name != 'workflow_dispatch' || inputs.route == 'all' needs: pick-releases strategy: - # One tag breaking is worth knowing about even if another already - # failed, so let every sub-graph report. fail-fast: false matrix: install-ref: ${{ fromJSON(needs.pick-releases.outputs.tags) }} - uses: ./.github/workflows/install-e2e-tag.yml + uses: ./.github/workflows/install-e2e-macos-run.yml with: + install-method: packaged-app + update-method: app-update install-ref: ${{ matrix.install-ref }} - route: ${{ inputs.route || 'all' }} diff --git a/scripts/sandbox/generate-e2e-matrix.mjs b/scripts/sandbox/generate-e2e-matrix.mjs deleted file mode 100644 index 4dfcf76d2d3a..000000000000 --- a/scripts/sandbox/generate-e2e-matrix.mjs +++ /dev/null @@ -1,212 +0,0 @@ -#!/usr/bin/env node -/** - * Expand the install/update support matrix into concrete E2E combinations. - * - * One source of truth for every {os, install-method, update-method} pair a - * user could be on. This file only DECLARES and EXPANDS: it knows nothing - * about which combinations CI can drive. Every combination is dispatched to - * its OS's run workflow, and THAT workflow natively skips the method pairs - * its driver cannot run yet -- capability knowledge lives next to each - * driver (install-e2e-run.yml, install-e2e-windows-run.yml, - * install-e2e-macos-run.yml). - * - * Used by .github/workflows/install-e2e-tag.yml, which runs it once per - * starting release tag: - * - * node scripts/sandbox/generate-e2e-matrix.mjs \ - * --tags '["v2026.8.3"]' --route all - * - * Prints JSON: { linux: {include:[...]}, windows: {include:[...]}, - * macos: {include:[...]} } -- every entry is - * {name, install_method, update_method, install_ref}. - */ - -import path from 'node:path'; -import { parseArgs } from 'node:util'; -import { fileURLToPath } from 'node:url'; - -/** - * Method ids are strict machine strings -- workflows and the IMPLEMENTED - * table key off them, so an unknown id must fail loudly (see KNOWN_METHODS). - * - * `versions` on an entry expands it into one combination per version. Only - * "latest" is allowed today: it means "the artifact published on the website - * right now" (Hermes-Setup.exe has no versioned archive yet). When archived - * installer versions exist, widen ALLOWED_VERSIONS. - */ -export const SPEC = { - windows: { - install: [ - // irm https://hermes.nousresearch.com/install.ps1 | iex - { method: 'irm-iex' }, - // Website Hermes-Setup.exe, clicked through the GUI. - { method: 'desktop-installer', versions: ['latest'] }, - ], - update: [ - { method: 'irm-iex' }, - // Run the bootstrap exe again over an existing install (--update flow). - { method: 'desktop-installer-rerun', versions: ['latest'] }, - { method: 'hermes-update' }, - // Settings -> About -> "Update now" inside the running desktop app. - { method: 'desktop-app' }, - ], - }, - macos: { - install: [ - { method: 'curl-bash' }, - { method: 'packaged-app' }, - ], - update: [ - { method: 'curl-bash' }, - { method: 'hermes-update' }, - { method: 'app-update' }, - ], - }, - linux: { - install: [ - { method: 'curl-bash' }, - ], - update: [ - { method: 'curl-bash' }, - { method: 'hermes-update' }, - ], - }, -}; - -const KNOWN_METHODS = new Set([ - 'irm-iex', - 'desktop-installer', - 'desktop-installer-rerun', - 'hermes-update', - 'desktop-app', - 'curl-bash', - 'packaged-app', - 'app-update', -]); - -const ALLOWED_VERSIONS = new Set(['latest']); - -function validateEntry(os, kind, entry) { - if (typeof entry.method !== 'string' || !KNOWN_METHODS.has(entry.method)) { - throw new Error(`${os}.${kind}: unknown method id ${JSON.stringify(entry.method)} -- add it to KNOWN_METHODS if intentional`); - } - if ('versions' in entry) { - if (!Array.isArray(entry.versions) || entry.versions.length === 0) { - throw new Error(`${os}.${kind}.${entry.method}: versions must be a non-empty array`); - } - for (const v of entry.versions) { - if (!ALLOWED_VERSIONS.has(v)) { - throw new Error(`${os}.${kind}.${entry.method}: version ${JSON.stringify(v)} not allowed -- only ${[...ALLOWED_VERSIONS].join(', ')} until versioned installer archives exist`); - } - } - } - const unknown = Object.keys(entry).filter((k) => k !== 'method' && k !== 'versions'); - if (unknown.length) { - throw new Error(`${os}.${kind}.${entry.method}: unknown keys ${unknown.join(', ')}`); - } -} - -/** Expand one method entry into concrete ids ("desktop-installer@latest"). */ -export function expandMethod(os, kind, entry) { - validateEntry(os, kind, entry); - if (!entry.versions) return [entry.method]; - return entry.versions.map((v) => `${entry.method}@${v}`); -} - -/** Every {os, install, update, secondUpdate} combination in SPEC. */ -export function generateEnvironments(spec) { - const envs = []; - for (const [os, osSpec] of Object.entries(spec)) { - const { install, update, secondUpdate = [], ...unknown } = osSpec; - if (Object.keys(unknown).length) { - throw new Error(`${os}: unknown spec keys ${Object.keys(unknown).join(', ')}`); - } - // Chained second updates (install -> update -> update again) are a real - // axis -- the updater that RESULTS from an update must itself update -- - // but nothing implements them yet. Refuse a spec that declares them so - // the first implementation is forced to come through here. - if (!Array.isArray(secondUpdate) || secondUpdate.length !== 0) { - throw new Error(`${os}: secondUpdate must be empty until a second-update leg is implemented`); - } - const installs = install.flatMap((e) => expandMethod(os, 'install', e)); - const updates = update.flatMap((e) => expandMethod(os, 'update', e)); - for (const i of installs) { - for (const u of updates) { - envs.push({ os, install: i, update: u, secondUpdate: '' }); - } - } - } - return envs; -} - -/** Mirror of install-e2e.yml's dispatch `route` choice. */ -function routeWants(route, env) { - switch (route) { - case 'all': - return true; - case 'both': - return env.os === 'linux'; - case 'update': - return env.os === 'linux' && env.update === 'hermes-update'; - case 'installer': - return env.os === 'linux' && env.update === 'curl-bash'; - case 'windows-desktop': - return env.os === 'windows'; - default: - throw new Error(`unknown route filter: ${JSON.stringify(route)}`); - } -} - -/** - * Split the combinations into one matrix per OS. - * - * `tags` (the released versions we test updating FROM) is the OUTER axis: - * for each tag, for each combination, one dispatch that installs the tag - * and updates to HEAD. No capability filtering happens here -- every - * declared combination is dispatched, and the OS's run workflow natively - * skips what its driver cannot run yet. - */ -export function buildMatrices(envs, { tags = [], route = 'all' } = {}) { - const byOs = { linux: [], windows: [], macos: [] }; - for (const env of envs) { - if (!routeWants(route, env)) continue; - const bucket = byOs[env.os]; - if (!bucket) { - throw new Error(`no matrix bucket for os ${JSON.stringify(env.os)}`); - } - if (tags.length === 0) { - throw new Error('combinations were selected but no --tags given'); - } - for (const tag of tags) { - bucket.push({ - name: `${env.install} -> ${env.update}`, - install_method: env.install, - update_method: env.update, - install_ref: tag, - }); - } - } - return Object.fromEntries( - Object.entries(byOs).map(([os, include]) => [os, { include }]), - ); -} - -function main() { - const { values } = parseArgs({ - options: { - tags: { type: 'string', default: '[]' }, - route: { type: 'string', default: 'all' }, - }, - }); - const tags = JSON.parse(values.tags); - if (!Array.isArray(tags) || !tags.every((t) => typeof t === 'string')) { - throw new Error('--tags must be a JSON array of strings'); - } - const envs = generateEnvironments(SPEC); - const matrices = buildMatrices(envs, { tags, route: values.route }); - process.stdout.write(`${JSON.stringify(matrices, null, 2)}\n`); -} - -if (process.argv[1] && fileURLToPath(import.meta.url) === path.resolve(process.argv[1])) { - main(); -} diff --git a/tests/install/windows-desktop-gui-e2e.ps1 b/tests/install/windows-desktop-gui-e2e.ps1 index e95aa783d0a2..0813903f3646 100644 --- a/tests/install/windows-desktop-gui-e2e.ps1 +++ b/tests/install/windows-desktop-gui-e2e.ps1 @@ -72,10 +72,9 @@ param( [string]$Phase = "all", # Update method to exercise in the update-gui phase, named by the same - # ids the combination generator (scripts/sandbox/generate-e2e-matrix - # .mjs) uses. Only "desktop-app" (the app's own Update button) is - # implemented; the others are declared arms so the surface is stable - # when they land. + # ids install-e2e.yml's combination jobs use. Only "desktop-app" (the + # app's own Update button) is implemented; the others are declared arms + # so the surface is stable when they land. [ValidateSet("desktop-app", "hermes-update", "desktop-installer-rerun@latest", "irm-iex")] [string]$Route = "desktop-app", From 83e9f883d8029d47a07f13cdce73fbf909484b9f Mon Sep 17 00:00:00 2001 From: ethernet Date: Tue, 11 Aug 2026 16:43:26 -0400 Subject: [PATCH 049/227] ci(install-e2e): per-leg names are just the transition; tag capability annotated at pick time Graph polish + one structural simplification, after the first render of the combo-box layout: * Leg names: every combination job's display name is now '${{ matrix.tag.ref }} -> HEAD' - the box title (job id) already carries os+methods, so repeating them per leg was noise. The inner job renders as a short static 'e2e' tail (dynamic names render unexpanded on skipped jobs, so it must stay static). * The windows probe job is gone: pick-releases now annotates each picked tag with whether its tree ships apps/desktop ({ref, desktop} objects in the matrix), and the windows run workflow gates on the new tag-has-desktop boolean input directly. One tree listing at pick time replaces N probe jobs, and the 'probe tag' noise disappears from the graph. Annotation loop verified against the real tag set (pre/post-desktop split lands exactly at the app's introduction); 16-combo inventory re-asserted; all four workflows pass actionlint. --- .github/workflows/install-e2e-macos-run.yml | 2 +- .github/workflows/install-e2e-run.yml | 10 +- .github/workflows/install-e2e-windows-run.yml | 52 ++------ .github/workflows/install-e2e.yml | 116 ++++++++++-------- 4 files changed, 84 insertions(+), 96 deletions(-) diff --git a/.github/workflows/install-e2e-macos-run.yml b/.github/workflows/install-e2e-macos-run.yml index e23dcab79ebc..450cf42bc0a2 100644 --- a/.github/workflows/install-e2e-macos-run.yml +++ b/.github/workflows/install-e2e-macos-run.yml @@ -39,7 +39,7 @@ jobs: # Static name on purpose: the caller's job name already carries the # method pair, and GitHub renders name expressions UNEXPANDED (literal # "${{ inputs... }}") on natively skipped jobs. - name: run + name: e2e # No macOS driver exists yet: every pair is a declared TODO, so this is # constant-false until the first method lands. Written as an impossible # input comparison rather than `if: false` because actionlint rejects diff --git a/.github/workflows/install-e2e-run.yml b/.github/workflows/install-e2e-run.yml index 1d73af329eb3..c7077870ca90 100644 --- a/.github/workflows/install-e2e-run.yml +++ b/.github/workflows/install-e2e-run.yml @@ -60,11 +60,11 @@ jobs: e2e: # Static name on purpose: the caller's job name already carries the # method pair, and GitHub renders name expressions UNEXPANDED (literal - # "${{ inputs... }}") on natively skipped jobs. - name: run - # The pairs the sandbox driver can run today. Anything else is a - # declared TODO: native skip, so the coverage gap is a grey check on - # every run. + # "${{ inputs... }}") on natively skipped jobs. Short because it is + # only a rendered tail (" -> HEAD / e2e"). + name: e2e + # The pairs the sandbox driver can run today; anything else is a + # declared TODO and natively skips. if: inputs.install-method == 'curl-bash' && contains(fromJSON('["hermes-update", "curl-bash"]'), inputs.update-method) runs-on: ${{ inputs.runner }} timeout-minutes: ${{ inputs.timeout-minutes }} diff --git a/.github/workflows/install-e2e-windows-run.yml b/.github/workflows/install-e2e-windows-run.yml index c1e9a63ffd0a..fdac6d6ce0ab 100644 --- a/.github/workflows/install-e2e-windows-run.yml +++ b/.github/workflows/install-e2e-windows-run.yml @@ -57,6 +57,11 @@ on: required: false type: string default: auto + tag-has-desktop: + description: 'Whether install-ref ships the desktop app (apps/desktop). The caller annotates this from the tag''s own tree; desktop-method legs from pre-desktop releases natively skip.' + required: false + type: boolean + default: true setup-exe-url: description: 'Bootstrap installer to install OLD with. Default: the latest published one — what a user downloads today.' required: false @@ -72,52 +77,17 @@ permissions: contents: read jobs: - # Can this starting version run this flow at all? The desktop app - # (apps/desktop) only exists in releases from v2026.5.31 on (#20059): - # older tags have no window to launch and no Update button to click, so - # desktop-method legs from them natively skip below. Data-driven from the - # tag's own tree -- no hardcoded version list to rot. - probe: - name: probe tag - if: inputs.install-method == 'desktop-installer@latest' && inputs.update-method == 'desktop-app' - runs-on: ubuntu-latest - timeout-minutes: 5 - outputs: - has-desktop: ${{ steps.probe.outputs.has-desktop }} - steps: - # Tree listings only: no blobs, full history + tags so any ref - # resolves. - - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2 - with: - fetch-depth: 0 - filter: blob:none - - id: probe - run: | - set -euo pipefail - ref='${{ inputs.install-ref }}' - if [ "$ref" = "auto" ]; then - # Same choice the driver's stage phase makes: newest release tag. - ref="$(git tag --list 'v[0-9]*' --sort=-creatordate | head -1)" - fi - if git ls-tree -d "$ref" apps/desktop | grep -q .; then - has=true - else - has=false - fi - echo "ref $ref: has-desktop=$has" - echo "has-desktop=$has" >> "$GITHUB_OUTPUT" - e2e: # Static name on purpose: the caller's job name already carries the # method pair, and GitHub renders name expressions UNEXPANDED (literal # "${{ inputs... }}") on natively skipped jobs. - name: run - needs: probe + name: e2e # The one pair the driver can run today, and only from a starting - # version that ships the desktop app. Anything else is a native skip: - # a declared-TODO method pair, or a tag from before the app existed. - # (When probe itself skipped, its output is empty and this is false.) - if: inputs.install-method == 'desktop-installer@latest' && inputs.update-method == 'desktop-app' && needs.probe.outputs.has-desktop == 'true' + # version that ships the desktop app (the caller annotates + # tag-has-desktop from the tag's own tree; releases before #20059 have + # no window to launch and no Update button to click). Anything else is + # a native skip: a declared-TODO method pair, or a pre-desktop tag. + if: inputs.install-method == 'desktop-installer@latest' && inputs.update-method == 'desktop-app' && inputs.tag-has-desktop runs-on: windows-latest timeout-minutes: ${{ inputs.timeout-minutes }} diff --git a/.github/workflows/install-e2e.yml b/.github/workflows/install-e2e.yml index c571186805e5..2a92db7578ee 100644 --- a/.github/workflows/install-e2e.yml +++ b/.github/workflows/install-e2e.yml @@ -95,12 +95,22 @@ jobs: set -euo pipefail tags="$(scripts/sandbox/pick-release-tags.sh --count '${{ inputs.tag-count || 5 }}')" echo "Testing updates from: $tags" - echo "tags=$tags" >> "$GITHUB_OUTPUT" + # Annotate each tag with what its own tree supports, so legs can + # natively skip surfaces the starting version does not have. + # Today: does the release ship the desktop app (apps/desktop, + # #20059)? Cheaper and simpler here -- the tags are already + # checked out -- than a probe job per windows leg. + enriched="$(for t in $(echo "$tags" | jq -r '.[]'); do + if git ls-tree -d "$t" apps/desktop | grep -q .; then d=true; else d=false; fi + echo "{\"ref\":\"$t\",\"desktop\":$d}" + done | jq -sc .)" + echo "Annotated: $enriched" + echo "tags=$enriched" >> "$GITHUB_OUTPUT" # ---- linux: install via curl-bash --------------------------------------- linux-curl-bash-to-hermes-update: - name: "linux: curl-bash -> hermes-update" + name: "${{ matrix.tag.ref }} -> HEAD" if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "both", "update"]'), inputs.route) needs: pick-releases strategy: @@ -108,32 +118,32 @@ jobs: # failed, so let every leg report. fail-fast: false matrix: - install-ref: ${{ fromJSON(needs.pick-releases.outputs.tags) }} + tag: ${{ fromJSON(needs.pick-releases.outputs.tags) }} uses: ./.github/workflows/install-e2e-run.yml with: install-method: curl-bash update-method: hermes-update - install-ref: ${{ matrix.install-ref }} + install-ref: ${{ matrix.tag.ref }} linux-curl-bash-to-curl-bash: - name: "linux: curl-bash -> curl-bash" + name: "${{ matrix.tag.ref }} -> HEAD" if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "both", "installer"]'), inputs.route) needs: pick-releases strategy: fail-fast: false max-parallel: 3 matrix: - install-ref: ${{ fromJSON(needs.pick-releases.outputs.tags) }} + tag: ${{ fromJSON(needs.pick-releases.outputs.tags) }} uses: ./.github/workflows/install-e2e-run.yml with: install-method: curl-bash update-method: curl-bash - install-ref: ${{ matrix.install-ref }} + install-ref: ${{ matrix.tag.ref }} # ---- windows: install via the website's Hermes-Setup.exe ---------------- windows-desktop-installer-to-desktop-app: - name: "windows: desktop-installer@latest -> desktop-app" + name: "${{ matrix.tag.ref }} -> HEAD" if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "windows-desktop"]'), inputs.route) needs: pick-releases strategy: @@ -142,195 +152,203 @@ jobs: # enough for a 5-tag axis without hogging the Windows pool. max-parallel: 2 matrix: - install-ref: ${{ fromJSON(needs.pick-releases.outputs.tags) }} + tag: ${{ fromJSON(needs.pick-releases.outputs.tags) }} uses: ./.github/workflows/install-e2e-windows-run.yml with: + tag-has-desktop: ${{ matrix.tag.desktop }} install-method: desktop-installer@latest update-method: desktop-app - install-ref: ${{ matrix.install-ref }} + install-ref: ${{ matrix.tag.ref }} windows-desktop-installer-to-installer-rerun: - name: "windows: desktop-installer@latest -> desktop-installer-rerun@latest" + name: "${{ matrix.tag.ref }} -> HEAD" if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "windows-desktop"]'), inputs.route) needs: pick-releases strategy: fail-fast: false matrix: - install-ref: ${{ fromJSON(needs.pick-releases.outputs.tags) }} + tag: ${{ fromJSON(needs.pick-releases.outputs.tags) }} uses: ./.github/workflows/install-e2e-windows-run.yml with: + tag-has-desktop: ${{ matrix.tag.desktop }} install-method: desktop-installer@latest update-method: desktop-installer-rerun@latest - install-ref: ${{ matrix.install-ref }} + install-ref: ${{ matrix.tag.ref }} windows-desktop-installer-to-hermes-update: - name: "windows: desktop-installer@latest -> hermes-update" + name: "${{ matrix.tag.ref }} -> HEAD" if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "windows-desktop"]'), inputs.route) needs: pick-releases strategy: fail-fast: false matrix: - install-ref: ${{ fromJSON(needs.pick-releases.outputs.tags) }} + tag: ${{ fromJSON(needs.pick-releases.outputs.tags) }} uses: ./.github/workflows/install-e2e-windows-run.yml with: + tag-has-desktop: ${{ matrix.tag.desktop }} install-method: desktop-installer@latest update-method: hermes-update - install-ref: ${{ matrix.install-ref }} + install-ref: ${{ matrix.tag.ref }} windows-desktop-installer-to-irm-iex: - name: "windows: desktop-installer@latest -> irm-iex" + name: "${{ matrix.tag.ref }} -> HEAD" if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "windows-desktop"]'), inputs.route) needs: pick-releases strategy: fail-fast: false matrix: - install-ref: ${{ fromJSON(needs.pick-releases.outputs.tags) }} + tag: ${{ fromJSON(needs.pick-releases.outputs.tags) }} uses: ./.github/workflows/install-e2e-windows-run.yml with: + tag-has-desktop: ${{ matrix.tag.desktop }} install-method: desktop-installer@latest update-method: irm-iex - install-ref: ${{ matrix.install-ref }} + install-ref: ${{ matrix.tag.ref }} # ---- windows: install via irm | iex -------------------------------------- windows-irm-iex-to-desktop-app: - name: "windows: irm-iex -> desktop-app" + name: "${{ matrix.tag.ref }} -> HEAD" if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "windows-desktop"]'), inputs.route) needs: pick-releases strategy: fail-fast: false matrix: - install-ref: ${{ fromJSON(needs.pick-releases.outputs.tags) }} + tag: ${{ fromJSON(needs.pick-releases.outputs.tags) }} uses: ./.github/workflows/install-e2e-windows-run.yml with: + tag-has-desktop: ${{ matrix.tag.desktop }} install-method: irm-iex update-method: desktop-app - install-ref: ${{ matrix.install-ref }} + install-ref: ${{ matrix.tag.ref }} windows-irm-iex-to-installer-rerun: - name: "windows: irm-iex -> desktop-installer-rerun@latest" + name: "${{ matrix.tag.ref }} -> HEAD" if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "windows-desktop"]'), inputs.route) needs: pick-releases strategy: fail-fast: false matrix: - install-ref: ${{ fromJSON(needs.pick-releases.outputs.tags) }} + tag: ${{ fromJSON(needs.pick-releases.outputs.tags) }} uses: ./.github/workflows/install-e2e-windows-run.yml with: + tag-has-desktop: ${{ matrix.tag.desktop }} install-method: irm-iex update-method: desktop-installer-rerun@latest - install-ref: ${{ matrix.install-ref }} + install-ref: ${{ matrix.tag.ref }} windows-irm-iex-to-hermes-update: - name: "windows: irm-iex -> hermes-update" + name: "${{ matrix.tag.ref }} -> HEAD" if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "windows-desktop"]'), inputs.route) needs: pick-releases strategy: fail-fast: false matrix: - install-ref: ${{ fromJSON(needs.pick-releases.outputs.tags) }} + tag: ${{ fromJSON(needs.pick-releases.outputs.tags) }} uses: ./.github/workflows/install-e2e-windows-run.yml with: + tag-has-desktop: ${{ matrix.tag.desktop }} install-method: irm-iex update-method: hermes-update - install-ref: ${{ matrix.install-ref }} + install-ref: ${{ matrix.tag.ref }} windows-irm-iex-to-irm-iex: - name: "windows: irm-iex -> irm-iex" + name: "${{ matrix.tag.ref }} -> HEAD" if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "windows-desktop"]'), inputs.route) needs: pick-releases strategy: fail-fast: false matrix: - install-ref: ${{ fromJSON(needs.pick-releases.outputs.tags) }} + tag: ${{ fromJSON(needs.pick-releases.outputs.tags) }} uses: ./.github/workflows/install-e2e-windows-run.yml with: + tag-has-desktop: ${{ matrix.tag.desktop }} install-method: irm-iex update-method: irm-iex - install-ref: ${{ matrix.install-ref }} + install-ref: ${{ matrix.tag.ref }} # ---- macos: no driver yet (every leg natively skips in the run workflow) - macos-curl-bash-to-curl-bash: - name: "macos: curl-bash -> curl-bash" + name: "${{ matrix.tag.ref }} -> HEAD" if: github.event_name != 'workflow_dispatch' || inputs.route == 'all' needs: pick-releases strategy: fail-fast: false matrix: - install-ref: ${{ fromJSON(needs.pick-releases.outputs.tags) }} + tag: ${{ fromJSON(needs.pick-releases.outputs.tags) }} uses: ./.github/workflows/install-e2e-macos-run.yml with: install-method: curl-bash update-method: curl-bash - install-ref: ${{ matrix.install-ref }} + install-ref: ${{ matrix.tag.ref }} macos-curl-bash-to-hermes-update: - name: "macos: curl-bash -> hermes-update" + name: "${{ matrix.tag.ref }} -> HEAD" if: github.event_name != 'workflow_dispatch' || inputs.route == 'all' needs: pick-releases strategy: fail-fast: false matrix: - install-ref: ${{ fromJSON(needs.pick-releases.outputs.tags) }} + tag: ${{ fromJSON(needs.pick-releases.outputs.tags) }} uses: ./.github/workflows/install-e2e-macos-run.yml with: install-method: curl-bash update-method: hermes-update - install-ref: ${{ matrix.install-ref }} + install-ref: ${{ matrix.tag.ref }} macos-curl-bash-to-app-update: - name: "macos: curl-bash -> app-update" + name: "${{ matrix.tag.ref }} -> HEAD" if: github.event_name != 'workflow_dispatch' || inputs.route == 'all' needs: pick-releases strategy: fail-fast: false matrix: - install-ref: ${{ fromJSON(needs.pick-releases.outputs.tags) }} + tag: ${{ fromJSON(needs.pick-releases.outputs.tags) }} uses: ./.github/workflows/install-e2e-macos-run.yml with: install-method: curl-bash update-method: app-update - install-ref: ${{ matrix.install-ref }} + install-ref: ${{ matrix.tag.ref }} macos-packaged-app-to-curl-bash: - name: "macos: packaged-app -> curl-bash" + name: "${{ matrix.tag.ref }} -> HEAD" if: github.event_name != 'workflow_dispatch' || inputs.route == 'all' needs: pick-releases strategy: fail-fast: false matrix: - install-ref: ${{ fromJSON(needs.pick-releases.outputs.tags) }} + tag: ${{ fromJSON(needs.pick-releases.outputs.tags) }} uses: ./.github/workflows/install-e2e-macos-run.yml with: install-method: packaged-app update-method: curl-bash - install-ref: ${{ matrix.install-ref }} + install-ref: ${{ matrix.tag.ref }} macos-packaged-app-to-hermes-update: - name: "macos: packaged-app -> hermes-update" + name: "${{ matrix.tag.ref }} -> HEAD" if: github.event_name != 'workflow_dispatch' || inputs.route == 'all' needs: pick-releases strategy: fail-fast: false matrix: - install-ref: ${{ fromJSON(needs.pick-releases.outputs.tags) }} + tag: ${{ fromJSON(needs.pick-releases.outputs.tags) }} uses: ./.github/workflows/install-e2e-macos-run.yml with: install-method: packaged-app update-method: hermes-update - install-ref: ${{ matrix.install-ref }} + install-ref: ${{ matrix.tag.ref }} macos-packaged-app-to-app-update: - name: "macos: packaged-app -> app-update" + name: "${{ matrix.tag.ref }} -> HEAD" if: github.event_name != 'workflow_dispatch' || inputs.route == 'all' needs: pick-releases strategy: fail-fast: false matrix: - install-ref: ${{ fromJSON(needs.pick-releases.outputs.tags) }} + tag: ${{ fromJSON(needs.pick-releases.outputs.tags) }} uses: ./.github/workflows/install-e2e-macos-run.yml with: install-method: packaged-app update-method: app-update - install-ref: ${{ matrix.install-ref }} + install-ref: ${{ matrix.tag.ref }} From 4e886166ace71290e18abb42d30a2e8c52dca3e3 Mon Sep 17 00:00:00 2001 From: ethernet Date: Tue, 11 Aug 2026 16:51:28 -0400 Subject: [PATCH 050/227] ci(install-e2e): leg names carry everything - os, method pair, tag transition --- .github/workflows/install-e2e.yml | 32 +++++++++++++++---------------- 1 file changed, 16 insertions(+), 16 deletions(-) diff --git a/.github/workflows/install-e2e.yml b/.github/workflows/install-e2e.yml index 2a92db7578ee..bf29956dd231 100644 --- a/.github/workflows/install-e2e.yml +++ b/.github/workflows/install-e2e.yml @@ -110,7 +110,7 @@ jobs: # ---- linux: install via curl-bash --------------------------------------- linux-curl-bash-to-hermes-update: - name: "${{ matrix.tag.ref }} -> HEAD" + name: "linux: curl-bash -> hermes-update (${{ matrix.tag.ref }} -> HEAD)" if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "both", "update"]'), inputs.route) needs: pick-releases strategy: @@ -126,7 +126,7 @@ jobs: install-ref: ${{ matrix.tag.ref }} linux-curl-bash-to-curl-bash: - name: "${{ matrix.tag.ref }} -> HEAD" + name: "linux: curl-bash -> curl-bash (${{ matrix.tag.ref }} -> HEAD)" if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "both", "installer"]'), inputs.route) needs: pick-releases strategy: @@ -143,7 +143,7 @@ jobs: # ---- windows: install via the website's Hermes-Setup.exe ---------------- windows-desktop-installer-to-desktop-app: - name: "${{ matrix.tag.ref }} -> HEAD" + name: "windows: desktop-installer@latest -> desktop-app (${{ matrix.tag.ref }} -> HEAD)" if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "windows-desktop"]'), inputs.route) needs: pick-releases strategy: @@ -161,7 +161,7 @@ jobs: install-ref: ${{ matrix.tag.ref }} windows-desktop-installer-to-installer-rerun: - name: "${{ matrix.tag.ref }} -> HEAD" + name: "windows: desktop-installer@latest -> desktop-installer-rerun@latest (${{ matrix.tag.ref }} -> HEAD)" if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "windows-desktop"]'), inputs.route) needs: pick-releases strategy: @@ -176,7 +176,7 @@ jobs: install-ref: ${{ matrix.tag.ref }} windows-desktop-installer-to-hermes-update: - name: "${{ matrix.tag.ref }} -> HEAD" + name: "windows: desktop-installer@latest -> hermes-update (${{ matrix.tag.ref }} -> HEAD)" if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "windows-desktop"]'), inputs.route) needs: pick-releases strategy: @@ -191,7 +191,7 @@ jobs: install-ref: ${{ matrix.tag.ref }} windows-desktop-installer-to-irm-iex: - name: "${{ matrix.tag.ref }} -> HEAD" + name: "windows: desktop-installer@latest -> irm-iex (${{ matrix.tag.ref }} -> HEAD)" if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "windows-desktop"]'), inputs.route) needs: pick-releases strategy: @@ -208,7 +208,7 @@ jobs: # ---- windows: install via irm | iex -------------------------------------- windows-irm-iex-to-desktop-app: - name: "${{ matrix.tag.ref }} -> HEAD" + name: "windows: irm-iex -> desktop-app (${{ matrix.tag.ref }} -> HEAD)" if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "windows-desktop"]'), inputs.route) needs: pick-releases strategy: @@ -223,7 +223,7 @@ jobs: install-ref: ${{ matrix.tag.ref }} windows-irm-iex-to-installer-rerun: - name: "${{ matrix.tag.ref }} -> HEAD" + name: "windows: irm-iex -> desktop-installer-rerun@latest (${{ matrix.tag.ref }} -> HEAD)" if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "windows-desktop"]'), inputs.route) needs: pick-releases strategy: @@ -238,7 +238,7 @@ jobs: install-ref: ${{ matrix.tag.ref }} windows-irm-iex-to-hermes-update: - name: "${{ matrix.tag.ref }} -> HEAD" + name: "windows: irm-iex -> hermes-update (${{ matrix.tag.ref }} -> HEAD)" if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "windows-desktop"]'), inputs.route) needs: pick-releases strategy: @@ -253,7 +253,7 @@ jobs: install-ref: ${{ matrix.tag.ref }} windows-irm-iex-to-irm-iex: - name: "${{ matrix.tag.ref }} -> HEAD" + name: "windows: irm-iex -> irm-iex (${{ matrix.tag.ref }} -> HEAD)" if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "windows-desktop"]'), inputs.route) needs: pick-releases strategy: @@ -270,7 +270,7 @@ jobs: # ---- macos: no driver yet (every leg natively skips in the run workflow) - macos-curl-bash-to-curl-bash: - name: "${{ matrix.tag.ref }} -> HEAD" + name: "macos: curl-bash -> curl-bash (${{ matrix.tag.ref }} -> HEAD)" if: github.event_name != 'workflow_dispatch' || inputs.route == 'all' needs: pick-releases strategy: @@ -284,7 +284,7 @@ jobs: install-ref: ${{ matrix.tag.ref }} macos-curl-bash-to-hermes-update: - name: "${{ matrix.tag.ref }} -> HEAD" + name: "macos: curl-bash -> hermes-update (${{ matrix.tag.ref }} -> HEAD)" if: github.event_name != 'workflow_dispatch' || inputs.route == 'all' needs: pick-releases strategy: @@ -298,7 +298,7 @@ jobs: install-ref: ${{ matrix.tag.ref }} macos-curl-bash-to-app-update: - name: "${{ matrix.tag.ref }} -> HEAD" + name: "macos: curl-bash -> app-update (${{ matrix.tag.ref }} -> HEAD)" if: github.event_name != 'workflow_dispatch' || inputs.route == 'all' needs: pick-releases strategy: @@ -312,7 +312,7 @@ jobs: install-ref: ${{ matrix.tag.ref }} macos-packaged-app-to-curl-bash: - name: "${{ matrix.tag.ref }} -> HEAD" + name: "macos: packaged-app -> curl-bash (${{ matrix.tag.ref }} -> HEAD)" if: github.event_name != 'workflow_dispatch' || inputs.route == 'all' needs: pick-releases strategy: @@ -326,7 +326,7 @@ jobs: install-ref: ${{ matrix.tag.ref }} macos-packaged-app-to-hermes-update: - name: "${{ matrix.tag.ref }} -> HEAD" + name: "macos: packaged-app -> hermes-update (${{ matrix.tag.ref }} -> HEAD)" if: github.event_name != 'workflow_dispatch' || inputs.route == 'all' needs: pick-releases strategy: @@ -340,7 +340,7 @@ jobs: install-ref: ${{ matrix.tag.ref }} macos-packaged-app-to-app-update: - name: "${{ matrix.tag.ref }} -> HEAD" + name: "macos: packaged-app -> app-update (${{ matrix.tag.ref }} -> HEAD)" if: github.event_name != 'workflow_dispatch' || inputs.route == 'all' needs: pick-releases strategy: From 1c60bdc30d07ca68c24aabebada7d1c43df8f371 Mon Sep 17 00:00:00 2001 From: ethernet Date: Tue, 11 Aug 2026 16:54:54 -0400 Subject: [PATCH 051/227] ci(install-e2e): back to the generator - workflows stay generic, names carry everything Revert the hardcoded 16-job experiment: the combination spec belongs in scripts/sandbox/generate-e2e-matrix.mjs (restored), not copy-pasted YAML blocks. What survives from the experiment: * leg names carry everything - 'os: install -> update (tag -> HEAD)' - generated per entry, since slash-joined names are all the graph renders; * pick-releases annotates each tag ({ref, desktop}) and the generator threads tag_has_desktop onto windows entries, so the windows run workflow still gates pre-desktop tags without a probe job; * the per-OS run workflows are untouched: single job, static 'e2e' name, native skip gates own all capability knowledge. install-e2e.yml is one generate job + three per-OS matrix fanouts. Generator shape (4/16/12 legs for 2 tags), annotation threading, and all six error paths verified; all four workflows pass actionlint; driver parses clean pure-ASCII. --- .github/workflows/install-e2e-macos-run.yml | 2 +- .github/workflows/install-e2e-run.yml | 2 +- .github/workflows/install-e2e.yml | 333 +++++--------------- scripts/sandbox/generate-e2e-matrix.mjs | 223 +++++++++++++ tests/install/windows-desktop-gui-e2e.ps1 | 7 +- 5 files changed, 309 insertions(+), 258 deletions(-) create mode 100644 scripts/sandbox/generate-e2e-matrix.mjs diff --git a/.github/workflows/install-e2e-macos-run.yml b/.github/workflows/install-e2e-macos-run.yml index 450cf42bc0a2..51e28efb76ac 100644 --- a/.github/workflows/install-e2e-macos-run.yml +++ b/.github/workflows/install-e2e-macos-run.yml @@ -10,7 +10,7 @@ name: Install & Update E2E — macOS (reusable) # next to where the driver will be, and implementing a method flips this # workflow's job-level `if`. # -# Declared methods (see install-e2e.yml's macos-* jobs): +# Declared methods (see the generator's SPEC): # install: curl-bash, packaged-app # update: curl-bash, hermes-update, app-update diff --git a/.github/workflows/install-e2e-run.yml b/.github/workflows/install-e2e-run.yml index c7077870ca90..db88cfbcd001 100644 --- a/.github/workflows/install-e2e-run.yml +++ b/.github/workflows/install-e2e-run.yml @@ -10,7 +10,7 @@ name: Install & Update E2E (reusable) # is independent: its own sandbox, its own install, nothing rewound or # shared. # -# Method ids come from install-e2e.yml's combination jobs. Supported +# Method ids come from scripts/sandbox/generate-e2e-matrix.mjs. Supported # today: install via curl-bash, update via hermes-update or curl-bash. # Anything else NATIVELY SKIPS (grey check, no runner): capability # knowledge lives here, next to the driver, so the caller can dispatch diff --git a/.github/workflows/install-e2e.yml b/.github/workflows/install-e2e.yml index bf29956dd231..bb39cf48d9c9 100644 --- a/.github/workflows/install-e2e.yml +++ b/.github/workflows/install-e2e.yml @@ -2,28 +2,26 @@ name: Install & Update E2E # Can a user on a released version get to this commit? # -# One matrix job per {os, install-method -> update-method} combination a -# user could be on, hardcoded HERE so the workflow is the support matrix -# and the Actions graph draws one box per combination. Each box fans out -# over the release tags picked at runtime, so every leg reads -# " -> HEAD": install that tag the combo's way, update to this commit -# the combo's way. +# The support matrix -- every {os, install-method, update-method} combination +# a user could be on -- lives in scripts/sandbox/generate-e2e-matrix.mjs. +# generate-matrix expands it against the picked release tags into one leg +# per {combination, tag}, split into one matrix job per OS: # -# Matrix: linux-* a real curl|bash install in the bubblewrap sandbox -# (install-e2e-run.yml) -# Matrix: windows-* the real desktop user flow: website Hermes-Setup.exe -# clicked by AutoHotkey, update via the app, -# Playwright clicking "Update now" -# (install-e2e-windows-run.yml) -# Matrix: macos-* no driver yet (install-e2e-macos-run.yml) +# Matrix: linux a real curl|bash install in the bubblewrap sandbox +# (install-e2e-run.yml) +# Matrix: windows the real desktop user flow: website Hermes-Setup.exe +# clicked by AutoHotkey, update via the app, Playwright +# clicking "Update now" (install-e2e-windows-run.yml) +# Matrix: macos no driver yet (install-e2e-macos-run.yml) # # Every combination is dispatched to its OS's run workflow; the run # workflow natively skips (grey) what its driver cannot run yet -- an # unimplemented method pair, or a starting tag that predates the surface -# under test (e.g. windows desktop legs from releases before the desktop -# app existed). Capability knowledge lives next to each driver, never -# here: adding a method = adding one job below; implementing one = flipping -# the run workflow's gate. +# under test (pick-releases annotates each tag with what its tree ships, +# e.g. whether the desktop app exists yet). Capability knowledge lives +# next to each driver, never here and never in the generator: declaring a +# method is a spec edit, implementing one is flipping the run workflow's +# gate. # # The starting versions are chosen at runtime from the repo's release tags # (scripts/sandbox/pick-release-tags.sh): newest, oldest, and a spread @@ -72,8 +70,9 @@ concurrency: cancel-in-progress: true jobs: - # Which released versions do we test updating FROM? Resolved once and - # shared by every combination's matrix, so all combos cover the same set. + # Which released versions do we test updating FROM? Resolved once, + # annotated with what each tag's own tree supports, and shared by every + # OS's matrix so all combos cover the same set. pick-releases: name: Pick release tags runs-on: ubuntu-latest @@ -81,9 +80,10 @@ jobs: outputs: tags: ${{ steps.pick.outputs.tags }} steps: - # This job only reads tag names and runs one script, so take the cheap - # checkout: no blobs (filter), no other files (sparse), but DO fetch tags - # -- they are the whole input, and the default shallow checkout has none. + # This job only reads tag names and trees, so take the cheap + # checkout: no blobs (filter), no other files (sparse), but DO fetch + # tags -- they are the whole input, and the default shallow checkout + # has none. - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2 with: filter: blob:none @@ -95,11 +95,11 @@ jobs: set -euo pipefail tags="$(scripts/sandbox/pick-release-tags.sh --count '${{ inputs.tag-count || 5 }}')" echo "Testing updates from: $tags" - # Annotate each tag with what its own tree supports, so legs can - # natively skip surfaces the starting version does not have. - # Today: does the release ship the desktop app (apps/desktop, - # #20059)? Cheaper and simpler here -- the tags are already - # checked out -- than a probe job per windows leg. + # Annotate each tag with what its own tree supports, so run + # workflows can natively skip surfaces the starting version does + # not have. Today: does the release ship the desktop app + # (apps/desktop, #20059)? Cheaper here -- the tags are already + # fetched -- than a probe job per leg. enriched="$(for t in $(echo "$tags" | jq -r '.[]'); do if git ls-tree -d "$t" apps/desktop | grep -q .; then d=true; else d=false; fi echo "{\"ref\":\"$t\",\"desktop\":$d}" @@ -107,248 +107,75 @@ jobs: echo "Annotated: $enriched" echo "tags=$enriched" >> "$GITHUB_OUTPUT" - # ---- linux: install via curl-bash --------------------------------------- - - linux-curl-bash-to-hermes-update: - name: "linux: curl-bash -> hermes-update (${{ matrix.tag.ref }} -> HEAD)" - if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "both", "update"]'), inputs.route) + # Expand the support matrix against the picked tags: one leg per + # {os, install-method, update-method, tag}, split into a matrix per OS. + generate-matrix: + name: Expand combinations needs: pick-releases + runs-on: ubuntu-latest + timeout-minutes: 5 + outputs: + linux: ${{ steps.gen.outputs.linux }} + windows: ${{ steps.gen.outputs.windows }} + macos: ${{ steps.gen.outputs.macos }} + steps: + - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2 + with: + sparse-checkout: scripts/sandbox/generate-e2e-matrix.mjs + sparse-checkout-cone-mode: false + - id: gen + run: | + set -euo pipefail + matrices="$(node scripts/sandbox/generate-e2e-matrix.mjs \ + --tags '${{ needs.pick-releases.outputs.tags }}' \ + --route '${{ inputs.route || 'all' }}')" + echo "$matrices" + for key in linux windows macos; do + echo "$key=$(echo "$matrices" | node -e 'let d="";process.stdin.on("data",c=>d+=c).on("end",()=>console.log(JSON.stringify(JSON.parse(d)[process.argv[1]])))' "$key")" >> "$GITHUB_OUTPUT" + done + + linux: + name: ${{ matrix.name }} + if: fromJSON(needs.generate-matrix.outputs.linux).include[0] != null + needs: generate-matrix strategy: # One leg breaking is worth knowing about even if another already # failed, so let every leg report. fail-fast: false - matrix: - tag: ${{ fromJSON(needs.pick-releases.outputs.tags) }} - uses: ./.github/workflows/install-e2e-run.yml - with: - install-method: curl-bash - update-method: hermes-update - install-ref: ${{ matrix.tag.ref }} - - linux-curl-bash-to-curl-bash: - name: "linux: curl-bash -> curl-bash (${{ matrix.tag.ref }} -> HEAD)" - if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "both", "installer"]'), inputs.route) - needs: pick-releases - strategy: - fail-fast: false - max-parallel: 3 - matrix: - tag: ${{ fromJSON(needs.pick-releases.outputs.tags) }} + max-parallel: 4 + matrix: ${{ fromJSON(needs.generate-matrix.outputs.linux) }} uses: ./.github/workflows/install-e2e-run.yml with: - install-method: curl-bash - update-method: curl-bash - install-ref: ${{ matrix.tag.ref }} - - # ---- windows: install via the website's Hermes-Setup.exe ---------------- + install-method: ${{ matrix.install_method }} + update-method: ${{ matrix.update_method }} + install-ref: ${{ matrix.install_ref }} - windows-desktop-installer-to-desktop-app: - name: "windows: desktop-installer@latest -> desktop-app (${{ matrix.tag.ref }} -> HEAD)" - if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "windows-desktop"]'), inputs.route) - needs: pick-releases + windows: + name: ${{ matrix.name }} + if: fromJSON(needs.generate-matrix.outputs.windows).include[0] != null + needs: generate-matrix strategy: fail-fast: false # 2x-cost runners at ~14 min a leg: two at a time is throughput - # enough for a 5-tag axis without hogging the Windows pool. + # enough without hogging the Windows pool. max-parallel: 2 - matrix: - tag: ${{ fromJSON(needs.pick-releases.outputs.tags) }} - uses: ./.github/workflows/install-e2e-windows-run.yml - with: - tag-has-desktop: ${{ matrix.tag.desktop }} - install-method: desktop-installer@latest - update-method: desktop-app - install-ref: ${{ matrix.tag.ref }} - - windows-desktop-installer-to-installer-rerun: - name: "windows: desktop-installer@latest -> desktop-installer-rerun@latest (${{ matrix.tag.ref }} -> HEAD)" - if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "windows-desktop"]'), inputs.route) - needs: pick-releases - strategy: - fail-fast: false - matrix: - tag: ${{ fromJSON(needs.pick-releases.outputs.tags) }} + matrix: ${{ fromJSON(needs.generate-matrix.outputs.windows) }} uses: ./.github/workflows/install-e2e-windows-run.yml with: - tag-has-desktop: ${{ matrix.tag.desktop }} - install-method: desktop-installer@latest - update-method: desktop-installer-rerun@latest - install-ref: ${{ matrix.tag.ref }} + install-method: ${{ matrix.install_method }} + update-method: ${{ matrix.update_method }} + install-ref: ${{ matrix.install_ref }} + tag-has-desktop: ${{ matrix.tag_has_desktop }} - windows-desktop-installer-to-hermes-update: - name: "windows: desktop-installer@latest -> hermes-update (${{ matrix.tag.ref }} -> HEAD)" - if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "windows-desktop"]'), inputs.route) - needs: pick-releases - strategy: - fail-fast: false - matrix: - tag: ${{ fromJSON(needs.pick-releases.outputs.tags) }} - uses: ./.github/workflows/install-e2e-windows-run.yml - with: - tag-has-desktop: ${{ matrix.tag.desktop }} - install-method: desktop-installer@latest - update-method: hermes-update - install-ref: ${{ matrix.tag.ref }} - - windows-desktop-installer-to-irm-iex: - name: "windows: desktop-installer@latest -> irm-iex (${{ matrix.tag.ref }} -> HEAD)" - if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "windows-desktop"]'), inputs.route) - needs: pick-releases - strategy: - fail-fast: false - matrix: - tag: ${{ fromJSON(needs.pick-releases.outputs.tags) }} - uses: ./.github/workflows/install-e2e-windows-run.yml - with: - tag-has-desktop: ${{ matrix.tag.desktop }} - install-method: desktop-installer@latest - update-method: irm-iex - install-ref: ${{ matrix.tag.ref }} - - # ---- windows: install via irm | iex -------------------------------------- - - windows-irm-iex-to-desktop-app: - name: "windows: irm-iex -> desktop-app (${{ matrix.tag.ref }} -> HEAD)" - if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "windows-desktop"]'), inputs.route) - needs: pick-releases - strategy: - fail-fast: false - matrix: - tag: ${{ fromJSON(needs.pick-releases.outputs.tags) }} - uses: ./.github/workflows/install-e2e-windows-run.yml - with: - tag-has-desktop: ${{ matrix.tag.desktop }} - install-method: irm-iex - update-method: desktop-app - install-ref: ${{ matrix.tag.ref }} - - windows-irm-iex-to-installer-rerun: - name: "windows: irm-iex -> desktop-installer-rerun@latest (${{ matrix.tag.ref }} -> HEAD)" - if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "windows-desktop"]'), inputs.route) - needs: pick-releases - strategy: - fail-fast: false - matrix: - tag: ${{ fromJSON(needs.pick-releases.outputs.tags) }} - uses: ./.github/workflows/install-e2e-windows-run.yml - with: - tag-has-desktop: ${{ matrix.tag.desktop }} - install-method: irm-iex - update-method: desktop-installer-rerun@latest - install-ref: ${{ matrix.tag.ref }} - - windows-irm-iex-to-hermes-update: - name: "windows: irm-iex -> hermes-update (${{ matrix.tag.ref }} -> HEAD)" - if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "windows-desktop"]'), inputs.route) - needs: pick-releases - strategy: - fail-fast: false - matrix: - tag: ${{ fromJSON(needs.pick-releases.outputs.tags) }} - uses: ./.github/workflows/install-e2e-windows-run.yml - with: - tag-has-desktop: ${{ matrix.tag.desktop }} - install-method: irm-iex - update-method: hermes-update - install-ref: ${{ matrix.tag.ref }} - - windows-irm-iex-to-irm-iex: - name: "windows: irm-iex -> irm-iex (${{ matrix.tag.ref }} -> HEAD)" - if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "windows-desktop"]'), inputs.route) - needs: pick-releases - strategy: - fail-fast: false - matrix: - tag: ${{ fromJSON(needs.pick-releases.outputs.tags) }} - uses: ./.github/workflows/install-e2e-windows-run.yml - with: - tag-has-desktop: ${{ matrix.tag.desktop }} - install-method: irm-iex - update-method: irm-iex - install-ref: ${{ matrix.tag.ref }} - - # ---- macos: no driver yet (every leg natively skips in the run workflow) - - - macos-curl-bash-to-curl-bash: - name: "macos: curl-bash -> curl-bash (${{ matrix.tag.ref }} -> HEAD)" - if: github.event_name != 'workflow_dispatch' || inputs.route == 'all' - needs: pick-releases - strategy: - fail-fast: false - matrix: - tag: ${{ fromJSON(needs.pick-releases.outputs.tags) }} - uses: ./.github/workflows/install-e2e-macos-run.yml - with: - install-method: curl-bash - update-method: curl-bash - install-ref: ${{ matrix.tag.ref }} - - macos-curl-bash-to-hermes-update: - name: "macos: curl-bash -> hermes-update (${{ matrix.tag.ref }} -> HEAD)" - if: github.event_name != 'workflow_dispatch' || inputs.route == 'all' - needs: pick-releases - strategy: - fail-fast: false - matrix: - tag: ${{ fromJSON(needs.pick-releases.outputs.tags) }} - uses: ./.github/workflows/install-e2e-macos-run.yml - with: - install-method: curl-bash - update-method: hermes-update - install-ref: ${{ matrix.tag.ref }} - - macos-curl-bash-to-app-update: - name: "macos: curl-bash -> app-update (${{ matrix.tag.ref }} -> HEAD)" - if: github.event_name != 'workflow_dispatch' || inputs.route == 'all' - needs: pick-releases - strategy: - fail-fast: false - matrix: - tag: ${{ fromJSON(needs.pick-releases.outputs.tags) }} - uses: ./.github/workflows/install-e2e-macos-run.yml - with: - install-method: curl-bash - update-method: app-update - install-ref: ${{ matrix.tag.ref }} - - macos-packaged-app-to-curl-bash: - name: "macos: packaged-app -> curl-bash (${{ matrix.tag.ref }} -> HEAD)" - if: github.event_name != 'workflow_dispatch' || inputs.route == 'all' - needs: pick-releases - strategy: - fail-fast: false - matrix: - tag: ${{ fromJSON(needs.pick-releases.outputs.tags) }} - uses: ./.github/workflows/install-e2e-macos-run.yml - with: - install-method: packaged-app - update-method: curl-bash - install-ref: ${{ matrix.tag.ref }} - - macos-packaged-app-to-hermes-update: - name: "macos: packaged-app -> hermes-update (${{ matrix.tag.ref }} -> HEAD)" - if: github.event_name != 'workflow_dispatch' || inputs.route == 'all' - needs: pick-releases - strategy: - fail-fast: false - matrix: - tag: ${{ fromJSON(needs.pick-releases.outputs.tags) }} - uses: ./.github/workflows/install-e2e-macos-run.yml - with: - install-method: packaged-app - update-method: hermes-update - install-ref: ${{ matrix.tag.ref }} - - macos-packaged-app-to-app-update: - name: "macos: packaged-app -> app-update (${{ matrix.tag.ref }} -> HEAD)" - if: github.event_name != 'workflow_dispatch' || inputs.route == 'all' - needs: pick-releases + macos: + name: ${{ matrix.name }} + if: fromJSON(needs.generate-matrix.outputs.macos).include[0] != null + needs: generate-matrix strategy: fail-fast: false - matrix: - tag: ${{ fromJSON(needs.pick-releases.outputs.tags) }} + matrix: ${{ fromJSON(needs.generate-matrix.outputs.macos) }} uses: ./.github/workflows/install-e2e-macos-run.yml with: - install-method: packaged-app - update-method: app-update - install-ref: ${{ matrix.tag.ref }} + install-method: ${{ matrix.install_method }} + update-method: ${{ matrix.update_method }} + install-ref: ${{ matrix.install_ref }} diff --git a/scripts/sandbox/generate-e2e-matrix.mjs b/scripts/sandbox/generate-e2e-matrix.mjs new file mode 100644 index 000000000000..85e668e35cd6 --- /dev/null +++ b/scripts/sandbox/generate-e2e-matrix.mjs @@ -0,0 +1,223 @@ +#!/usr/bin/env node +/** + * Expand the install/update support matrix into concrete E2E combinations. + * + * One source of truth for every {os, install-method, update-method} pair a + * user could be on. This file only DECLARES and EXPANDS: it knows nothing + * about which combinations CI can drive. Every combination is dispatched to + * its OS's run workflow, and THAT workflow natively skips the method pairs + * its driver cannot run yet -- capability knowledge lives next to each + * driver (install-e2e-run.yml, install-e2e-windows-run.yml, + * install-e2e-macos-run.yml). + * + * Used by .github/workflows/install-e2e.yml, which runs it with the picked + * release tags (annotated at pick time with what each tag's tree ships): + * + * node scripts/sandbox/generate-e2e-matrix.mjs \ + * --tags '[{"ref":"v2026.8.3","desktop":true}]' --route all + * + * Prints JSON: { linux: {include:[...]}, windows: {include:[...]}, + * macos: {include:[...]} } -- every entry is {name, install_method, + * update_method, install_ref}, and windows entries add tag_has_desktop + * (from the tag annotation) so the run workflow can natively skip + * desktop-surface legs from releases that predate the desktop app. + */ + +import path from 'node:path'; +import { parseArgs } from 'node:util'; +import { fileURLToPath } from 'node:url'; + +/** + * Method ids are strict machine strings -- workflows and the IMPLEMENTED + * table key off them, so an unknown id must fail loudly (see KNOWN_METHODS). + * + * `versions` on an entry expands it into one combination per version. Only + * "latest" is allowed today: it means "the artifact published on the website + * right now" (Hermes-Setup.exe has no versioned archive yet). When archived + * installer versions exist, widen ALLOWED_VERSIONS. + */ +export const SPEC = { + windows: { + install: [ + // irm https://hermes.nousresearch.com/install.ps1 | iex + { method: 'irm-iex' }, + // Website Hermes-Setup.exe, clicked through the GUI. + { method: 'desktop-installer', versions: ['latest'] }, + ], + update: [ + { method: 'irm-iex' }, + // Run the bootstrap exe again over an existing install (--update flow). + { method: 'desktop-installer-rerun', versions: ['latest'] }, + { method: 'hermes-update' }, + // Settings -> About -> "Update now" inside the running desktop app. + { method: 'desktop-app' }, + ], + }, + macos: { + install: [ + { method: 'curl-bash' }, + { method: 'packaged-app' }, + ], + update: [ + { method: 'curl-bash' }, + { method: 'hermes-update' }, + { method: 'app-update' }, + ], + }, + linux: { + install: [ + { method: 'curl-bash' }, + ], + update: [ + { method: 'curl-bash' }, + { method: 'hermes-update' }, + ], + }, +}; + +const KNOWN_METHODS = new Set([ + 'irm-iex', + 'desktop-installer', + 'desktop-installer-rerun', + 'hermes-update', + 'desktop-app', + 'curl-bash', + 'packaged-app', + 'app-update', +]); + +const ALLOWED_VERSIONS = new Set(['latest']); + +function validateEntry(os, kind, entry) { + if (typeof entry.method !== 'string' || !KNOWN_METHODS.has(entry.method)) { + throw new Error(`${os}.${kind}: unknown method id ${JSON.stringify(entry.method)} -- add it to KNOWN_METHODS if intentional`); + } + if ('versions' in entry) { + if (!Array.isArray(entry.versions) || entry.versions.length === 0) { + throw new Error(`${os}.${kind}.${entry.method}: versions must be a non-empty array`); + } + for (const v of entry.versions) { + if (!ALLOWED_VERSIONS.has(v)) { + throw new Error(`${os}.${kind}.${entry.method}: version ${JSON.stringify(v)} not allowed -- only ${[...ALLOWED_VERSIONS].join(', ')} until versioned installer archives exist`); + } + } + } + const unknown = Object.keys(entry).filter((k) => k !== 'method' && k !== 'versions'); + if (unknown.length) { + throw new Error(`${os}.${kind}.${entry.method}: unknown keys ${unknown.join(', ')}`); + } +} + +/** Expand one method entry into concrete ids ("desktop-installer@latest"). */ +export function expandMethod(os, kind, entry) { + validateEntry(os, kind, entry); + if (!entry.versions) return [entry.method]; + return entry.versions.map((v) => `${entry.method}@${v}`); +} + +/** Every {os, install, update, secondUpdate} combination in SPEC. */ +export function generateEnvironments(spec) { + const envs = []; + for (const [os, osSpec] of Object.entries(spec)) { + const { install, update, secondUpdate = [], ...unknown } = osSpec; + if (Object.keys(unknown).length) { + throw new Error(`${os}: unknown spec keys ${Object.keys(unknown).join(', ')}`); + } + // Chained second updates (install -> update -> update again) are a real + // axis -- the updater that RESULTS from an update must itself update -- + // but nothing implements them yet. Refuse a spec that declares them so + // the first implementation is forced to come through here. + if (!Array.isArray(secondUpdate) || secondUpdate.length !== 0) { + throw new Error(`${os}: secondUpdate must be empty until a second-update leg is implemented`); + } + const installs = install.flatMap((e) => expandMethod(os, 'install', e)); + const updates = update.flatMap((e) => expandMethod(os, 'update', e)); + for (const i of installs) { + for (const u of updates) { + envs.push({ os, install: i, update: u, secondUpdate: '' }); + } + } + } + return envs; +} + +/** Mirror of install-e2e.yml's dispatch `route` choice. */ +function routeWants(route, env) { + switch (route) { + case 'all': + return true; + case 'both': + return env.os === 'linux'; + case 'update': + return env.os === 'linux' && env.update === 'hermes-update'; + case 'installer': + return env.os === 'linux' && env.update === 'curl-bash'; + case 'windows-desktop': + return env.os === 'windows'; + default: + throw new Error(`unknown route filter: ${JSON.stringify(route)}`); + } +} + +/** + * Split the combinations into one matrix per OS. + * + * `tags` (the released versions we test updating FROM, as {ref, desktop} + * annotation objects) is the OUTER axis: for each tag, for each + * combination, one dispatch that installs the tag and updates to HEAD. + * No capability filtering happens here -- every declared combination is + * dispatched, and the OS's run workflow natively skips what its driver + * cannot run yet. Entry names carry everything (os, method pair, tag + * transition) because slash-joined leg names are all the graph renders. + */ +export function buildMatrices(envs, { tags = [], route = 'all' } = {}) { + for (const t of tags) { + if (typeof t?.ref !== 'string' || typeof t?.desktop !== 'boolean') { + throw new Error(`tags must be {ref, desktop} annotation objects, got ${JSON.stringify(t)}`); + } + } + const byOs = { linux: [], windows: [], macos: [] }; + for (const env of envs) { + if (!routeWants(route, env)) continue; + const bucket = byOs[env.os]; + if (!bucket) { + throw new Error(`no matrix bucket for os ${JSON.stringify(env.os)}`); + } + if (tags.length === 0) { + throw new Error('combinations were selected but no --tags given'); + } + for (const tag of tags) { + const entry = { + name: `${env.os}: ${env.install} -> ${env.update} (${tag.ref} -> HEAD)`, + install_method: env.install, + update_method: env.update, + install_ref: tag.ref, + }; + if (env.os === 'windows') entry.tag_has_desktop = tag.desktop; + bucket.push(entry); + } + } + return Object.fromEntries( + Object.entries(byOs).map(([os, include]) => [os, { include }]), + ); +} + +function main() { + const { values } = parseArgs({ + options: { + tags: { type: 'string', default: '[]' }, + route: { type: 'string', default: 'all' }, + }, + }); + const tags = JSON.parse(values.tags); + if (!Array.isArray(tags)) { + throw new Error('--tags must be a JSON array of {ref, desktop} annotation objects'); + } + const envs = generateEnvironments(SPEC); + const matrices = buildMatrices(envs, { tags, route: values.route }); + process.stdout.write(`${JSON.stringify(matrices, null, 2)}\n`); +} + +if (process.argv[1] && fileURLToPath(import.meta.url) === path.resolve(process.argv[1])) { + main(); +} diff --git a/tests/install/windows-desktop-gui-e2e.ps1 b/tests/install/windows-desktop-gui-e2e.ps1 index 0813903f3646..afb8b5c29fc9 100644 --- a/tests/install/windows-desktop-gui-e2e.ps1 +++ b/tests/install/windows-desktop-gui-e2e.ps1 @@ -72,9 +72,10 @@ param( [string]$Phase = "all", # Update method to exercise in the update-gui phase, named by the same - # ids install-e2e.yml's combination jobs use. Only "desktop-app" (the - # app's own Update button) is implemented; the others are declared arms - # so the surface is stable when they land. + # ids the combination generator (scripts/sandbox/generate-e2e-matrix + # .mjs) declares. Only "desktop-app" (the app's own Update button) is + # implemented; the others are declared arms so the surface is stable + # when they land. [ValidateSet("desktop-app", "hermes-update", "desktop-installer-rerun@latest", "irm-iex")] [string]$Route = "desktop-app", From 17dea59026caede152bc147c579a0d642872062d Mon Sep 17 00:00:00 2001 From: ethernet Date: Tue, 11 Aug 2026 17:11:07 -0400 Subject: [PATCH 052/227] ci(install-e2e): generator drops runtime validation for jsdoc type unions Per review: the method/version vocabulary is now closed TYPE unions (@ts-check + jsdoc typedefs - InstallerVersion, InstallMethod, UpdateMethod - checked with tsc --checkJs, which rejects a SPEC entry outside the unions; verified by corrupting a copy: 6 errors) instead of runtime KNOWN_METHODS/ALLOWED_VERSIONS sets. validateEntry and routeWants are deleted with all the paranoia: the generator always emits every OS matrix and the dispatch route filter moved to plain job-level ifs in install-e2e.yml, where the OS jobs already live. secondUpdate is typed never[] so declaring one is a type error until a leg implements it. Anything types cannot catch is self-evident on the next CI run. --- .github/workflows/install-e2e.yml | 12 +- scripts/sandbox/generate-e2e-matrix.mjs | 179 +++++++++--------------- 2 files changed, 73 insertions(+), 118 deletions(-) diff --git a/.github/workflows/install-e2e.yml b/.github/workflows/install-e2e.yml index bb39cf48d9c9..229f46dda31c 100644 --- a/.github/workflows/install-e2e.yml +++ b/.github/workflows/install-e2e.yml @@ -127,8 +127,7 @@ jobs: run: | set -euo pipefail matrices="$(node scripts/sandbox/generate-e2e-matrix.mjs \ - --tags '${{ needs.pick-releases.outputs.tags }}' \ - --route '${{ inputs.route || 'all' }}')" + --tags '${{ needs.pick-releases.outputs.tags }}')" echo "$matrices" for key in linux windows macos; do echo "$key=$(echo "$matrices" | node -e 'let d="";process.stdin.on("data",c=>d+=c).on("end",()=>console.log(JSON.stringify(JSON.parse(d)[process.argv[1]])))' "$key")" >> "$GITHUB_OUTPUT" @@ -136,7 +135,10 @@ jobs: linux: name: ${{ matrix.name }} - if: fromJSON(needs.generate-matrix.outputs.linux).include[0] != null + # The update/installer route choices map to the linux update methods; + # either way the whole linux matrix runs (legs are cheap and the + # distinction wasn't worth a filter layer in the generator). + if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "both", "update", "installer"]'), inputs.route) needs: generate-matrix strategy: # One leg breaking is worth knowing about even if another already @@ -152,7 +154,7 @@ jobs: windows: name: ${{ matrix.name }} - if: fromJSON(needs.generate-matrix.outputs.windows).include[0] != null + if: github.event_name != 'workflow_dispatch' || contains(fromJSON('["all", "windows-desktop"]'), inputs.route) needs: generate-matrix strategy: fail-fast: false @@ -169,7 +171,7 @@ jobs: macos: name: ${{ matrix.name }} - if: fromJSON(needs.generate-matrix.outputs.macos).include[0] != null + if: github.event_name != 'workflow_dispatch' || inputs.route == 'all' needs: generate-matrix strategy: fail-fast: false diff --git a/scripts/sandbox/generate-e2e-matrix.mjs b/scripts/sandbox/generate-e2e-matrix.mjs index 85e668e35cd6..ad836169ef95 100644 --- a/scripts/sandbox/generate-e2e-matrix.mjs +++ b/scripts/sandbox/generate-e2e-matrix.mjs @@ -1,4 +1,5 @@ #!/usr/bin/env node +// @ts-check /** * Expand the install/update support matrix into concrete E2E combinations. * @@ -8,13 +9,15 @@ * its OS's run workflow, and THAT workflow natively skips the method pairs * its driver cannot run yet -- capability knowledge lives next to each * driver (install-e2e-run.yml, install-e2e-windows-run.yml, - * install-e2e-macos-run.yml). + * install-e2e-macos-run.yml). Correctness here is enforced by the type + * unions below (checked via `tsc --checkJs`), not by runtime validation -- + * anything the types can't catch is self-evident on the next CI run. * * Used by .github/workflows/install-e2e.yml, which runs it with the picked * release tags (annotated at pick time with what each tag's tree ships): * * node scripts/sandbox/generate-e2e-matrix.mjs \ - * --tags '[{"ref":"v2026.8.3","desktop":true}]' --route all + * --tags '[{"ref":"v2026.8.3","desktop":true}]' * * Prints JSON: { linux: {include:[...]}, windows: {include:[...]}, * macos: {include:[...]} } -- every entry is {name, install_method, @@ -28,14 +31,34 @@ import { parseArgs } from 'node:util'; import { fileURLToPath } from 'node:url'; /** - * Method ids are strict machine strings -- workflows and the IMPLEMENTED - * table key off them, so an unknown id must fail loudly (see KNOWN_METHODS). + * The closed method/version vocabulary. Workflows key off these exact + * strings, so they are types, not conventions. * - * `versions` on an entry expands it into one combination per version. Only - * "latest" is allowed today: it means "the artifact published on the website - * right now" (Hermes-Setup.exe has no versioned archive yet). When archived - * installer versions exist, widen ALLOWED_VERSIONS. + * @typedef {'latest'} InstallerVersion + * The artifact published on the website right now -- Hermes-Setup.exe has + * no versioned archive yet. Widen this union when one exists. + * @typedef {'irm-iex' | 'desktop-installer' | 'curl-bash' | 'packaged-app'} InstallMethod + * @typedef {InstallMethod | 'desktop-installer-rerun' | 'hermes-update' | 'app-update' | 'desktop-app'} UpdateMethod + * @typedef {'linux' | 'windows' | 'macos'} Os + * + * @typedef {{method: InstallMethod, versions?: InstallerVersion[]}} InstallEntry + * @typedef {{method: UpdateMethod, versions?: InstallerVersion[]}} UpdateEntry + * `versions` expands the entry into one combination per version + * ("desktop-installer@latest"). + * @typedef {{install: InstallEntry[], update: UpdateEntry[], secondUpdate?: never[]}} OsSpec + * secondUpdate (install -> update -> update again) is a real future axis + * -- the updater that RESULTS from an update must itself update -- typed + * `never[]` so declaring one is a type error until a leg implements it. + * + * @typedef {{ref: string, desktop: boolean}} TagAnnotation + * A picked release tag plus what its own tree ships (annotated by + * pick-releases in install-e2e.yml). + * + * @typedef {{name: string, install_method: string, update_method: string, + * install_ref: string, tag_has_desktop?: boolean}} MatrixEntry */ + +/** @type {Record} */ export const SPEC = { windows: { install: [ @@ -75,118 +98,55 @@ export const SPEC = { }, }; -const KNOWN_METHODS = new Set([ - 'irm-iex', - 'desktop-installer', - 'desktop-installer-rerun', - 'hermes-update', - 'desktop-app', - 'curl-bash', - 'packaged-app', - 'app-update', -]); - -const ALLOWED_VERSIONS = new Set(['latest']); - -function validateEntry(os, kind, entry) { - if (typeof entry.method !== 'string' || !KNOWN_METHODS.has(entry.method)) { - throw new Error(`${os}.${kind}: unknown method id ${JSON.stringify(entry.method)} -- add it to KNOWN_METHODS if intentional`); - } - if ('versions' in entry) { - if (!Array.isArray(entry.versions) || entry.versions.length === 0) { - throw new Error(`${os}.${kind}.${entry.method}: versions must be a non-empty array`); - } - for (const v of entry.versions) { - if (!ALLOWED_VERSIONS.has(v)) { - throw new Error(`${os}.${kind}.${entry.method}: version ${JSON.stringify(v)} not allowed -- only ${[...ALLOWED_VERSIONS].join(', ')} until versioned installer archives exist`); - } - } - } - const unknown = Object.keys(entry).filter((k) => k !== 'method' && k !== 'versions'); - if (unknown.length) { - throw new Error(`${os}.${kind}.${entry.method}: unknown keys ${unknown.join(', ')}`); - } -} - -/** Expand one method entry into concrete ids ("desktop-installer@latest"). */ -export function expandMethod(os, kind, entry) { - validateEntry(os, kind, entry); +/** + * Expand one method entry into concrete ids ("desktop-installer@latest"). + * @param {InstallEntry | UpdateEntry} entry + * @returns {string[]} + */ +export function expandMethod(entry) { if (!entry.versions) return [entry.method]; return entry.versions.map((v) => `${entry.method}@${v}`); } -/** Every {os, install, update, secondUpdate} combination in SPEC. */ +/** + * Every {os, install, update} combination in SPEC. + * @param {Record} spec + * @returns {{os: Os, install: string, update: string}[]} + */ export function generateEnvironments(spec) { + /** @type {{os: Os, install: string, update: string}[]} */ const envs = []; - for (const [os, osSpec] of Object.entries(spec)) { - const { install, update, secondUpdate = [], ...unknown } = osSpec; - if (Object.keys(unknown).length) { - throw new Error(`${os}: unknown spec keys ${Object.keys(unknown).join(', ')}`); - } - // Chained second updates (install -> update -> update again) are a real - // axis -- the updater that RESULTS from an update must itself update -- - // but nothing implements them yet. Refuse a spec that declares them so - // the first implementation is forced to come through here. - if (!Array.isArray(secondUpdate) || secondUpdate.length !== 0) { - throw new Error(`${os}: secondUpdate must be empty until a second-update leg is implemented`); - } - const installs = install.flatMap((e) => expandMethod(os, 'install', e)); - const updates = update.flatMap((e) => expandMethod(os, 'update', e)); - for (const i of installs) { - for (const u of updates) { - envs.push({ os, install: i, update: u, secondUpdate: '' }); + for (const [os, osSpec] of /** @type {[Os, OsSpec][]} */ (Object.entries(spec))) { + for (const install of osSpec.install.flatMap(expandMethod)) { + for (const update of osSpec.update.flatMap(expandMethod)) { + envs.push({ os, install, update }); } } } return envs; } -/** Mirror of install-e2e.yml's dispatch `route` choice. */ -function routeWants(route, env) { - switch (route) { - case 'all': - return true; - case 'both': - return env.os === 'linux'; - case 'update': - return env.os === 'linux' && env.update === 'hermes-update'; - case 'installer': - return env.os === 'linux' && env.update === 'curl-bash'; - case 'windows-desktop': - return env.os === 'windows'; - default: - throw new Error(`unknown route filter: ${JSON.stringify(route)}`); - } -} - /** * Split the combinations into one matrix per OS. * - * `tags` (the released versions we test updating FROM, as {ref, desktop} - * annotation objects) is the OUTER axis: for each tag, for each - * combination, one dispatch that installs the tag and updates to HEAD. - * No capability filtering happens here -- every declared combination is - * dispatched, and the OS's run workflow natively skips what its driver - * cannot run yet. Entry names carry everything (os, method pair, tag - * transition) because slash-joined leg names are all the graph renders. + * `tags` (the released versions we test updating FROM) is the OUTER axis: + * for each tag, for each combination, one dispatch that installs the tag + * and updates to HEAD. No capability filtering happens here -- every + * declared combination is dispatched, and the OS's run workflow natively + * skips what its driver cannot run yet. Entry names carry everything (os, + * method pair, tag transition) because slash-joined leg names are all the + * graph renders. + * + * @param {{os: Os, install: string, update: string}[]} envs + * @param {TagAnnotation[]} tags + * @returns {Record} */ -export function buildMatrices(envs, { tags = [], route = 'all' } = {}) { - for (const t of tags) { - if (typeof t?.ref !== 'string' || typeof t?.desktop !== 'boolean') { - throw new Error(`tags must be {ref, desktop} annotation objects, got ${JSON.stringify(t)}`); - } - } - const byOs = { linux: [], windows: [], macos: [] }; +export function buildMatrices(envs, tags) { + /** @type {Record} */ + const byOs = { linux: { include: [] }, windows: { include: [] }, macos: { include: [] } }; for (const env of envs) { - if (!routeWants(route, env)) continue; - const bucket = byOs[env.os]; - if (!bucket) { - throw new Error(`no matrix bucket for os ${JSON.stringify(env.os)}`); - } - if (tags.length === 0) { - throw new Error('combinations were selected but no --tags given'); - } for (const tag of tags) { + /** @type {MatrixEntry} */ const entry = { name: `${env.os}: ${env.install} -> ${env.update} (${tag.ref} -> HEAD)`, install_method: env.install, @@ -194,27 +154,20 @@ export function buildMatrices(envs, { tags = [], route = 'all' } = {}) { install_ref: tag.ref, }; if (env.os === 'windows') entry.tag_has_desktop = tag.desktop; - bucket.push(entry); + byOs[env.os].include.push(entry); } } - return Object.fromEntries( - Object.entries(byOs).map(([os, include]) => [os, { include }]), - ); + return byOs; } function main() { const { values } = parseArgs({ options: { tags: { type: 'string', default: '[]' }, - route: { type: 'string', default: 'all' }, }, }); - const tags = JSON.parse(values.tags); - if (!Array.isArray(tags)) { - throw new Error('--tags must be a JSON array of {ref, desktop} annotation objects'); - } - const envs = generateEnvironments(SPEC); - const matrices = buildMatrices(envs, { tags, route: values.route }); + const tags = /** @type {TagAnnotation[]} */ (JSON.parse(values.tags)); + const matrices = buildMatrices(generateEnvironments(SPEC), tags); process.stdout.write(`${JSON.stringify(matrices, null, 2)}\n`); } From 6d7ba869630ab20b2f8d5c9dc31800b914c890ed Mon Sep 17 00:00:00 2001 From: ethernet Date: Tue, 11 Aug 2026 17:14:17 -0400 Subject: [PATCH 053/227] ci(install-e2e): collapse the method vocabulary - 3 install ids, install+2 update ids Per review the unions were overcomplicated. Install methods are now just: installer-script (the platform one-liner - curl | bash on linux/macos, irm | iex on windows), desktop-installer, and packaged-app (declared, unused). Update methods are every install method (re-run it over the existing install) plus hermes-update and app-update. desktop-installer-rerun, desktop-app, curl-bash, and irm-iex are gone as ids; the windows driver's ValidateSet, switch arms, and both run workflows' gates renamed to match. tsc --checkJs clean; generator output re-verified (4 linux / 16 windows / 6 macos legs for 2 tags). --- .github/workflows/install-e2e-macos-run.yml | 4 +-- .github/workflows/install-e2e-run.yml | 15 ++++++----- .github/workflows/install-e2e-windows-run.yml | 12 ++++----- scripts/sandbox/generate-e2e-matrix.mjs | 27 +++++++++++-------- tests/install/windows-desktop-gui-e2e.ps1 | 16 +++++------ 5 files changed, 40 insertions(+), 34 deletions(-) diff --git a/.github/workflows/install-e2e-macos-run.yml b/.github/workflows/install-e2e-macos-run.yml index 51e28efb76ac..5767986bf0bb 100644 --- a/.github/workflows/install-e2e-macos-run.yml +++ b/.github/workflows/install-e2e-macos-run.yml @@ -11,8 +11,8 @@ name: Install & Update E2E — macOS (reusable) # workflow's job-level `if`. # # Declared methods (see the generator's SPEC): -# install: curl-bash, packaged-app -# update: curl-bash, hermes-update, app-update +# install: installer-script +# update: installer-script, hermes-update, app-update on: workflow_call: diff --git a/.github/workflows/install-e2e-run.yml b/.github/workflows/install-e2e-run.yml index db88cfbcd001..021ce45ddf6c 100644 --- a/.github/workflows/install-e2e-run.yml +++ b/.github/workflows/install-e2e-run.yml @@ -11,7 +11,8 @@ name: Install & Update E2E (reusable) # shared. # # Method ids come from scripts/sandbox/generate-e2e-matrix.mjs. Supported -# today: install via curl-bash, update via hermes-update or curl-bash. +# today: install via installer-script, update via hermes-update or +# installer-script (re-run the one-liner). # Anything else NATIVELY SKIPS (grey check, no runner): capability # knowledge lives here, next to the driver, so the caller can dispatch # every declared combination without knowing which ones work. @@ -22,7 +23,7 @@ name: Install & Update E2E (reusable) # tip: # uses: ./.github/workflows/install-e2e-run.yml # with: -# install-method: curl-bash +# install-method: installer-script # update-method: hermes-update # install-ref: refs/heads/main @@ -30,11 +31,11 @@ on: workflow_call: inputs: install-method: - description: 'How the starting version gets installed. Supported: curl-bash (the real curl | install.sh one-liner).' + description: 'How the starting version gets installed. Supported: installer-script (the real curl | install.sh one-liner).' required: true type: string update-method: - description: 'How the install updates to HEAD. Supported: hermes-update (the updater) or curl-bash (re-run the one-liner). Declared-but-TODO methods skip.' + description: 'How the install updates to HEAD. Supported: hermes-update (the updater) or installer-script (re-run the one-liner). Declared-but-TODO methods skip.' required: true type: string install-ref: @@ -65,7 +66,7 @@ jobs: name: e2e # The pairs the sandbox driver can run today; anything else is a # declared TODO and natively skips. - if: inputs.install-method == 'curl-bash' && contains(fromJSON('["hermes-update", "curl-bash"]'), inputs.update-method) + if: inputs.install-method == 'installer-script' && contains(fromJSON('["hermes-update", "installer-script"]'), inputs.update-method) runs-on: ${{ inputs.runner }} timeout-minutes: ${{ inputs.timeout-minutes }} @@ -107,8 +108,8 @@ jobs: # Method id -> the driver script's --route vocabulary. The one # place that knows both names. case '${{ inputs.update-method }}' in - hermes-update) route=update ;; - curl-bash) route=installer ;; + hermes-update) route=update ;; + installer-script) route=installer ;; *) echo "unreachable: update-method passed the job-level gate but has no route mapping" >&2; exit 1 ;; esac tests/install/install-update-e2e.sh \ diff --git a/.github/workflows/install-e2e-windows-run.yml b/.github/workflows/install-e2e-windows-run.yml index fdac6d6ce0ab..90a010f3142a 100644 --- a/.github/workflows/install-e2e-windows-run.yml +++ b/.github/workflows/install-e2e-windows-run.yml @@ -16,16 +16,16 @@ name: Install & Update E2E — Windows desktop (reusable) # real user. # # Routes (from the update-method input): -# desktop-app the app's own Update button: the installed +# app-update the app's own Update button: the installed # Hermes.exe runs under Playwright's Electron # driver, which clicks Settings -> About -> # "Update now"; the production hand-off chain runs # untouched (marker, app quit, detached updater, # hermes update, desktop rebuild, relaunch). # hermes-update TODO: `hermes update` from the installed venv. -# desktop-installer-rerun@latest +# desktop-installer@latest # TODO: re-run the bootstrap exe over the install. -# irm-iex TODO: re-run the irm | iex one-liner. +# installer-script TODO: re-run the irm | iex one-liner. # # Method pairs without a driver yet NATIVELY SKIP (grey check, no runner): # the capability knowledge lives here, next to the driver, so the caller @@ -38,7 +38,7 @@ name: Install & Update E2E — Windows desktop (reusable) # uses: ./.github/workflows/install-e2e-windows-run.yml # with: # install-method: desktop-installer@latest -# update-method: desktop-app +# update-method: app-update # install-ref: v2026.8.3 on: @@ -49,7 +49,7 @@ on: required: true type: string update-method: - description: 'How the install updates to HEAD. Supported: desktop-app (Update button under Playwright). Declared-but-TODO methods skip.' + description: 'How the install updates to HEAD. Supported: app-update (Update button under Playwright). Declared-but-TODO methods skip.' required: true type: string install-ref: @@ -87,7 +87,7 @@ jobs: # tag-has-desktop from the tag's own tree; releases before #20059 have # no window to launch and no Update button to click). Anything else is # a native skip: a declared-TODO method pair, or a pre-desktop tag. - if: inputs.install-method == 'desktop-installer@latest' && inputs.update-method == 'desktop-app' && inputs.tag-has-desktop + if: inputs.install-method == 'desktop-installer@latest' && inputs.update-method == 'app-update' && inputs.tag-has-desktop runs-on: windows-latest timeout-minutes: ${{ inputs.timeout-minutes }} diff --git a/scripts/sandbox/generate-e2e-matrix.mjs b/scripts/sandbox/generate-e2e-matrix.mjs index ad836169ef95..31297e8a2369 100644 --- a/scripts/sandbox/generate-e2e-matrix.mjs +++ b/scripts/sandbox/generate-e2e-matrix.mjs @@ -37,8 +37,14 @@ import { fileURLToPath } from 'node:url'; * @typedef {'latest'} InstallerVersion * The artifact published on the website right now -- Hermes-Setup.exe has * no versioned archive yet. Widen this union when one exists. - * @typedef {'irm-iex' | 'desktop-installer' | 'curl-bash' | 'packaged-app'} InstallMethod - * @typedef {InstallMethod | 'desktop-installer-rerun' | 'hermes-update' | 'app-update' | 'desktop-app'} UpdateMethod + * @typedef {'installer-script' | 'desktop-installer' | 'packaged-app'} InstallMethod + * installer-script is the platform's one-liner (curl | bash on + * linux/macos, irm | iex on windows); packaged-app is declared but not + * used by any OS spec yet. + * @typedef {InstallMethod | 'hermes-update' | 'app-update'} UpdateMethod + * Every install method doubles as an update method (re-run it over the + * existing install), plus the updater CLI and the running app's own + * Update button. * @typedef {'linux' | 'windows' | 'macos'} Os * * @typedef {{method: InstallMethod, versions?: InstallerVersion[]}} InstallEntry @@ -63,36 +69,35 @@ export const SPEC = { windows: { install: [ // irm https://hermes.nousresearch.com/install.ps1 | iex - { method: 'irm-iex' }, + { method: 'installer-script' }, // Website Hermes-Setup.exe, clicked through the GUI. { method: 'desktop-installer', versions: ['latest'] }, ], update: [ - { method: 'irm-iex' }, + { method: 'installer-script' }, // Run the bootstrap exe again over an existing install (--update flow). - { method: 'desktop-installer-rerun', versions: ['latest'] }, + { method: 'desktop-installer', versions: ['latest'] }, { method: 'hermes-update' }, // Settings -> About -> "Update now" inside the running desktop app. - { method: 'desktop-app' }, + { method: 'app-update' }, ], }, macos: { install: [ - { method: 'curl-bash' }, - { method: 'packaged-app' }, + { method: 'installer-script' }, ], update: [ - { method: 'curl-bash' }, + { method: 'installer-script' }, { method: 'hermes-update' }, { method: 'app-update' }, ], }, linux: { install: [ - { method: 'curl-bash' }, + { method: 'installer-script' }, ], update: [ - { method: 'curl-bash' }, + { method: 'installer-script' }, { method: 'hermes-update' }, ], }, diff --git a/tests/install/windows-desktop-gui-e2e.ps1 b/tests/install/windows-desktop-gui-e2e.ps1 index afb8b5c29fc9..ea0aff0d9e7e 100644 --- a/tests/install/windows-desktop-gui-e2e.ps1 +++ b/tests/install/windows-desktop-gui-e2e.ps1 @@ -73,11 +73,11 @@ param( # Update method to exercise in the update-gui phase, named by the same # ids the combination generator (scripts/sandbox/generate-e2e-matrix - # .mjs) declares. Only "desktop-app" (the app's own Update button) is + # .mjs) declares. Only "app-update" (the app's own Update button) is # implemented; the others are declared arms so the surface is stable # when they land. - [ValidateSet("desktop-app", "hermes-update", "desktop-installer-rerun@latest", "irm-iex")] - [string]$Route = "desktop-app", + [ValidateSet("app-update", "hermes-update", "desktop-installer@latest", "installer-script")] + [string]$Route = "app-update", # The OLD version: the ref served as `main` while the installer runs, # i.e. what the user starts on. The published Hermes-Setup.exe carries @@ -658,7 +658,7 @@ function Invoke-GuiUpdateDesktopRoute([string]$TargetSha) { function Invoke-PhaseUpdateGui { $state = Read-State switch ($Route) { - "desktop-app" { + "app-update" { Invoke-GuiUpdateDesktopRoute $state.current } "hermes-update" { @@ -667,16 +667,16 @@ function Invoke-PhaseUpdateGui { # app-quit dance. throw "update method 'hermes-update' is not implemented yet" } - "desktop-installer-rerun@latest" { + "desktop-installer@latest" { # TODO: re-run the bootstrap Hermes-Setup.exe over the existing # install (its --update flow jumps straight to progress and # runs unattended). - throw "update method 'desktop-installer-rerun@latest' is not implemented yet" + throw "update method 'desktop-installer@latest' is not implemented yet" } - "irm-iex" { + "installer-script" { # TODO: re-run the irm | iex one-liner over the existing # install. - throw "update method 'irm-iex' is not implemented yet" + throw "update method 'installer-script' is not implemented yet" } } } From e6e056f854596db8ede6d32c5c79ec9caad37c68 Mon Sep 17 00:00:00 2001 From: ethernet Date: Tue, 11 Aug 2026 20:44:27 -0400 Subject: [PATCH 054/227] ci(windows-e2e): step names carry the actual install-ref --- .github/workflows/install-e2e-windows-run.yml | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/.github/workflows/install-e2e-windows-run.yml b/.github/workflows/install-e2e-windows-run.yml index 90a010f3142a..27b75e67df87 100644 --- a/.github/workflows/install-e2e-windows-run.yml +++ b/.github/workflows/install-e2e-windows-run.yml @@ -58,7 +58,7 @@ on: type: string default: auto tag-has-desktop: - description: 'Whether install-ref ships the desktop app (apps/desktop). The caller annotates this from the tag''s own tree; desktop-method legs from pre-desktop releases natively skip.' + description: "Whether install-ref ships the desktop app (apps/desktop). The caller annotates this from the tag's own tree; desktop-method legs from pre-desktop releases natively skip." required: false type: boolean default: true @@ -134,15 +134,15 @@ jobs: shell: pwsh run: Add-Content -Path $env:GITHUB_PATH -Value "$env:RUNNER_TEMP\test-bins\ffmpeg\bin" - - name: Stage serve repo (main -> OLD) + - name: Stage serve repo (main -> ${{ inputs.install-ref }}) shell: powershell run: powershell -NoProfile -ExecutionPolicy Bypass -File tests\install\windows-desktop-gui-e2e.ps1 -Phase stage -Route "${{ inputs.update-method }}" -InstallRef "${{ inputs.install-ref }}" -SetupExeUrl ${{ inputs.setup-exe-url }} - - name: Install OLD via website Hermes-Setup.exe (headed, AHK-clicked) + - name: Install ${{ inputs.install-ref }} via website Hermes-Setup.exe (headed, AHK-clicked) shell: powershell run: powershell -NoProfile -ExecutionPolicy Bypass -File tests\install\windows-desktop-gui-e2e.ps1 -Phase install-gui -Route "${{ inputs.update-method }}" -InstallRef "${{ inputs.install-ref }}" -SetupExeUrl ${{ inputs.setup-exe-url }} - - name: Update OLD -> HEAD (${{ inputs.update-method }}) + - name: Update ${{ inputs.install-ref }} -> HEAD (${{ inputs.update-method }}) shell: powershell run: powershell -NoProfile -ExecutionPolicy Bypass -File tests\install\windows-desktop-gui-e2e.ps1 -Phase update-gui -Route "${{ inputs.update-method }}" -InstallRef "${{ inputs.install-ref }}" -SetupExeUrl ${{ inputs.setup-exe-url }} From 718722ae5e359fd490ce47d36c750f2accf4ab40 Mon Sep 17 00:00:00 2001 From: ethernet Date: Tue, 11 Aug 2026 21:10:42 -0400 Subject: [PATCH 055/227] test(install): installer-script e2e driver - git redirect, no sandbox The POSIX sibling of windows-desktop-gui-e2e.ps1, sharing its staging trick: bare-clone the checkout to serve.git, park main at OLD, point every git process at it with url..insteadOf in a driver-owned GIT_CONFIG_GLOBAL. The installer and updater run byte-for-byte against their real URLs; no bwrap, no MITM proxy, no TLS interception - a disposable CI runner IS the sandbox, so the same driver can run on macos-latest unchanged. install.sh is not curl'd: the install leg runs the copy shipped AT the OLD ref (what a user who installed then actually executed), the installer-script update leg runs HEAD's copy (what the website serves at update time). Flags are probed per-ref (--skip-browser is newer than sampled tags); HOME is isolated because old installers hardcode ~/.hermes; .skip_upstream_prompt suppresses the updater's fork prompt on the file:// origin; the dirty-tree guard checks tracked files only (-uno) since untracked files cannot leak into a bare clone. Verified locally end-to-end: v0.20.2 installed via its own install.sh (uv, managed Python, Node, venv; hermes --version OK), served main advanced, hermes update landed the checkout on HEAD with a working hermes. bash -n + shellcheck clean. --- tests/install/installer-script-e2e.sh | 207 ++++++++++++++++++++++++++ 1 file changed, 207 insertions(+) create mode 100755 tests/install/installer-script-e2e.sh diff --git a/tests/install/installer-script-e2e.sh b/tests/install/installer-script-e2e.sh new file mode 100755 index 000000000000..88c7327e88e9 --- /dev/null +++ b/tests/install/installer-script-e2e.sh @@ -0,0 +1,207 @@ +#!/usr/bin/env bash +# Prove a user who installed OLD via the installer script can reach HEAD. +# +# The POSIX sibling of tests/install/windows-desktop-gui-e2e.ps1, sharing its +# staging trick and replacing the old bubblewrap sandbox: instead of a fake +# Internet (MITM proxy + upload-pack shim), every git process is pointed at a +# local bare clone with url..insteadOf rewrites for both +# canonical repo URLs in a driver-owned GIT_CONFIG_GLOBAL. The installer and +# updater run byte-for-byte against their real URLs and land on serve.git; +# `main` serves OLD during the install, then advances to HEAD for the update +# leg -- an update becomes available exactly the way it does for a real user. +# No bwrap, no slirp4netns, no TLS interception; the CI runner is disposable, +# so the host IS the sandbox. +# +# install.sh itself is not curl'd: the install leg runs the copy shipped AT +# the OLD ref (what a user who installed then actually executed), and the +# installer-script update leg runs HEAD's copy (what the website serves at +# update time). +# +# Phases (mirroring the windows driver): +# stage bare-clone this checkout to serve.git, park main at OLD +# install run OLD's scripts/install.sh under the redirect; assert the +# install landed on OLD with a working `hermes` +# update advance served main to HEAD, apply ONE update method, assert +# the checkout landed on HEAD with a working `hermes` +# +# Usage: +# tests/install/installer-script-e2e.sh --update-method hermes-update|installer-script +# [--install-ref REF] +# +# --update-method hermes-update `hermes update` +# installer-script re-run install.sh (HEAD's copy) +# --install-ref what to install first; anything git resolves. Default: +# the newest release tag in the checkout. +# +# Requires a clean full-history checkout with release tags fetched. + +set -euo pipefail + +UPDATE_METHOD="" +INSTALL_REF="" +while [ "$#" -gt 0 ]; do + case "$1" in + --update-method) + [ "$#" -ge 2 ] || { echo 'error: --update-method needs a value' >&2; exit 1; } + UPDATE_METHOD="$2"; shift 2 ;; + --install-ref) + [ "$#" -ge 2 ] || { echo 'error: --install-ref needs a value' >&2; exit 1; } + INSTALL_REF="$2"; shift 2 ;; + -h|--help) sed -n '2,37p' "$0"; exit 0 ;; + *) echo "error: unknown argument: $1" >&2; exit 1 ;; + esac +done +case "$UPDATE_METHOD" in + hermes-update|installer-script) ;; + *) echo "error: --update-method must be hermes-update or installer-script, got '$UPDATE_METHOD'" >&2; exit 1 ;; +esac + +REPO_ROOT="$(cd "$(dirname "$0")/../.." && pwd)" +REPO_URL_SSH="git@github.com:NousResearch/hermes-agent.git" +REPO_URL_HTTPS="https://github.com/NousResearch/hermes-agent.git" + +# Everything lives OUTSIDE the checkout; an untracked dir inside the repo +# would make later dirty-tree checks lie. +WORK_ROOT="${RUNNER_TEMP:-${TMPDIR:-/tmp}}/hermes-installer-script-e2e" +LOG_DIR="${HERMES_E2E_LOG_DIR:-$WORK_ROOT/logs}" +SERVE_REPO="$WORK_ROOT/serve.git" + +step() { printf '\n=== %s ===\n' "$*"; } +ok() { printf ' OK %s\n' "$*"; } +fail() { printf 'E2E ASSERTION FAILED: %s\n' "$*" >&2; exit 1; } + +rm -rf "$WORK_ROOT" +mkdir -p "$WORK_ROOT" "$LOG_DIR" + +# --- stage: serve.git with main parked at OLD -------------------------------- + +step "staging serve.git (main -> OLD)" +# Tracked changes only (-uno): the bare clone serves committed objects, so a +# modified tracked file means HEAD is not the code being reviewed -- but an +# untracked file (scratch notes, this driver before it lands) cannot leak +# into the clone at all. +[ -z "$(git -C "$REPO_ROOT" status --porcelain -uno)" ] \ + || fail "checkout has uncommitted tracked changes; the staged clone must be a reviewable commit" + +if [ -z "$INSTALL_REF" ]; then + INSTALL_REF="$(git -C "$REPO_ROOT" tag --list 'v[0-9]*' --sort=-creatordate | head -1)" + [ -n "$INSTALL_REF" ] || fail "no release tags in the checkout to use as OLD" +fi +OLD_SHA="$(git -C "$REPO_ROOT" rev-parse "${INSTALL_REF}^{commit}")" +HEAD_SHA="$(git -C "$REPO_ROOT" rev-parse HEAD)" +[ "$OLD_SHA" != "$HEAD_SHA" ] || fail "OLD ($INSTALL_REF) IS HEAD; no update would be available" + +git clone --bare --quiet "$REPO_ROOT" "$SERVE_REPO" +git -C "$SERVE_REPO" update-ref refs/heads/main "$OLD_SHA" +git -C "$SERVE_REPO" symbolic-ref HEAD refs/heads/main +# The installer may pin a commit that is reachable but not at a ref tip. +git -C "$SERVE_REPO" config uploadpack.allowAnySHA1InWant true +ok "serve.git main = $OLD_SHA ($INSTALL_REF), update target $HEAD_SHA" + +# --- the git URL redirect ----------------------------------------------------- + +# A driver-owned global gitconfig, NOT GIT_CONFIG_COUNT/KEY_n/VALUE_n env +# config: install.sh sets those itself and would clobber ours. +GIT_CFG="$WORK_ROOT/gitconfig" +cat > "$GIT_CFG" < "$script" + chmod +x "$script" + # Installer flags have to match the installer being run, not this + # checkout's: older releases reject options added later. --skip-setup goes + # back further than any tag we sample; anything newer is probed for. + local flags=(--skip-setup) + if installer_supports "$1" "--skip-browser"; then + flags+=(--skip-browser) + fi + # "$LOG_DIR/install-$2.log" 2>&1; then + tail -50 "$LOG_DIR/install-$2.log" >&2 + fail "install.sh ($2) exited non-zero; full log in $LOG_DIR/install-$2.log" + fi +} + +assert_checkout() { + # $1: expected sha, $2: label + local got + got="$(git -C "$INSTALL_DIR" rev-parse HEAD)" + [ "$got" = "$1" ] || fail "installed checkout is $got, expected $2 ($1)" + ok "checkout is $2 ($1)" + local hermes="$INSTALL_DIR/venv/bin/hermes" + [ -x "$hermes" ] || fail "no hermes console script at $hermes" + "$hermes" --version > "$LOG_DIR/version-$2.log" 2>&1 \ + || fail "hermes --version failed after $2; log in $LOG_DIR/version-$2.log" + ok "hermes --version works: $(head -c 120 "$LOG_DIR/version-$2.log" | tr -d '\n')" +} + +# --- install OLD --------------------------------------------------------------- + +step "installing OLD ($INSTALL_REF) via its own scripts/install.sh" +run_installer "$OLD_SHA" old +assert_checkout "$OLD_SHA" OLD + +# --- update OLD -> HEAD ---------------------------------------------------------- + +step "advancing served main to HEAD" +git -C "$SERVE_REPO" update-ref refs/heads/main "$HEAD_SHA" +ok "serve.git main = $HEAD_SHA" + +step "updating via $UPDATE_METHOD" +case "$UPDATE_METHOD" in + hermes-update) + # `--yes` reaches the update subcommand only in later releases, and + # argparse rejects the whole invocation when it does not exist. Ask the + # installed hermes; older ones read the prompt from stdin, so close it. + HERMES="$INSTALL_DIR/venv/bin/hermes" + if "$HERMES" update --help 2>&1 | grep -qF -- --yes; then + update_cmd=("$HERMES" update --yes) + else + update_cmd=("$HERMES" update) + fi + if ! (cd "$INSTALL_DIR" && "${update_cmd[@]}" < /dev/null > "$LOG_DIR/update.log" 2>&1); then + tail -50 "$LOG_DIR/update.log" >&2 + fail "hermes update exited non-zero; full log in $LOG_DIR/update.log" + fi + ;; + installer-script) + # A user re-running the one-liner today gets the CURRENT script. + run_installer "$HEAD_SHA" head + ;; +esac +assert_checkout "$HEAD_SHA" HEAD + +step "PASS: $INSTALL_REF -> HEAD via $UPDATE_METHOD" From ea4cd375f818134ada420c3c5ea53c65b0c9a223 Mon Sep 17 00:00:00 2001 From: ethernet Date: Tue, 11 Aug 2026 21:20:42 -0400 Subject: [PATCH 056/227] ci(install-e2e): retire the bubblewrap sandbox - git redirect everywhere, macos legs live The fake Internet (bubblewrap + slirp4netns + MITM proxy + upload-pack shim, 883 lines across dev-sandbox.sh, stage2-run.sh, proxy.py, ssh-shim.sh, openssl.cnf, install-update-e2e.sh) existed to isolate install.sh's network. The GIT_CONFIG_GLOBAL insteadOf redirect the windows driver introduced does the same job with a gitconfig file and works on any OS, so: * install-e2e-run.yml now runs tests/install/installer-script-e2e.sh directly on the bare runner - no sandbox deps, no userns sysctls - and takes a runner input; * the macos matrix calls the SAME workflow on macos-latest, deleting install-e2e-macos-run.yml: installer-script -> installer-script / hermes-update flip from grey to live, app-update pairs stay TODO inside the shared gate; * install.sh is no longer curl'd through a fake CA - each leg runs the copy from the ref a user of that version actually executed; * scripts/dev-sandbox.sh becomes the minimal isolation sandbox from ab6b9492f (separate HERMES_HOME / Electron userData / app name, same CLI surface: --persistent, --from, --delete), keeping its .hermes-sandbox dir name so gitignore and docs hold; * nix/sandbox.nix drops the bwrap/proxy closure and keeps only the Electron runtime LD_LIBRARY_PATH the desktop app needs. Verified: nix build .#sandbox + smoke run (isolated HERMES_HOME created, ephemeral cleanup), shellcheck/bash -n on both scripts, actionlint on all three workflows, and the new driver ran the full v0.20.2 -> HEAD hermes-update pass locally before this commit. --- .github/workflows/install-e2e-macos-run.yml | 51 -- .github/workflows/install-e2e-run.yml | 71 +-- .github/workflows/install-e2e.yml | 12 +- nix/sandbox.nix | 57 +- scripts/dev-sandbox.sh | 672 ++++---------------- scripts/sandbox/generate-e2e-matrix.mjs | 4 +- scripts/sandbox/openssl.cnf | 43 -- scripts/sandbox/proxy.py | 237 ------- scripts/sandbox/ssh-shim.sh | 13 - scripts/sandbox/stage2-run.sh | 251 -------- tests/install/install-update-e2e.sh | 293 --------- 11 files changed, 177 insertions(+), 1527 deletions(-) delete mode 100644 .github/workflows/install-e2e-macos-run.yml delete mode 100644 scripts/sandbox/openssl.cnf delete mode 100644 scripts/sandbox/proxy.py delete mode 100644 scripts/sandbox/ssh-shim.sh delete mode 100755 scripts/sandbox/stage2-run.sh delete mode 100755 tests/install/install-update-e2e.sh diff --git a/.github/workflows/install-e2e-macos-run.yml b/.github/workflows/install-e2e-macos-run.yml deleted file mode 100644 index 5767986bf0bb..000000000000 --- a/.github/workflows/install-e2e-macos-run.yml +++ /dev/null @@ -1,51 +0,0 @@ -name: Install & Update E2E — macOS (reusable) - -# Runs ONE {install-method, update-method} combination on macOS, installing -# a starting version and updating it to HEAD. -# -# NOTHING is implemented yet: every method pair NATIVELY SKIPS (grey check, -# no runner) until a macOS driver exists. The workflow exists now so -# install-e2e.yml can dispatch every declared macOS combination the same -# way it does for linux and windows -- capability knowledge lives here, -# next to where the driver will be, and implementing a method flips this -# workflow's job-level `if`. -# -# Declared methods (see the generator's SPEC): -# install: installer-script -# update: installer-script, hermes-update, app-update - -on: - workflow_call: - inputs: - install-method: - description: 'How the starting version gets installed. All macOS methods are TODO and skip.' - required: true - type: string - update-method: - description: 'How the install updates to HEAD. All macOS methods are TODO and skip.' - required: true - type: string - install-ref: - description: 'What to install before updating: a branch, a tag (v2026.7.7), or a SHA reachable from main.' - required: false - type: string - default: refs/heads/main - -permissions: - contents: read - -jobs: - e2e: - # Static name on purpose: the caller's job name already carries the - # method pair, and GitHub renders name expressions UNEXPANDED (literal - # "${{ inputs... }}") on natively skipped jobs. - name: e2e - # No macOS driver exists yet: every pair is a declared TODO, so this is - # constant-false until the first method lands. Written as an impossible - # input comparison rather than `if: false` because actionlint rejects - # constant conditions. - if: inputs.install-method == inputs.update-method && inputs.install-method == 'implemented' - runs-on: macos-latest - timeout-minutes: 5 - steps: - - run: 'true' diff --git a/.github/workflows/install-e2e-run.yml b/.github/workflows/install-e2e-run.yml index 021ce45ddf6c..559601cdcd39 100644 --- a/.github/workflows/install-e2e-run.yml +++ b/.github/workflows/install-e2e-run.yml @@ -1,14 +1,21 @@ name: Install & Update E2E (reusable) # Runs ONE {install-method, update-method} combination against ONE starting -# commit, in the dev sandbox, with a real install (uv, a managed Python, -# Node, the venv) behind it. +# commit, with a real install (uv, a managed Python, Node, the venv) behind +# it. # # Reusable so callers can fan out over the combinations that matter -- # update from the tip vs. from an older release, `hermes update` vs. # re-running the installer -- without duplicating the runner setup. Each leg -# is independent: its own sandbox, its own install, nothing rewound or -# shared. +# is independent: its own isolated HOME, its own install, nothing rewound +# or shared. +# +# No sandbox: tests/install/installer-script-e2e.sh points every git +# process at a local bare clone (url..insteadOf in a +# driver-owned GIT_CONFIG_GLOBAL) and isolates HOME, so the installer and +# updater run byte-for-byte against their real URLs on the bare runner -- +# which is disposable, and therefore IS the sandbox. That also makes this +# workflow OS-agnostic: the same driver runs on ubuntu and macos runners. # # Method ids come from scripts/sandbox/generate-e2e-matrix.mjs. Supported # today: install via installer-script, update via hermes-update or @@ -62,65 +69,31 @@ jobs: # Static name on purpose: the caller's job name already carries the # method pair, and GitHub renders name expressions UNEXPANDED (literal # "${{ inputs... }}") on natively skipped jobs. Short because it is - # only a rendered tail (" -> HEAD / e2e"). + # only a rendered tail (" / e2e"). name: e2e - # The pairs the sandbox driver can run today; anything else is a - # declared TODO and natively skips. + # The pairs the driver can run today; anything else is a declared TODO + # and natively skips. if: inputs.install-method == 'installer-script' && contains(fromJSON('["hermes-update", "installer-script"]'), inputs.update-method) runs-on: ${{ inputs.runner }} timeout-minutes: ${{ inputs.timeout-minutes }} steps: - # Full history: the sandbox fetches the starting commit and the test - # compares against this commit, so a shallow clone is not enough. + # Full history: the driver bare-clones this checkout as the repo the + # installer/updater talk to, and both OLD and HEAD must be reachable + # in that clone. A shallow clone cannot serve either need. - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2 with: fetch-depth: 0 - # bubblewrap + slirp4netns are what the sandbox is built on; util-linux - # supplies the `unshare` that builds the multi-uid userns for the - # user-level (non-root) install. - - name: Install sandbox dependencies - run: | - set -euo pipefail - sudo apt-get update -qq - sudo apt-get install -y -qq bubblewrap slirp4netns uidmap util-linux - - # Ubuntu 24.04 restricts unprivileged user namespaces through AppArmor, - # which is exactly what bwrap needs. Report the state before touching it - # so a future runner-image change is visible in the log rather than - # silently altering what this job proves. - - name: Permit unprivileged user namespaces - run: | - set -euo pipefail - echo "--- kernel userns settings (before)" - sysctl kernel.unprivileged_userns_clone 2>/dev/null || echo " (sysctl absent)" - sysctl kernel.apparmor_restrict_unprivileged_userns 2>/dev/null || echo " (sysctl absent)" - if sysctl -n kernel.apparmor_restrict_unprivileged_userns >/dev/null 2>&1; then - sudo sysctl -w kernel.apparmor_restrict_unprivileged_userns=0 - fi - echo "--- subuid/subgid for $(id -un)" - grep "^$(id -un):" /etc/subuid /etc/subgid || echo " (none — sandbox will say so)" - - name: Run install + update E2E run: | set -euo pipefail - # Method id -> the driver script's --route vocabulary. The one - # place that knows both names. - case '${{ inputs.update-method }}' in - hermes-update) route=update ;; - installer-script) route=installer ;; - *) echo "unreachable: update-method passed the job-level gate but has no route mapping" >&2; exit 1 ;; - esac - tests/install/install-update-e2e.sh \ - --route "$route" \ + tests/install/installer-script-e2e.sh \ + --update-method '${{ inputs.update-method }}' \ --install-ref '${{ inputs.install-ref }}' env: - # Outside the workspace on purpose: the script creates this directory - # up front, and an untracked dir inside the repo makes the worktree - # dirty -- which dev-sandbox reacts to by snapshotting the working - # copy into a fresh fake-main commit on every invocation, moving the - # update target mid-run. + # Outside the workspace on purpose: logs written into the repo + # would trip the driver's own dirty-tree guard. HERMES_E2E_LOG_DIR: ${{ runner.temp }}/e2e-logs # Artifact names cannot contain '/', and install-ref may be a full ref @@ -134,7 +107,7 @@ jobs: set -euo pipefail safe_ref='${{ inputs.install-ref }}' safe_ref="${safe_ref//\//-}" - echo "name=install-e2e-${{ inputs.update-method }}-${safe_ref}" >> "$GITHUB_OUTPUT" + echo "name=install-e2e-${{ runner.os }}-${{ inputs.update-method }}-${safe_ref}" >> "$GITHUB_OUTPUT" # The installer's own transcripts say far more than the assertion that # tripped when a real install breaks. diff --git a/.github/workflows/install-e2e.yml b/.github/workflows/install-e2e.yml index 229f46dda31c..8f5fb3aa9ecd 100644 --- a/.github/workflows/install-e2e.yml +++ b/.github/workflows/install-e2e.yml @@ -7,12 +7,14 @@ name: Install & Update E2E # generate-matrix expands it against the picked release tags into one leg # per {combination, tag}, split into one matrix job per OS: # -# Matrix: linux a real curl|bash install in the bubblewrap sandbox +# Matrix: linux the real curl|bash install one-liner, isolated by a +# git URL redirect to a local bare clone # (install-e2e-run.yml) # Matrix: windows the real desktop user flow: website Hermes-Setup.exe # clicked by AutoHotkey, update via the app, Playwright # clicking "Update now" (install-e2e-windows-run.yml) -# Matrix: macos no driver yet (install-e2e-macos-run.yml) +# Matrix: macos the same OS-agnostic driver as linux, on macos-latest +# (install-e2e-run.yml; app-update pairs are TODO) # # Every combination is dispatched to its OS's run workflow; the run # workflow natively skips (grey) what its driver cannot run yet -- an @@ -175,9 +177,13 @@ jobs: needs: generate-matrix strategy: fail-fast: false + # 10x-cost runners: keep concurrency low. + max-parallel: 2 matrix: ${{ fromJSON(needs.generate-matrix.outputs.macos) }} - uses: ./.github/workflows/install-e2e-macos-run.yml + # The same OS-agnostic driver as linux -- only the runner differs. + uses: ./.github/workflows/install-e2e-run.yml with: install-method: ${{ matrix.install_method }} update-method: ${{ matrix.update_method }} install-ref: ${{ matrix.install_ref }} + runner: macos-latest diff --git a/nix/sandbox.nix b/nix/sandbox.nix index cec5dc986807..c727274214ca 100644 --- a/nix/sandbox.nix +++ b/nix/sandbox.nix @@ -1,5 +1,7 @@ { - # electron deps + # Electron needs its native runtime libraries on LD_LIBRARY_PATH when the + # sandboxed command launches the desktop app (`sandbox hermes desktop`, + # `sandbox npm run dev`); nothing else in the sandbox is nix-specific. alsa-lib, at-spi2-atk, atk, @@ -29,28 +31,6 @@ libXtst, libxcb, - # sandbox deps - bash, - bubblewrap, - cacert, - coreutils, - curl, - gawk, - git, - glibc, - gnumake, - gnugrep, - gnused, - gzip, - nodejs_22, - openssl, - python3, - slirp4netns, - stdenv, - gnutar, - util-linux, - - # etc writeShellApplication, lib, }: @@ -88,37 +68,8 @@ let in writeShellApplication { name = "sandbox"; - runtimeInputs = [ - bash - bubblewrap - cacert - coreutils - curl - gawk - git - glibc.bin - gnumake - gnugrep - gnused - gzip - nodejs_22 - openssl - python3 - slirp4netns - stdenv.cc - gnutar - util-linux - ] - ++ electronRuntime; text = '' - export DEV_SANDBOX_REAL_CA_CERT=${cacert}/etc/ssl/certs/ca-bundle.crt - export DEV_SANDBOX_DYNAMIC_LINKER=${stdenv.cc.bintools.dynamicLinker} - export DEV_SANDBOX_NODE_DIR=${nodejs_22} - export DEV_SANDBOX_ELECTRON_LD_LIBRARY_PATH=${lib.makeLibraryPath electronRuntime} - # The script is imported into the store as a single file, so its own - # directory has no scripts/sandbox/ beside it. Point it at the assets - # (fake-internet proxy, ssh shim) explicitly. - export DEV_SANDBOX_ASSETS=${../scripts/sandbox} + export LD_LIBRARY_PATH=${lib.makeLibraryPath electronRuntime}''${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH} exec ${../scripts/dev-sandbox.sh} "$@" ''; } diff --git a/scripts/dev-sandbox.sh b/scripts/dev-sandbox.sh index dca11a72f3aa..368f4874c1ed 100755 --- a/scripts/dev-sandbox.sh +++ b/scripts/dev-sandbox.sh @@ -1,591 +1,199 @@ #!/usr/bin/env bash -# Run a command in a disposable, network-isolated fake Internet. +# Run a Hermes instance in an isolated sandbox — separate HERMES_HOME, +# separate Electron userData, and a distinct Desktop app name so it doesn't compete +# with your main desktop instance's single-instance lock. # -# The command runs in private user, mount, PID, and network namespaces. This -# script is stage 1: it builds the sandbox tree, mints the fake CA, and creates -# the user+network namespaces with `unshare` (see the namespace plan further -# down), then re-execs into scripts/sandbox/stage2-run.sh, which adds the -# mount/pid namespaces with bubblewrap and runs the payload. Its only writable -# filesystem is SANDBOX_ROOT. HTTP(S) goes to a local static MITM proxy; -# github.com SSH uses a sandbox-local git-upload-pack shim; neither transport -# can reach the host network. +# By default the sandbox is throwaway: a temp dir is created and removed on +# exit. Use --persistent to keep the sandbox across restarts (stored under +# .hermes-sandbox/ in the worktree git root). +# +# Usage: +# scripts/dev-sandbox.sh python -m hermes_cli.main +# scripts/dev-sandbox.sh hermes desktop +# scripts/dev-sandbox.sh electron . +# scripts/dev-sandbox.sh -- npm run dev # from apps/desktop/ +# scripts/dev-sandbox.sh --persistent hermes desktop +# scripts/dev-sandbox.sh --persistent -- npm run dev +# +# Seed the sandbox HERMES_HOME from an existing directory (e.g. your main +# ~/.hermes) so config, sessions, skills, etc. are pre-populated: +# scripts/dev-sandbox.sh --from ~/.hermes hermes desktop +# +# Override the app name (default: HermesSandbox): +# HERMES_DEV_SANDBOX_NAME=Staging scripts/dev-sandbox.sh hermes desktop +# +# Override the persistent sandbox dir name (default: .hermes-sandbox): +# HERMES_DEV_SANDBOX_DIR=.staging-sandbox scripts/dev-sandbox.sh --persistent hermes desktop set -euo pipefail SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)" -# Helper files the sandbox needs: the stage-2 script it re-execs into, plus the -# files it copies in (the fake-internet proxy, the ssh shim, the openssl config). -# They sit next to this script in the repo, but the Nix wrapper installs the -# script into the store on its own, so it exports DEV_SANDBOX_ASSETS to point -# here. -SANDBOX_ASSETS="${DEV_SANDBOX_ASSETS:-$SCRIPT_DIR/sandbox}" -for asset in proxy.py ssh-shim.sh openssl.cnf stage2-run.sh; do - [ -f "$SANDBOX_ASSETS/$asset" ] || { - echo "error: missing sandbox asset: $SANDBOX_ASSETS/$asset" >&2 - exit 1 - } -done - print_help() { cat <<'EOF' -Usage: dev-sandbox.sh [options] [--] - dev-sandbox.sh install [options] [--] [installer arguments...] +Usage: dev-sandbox.sh [--persistent] [--from DIR] [--] -Run COMMAND in a throwaway chroot-like bubblewrap sandbox. The sandbox has no -writable host mounts: only its own root, mounted at /work, is writable. +Run a Hermes instance in an isolated sandbox. Options: - --persistent Keep the whole sandbox under .hermes-sandbox/. - --delete Delete the persistent sandbox (asks first). - --root Install as uid 0 with the root FHS layout: code in - /usr/local/lib/hermes-agent, command in - /usr/local/bin. Default is the user-level layout. - --from DIR One-time copy of DIR into the sandbox's $HOME. - Existing persistent sandboxes are never overwritten. - --http-root DIR Copy DIR into the fake web server root for this run. - Requests map to DIR//; no URL is forwarded. - --installer PATH With `install`, serve PATH at the canonical install.sh - URL. Default: scripts/install.sh in this worktree. - --from-main With `install`, fetch the real upstream main installer - and repository, then advance fake main to this folder - after a successful install for update testing. - Shorthand for --install-ref refs/heads/main. - --install-ref REF Like --from-main, but installs REF instead of main: - a branch, a tag (v2026.7.7), or a SHA reachable from main. - Use it to test updating from an older release, not just - from the tip. - -h, --help Show this help. - -Option order matters: every option above is consumed by THIS script, and -parsing stops at the first argument it does not recognize. Everything from -that point on is passed through to the command (or, with `install`, to the -installer). Put sandbox options first and separate installer arguments with -`--`, otherwise they arrive here and fail: - - # WRONG — --from-main reaches install.sh, which rejects it - scripts/dev-sandbox.sh install --skip-setup --from-main - - # RIGHT - scripts/dev-sandbox.sh install --from-main -- --skip-setup - -Install layout: `install.sh` picks its layout from `id -u` alone, so uid is what -separates the two real-world Linux installs. By default the sandbox runs as an -unprivileged `hermes` user, giving the layout most people have — -$HERMES_HOME/hermes-agent plus a ~/.local/bin launcher. Pass --root for the FHS -one. Both are worth testing; they differ in more than paths (root also relocates -uv's Python to /usr/local/share for world-readability). - -The fake web server signs certificates with a CA trusted only inside this -sandbox. HTTP_PROXY/HTTPS_PROXY send fixture URLs there first; other HTTP(S) -requests pass through the sandbox's rootless outbound network. SSH to github.com -runs a sandbox-local upload-pack shim, never your SSH config, agent, -known-hosts file, or authorized keys. - -Fake github main always comes from this folder. If it has staged, unstaged, or -non-ignored untracked changes, the sandbox warns and creates a temporary local -commit containing them; it never stages or commits the real worktree. + --persistent Keep the sandbox dir across restarts (under the worktree + git root, in .hermes-sandbox/). Without this flag the + sandbox is a temp dir that is removed on exit. + --from DIR Copy DIR into the sandbox HERMES_HOME as the starting + point (config, sessions, skills, etc.). + Ignored if the sandbox HERMES_HOME already has content + (e.g. reusing a --persistent sandbox) to avoid clobbering. + --delete Delete the existing persistent sandbox in .hermes-sandbox. + -h, --help Show this help message. Environment: - HERMES_DEV_SANDBOX_DIR Sandbox directory name, relative to the repo root - (default: .hermes-sandbox). + HERMES_DEV_SANDBOX_NAME Override the app name (default: HermesSandbox) + HERMES_DEV_SANDBOX_DIR Override the persistent dir name (default: .hermes-sandbox) Examples: - # create a sandbox, install this branch as `main`, and then drop to a shell, - # skipping `hermes setup` & the browser tools for speed. - scripts/dev-sandbox.sh install --persistent -- --skip-setup --skip-browser - - # Install the official upstream main. You're dropped into a shell where - # you can run `hermes update`. - scripts/dev-sandbox.sh install --persistent --from-main - + dev-sandbox.sh hermes desktop + dev-sandbox.sh --persistent hermes desktop + dev-sandbox.sh --from ~/.hermes hermes desktop + dev-sandbox.sh -- npm run dev EOF } PERSISTENT=false DELETE=false -RUN_AS_USER=true SEED_DIR="" -HTTP_ROOT="" -INSTALL_SHORTCUT=false -INSTALLER_PATH="" -# Which upstream commit the sandbox installs before the update routes run. -# Empty means "install this worktree's own installer" (no upstream fetch); set, -# it is anything git can resolve -- a branch, a tag (v2026.7.7), or a SHA -# reachable from main -- so "can a user two releases back still update?" is -# expressible. --from-main is shorthand for refs/heads/main. -INSTALL_REF="" -UPSTREAM_URL="${HERMES_DEV_SANDBOX_UPSTREAM:-https://github.com/NousResearch/hermes-agent.git}" - -if [ "${1:-}" = install ]; then - INSTALL_SHORTCUT=true - shift -fi while [ "$#" -gt 0 ]; do case "$1" in - --persistent) PERSISTENT=true; shift ;; - --delete) DELETE=true; shift ;; - --root) RUN_AS_USER=false; shift ;; - --user) RUN_AS_USER=true; shift ;; # the default; accepted for symmetry + --persistent) + PERSISTENT=true + shift + ;; --from) - [ "$#" -ge 2 ] || { echo 'error: --from needs a directory' >&2; exit 1; } - SEED_DIR="$2"; shift 2 ;; - --http-root) - [ "$#" -ge 2 ] || { echo 'error: --http-root needs a directory' >&2; exit 1; } - HTTP_ROOT="$2"; shift 2 ;; - --installer) - [ "$#" -ge 2 ] || { echo 'error: --installer needs a file' >&2; exit 1; } - INSTALLER_PATH="$2"; shift 2 ;; - --from-main) INSTALL_REF="refs/heads/main"; shift ;; - --install-ref) - [ "$#" -ge 2 ] || { echo 'error: --install-ref needs a value' >&2; exit 1; } - INSTALL_REF="$2" - shift 2 ;; - --from=*|--http-root=*|--installer=*|--install-ref=*) - key="${1%%=*}"; value="${1#*=}" - [ -n "$value" ] || { echo "error: $key needs a value" >&2; exit 1; } - case "$key" in - --from) SEED_DIR="$value" ;; - --http-root) HTTP_ROOT="$value" ;; - --installer) INSTALLER_PATH="$value" ;; - --install-ref) INSTALL_REF="$value" ;; - esac - shift ;; - -h|--help) print_help; exit 0 ;; - --) shift; break ;; - *) break ;; + if [ "$#" -lt 2 ] || [[ "$2" == -* ]]; then + echo "error: --from requires a directory argument" >&2 + exit 1 + fi + SEED_DIR="$2" + shift 2 + ;; + --from=*) + SEED_DIR="${1#--from=}" + if [ -z "$SEED_DIR" ]; then + echo "error: --from requires a directory argument" >&2 + exit 1 + fi + shift + ;; + --delete) + DELETE=true + shift + ;; + -h|--help) + print_help + exit 0 + ;; + --) + shift + break + ;; + *) + break + ;; esac done -if [ "$INSTALL_SHORTCUT" = false ] && [ "$#" -eq 0 ]; then - print_help >&2 - exit 1 +if [ -n "$SEED_DIR" ]; then + if [ ! -d "$SEED_DIR" ]; then + echo "error: --from dir '$SEED_DIR' does not exist" >&2 + exit 1 + fi + # Resolve to absolute path so it's valid after we cd later. + SEED_DIR="$(cd "$SEED_DIR" && pwd)" fi -if [ -n "$INSTALLER_PATH" ] && [ "$INSTALL_SHORTCUT" = false ]; then - echo 'error: --installer is only valid with the install shortcut' >&2 - exit 1 -fi -if [ -n "$INSTALL_REF" ] && [ "$INSTALL_SHORTCUT" = false ]; then - echo 'error: --from-main / --install-ref are only valid with the install shortcut' >&2 - exit 1 -fi -if [ -n "$INSTALL_REF" ] && [ -n "$INSTALLER_PATH" ]; then - echo 'error: --from-main / --install-ref cannot be combined with --installer' >&2 +if [ "$#" -eq 0 ]; then + print_help >&2 exit 1 fi -for dir in "$SEED_DIR" "$HTTP_ROOT"; do - [ -z "$dir" ] || [ -d "$dir" ] || { echo "error: directory '$dir' does not exist" >&2; exit 1; } -done -GIT_ROOT="${HERMES_SANDBOX_SOURCE_ROOT:-$(git rev-parse --show-toplevel)}" -GIT_ROOT="$(cd "$GIT_ROOT" && pwd)" -if [ "$INSTALL_SHORTCUT" = true ] && [ -z "$INSTALL_REF" ] && [ -z "$INSTALLER_PATH" ]; then - INSTALLER_PATH="$GIT_ROOT/scripts/install.sh" -fi -if [ -n "$INSTALLER_PATH" ] && [ ! -f "$INSTALLER_PATH" ]; then - echo "error: installer '$INSTALLER_PATH' does not exist" >&2 - exit 1 -fi -COMMIT="$(git -C "$GIT_ROOT" rev-parse --verify 'HEAD^{commit}')" || { - echo "error: current folder has no HEAD commit" >&2 - exit 1 -} SANDBOX_DIR_NAME="${HERMES_DEV_SANDBOX_DIR:-.hermes-sandbox}" -PERSISTENT_ROOT="$GIT_ROOT/$SANDBOX_DIR_NAME" +GIT_ROOT="$(git rev-parse --show-toplevel 2>/dev/null || echo "$SCRIPT_DIR/..")" +GIT_ROOT="$(cd "$GIT_ROOT" && pwd)" +PERSISTENT_SANDBOX_ROOT="$GIT_ROOT/$SANDBOX_DIR_NAME" if [ "$DELETE" = true ]; then - if [ ! -d "$PERSISTENT_ROOT" ]; then - echo "[sandbox] nothing to delete at $PERSISTENT_ROOT" >&2 - exit 0 + if [ -d "$PERSISTENT_SANDBOX_ROOT" ]; then + read -r -p "[sandbox] delete $PERSISTENT_SANDBOX_ROOT? [y/N] " REPLY + case "$REPLY" in + [yY]|[yY][eE][sS]) + echo "[sandbox] deleting $PERSISTENT_SANDBOX_ROOT" >&2 + rm -rf -- "$PERSISTENT_SANDBOX_ROOT" + ;; + *) + echo "[sandbox] aborted" >&2 + exit 1 + ;; + esac + else + echo "[sandbox] nothing to delete at $PERSISTENT_SANDBOX_ROOT" >&2 fi - read -r -p "[sandbox] delete $PERSISTENT_ROOT? [y/N] " reply - case "$reply" in - y|Y|yes|YES) rm -rf -- "$PERSISTENT_ROOT" ;; - *) echo '[sandbox] aborted' >&2; exit 1 ;; - esac exit 0 fi +# Derive a per-worktree app name so multiple checkouts don't collide. +# Each worktree has its own toplevel path even though they share one repo, +# so we hash that path into a short, stable suffix. +WORKTREE_ROOT="$(git rev-parse --show-toplevel 2>/dev/null || echo "$SCRIPT_DIR/..")" +WORKTREE_ROOT="$(cd "$WORKTREE_ROOT" && pwd)" +WORKTREE_HASH="$(printf '%s' "$WORKTREE_ROOT" | cksum | cut -d' ' -f1)" +WORKTREE_NAME="$(basename "$WORKTREE_ROOT")" +DEFAULT_SANDBOX_NAME="HermesSandbox-${WORKTREE_NAME}-${WORKTREE_HASH}" + +SANDBOX_NAME="${HERMES_DEV_SANDBOX_NAME:-$DEFAULT_SANDBOX_NAME}" + if [ "$PERSISTENT" = true ]; then - SANDBOX_ROOT="$PERSISTENT_ROOT" + SANDBOX_ROOT="$PERSISTENT_SANDBOX_ROOT" else SANDBOX_ROOT="$(mktemp -d -t hermes-sandbox.XXXXXX)" - cleanup() { chmod -R u+w "$SANDBOX_ROOT"; rm -rf -- "$SANDBOX_ROOT"; } - trap cleanup EXIT INT TERM fi -mkdir -p "$SANDBOX_ROOT"/{root,home,etc} -UPSTREAM_REPO="" -UPSTREAM_COMMIT="" -if [ -n "$INSTALL_REF" ]; then - echo "[sandbox] fetching upstream $INSTALL_REF for installer/update test" >&2 - UPSTREAM_REPO="$(mktemp -d -t hermes-sandbox-upstream.XXXXXX)" - git -C "$UPSTREAM_REPO" init -q - # Fetch the ref as given. A branch or tag name resolves on its own; a raw SHA - # needs the remote to allow fetching it directly, so fall back to fetching - # main and resolving the SHA locally (which works for any commit that is an - # ancestor of main -- the interesting case for "update from N versions ago"). - # - # Peel to ^{commit} in both cases: an annotated tag fetches as a tag OBJECT, - # and using it directly fails later with "trying to write non-commit object - # ... to branch 'refs/heads/main'". - if git -C "$UPSTREAM_REPO" fetch -q "$UPSTREAM_URL" "$INSTALL_REF" 2>/dev/null; then - UPSTREAM_COMMIT="$(git -C "$UPSTREAM_REPO" rev-parse "FETCH_HEAD^{commit}")" - elif git -C "$UPSTREAM_REPO" fetch -q "$UPSTREAM_URL" refs/heads/main \ - && UPSTREAM_COMMIT="$(git -C "$UPSTREAM_REPO" rev-parse --verify -q "$INSTALL_REF^{commit}")"; then - : - else - rm -rf -- "$UPSTREAM_REPO" - echo "error: could not resolve upstream ref: $INSTALL_REF" >&2 - echo ' Use a branch (main), a tag (v2026.7.7), or a SHA reachable from main.' >&2 - exit 1 - fi -fi -if [ ! -e "$SANDBOX_ROOT/root/repo/.sandbox-source" ]; then - mkdir -p "$SANDBOX_ROOT/root/repo" - # Persistent roots live under the worktree, so copying with cp would recurse - # into the sandbox itself. tar also lets us exclude a worktree's .git file, - # which can point at the host's shared worktree metadata. - tar -C "$GIT_ROOT" --exclude='./.git' --exclude="./$SANDBOX_DIR_NAME" -cf - . \ - | tar -C "$SANDBOX_ROOT/root/repo" -xf - - : > "$SANDBOX_ROOT/root/repo/.sandbox-source" -fi +export HERMES_HOME="$SANDBOX_ROOT/hermes-home" +export HERMES_DESKTOP_USER_DATA_DIR="$SANDBOX_ROOT/user-data" +export HERMES_DESKTOP_APP_NAME="$SANDBOX_NAME" -if [ -n "$SEED_DIR" ] && [ ! -e "$SANDBOX_ROOT/.seeded" ]; then - echo "[sandbox] seeding home from $SEED_DIR" >&2 - cp -a "$SEED_DIR/." "$SANDBOX_ROOT/home/" - : > "$SANDBOX_ROOT/.seeded" -fi +mkdir -p "$HERMES_HOME" "$HERMES_DESKTOP_USER_DATA_DIR" -rm -rf "$SANDBOX_ROOT/root/http" -mkdir -p "$SANDBOX_ROOT/root/http" -if [ -n "$HTTP_ROOT" ]; then - cp -a "$HTTP_ROOT/." "$SANDBOX_ROOT/root/http/" -fi -if [ "$INSTALL_SHORTCUT" = true ]; then - mkdir -p "$SANDBOX_ROOT/root/http/hermes-agent.nousresearch.com" - if [ -n "$INSTALL_REF" ]; then - git -C "$UPSTREAM_REPO" show "$UPSTREAM_COMMIT:scripts/install.sh" \ - > "$SANDBOX_ROOT/root/http/hermes-agent.nousresearch.com/install.sh" +if [ -n "$SEED_DIR" ]; then + # Only seed when the sandbox HERMES_HOME is empty — avoids clobbering an + # existing persistent sandbox on re-run. + if [ -z "$(ls -A "$HERMES_HOME" 2>/dev/null)" ]; then + echo "[sandbox] seeding HERMES_HOME from $SEED_DIR" >&2 + cp -a "$SEED_DIR/." "$HERMES_HOME/" else - cp -a "$INSTALLER_PATH" "$SANDBOX_ROOT/root/http/hermes-agent.nousresearch.com/install.sh" + echo "[sandbox] --from ignored: $HERMES_HOME already has content" >&2 fi - set -- bash -c ' - set +e - curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash -s -- "$@" - install_status=$? - if [ "$install_status" -eq 0 ] && [ -f /work/promote-main ]; then - next_main=$(cat /work/promote-main) - if git --git-dir=/work/repos/hermes-agent.git update-ref refs/heads/main "$next_main"; then - rm -f /work/promote-main - printf "[sandbox] fake main advanced to this folder for update testing\n" >&2 - else - printf "[sandbox] failed to advance fake main after install\n" >&2 - install_status=1 - fi - fi - if [ "$DEV_SANDBOX_INTERACTIVE" = true ]; then - printf "\n[sandbox] installer exited %s; entering sandbox shell\n" "$install_status" >&2 - exec /dev/tty 2>&1 - exec bash -i - fi - exit "$install_status" - ' sandbox-installer "$@" fi -mkdir -p "$SANDBOX_ROOT/root"/{bin,certs,lib64,logs,repos,ssh,usr/bin,usr/local} -REAL_CA_CERT="${DEV_SANDBOX_REAL_CA_CERT:-}" -if [ -z "$REAL_CA_CERT" ]; then - for candidate in /etc/ssl/certs/ca-certificates.crt /etc/ssl/cert.pem; do - if [ -f "$candidate" ]; then - REAL_CA_CERT="$candidate" - break - fi - done -fi -if [ ! -f "$REAL_CA_CERT" ]; then - echo 'error: no system CA bundle found for outbound sandbox HTTPS' >&2 - exit 1 -fi -if [ ! -f "$SANDBOX_ROOT/root/certs/real-ca.pem" ]; then - cp "$REAL_CA_CERT" "$SANDBOX_ROOT/root/certs/real-ca.pem" -fi -printf 'nameserver 10.0.2.3\n' > "$SANDBOX_ROOT/etc/resolv.conf" -SANDBOX_SHELL="$(command -v bash)" -DYNAMIC_LINKER="${DEV_SANDBOX_DYNAMIC_LINKER:-}" -if [ -z "$DYNAMIC_LINKER" ]; then - # Nix store first: NixOS also ships a /lib64/ld-linux-x86-64.so.2 compat stub, - # so probing FHS paths first would quietly switch which loader a bare script - # invocation uses on this host. Globs that match nothing expand to themselves, - # so every candidate is -f tested. The FHS paths cover Debian/Ubuntu (where - # the loader is under /lib64 or a multiarch /lib dir), which is what CI runs. - for candidate in \ - /nix/store/*-glibc-*/lib/ld-linux-*.so.* \ - /lib64/ld-linux-x86-64.so.2 \ - /lib/ld-linux-aarch64.so.1 \ - /lib/x86_64-linux-gnu/ld-linux-x86-64.so.2 \ - /lib/aarch64-linux-gnu/ld-linux-aarch64.so.1 - do - if [ -f "$candidate" ]; then - DYNAMIC_LINKER="$candidate" - break - fi - done -fi -if [ ! -f "$DYNAMIC_LINKER" ]; then - echo 'error: no glibc dynamic linker found for sandboxed release binaries' >&2 - echo ' Set DEV_SANDBOX_DYNAMIC_LINKER to its path.' >&2 - exit 1 -fi -ln -sf "$SANDBOX_SHELL" "$SANDBOX_ROOT/root/bin/sh" -ln -sf "$(command -v ls)" "$SANDBOX_ROOT/root/bin/ls" -ln -sf "$(command -v env)" "$SANDBOX_ROOT/root/usr/bin/env" -ln -sf "$DYNAMIC_LINKER" "$SANDBOX_ROOT/root/lib64/$(basename "$DYNAMIC_LINKER")" -# Identity inside the sandbox. install.sh chooses its layout from `id -u` -# alone (see resolve_install_layout), so the uid here is what decides between -# the root FHS install and a user-level one. -if [ "$RUN_AS_USER" = true ]; then - SANDBOX_UID=1000 - SANDBOX_GID=1000 - SANDBOX_USER=hermes - SANDBOX_HOME=/home/hermes -else - SANDBOX_UID=0 - SANDBOX_GID=0 - SANDBOX_USER=root - SANDBOX_HOME=/root -fi -{ - printf 'root:x:0:0:Sandbox Root:/root:%s\n' "$SANDBOX_SHELL" - if [ "$RUN_AS_USER" = true ]; then - printf '%s:x:%s:%s:Sandbox User:%s:%s\n' \ - "$SANDBOX_USER" "$SANDBOX_UID" "$SANDBOX_GID" "$SANDBOX_HOME" "$SANDBOX_SHELL" - fi -} > "$SANDBOX_ROOT/etc/passwd" -{ - printf 'root:x:0:\n' - if [ "$RUN_AS_USER" = true ]; then - printf '%s:x:%s:\n' "$SANDBOX_USER" "$SANDBOX_GID" - fi -} > "$SANDBOX_ROOT/etc/group" -# A user-level install writes the `hermes` launcher to ~/.local/bin and the -# checkout to $HERMES_HOME; both live under the sandbox HOME, which is bound -# from $SANDBOX_ROOT/home. bwrap maps our real uid to $SANDBOX_UID, so the -# host-side ownership of that directory is what the sandbox sees as its own. -printf 'hosts: files dns\n' > "$SANDBOX_ROOT/etc/nsswitch.conf" -printf '127.0.0.1 localhost\n' > "$SANDBOX_ROOT/etc/hosts" - -SOURCE_REPO="$GIT_ROOT" -SOURCE_REF="$COMMIT" -SNAPSHOT_REPO="" -FAKE_REPO="$SANDBOX_ROOT/root/repos/hermes-agent.git" -git -C "$SANDBOX_ROOT/root/repos" init --bare -q hermes-agent.git -if [ -n "$INSTALL_REF" ]; then - git --git-dir="$FAKE_REPO" fetch -q --force "$UPSTREAM_REPO" \ - "$UPSTREAM_COMMIT:refs/heads/main" -fi -if [ -n "$(git -C "$GIT_ROOT" status --porcelain)" ]; then - echo '[sandbox] warning: current folder is dirty; creating a temporary fake commit for main' >&2 - SNAPSHOT_REPO="$(mktemp -d -t hermes-sandbox-snapshot.XXXXXX)" - git -C "$SNAPSHOT_REPO" init -q - git -C "$SNAPSHOT_REPO" fetch -q "$GIT_ROOT" "$COMMIT" - git -C "$SNAPSHOT_REPO" config user.name 'Hermes sandbox' - git -C "$SNAPSHOT_REPO" config user.email 'sandbox@invalid' - GIT_DIR="$SNAPSHOT_REPO/.git" GIT_WORK_TREE="$GIT_ROOT" git read-tree "$COMMIT" - GIT_DIR="$SNAPSHOT_REPO/.git" GIT_WORK_TREE="$GIT_ROOT" \ - git add -A -- . - SNAPSHOT_TREE="$(GIT_DIR="$SNAPSHOT_REPO/.git" git write-tree)" - SNAPSHOT_PARENT="$COMMIT" - if EXISTING_MAIN="$(git --git-dir="$FAKE_REPO" rev-parse --verify refs/heads/main 2>/dev/null)"; then - git -C "$SNAPSHOT_REPO" fetch -q "$FAKE_REPO" "$EXISTING_MAIN" - SNAPSHOT_PARENT="$EXISTING_MAIN" - fi - SOURCE_REF="$(GIT_DIR="$SNAPSHOT_REPO/.git" git commit-tree "$SNAPSHOT_TREE" -p "$SNAPSHOT_PARENT" \ - -m 'sandbox snapshot of dirty worktree')" - SOURCE_REPO="$SNAPSHOT_REPO" -fi - -if [ -n "$INSTALL_REF" ]; then - git --git-dir="$FAKE_REPO" fetch -q --force "$SOURCE_REPO" \ - "$SOURCE_REF:refs/hermes-sandbox/next" - printf '%s\n' "$SOURCE_REF" > "$SANDBOX_ROOT/root/promote-main" -else - git --git-dir="$FAKE_REPO" fetch -q --force "$SOURCE_REPO" \ - "$SOURCE_REF:refs/heads/main" -fi -git --git-dir="$FAKE_REPO" symbolic-ref HEAD refs/heads/main -if [ -n "$SNAPSHOT_REPO" ]; then - # Best-effort: it is a mktemp directory the OS will reap, and failing the whole - # run over a leftover object file would be worse than leaking it. Concurrent - # git activity in the worktree can still be writing here as we delete. - rm -rf -- "$SNAPSHOT_REPO" 2>/dev/null || true -fi -if [ -n "$UPSTREAM_REPO" ]; then - rm -rf -- "$UPSTREAM_REPO" -fi - -# openssl reads a config even for `req -addext`, and its compiled-in path is a -# symlink into /etc/ssl on Debian/Ubuntu -- which the sandbox replaces. Ship our -# own and point OPENSSL_CONF at it, both here and inside the sandbox. -cp "$SANDBOX_ASSETS/openssl.cnf" "$SANDBOX_ROOT/root/certs/openssl.cnf" - -if [ ! -f "$SANDBOX_ROOT/root/certs/ca.pem" ]; then - if ! ca_error="$(OPENSSL_CONF="$SANDBOX_ROOT/root/certs/openssl.cnf" \ - openssl req -x509 -newkey rsa:2048 -nodes -days 2 \ - -subj '/CN=Hermes dev sandbox CA' \ - -extensions sandbox_ca_ext \ - -keyout "$SANDBOX_ROOT/root/certs/ca.key" \ - -out "$SANDBOX_ROOT/root/certs/ca.pem" 2>&1 >/dev/null)"; then - echo 'error: could not create the sandbox CA:' >&2 - printf '%s\n' "$ca_error" >&2 - exit 1 - fi -fi -GIT_UPLOAD_PACK="$(command -v git-upload-pack)" -sed "s|@GIT_UPLOAD_PACK@|$GIT_UPLOAD_PACK|" "$SANDBOX_ASSETS/ssh-shim.sh" \ - > "$SANDBOX_ROOT/root/usr/bin/ssh" -chmod 700 "$SANDBOX_ROOT/root/usr/bin/ssh" - -# The fake-internet proxy and the ssh shim are real files under -# scripts/sandbox/ rather than heredocs, so they can be linted, syntax-checked -# and diffed like any other source. Copy them into the sandbox tree. -cp "$SANDBOX_ASSETS/proxy.py" "$SANDBOX_ROOT/root/proxy.py" - -if [ -n "$INSTALL_REF" ]; then - echo "[sandbox] fake main: upstream $INSTALL_REF ($UPSTREAM_COMMIT)" >&2 - echo "[sandbox] prepared update: current folder ($SOURCE_REF)" >&2 -else - echo "[sandbox] fake main: current folder ($SOURCE_REF)" >&2 -fi -echo "[sandbox] root: $SANDBOX_ROOT" >&2 -echo "[sandbox] http root: $SANDBOX_ROOT/root/http" >&2 -if [ "$RUN_AS_USER" = true ]; then - echo "[sandbox] identity: $SANDBOX_USER (uid $SANDBOX_UID) — installs are user-level under $SANDBOX_HOME" >&2 +echo "[sandbox] HERMES_HOME=$HERMES_HOME" >&2 +echo "[sandbox] userData=$HERMES_DESKTOP_USER_DATA_DIR" >&2 +echo "[sandbox] appName=$HERMES_DESKTOP_APP_NAME" >&2 +if [ "$PERSISTENT" = true ]; then + echo "[sandbox] persistent: $SANDBOX_ROOT" >&2 else - echo '[sandbox] identity: root (uid 0) — installs use the /usr/local FHS layout' >&2 + echo "[sandbox] ephemeral (will be cleaned up on exit)" >&2 fi -[ "$PERSISTENT" = true ] && echo '[sandbox] persistent' >&2 || echo '[sandbox] ephemeral' >&2 -for command in awk bash bwrap curl git openssl python3 slirp4netns tar unshare; do - command -v "$command" >/dev/null || { - echo "error: missing required command: $command" >&2 - exit 1 +if [ "$PERSISTENT" = false ]; then + # shellcheck disable=SC2329 # invoked via trap + cleanup() { + chmod -R u+w "$SANDBOX_ROOT" + rm -rf -- "$SANDBOX_ROOT" } -done - -INTERACTIVE=false -if [ -t 0 ] && [ -t 1 ]; then - INTERACTIVE=true -fi -NODE_DIR="${DEV_SANDBOX_NODE_DIR:-}" -if [ -z "$NODE_DIR" ] && command -v node >/dev/null; then - NODE_DIR="$(dirname "$(dirname "$(command -v node)")")" -fi -WAYLAND_SOCKET="" -if [ -n "${XDG_RUNTIME_DIR:-}" ] && [ -n "${WAYLAND_DISPLAY:-}" ] \ - && [ -S "$XDG_RUNTIME_DIR/$WAYLAND_DISPLAY" ]; then - WAYLAND_SOCKET="$XDG_RUNTIME_DIR/$WAYLAND_DISPLAY" -fi - -# Namespace plan (stage 1 -> stage 2). -# -# slirp4netns joins the target's userns and setuids to root before configuring -# the netns, so the userns MUST map a uid 0. bwrap's own --unshare-user maps -# exactly one uid, so it cannot both run the payload as uid 1000 and offer slirp -# a root to become: that combination fails with -# setns(CLONE_NEWNET): Operation not permitted. -# -# So stage 1 builds the namespaces here with two ranges: -# inner 0 <- a subuid, unused by the payload, present only so slirp can -# become root inside the namespace -# inner $SANDBOX_UID <- our real host uid, so everything the sandbox writes -# stays owned by us and `rm -rf` on a persistent sandbox needs -# no privileges or chown dance -# The payload then runs in stage 2, where bwrap adds the mount/pid namespaces -# without creating a userns at all. -# -# The root layout needs no subuid at all: inner 0 IS the host uid there. -netns_args=(--user --net) -if [ "$RUN_AS_USER" = true ]; then - host_user="$(id -un)" - subuid_base="$(awk -F: -v u="$host_user" '$1 == u {print $2; exit}' /etc/subuid)" - subgid_base="$(awk -F: -v u="$host_user" '$1 == u {print $2; exit}' /etc/subgid)" - if [ -z "$subuid_base" ] || [ -z "$subgid_base" ]; then - echo "error: no /etc/subuid or /etc/subgid range for $host_user" >&2 - echo ' A user-level sandbox needs one spare subordinate id to host' >&2 - echo " its internal root. Add e.g. '$host_user:100000:65536' to both," >&2 - echo ' or use --root.' >&2 - exit 1 - fi - netns_args+=( - --map-users="0:$subuid_base:1" --map-users="$SANDBOX_UID:$(id -u):1" - --map-groups="0:$subgid_base:1" --map-groups="$SANDBOX_GID:$(id -g):1" - ) -else - netns_args+=(--map-root-user) -fi - -sandbox_pid_file="$SANDBOX_ROOT/root/logs/sandbox.pid" -slirp_ready="$SANDBOX_ROOT/root/logs/slirp.ready" -slirp_log="$SANDBOX_ROOT/root/logs/slirp.log" -: > "$sandbox_pid_file" -: > "$slirp_ready" - -env \ - DEV_SANDBOX_ROOT="$SANDBOX_ROOT" \ - DEV_SANDBOX_BASH="$(command -v bash)" \ - DEV_SANDBOX_REAL_CA_CERT="$REAL_CA_CERT" \ - DEV_SANDBOX_INTERACTIVE="$INTERACTIVE" \ - DEV_SANDBOX_USER="$SANDBOX_USER" \ - DEV_SANDBOX_HOME="$SANDBOX_HOME" \ - DEV_SANDBOX_NODE_DIR="$NODE_DIR" \ - DEV_SANDBOX_ELECTRON_LD_LIBRARY_PATH="${DEV_SANDBOX_ELECTRON_LD_LIBRARY_PATH:-}" \ - DEV_SANDBOX_XDG_RUNTIME_DIR="${XDG_RUNTIME_DIR:-}" \ - DEV_SANDBOX_WAYLAND_DISPLAY="${WAYLAND_DISPLAY:-}" \ - DEV_SANDBOX_WAYLAND_SOCKET="$WAYLAND_SOCKET" \ - unshare "${netns_args[@]}" \ - "$SANDBOX_ASSETS/stage2-run.sh" "$@" & -sandbox_launcher=$! - -for _ in $(seq 1 200); do - [ -s "$sandbox_pid_file" ] && break - if ! kill -0 "$sandbox_launcher" 2>/dev/null; then - wait "$sandbox_launcher" - exit $? - fi - sleep 0.05 -done -sandbox_pid="$(tr -dc '0-9' < "$sandbox_pid_file")" -if [ -z "$sandbox_pid" ]; then - echo 'error: sandbox did not report its PID' >&2 - exit 1 -fi - -slirp4netns --configure --disable-host-loopback --ready-fd=3 \ - --userns-path="/proc/$sandbox_pid/ns/user" "$sandbox_pid" tap0 \ - 3>"$slirp_ready" >"$slirp_log" 2>&1 & -slirp_pid=$! -cleanup_slirp() { - kill "$slirp_pid" 2>/dev/null || true - wait "$slirp_pid" 2>/dev/null || true -} -trap cleanup_slirp EXIT INT TERM - -for _ in $(seq 1 200); do - [ -s "$slirp_ready" ] && break - if ! kill -0 "$slirp_pid" 2>/dev/null; then - cat "$slirp_log" >&2 || true - exit 1 - fi - sleep 0.05 -done -if [ ! -s "$slirp_ready" ]; then - echo 'error: timed out waiting for sandbox network setup' >&2 - exit 1 + trap cleanup EXIT + trap 'cleanup; exit 130' INT TERM fi -wait "$sandbox_launcher" -exit $? \ No newline at end of file +"$@" +rc=$? +exit $rc \ No newline at end of file diff --git a/scripts/sandbox/generate-e2e-matrix.mjs b/scripts/sandbox/generate-e2e-matrix.mjs index 31297e8a2369..6abbdabb3881 100644 --- a/scripts/sandbox/generate-e2e-matrix.mjs +++ b/scripts/sandbox/generate-e2e-matrix.mjs @@ -8,8 +8,8 @@ * about which combinations CI can drive. Every combination is dispatched to * its OS's run workflow, and THAT workflow natively skips the method pairs * its driver cannot run yet -- capability knowledge lives next to each - * driver (install-e2e-run.yml, install-e2e-windows-run.yml, - * install-e2e-macos-run.yml). Correctness here is enforced by the type + * driver (install-e2e-run.yml for linux AND macos, + * install-e2e-windows-run.yml). Correctness here is enforced by the type * unions below (checked via `tsc --checkJs`), not by runtime validation -- * anything the types can't catch is self-evident on the next CI run. * diff --git a/scripts/sandbox/openssl.cnf b/scripts/sandbox/openssl.cnf deleted file mode 100644 index 04884355bd16..000000000000 --- a/scripts/sandbox/openssl.cnf +++ /dev/null @@ -1,43 +0,0 @@ -# Minimal openssl config for the dev sandbox. -# -# The sandbox replaces /etc wholesale, and on Debian/Ubuntu -# /usr/lib/ssl/openssl.cnf (openssl's compiled-in OPENSSLDIR) is a symlink into -# /etc/ssl -- so the config openssl insists on reading disappears and every -# `openssl req` fails with: -# -# Can't open "/usr/lib/ssl/openssl.cnf" for reading -# -# which surfaces to the payload as a bare `curl: (35) Recv failure`. Rather than -# reconstruct each distro's /etc/ssl, point OPENSSL_CONF at this file: the proxy -# only needs enough config for `req -addext` and `x509 -copy_extensions`. - -[ req ] -distinguished_name = req_distinguished_name - -[ req_distinguished_name ] - -# Used by `req -x509` for the sandbox's own CA. Without an explicit -# basicConstraints the generated certificate is not a CA, and every leaf it -# signs is rejected by the client with "invalid CA certificate (79)". -[ sandbox_ca_ext ] -basicConstraints = critical,CA:true -keyUsage = critical,keyCertSign,cRLSign -subjectKeyIdentifier = hash - -[ ca ] -default_ca = sandbox_ca - -[ sandbox_ca ] -default_md = sha256 -policy = policy_anything -email_in_dn = no -preserve = no - -[ policy_anything ] -commonName = optional -countryName = optional -stateOrProvinceName = optional -localityName = optional -organizationName = optional -organizationalUnitName = optional -emailAddress = optional diff --git a/scripts/sandbox/proxy.py b/scripts/sandbox/proxy.py deleted file mode 100644 index f34c9a843a9a..000000000000 --- a/scripts/sandbox/proxy.py +++ /dev/null @@ -1,237 +0,0 @@ -"""MITM proxy backing the dev sandbox's fake Internet. - -Listens on 127.0.0.1:8080 and is pointed at by http_proxy/https_proxy inside -the sandbox. For each request it either serves a fixture from the filesystem or -forwards to the real host: - -* ``//`` exists -> serve it. This is how the sandbox answers - the canonical install URL with the installer under test, so the payload can - run the true ``curl -fsSL https://…/install.sh | bash`` one-liner. -* otherwise -> forward upstream, verifying against the real CA bundle. The - sandbox is isolated from the *host*, not from the internet: a real install - still has to reach PyPI and npm. - -HTTPS is intercepted by minting a per-host certificate from the sandbox's own -throwaway CA, which the payload trusts via CURL_CA_BUNDLE / SSL_CERT_FILE. - -Usage: proxy.py -""" - -import os -import pathlib -import socket -import ssl -import subprocess -import sys -import threading -from urllib.parse import unquote, urlsplit - -ROOT, CERTS, REAL_CA = map(pathlib.Path, sys.argv[1:]) - -LISTEN_ADDRESS = ('127.0.0.1', 8080) -MAX_REQUEST_BYTES = 65536 -UPSTREAM_TIMEOUT_SECONDS = 30 -CERT_VALIDITY_DAYS = 2 - - -def read_request(conn): - data = b"" - while b"\r\n\r\n" not in data and len(data) < MAX_REQUEST_BYTES: - part = conn.recv(4096) - if not part: - return b"" - data += part - return data - - -def run_openssl(args): - """Run openssl, raising with its stderr when it fails. - - Discarding stderr here costs real debugging time: the caller sees only a - dropped connection (``curl: (35) Recv failure``) and the log holds nothing - but the argv, so an unwritable directory, a missing CA key, and an option - the host's openssl rejects all look identical. - """ - done = subprocess.run( - ['openssl', *args], stdout=subprocess.DEVNULL, stderr=subprocess.PIPE - ) - if done.returncode != 0: - detail = done.stderr.decode('utf-8', 'replace').strip() - raise RuntimeError( - f'openssl {args[0]} failed (exit {done.returncode}): {detail}' - ) - - -_CERT_LOCK = threading.Lock() - - -def cert_for(host): - """Return a (cert, key) pair for host, minting it from the sandbox CA. - - Minting is serialized and published atomically. The proxy is threaded, so - two concurrent requests for the same host would otherwise both run openssl - into the same paths, and a reader could pick up a finished certificate - beside a key from the other writer -- which TLS rejects as - ``[X509: KEY_VALUES_MISMATCH] key values mismatch``. - """ - safe = ''.join(char if char.isalnum() or char in '.-' else '_' for char in host) - cert, key = CERTS / f'{safe}.pem', CERTS / f'{safe}.key' - if cert.exists() and key.exists(): - return cert, key - with _CERT_LOCK: - # Re-check: another thread may have finished while we waited. - if cert.exists() and key.exists(): - return cert, key - # Build under unique temp names, then rename into place. os.replace is - # atomic, so a reader sees either the old pair or the new one, never a - # half-written mix. The key lands first: the certificate's existence is - # what everything else keys off. - stamp = f'{os.getpid()}.{threading.get_ident()}' - tmp_key = CERTS / f'{safe}.key.{stamp}' - tmp_cert = CERTS / f'{safe}.pem.{stamp}' - csr = CERTS / f'{safe}.csr.{stamp}' - run_openssl([ - 'req', '-newkey', 'rsa:2048', '-nodes', - '-subj', f'/CN={host}', - '-addext', f'subjectAltName=DNS:{host}', - '-keyout', str(tmp_key), '-out', str(csr), - ]) - run_openssl([ - 'x509', '-req', '-days', str(CERT_VALIDITY_DAYS), '-in', str(csr), - '-CA', str(CERTS / 'ca.pem'), '-CAkey', str(CERTS / 'ca.key'), - '-CAcreateserial', '-copy_extensions', 'copy', '-out', str(tmp_cert), - ]) - csr.unlink(missing_ok=True) - os.replace(tmp_key, key) - os.replace(tmp_cert, cert) - return cert, key - - -def file_for(host, target): - """Resolve a request to a fixture file, or None to forward upstream.""" - path = urlsplit(target).path or '/' - parts = pathlib.PurePosixPath(unquote(path)).parts - if '..' in parts: - return None - candidate = ROOT / host / pathlib.PurePosixPath(*[p for p in parts if p != '/']) - if candidate.is_dir(): - candidate /= 'index.html' - return candidate if candidate.is_file() else None - - -def respond_fixture(conn, found): - body = found.read_bytes() - headers = ( - f'Content-Length: {len(body)}\r\nConnection: close\r\n\r\n'.encode() - ) - conn.sendall(b'HTTP/1.1 200 OK\r\n' + headers + body) - - -def close_request(request, target=None): - """Rewrite a proxied request for a direct upstream connection.""" - headers, separator, body = request.partition(b'\r\n\r\n') - lines = headers.split(b'\r\n') - if target is not None: - method, _, version = lines[0].split(b' ', 2) - lines[0] = b' '.join((method, target.encode(), version)) - lines = [ - line for line in lines - if not line.lower().startswith(b'proxy-connection:') - ] - lines.append(b'Connection: close') - return b'\r\n'.join(lines) + separator + body - - -def relay(source, destination): - while True: - chunk = source.recv(MAX_REQUEST_BYTES) - if not chunk: - return - destination.sendall(chunk) - - -def forward_https(conn, host, port, request): - context = ssl.create_default_context(cafile=str(REAL_CA)) - with socket.create_connection((host, port), timeout=UPSTREAM_TIMEOUT_SECONDS) as raw: - with context.wrap_socket(raw, server_hostname=host) as upstream: - upstream.sendall(close_request(request)) - relay(upstream, conn) - - -def forward_http(conn, host, port, request, target): - parsed = urlsplit(target) - path = parsed.path or '/' - if parsed.query: - path += f'?{parsed.query}' - with socket.create_connection((host, port), timeout=UPSTREAM_TIMEOUT_SECONDS) as upstream: - upstream.sendall(close_request(request, path)) - relay(upstream, conn) - - -def handle_connect(conn, target): - """Intercept a CONNECT tunnel, terminating TLS with a minted cert.""" - host, _, port_text = target.rpartition(':') - port = int(port_text or '443') - conn.sendall(b'HTTP/1.1 200 Connection Established\r\n\r\n') - cert, key = cert_for(host) - context = ssl.SSLContext(ssl.PROTOCOL_TLS_SERVER) - context.load_cert_chain(cert, key) - with context.wrap_socket(conn, server_side=True) as tls: - nested = read_request(tls) - if not nested: - return - line = nested.split(b'\r\n', 1)[0].decode('iso-8859-1') - nested_target = line.split(' ', 2)[1] - found = file_for(host, nested_target) - if found is not None: - respond_fixture(tls, found) - else: - forward_https(tls, host, port, nested) - - -def host_from_headers(request): - for header in request.split(b'\r\n')[1:]: - if header.lower().startswith(b'host:'): - value = header.split(b':', 1)[1].strip().decode() - return value.split(':', 1)[0] - return None - - -def handle_request(conn): - with conn: - request = read_request(conn) - if not request: - return - line = request.split(b'\r\n', 1)[0].decode('iso-8859-1') - method, target, _ = line.split(' ', 2) - if method.upper() == 'CONNECT': - handle_connect(conn, target) - return - parsed = urlsplit(target) - host = parsed.hostname or host_from_headers(request) or 'unknown' - found = file_for(host, target) - if found is not None: - respond_fixture(conn, found) - else: - forward_http(conn, host, parsed.port or 80, request, target) - - -def handle(conn): - try: - handle_request(conn) - except Exception as error: - print(f'proxy request failed: {error!r}', file=sys.stderr, flush=True) - - -def main(): - with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as server: - server.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1) - server.bind(LISTEN_ADDRESS) - server.listen() - while True: - conn, _ = server.accept() - threading.Thread(target=handle, args=(conn,), daemon=True).start() - - -if __name__ == '__main__': - main() diff --git a/scripts/sandbox/ssh-shim.sh b/scripts/sandbox/ssh-shim.sh deleted file mode 100644 index 1b90035eea4c..000000000000 --- a/scripts/sandbox/ssh-shim.sh +++ /dev/null @@ -1,13 +0,0 @@ -#!/usr/bin/env bash -# Stand-in for ssh inside the dev sandbox. -# -# install.sh and `hermes update` clone over ssh first (git@github.com:...), so -# the sandbox needs an `ssh` that answers. Rather than run a real sshd, this -# ignores the host, user, and command git asked for and speaks the -# upload-pack protocol directly against the sandbox's bare repo -- which is -# what makes the ssh-first code path exercisable with no keys, no known_hosts, -# and no network. -# -# GIT_UPLOAD_PACK is substituted by dev-sandbox.sh when it installs this shim, -# because the host's git-upload-pack is not necessarily on the sandbox PATH. -exec @GIT_UPLOAD_PACK@ /work/repos/hermes-agent.git diff --git a/scripts/sandbox/stage2-run.sh b/scripts/sandbox/stage2-run.sh deleted file mode 100755 index 42d2800ec7eb..000000000000 --- a/scripts/sandbox/stage2-run.sh +++ /dev/null @@ -1,251 +0,0 @@ -#!/usr/bin/env bash -# Stage 2 of the dev sandbox: build the mounts and run the payload. -# -# Not called directly. scripts/dev-sandbox.sh (stage 1) creates the user and -# network namespaces with `unshare` and re-execs into this script inside them, -# so by the time this runs we are already at the target uid with a private -# netns. bwrap therefore does NOT create a userns here -- it only adds the -# mount and pid namespaces. (`unshare --user` grants its creator full -# capabilities in the new userns regardless of which uid it maps, which is what -# lets bwrap mount as a non-root uid.) -# -# The whole interface with stage 1 is the DEV_SANDBOX_* environment, asserted -# below: there are no shared functions or variables between the two stages. -# Stage 1 locates this script alongside the other sandbox assets (see -# DEV_SANDBOX_ASSETS in dev-sandbox.sh), so the Nix wrapper's store copy and a -# plain repo checkout both work. - -set -euo pipefail - -: "${DEV_SANDBOX_ROOT:?missing DEV_SANDBOX_ROOT}" -: "${DEV_SANDBOX_BASH:?missing DEV_SANDBOX_BASH}" -: "${DEV_SANDBOX_INTERACTIVE:?missing DEV_SANDBOX_INTERACTIVE}" -: "${DEV_SANDBOX_USER:?missing DEV_SANDBOX_USER}" -: "${DEV_SANDBOX_HOME:?missing DEV_SANDBOX_HOME}" - -# Announce our pid so stage 1 can point slirp4netns at these namespaces, -# then hold until it reports the network is up. -slirp_ready="$DEV_SANDBOX_ROOT/root/logs/slirp.ready" -printf '%s\n' "$$" > "$DEV_SANDBOX_ROOT/root/logs/sandbox.pid" -for _ in $(seq 1 200); do - [ -s "$slirp_ready" ] && break - sleep 0.05 -done -if [ ! -s "$slirp_ready" ]; then - echo 'error: timed out waiting for sandbox network setup' >&2 - cat "$DEV_SANDBOX_ROOT/root/logs/slirp.log" >&2 || true - exit 1 -fi - -# The sandbox HOME is /root for a root install and /home/ for a -# user-level one. Only the latter needs its parent created first; --dir / -# is not a thing bwrap accepts. -home_mounts=() -home_parent="$(dirname "$DEV_SANDBOX_HOME")" -if [ "$home_parent" != / ]; then - home_mounts+=(--dir "$home_parent") -fi -home_mounts+=(--bind "$DEV_SANDBOX_ROOT/home" "$DEV_SANDBOX_HOME") - -node_env=() -if [ -n "${DEV_SANDBOX_NODE_DIR:-}" ]; then - node_env+=(--setenv npm_config_nodedir "$DEV_SANDBOX_NODE_DIR") -fi -electron_env=() -if [ -n "${DEV_SANDBOX_ELECTRON_LD_LIBRARY_PATH:-}" ]; then - electron_env+=( - --setenv LD_LIBRARY_PATH "$DEV_SANDBOX_ELECTRON_LD_LIBRARY_PATH" - --setenv HERMES_DESKTOP_DISABLE_GPU 1 - ) -fi -gui_mounts=() -if [ -n "${DEV_SANDBOX_WAYLAND_SOCKET:-}" ]; then - runtime_dir="${DEV_SANDBOX_XDG_RUNTIME_DIR:?missing DEV_SANDBOX_XDG_RUNTIME_DIR}" - runtime_parent="$(dirname "$runtime_dir")" - runtime_grandparent="$(dirname "$runtime_parent")" - gui_mounts+=( - --dir "$runtime_grandparent" - --dir "$runtime_parent" - --dir "$runtime_dir" - --bind "$DEV_SANDBOX_WAYLAND_SOCKET" "$DEV_SANDBOX_WAYLAND_SOCKET" - --setenv XDG_RUNTIME_DIR "$runtime_dir" - --setenv WAYLAND_DISPLAY "${DEV_SANDBOX_WAYLAND_DISPLAY:?missing DEV_SANDBOX_WAYLAND_DISPLAY}" - ) -fi - -# How the sandbox gets a usable runtime, and where its own shims go. -# -# On Nix, every binary lives under /nix/store, so the sandbox can own /bin, -# /lib64 and /usr/bin outright and fill them with symlinks into the store. -# -# Elsewhere the runtime IS /usr, /bin, /lib, /lib64 -- so binding the -# sandbox's near-empty versions over them hides the real thing, and bwrap -# dies with `execvp /usr/bin/bash: No such file or directory`. Keep the host -# directories read-only and override only the individual files we shim. -# -# The same answer decides how /etc is handled further down. -if [ -d /nix ] && [[ "$(readlink -f "$DEV_SANDBOX_BASH")" == /nix/* ]]; then - USE_HOST_RUNTIME=false -else - USE_HOST_RUNTIME=true -fi - -runtime_mounts=() -shim_mounts=() -if [ "$USE_HOST_RUNTIME" = false ]; then - runtime_mounts+=(--ro-bind /nix /nix) - shim_mounts+=( - --dir /usr - --dir /bin - --dir /lib64 - --bind "$DEV_SANDBOX_ROOT/root/bin" /bin - --bind "$DEV_SANDBOX_ROOT/root/lib64" /lib64 - --bind "$DEV_SANDBOX_ROOT/root/usr/bin" /usr/bin - ) -else - for path in /usr /bin /sbin /lib /lib64; do - [ -e "$path" ] && runtime_mounts+=(--ro-bind "$path" "$path") - done - # The git-upload-pack shim standing in for github.com is the only file that - # must beat the host's copy; sh/ls/env are already there for real. - shim_mounts+=(--bind "$DEV_SANDBOX_ROOT/root/usr/bin/ssh" /usr/bin/ssh) -fi - -# /etc: start from a copy of the host's and overwrite only the files we fake. -# -# Replacing the whole directory with a five-file one is the tempting shortcut -# and it is wrong: a distro puts things under /etc that binaries outside /etc -# depend on, so hiding all of it breaks tools that look fine on PATH. Two real -# examples, both Debian/Ubuntu: openssl's compiled-in openssl.cnf is a symlink -# into /etc/ssl, and /usr/bin/awk is a symlink to /etc/alternatives/awk -- with -# /etc replaced, openssl cannot mint a certificate and awk reports "not found". -# Those are two symptoms of one cause, and nothing says there are only two. -# -# Copying rather than mount-overlaying the individual files, because several of -# these are symlinks in the wild (resolv.conf -> ../run/systemd/... on Ubuntu, -# hosts and nsswitch.conf -> /etc/static/... on NixOS) and bwrap cannot bind a -# file onto a symlink whose target does not exist inside the sandbox. -# -# Symlinks are copied as symlinks, never dereferenced: on NixOS /etc/static -# points into the store and following it would copy gigabytes per sandbox. The -# store is already mounted at /nix on that path, and the host runtime dirs are -# mounted at their own paths, so absolute symlinks still resolve. -# -# The five we override, and why each must differ from the host's: -# passwd, group the sandbox identity, which does not exist on the host -# resolv.conf slirp4netns's DNS, not the host resolver -# nsswitch.conf files+dns only, so nothing consults host NSS modules -# hosts minimal, so no host entry leaks in -# -# os-release is removed rather than replaced. Installers branch on it to reach -# for a package manager -- `install.sh` reads ID from it and, on debian/ubuntu, -# offers to apt-get build tools, prompting on /dev/tty when sudo exists but is -# not passwordless. That prompt cannot be satisfied here (no terminal) and it is -# fatal under `set -e`. Inheriting the host's file would make the sandbox claim -# to be a distro whose package manager it cannot actually use; absent means -# DISTRO="unknown" and the apt path is skipped, which is the truth. -etc_mounts=() -if [ "$USE_HOST_RUNTIME" = true ] && [ -d /etc ]; then - sandbox_etc="$DEV_SANDBOX_ROOT/etc-merged" - rm -rf -- "$sandbox_etc" - mkdir -p "$sandbox_etc" - # -a keeps symlinks as symlinks; unreadable entries (shadow, sudoers) are - # skipped rather than failing the run. - cp -a /etc/. "$sandbox_etc/" 2>/dev/null || true - for etc_file in passwd group resolv.conf nsswitch.conf hosts; do - [ -f "$DEV_SANDBOX_ROOT/etc/$etc_file" ] || continue - rm -f "$sandbox_etc/$etc_file" - cp "$DEV_SANDBOX_ROOT/etc/$etc_file" "$sandbox_etc/$etc_file" - done - rm -f "$sandbox_etc/os-release" "$sandbox_etc/lsb-release" - etc_mounts+=(--ro-bind "$sandbox_etc" /etc) -else - etc_mounts+=(--bind "$DEV_SANDBOX_ROOT/etc" /etc) -fi - -# /dev without a tty, so a script guarding on `[ -e /dev/tty ]` takes its -# no-terminal path. -# -# bwrap's --dev creates a /dev/tty NODE, but nothing in here has a controlling -# terminal, so opening it fails with "No such device or address". That is the -# worst of both: the guard passes and the read then fails. Under `set -e` -- -# which install.sh uses -- a failed read inside a function aborts the whole -# installer, which is exactly how older releases died here while prompting for -# sudo to install ripgrep/ffmpeg. -# -# Making the tty real is not the fix: with an openable terminal that prompt -# blocks forever waiting for input nobody will type. Absent is what a headless -# machine looks like, and what every prompt in here should assume. -# -# --dev cannot be used with the node removed afterwards (bwrap refuses to mount -# a directory over a device node), so /dev is assembled explicitly. -dev_mounts=( - --tmpfs /dev - --dev-bind /dev/null /dev/null - --dev-bind /dev/zero /dev/zero - --dev-bind /dev/full /dev/full - --dev-bind /dev/random /dev/random - --dev-bind /dev/urandom /dev/urandom - --symlink /proc/self/fd /dev/fd - --symlink /proc/self/fd/0 /dev/stdin - --symlink /proc/self/fd/1 /dev/stdout - --symlink /proc/self/fd/2 /dev/stderr -) -if [ "$DEV_SANDBOX_INTERACTIVE" = true ]; then - # An interactive shell is deliberately given a terminal; keep bwrap's /dev. - dev_mounts=(--dev /dev) -fi - -exec bwrap \ - --unshare-pid \ - --die-with-parent --proc /proc --tmpfs /tmp \ - "${dev_mounts[@]}" \ - "${gui_mounts[@]}" \ - "${runtime_mounts[@]}" \ - --bind "$DEV_SANDBOX_ROOT/root" /work \ - "${shim_mounts[@]}" \ - --bind "$DEV_SANDBOX_ROOT/root/usr/local" /usr/local \ - "${home_mounts[@]}" \ - "${etc_mounts[@]}" \ - --chdir /work/repo \ - --clearenv \ - --setenv PATH "$DEV_SANDBOX_HOME/.local/bin:/usr/local/bin:/usr/bin:$PATH" \ - --setenv HOME "$DEV_SANDBOX_HOME" \ - --setenv USER "$DEV_SANDBOX_USER" \ - --setenv LOGNAME "$DEV_SANDBOX_USER" \ - --setenv CURL_CA_BUNDLE /work/certs/ca.pem \ - --setenv SSL_CERT_FILE /work/certs/ca.pem \ - --setenv GIT_SSL_CAINFO /work/certs/ca.pem \ - --setenv NODE_EXTRA_CA_CERTS /work/certs/real-ca.pem \ - --setenv OPENSSL_CONF /work/certs/openssl.cnf \ - --setenv HTTP_PROXY http://127.0.0.1:8080 \ - --setenv HTTPS_PROXY http://127.0.0.1:8080 \ - --setenv ALL_PROXY http://127.0.0.1:8080 \ - --setenv NO_PROXY '' \ - --setenv DEV_SANDBOX_INTERACTIVE "$DEV_SANDBOX_INTERACTIVE" \ - --setenv ELECTRON_DISABLE_SANDBOX 1 \ - "${node_env[@]}" \ - "${electron_env[@]}" \ - -- "$DEV_SANDBOX_BASH" -ceu ' - python3 /work/proxy.py /work/http /work/certs /work/certs/real-ca.pem >/work/logs/proxy.log 2>&1 & - proxy_pid=$! - cleanup() { - kill "$proxy_pid" 2>/dev/null || true - wait "$proxy_pid" 2>/dev/null || true - } - trap cleanup EXIT INT TERM - # Bash opens /dev/tcp itself, so the readiness probe needs no netcat -- - # one less binary the sandbox has to find on the host (GitHub runners - # ship no `nc`). - proxy_up() { (exec 3<>/dev/tcp/127.0.0.1/8080) 2>/dev/null; } - for _ in $(seq 1 100); do - proxy_up && break - sleep 0.05 - done - if ! proxy_up; then - echo "error: the sandbox fake-internet proxy never came up" >&2 - cat /work/logs/proxy.log >&2 || true - exit 1 - fi - "$@" - ' sandbox-command "$@" diff --git a/tests/install/install-update-e2e.sh b/tests/install/install-update-e2e.sh deleted file mode 100755 index 80f257907ba7..000000000000 --- a/tests/install/install-update-e2e.sh +++ /dev/null @@ -1,293 +0,0 @@ -#!/usr/bin/env bash -# Prove a user on some earlier commit can reach this one. -# -# Installs a real, earlier Hermes the way a user does, applies ONE update route, -# and requires the checkout to land on this commit with a working `hermes`. -# -# Nothing here is mocked. scripts/dev-sandbox.sh provides the fake Internet -- -# a bubblewrap sandbox with no writable host mounts, a MITM proxy serving the -# canonical install.sh URL, and a git-upload-pack shim standing in for -# github.com -- so `install.sh` really installs uv, a managed Python, Node and -# the venv, cloning "github.com" over the ssh-first path a user hits. -# -# One route per run, on a sandbox built from scratch, because the routes are only -# meaningful from a pristine install. Sharing one install across routes -- or -# rewinding the checkout with `git reset --hard` between them -- leaves the -# second route running against a tree the first already updated (same venv, same -# installed console script, same __pycache__), which is not the state any real -# user is in: a route can then pass only because its predecessor did the work, -# and a failure in the first leaves the second exercising something undefined. -# If you add a route, give it its own run. -# -# Usage: -# tests/install/install-update-e2e.sh --route update|installer -# [--install-ref REF] [--keep] -# -# --route which update path to exercise (required): -# update `hermes update` -# installer re-running the curl one-liner over the checkout -# --install-ref what to install first; anything git resolves (a branch, a -# tag like v2026.7.7, or a SHA reachable from main). -# Default: refs/heads/main. -# -# Requires a CLEAN worktree: every dev-sandbox invocation re-derives fake main -# from the working copy, so uncommitted changes move the update target between -# the call that installs and the call that verifies. - -set -euo pipefail - -ROUTE="" -INSTALL_REF="refs/heads/main" -KEEP=false -while [ "$#" -gt 0 ]; do - case "$1" in - --route) - [ "$#" -ge 2 ] || { echo 'error: --route needs a value' >&2; exit 1; } - ROUTE="$2"; shift 2 ;; - --install-ref) - [ "$#" -ge 2 ] || { echo 'error: --install-ref needs a value' >&2; exit 1; } - INSTALL_REF="$2"; shift 2 ;; - --keep) KEEP=true; shift ;; - -h|--help) sed -n '2,35p' "$0"; exit 0 ;; - *) echo "error: unknown argument: $1" >&2; exit 1 ;; - esac -done -case "$ROUTE" in - update|installer) ;; - '') echo 'error: --route is required (update or installer)' >&2; exit 1 ;; - *) echo "error: unknown route: $ROUTE (want update or installer)" >&2; exit 1 ;; -esac - -REPO_ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/../.." && pwd)" -cd "$REPO_ROOT" - -# Keep sandbox state out of the default .hermes-sandbox so a run never clobbers -# a developer's own sandbox, and scope it per route so two routes can run -# concurrently (CI runs them as parallel matrix legs). dev-sandbox.sh joins this -# onto the worktree root and feeds it to `tar --exclude`, so it MUST be a -# relative directory name. -SANDBOX_DIR_NAME=".hermes-sandbox-e2e-$ROUTE" -export HERMES_DEV_SANDBOX_DIR="$SANDBOX_DIR_NAME" - -SANDBOX_ROOT="$REPO_ROOT/$SANDBOX_DIR_NAME" -INSTALL_DIR="/home/hermes/.hermes/hermes-agent" # user-level layout (sandbox default) -FAKE_REMOTE="/work/repos/hermes-agent.git" -# Only used to fetch an old install.sh for the flag probe below; the sandbox does -# its own fetching. Same override dev-sandbox.sh honours, so a fork can retarget -# both together. -UPSTREAM_URL="${HERMES_DEV_SANDBOX_UPSTREAM:-https://github.com/NousResearch/hermes-agent.git}" - -# Installer transcripts live outside the sandbox root: the sandbox is recreated -# and (unless --keep) deleted, and these logs are the most useful artifact when -# a real install breaks. Created after the dirty check below, so that a log dir -# pointed inside the repo cannot be the thing that makes the tree dirty. -LOG_DIR="${HERMES_E2E_LOG_DIR:-$(mktemp -d -t hermes-install-e2e-logs.XXXXXX)}" - -step() { printf '\n\033[1;36m▶ %s\033[0m\n' "$*"; } -ok() { printf '\033[1;32m ✓ %s\033[0m\n' "$*"; } -fail() { printf '\n\033[1;31m✗ %s\033[0m\n' "$*" >&2; exit 1; } - -# The sandbox's internal logs (fake-internet proxy, slirp) explain failures that -# happen BEFORE install.sh gets to say anything -- a TLS handshake the proxy -# rejected looks like a bare `curl: (35)` from outside. Copy them out where a CI -# artifact upload can find them, and echo the proxy log since it is the usual -# culprit. -collect_sandbox_logs() { - # Separate `local` statements on purpose: a single `local a=$1 b="$a"` does - # NOT see the earlier assignment, so under `set -u` the second expansion dies - # with "a: unbound variable". - local tag="$1" - local src="$SANDBOX_ROOT/root/logs" - local dest="$LOG_DIR/sandbox-$tag" - [ -d "$src" ] || return 0 - mkdir -p "$dest" - cp -a "$src/." "$dest/" 2>/dev/null || true - # Print it, not just archive it: a rejected TLS handshake here is the whole - # explanation for a failure that otherwise reads as a bare `curl: (35)`, and - # whoever is reading the job log should not have to download an artifact to - # see it. In full, not tailed -- the file is short, and the useful line is not - # reliably at the end. - if [ -s "$dest/proxy.log" ]; then - echo "--- sandbox proxy.log ---" >&2 - cat "$dest/proxy.log" >&2 - echo "--- end proxy.log ---" >&2 - fi -} - -# ── preflight ────────────────────────────────────────────────────────────── -# Prefer the `sandbox` wrapper from the Nix devShell: it supplies both the PATH -# (bwrap, slirp4netns, openssl, ...) and the DEV_SANDBOX_* variables the script -# needs -- notably DEV_SANDBOX_DYNAMIC_LINKER, without which it cannot find a -# glibc loader on NixOS. Off Nix, the script is the entry point and finds its -# dependencies on the system PATH. -if command -v sandbox >/dev/null 2>&1; then - SANDBOX=(sandbox) -elif command -v bwrap >/dev/null 2>&1; then - SANDBOX=("$REPO_ROOT/scripts/dev-sandbox.sh") -else - fail 'no usable sandbox: enter the Nix devShell (for `sandbox`) or install bubblewrap' -fi - -if [ -n "$(git status --porcelain)" ]; then - printf '\033[1;31m✗ working tree is dirty:\033[0m\n' >&2 - git status --porcelain | sed 's/^/ /' >&2 - fail 'Every sandbox invocation re-snapshots the working copy into a new - fake-main commit, so the update target would move mid-run. Commit or stash - first. (If a path above is build or log output, it needs gitignoring or to - live outside the repo.)' -fi - -mkdir -p "$LOG_DIR" - -if [ "$KEEP" = false ]; then - trap 'rm -rf -- "$SANDBOX_ROOT"' EXIT INT TERM -fi -rm -rf -- "$SANDBOX_ROOT" - -# ── helpers ──────────────────────────────────────────────────────────────── -# Does the INSTALLED hermes accept FLAG on `hermes update`? -# -# Asked of the installed binary rather than parsed out of a release's source: -# the update subcommand has lived in main.py, subcommands/update.py, and -# update_cmd.py across the releases we sample, so any static parse is a guess -# that silently rots. `hermes update --help` is the same surface a user meets, -# and argparse prints every option it accepts. -update_supports() { - local flag="$1" - in_sandbox "hermes update --help 2>&1" | grep -qF -- "$flag" -} - -# Does the installer at REF accept FLAG? Read it out of that ref's own -# install.sh rather than assuming this checkout's flag set: the point of the -# matrix is to install releases from months back, whose installers predate -# options we take for granted. (Unlike the updater, the installer runs before -# anything is installed, so there is no --help to ask yet.) -# -# The ref may not be local -- the sandbox does its own fetching -- so fall back -# to fetching just that blob. Unresolvable means "flag absent", which costs a -# more conservative invocation, never a wrong one. -installer_supports() { - local ref="$1" - local flag="$2" - local script="" - script="$(git show "$ref:scripts/install.sh" 2>/dev/null)" || { - git fetch -q --depth 1 "$UPSTREAM_URL" "$ref" 2>/dev/null || return 1 - script="$(git show FETCH_HEAD:scripts/install.sh 2>/dev/null)" || return 1 - } - printf '%s' "$script" | grep -qF -- "$flag" -} - -# Run the real install one-liner inside the sandbox. `ref` non-empty installs -# that upstream commit and promotes THIS checkout to fake main afterwards, -# leaving the state a user is in when an update is waiting; empty serves this -# worktree's own installer and points fake main here. -install_in_sandbox() { - local what="$1" - local ref="$2" - local tag="$3" - local log="$LOG_DIR/$tag.log" - local args=(install --persistent) - [ -n "$ref" ] && args+=(--install-ref "$ref") - - # Installer flags have to match the installer being run, not this checkout's. - # Older releases reject options added later ("Unknown option: --skip-browser"), - # and this test deliberately installs releases from months back. --skip-setup - # goes back further than any tag we sample; anything newer is probed for. - local installer_flags=(--skip-setup) - if [ -z "$ref" ] || installer_supports "$ref" --skip-browser; then - installer_flags+=(--skip-browser) - fi - # Sandbox flags must precede `--`; the rest goes to install.sh. - args+=(-- "${installer_flags[@]}") - - # Stream the installer's output to stdout AND keep a copy on disk. It is the - # substance of this test -- a real install of uv, a managed Python, Node and - # the venv -- so it belongs in the job log where anyone reading the run can - # see it, not only in an artifact they have to download. The file copy is what - # the artifact upload keeps and what the failure paths grep. - # - # `set -o pipefail` is load-bearing here: without it the pipeline reports - # tee's status and a failed install looks like a pass. - local status=0 - "${SANDBOX[@]}" "${args[@]}" 2>&1 | tee "$log" || status=$? - - if [ "$status" -ne 0 ]; then - collect_sandbox_logs "$tag" - fail "$what failed (exit $status)" - fi - grep -q 'Installation Complete' "$log" \ - || { collect_sandbox_logs "$tag"; \ - fail "$what did not report a completed install"; } - ok "$what completed (log: $log)" -} - -in_sandbox() { "${SANDBOX[@]}" --persistent bash -lc "$1"; } - -# fake main's SHA is read fresh whenever it is needed, never cached across a -# sandbox invocation: each invocation re-derives it from the worktree. -sandbox_target() { in_sandbox "git --git-dir=$FAKE_REMOTE rev-parse main" | tr -d '[:space:]'; } -sandbox_head() { in_sandbox "cd $INSTALL_DIR && git rev-parse HEAD" | tr -d '[:space:]'; } - -require_landed_on_target() { - local what="$1" head target - head="$(sandbox_head)" - target="$(sandbox_target)" - [ "$head" = "$target" ] || fail "$what left HEAD at $head, wanted $target" - ok "$what landed on ${head:0:12}" -} - -# The real smoke test: goes through the venv launcher and imports the app, so it -# fails if the venv, dependencies, or entry point are broken. -require_hermes_works() { - local when="$1" out - out="$(in_sandbox "hermes --version" 2>&1)" \ - || { printf '%s\n' "$out" >&2; fail "hermes --version failed $when"; } - printf '%s\n' "$out" | sed 's/^/ /' - ok "hermes runs $when" -} - -# ── install the earlier Hermes ───────────────────────────────────────────── -step "installing upstream $INSTALL_REF (real curl | install.sh: uv, Python, Node, venv)" -install_in_sandbox "install of upstream $INSTALL_REF" "$INSTALL_REF" install - -BASE="$(sandbox_head)" -TARGET="$(sandbox_target)" -[ -n "$BASE" ] || fail "could not read the installed commit" -[ "$BASE" != "$TARGET" ] \ - || fail "install landed on the update target ($BASE); base and target must differ" -ok "installed ${BASE:0:12}; update target is ${TARGET:0:12}" -require_hermes_works 'after install' - -# ── apply exactly one update route ───────────────────────────────────────── -case "$ROUTE" in - update) - step 'ROUTE: hermes update' - # `--yes` reaches the update subcommand only in later releases, and argparse - # rejects the whole invocation when it does not exist. Ask the installed - # hermes which it accepts; older ones read the prompt from stdin, so close it. - if update_supports --yes; then - update_cmd="hermes update --yes" - else - update_cmd="hermes update Date: Tue, 11 Aug 2026 22:45:00 -0400 Subject: [PATCH 057/227] test(install): windows installer-script e2e - irm | iex arm goes live The install.ps1 sibling of installer-script-e2e.sh: stage serve.git (main parked at OLD, GIT_CONFIG_GLOBAL insteadOf redirect - NOT env config, which install.ps1 clobbers), run the install.ps1 shipped AT the OLD ref headless (-SkipSetup -HermesHome/-InstallDir explicit because the oldest tags predate the HERMES_HOME env override; -NonInteractive probed from the ref's own script text), assert the checkout + venv hermes.exe, advance served main, update via hermes-update (--yes probed) or HEAD's install.ps1, assert HEAD. install-e2e-windows-run.yml grows a second job for the arm: the installer-script x {hermes-update, installer-script} pairs flip from grey to live, app-update from a script install stays a declared TODO. PS 5.1-safe pure ASCII. Stage logic verified behaviorally under pwsh (redirect resolves the canonical URL to serve.git at OLD, per-ref install.ps1 extraction parses, advance lands HEAD); the install legs themselves need a real Windows runner - dispatched next. --- .github/workflows/install-e2e-windows-run.yml | 98 ++++++--- .../install/windows-installer-script-e2e.ps1 | 194 ++++++++++++++++++ 2 files changed, 265 insertions(+), 27 deletions(-) create mode 100644 tests/install/windows-installer-script-e2e.ps1 diff --git a/.github/workflows/install-e2e-windows-run.yml b/.github/workflows/install-e2e-windows-run.yml index 27b75e67df87..40bf49fb3494 100644 --- a/.github/workflows/install-e2e-windows-run.yml +++ b/.github/workflows/install-e2e-windows-run.yml @@ -1,34 +1,43 @@ name: Install & Update E2E — Windows desktop (reusable) -# Runs the REAL Windows desktop user flow against ONE update route: -# install OLD via the website's Hermes-Setup.exe (headed, AutoHotkey clicks -# Install -> Launch, the real Electron window must appear), then update -# OLD -> HEAD through the route a user would take. +# Runs ONE Windows {install-method, update-method} combination: install +# OLD, then update OLD -> HEAD through the route a user would take. # -# The Windows sibling of install-e2e-run.yml. No bubblewrap here — the -# driver (tests/install/windows-desktop-gui-e2e.ps1) fakes GitHub with -# git's own transport rewrite: a driver-owned GIT_CONFIG_GLOBAL carrying -# url..insteadOf for both canonical repo URLs, so the -# installer and updater run byte-for-byte against their real URLs and land -# on a local bare repo. Its `main` serves OLD (install-ref, default: the -# newest release tag) during the install, then advances to HEAD for the -# update leg — an update becomes available exactly the way it does for a -# real user. +# The Windows sibling of install-e2e-run.yml. No bubblewrap anywhere -- the +# drivers fake GitHub with git's own transport rewrite: a driver-owned +# GIT_CONFIG_GLOBAL carrying url..insteadOf for both +# canonical repo URLs, so the installer and updater run byte-for-byte +# against their real URLs and land on a local bare repo. Its `main` serves +# OLD (install-ref, default: the newest release tag) during the install, +# then advances to HEAD for the update leg -- an update becomes available +# exactly the way it does for a real user. # -# Routes (from the update-method input): -# app-update the app's own Update button: the installed -# Hermes.exe runs under Playwright's Electron -# driver, which clicks Settings -> About -> -# "Update now"; the production hand-off chain runs -# untouched (marker, app quit, detached updater, -# hermes update, desktop rebuild, relaunch). -# hermes-update TODO: `hermes update` from the installed venv. -# desktop-installer@latest -# TODO: re-run the bootstrap exe over the install. -# installer-script TODO: re-run the irm | iex one-liner. +# Two install arms, two drivers: +# desktop-installer@latest the REAL desktop user flow +# (tests/install/windows-desktop-gui-e2e.ps1): +# the website's published Hermes-Setup.exe runs +# headed, AutoHotkey clicks Install -> Launch, +# the real Electron window must appear. Update +# methods: +# app-update the app's own Update button: +# the installed Hermes.exe runs +# under Playwright's Electron +# driver, which clicks Settings +# -> About -> "Update now"; the +# production hand-off chain runs +# untouched. +# (hermes-update, desktop-installer@latest, +# installer-script re-run: declared TODOs.) +# installer-script the irm | iex one-liner +# (tests/install/windows-installer-script-e2e +# .ps1): runs the install.ps1 shipped AT the +# OLD ref, headless. Update methods: +# hermes-update venv hermes.exe update +# installer-script re-run HEAD's install.ps1 +# (app-update from a script install: TODO.) # # Method pairs without a driver yet NATIVELY SKIP (grey check, no runner): -# the capability knowledge lives here, next to the driver, so the caller +# the capability knowledge lives here, next to the drivers, so the caller # can dispatch every declared combination without knowing which ones work. # # Call it: @@ -45,11 +54,11 @@ on: workflow_call: inputs: install-method: - description: 'How OLD gets installed. Supported: desktop-installer@latest (website exe, AHK-clicked). Declared-but-TODO methods skip.' + description: 'How OLD gets installed. Supported: desktop-installer@latest (website exe, AHK-clicked) and installer-script (irm | iex install.ps1). Declared-but-TODO methods skip.' required: true type: string update-method: - description: 'How the install updates to HEAD. Supported: app-update (Update button under Playwright). Declared-but-TODO methods skip.' + description: 'How the install updates to HEAD. Supported: app-update (Update button under Playwright, from a desktop install) and hermes-update / installer-script (from a script install). Declared-but-TODO pairs skip.' required: true type: string install-ref: @@ -77,6 +86,7 @@ permissions: contents: read jobs: + # ---- arm 1: install via the website's Hermes-Setup.exe (GUI) ------------- e2e: # Static name on purpose: the caller's job name already carries the # method pair, and GitHub renders name expressions UNEXPANDED (literal @@ -171,3 +181,37 @@ jobs: path: gui-e2e-proof retention-days: 14 if-no-files-found: ignore + + # ---- arm 2: install via the irm | iex one-liner (headless) --------------- + script-e2e: + # Same static-name reasoning as e2e above. + name: e2e + # The pairs the script driver can run today. app-update from a script + # install is a declared TODO (the desktop app also has to be BUILT by + # that path first). + if: inputs.install-method == 'installer-script' && contains(fromJSON('["hermes-update", "installer-script"]'), inputs.update-method) + runs-on: windows-latest + timeout-minutes: ${{ inputs.timeout-minutes }} + + steps: + # Full history: the driver bare-clones this checkout as the repo the + # installer/updater talk to, and both OLD and HEAD must be reachable + # in that clone. + - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2 + with: + fetch-depth: 0 + + - name: Run install + update E2E + shell: powershell + run: powershell -NoProfile -ExecutionPolicy Bypass -File tests\install\windows-installer-script-e2e.ps1 -UpdateMethod "${{ inputs.update-method }}" -InstallRef "${{ inputs.install-ref }}" + env: + HERMES_E2E_LOG_DIR: ${{ runner.temp }}\e2e-logs + + - name: Upload installer logs + if: always() + uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1 + with: + name: install-e2e-windows-script-${{ inputs.update-method }}-${{ inputs.install-ref }}-${{ github.sha }} + path: ${{ runner.temp }}\e2e-logs + retention-days: 14 + if-no-files-found: ignore diff --git a/tests/install/windows-installer-script-e2e.ps1 b/tests/install/windows-installer-script-e2e.ps1 new file mode 100644 index 000000000000..f011a5f16e98 --- /dev/null +++ b/tests/install/windows-installer-script-e2e.ps1 @@ -0,0 +1,194 @@ +# Prove a Windows user who installed OLD via the installer script (irm | +# iex) can reach HEAD. +# +# The install.ps1 sibling of tests/install/installer-script-e2e.sh, sharing +# its staging trick with the GUI driver (windows-desktop-gui-e2e.ps1): +# every git process is pointed at a local bare clone with +# url..insteadOf rewrites for both canonical repo URLs in +# a driver-owned GIT_CONFIG_GLOBAL. The installer and updater run +# byte-for-byte against their real URLs and land on serve.git; `main` +# serves OLD during the install, then advances to HEAD for the update leg. +# (NOT GIT_CONFIG_COUNT/KEY_n/VALUE_n env config -- install.ps1 SETS those +# itself and would clobber ours.) +# +# install.ps1 itself is not downloaded: the install leg runs the copy +# shipped AT the OLD ref (what a user who installed then actually +# executed), and the installer-script update leg runs HEAD's copy (what +# the website serves at update time). +# +# Usage: +# powershell -File tests\install\windows-installer-script-e2e.ps1 ` +# -UpdateMethod hermes-update -InstallRef v0.20.2 +# +# -UpdateMethod hermes-update venv\Scripts\hermes.exe update +# installer-script re-run install.ps1 (HEAD's copy) +# -InstallRef what to install first; anything git resolves. auto = +# the newest release tag in the checkout. +# +# Requires a clean full-history checkout with release tags fetched. +# PowerShell 5.1-safe, pure ASCII (OEM codepages explode on fancy dashes). + +#Requires -Version 5.1 + +param( + [ValidateSet("hermes-update", "installer-script")] + [string]$UpdateMethod = "hermes-update", + [string]$InstallRef = "auto" +) + +$ErrorActionPreference = "Stop" + +$RepoRoot = (Resolve-Path (Join-Path $PSScriptRoot "..\..")).Path +$RepoUrlSsh = "git@github.com:NousResearch/hermes-agent.git" +$RepoUrlHttps = "https://github.com/NousResearch/hermes-agent.git" + +# Everything lives OUTSIDE the checkout; an untracked dir inside the repo +# would trip the dirty-tree guard below on the next run. +$WorkRoot = Join-Path $(if ($env:RUNNER_TEMP) { $env:RUNNER_TEMP } else { $env:TEMP }) "hermes-installer-script-e2e" +$LogDir = if ($env:HERMES_E2E_LOG_DIR) { $env:HERMES_E2E_LOG_DIR } else { Join-Path $WorkRoot "logs" } +$ServeRepo = Join-Path $WorkRoot "serve.git" + +function Step([string]$Message) { Write-Host "`n=== $Message ===" } +function Ok([string]$Message) { Write-Host " OK $Message" } +function Fail([string]$Message) { + Write-Host "E2E ASSERTION FAILED: $Message" -ForegroundColor Red + exit 1 +} + +function Invoke-Git { + param([string[]]$GitArgs) + $out = & git @GitArgs 2>&1 + if ($LASTEXITCODE -ne 0) { + Fail "git $($GitArgs -join ' ') exited $LASTEXITCODE`: $out" + } + return $out +} + +if (Test-Path -LiteralPath $WorkRoot) { Remove-Item -Recurse -Force -LiteralPath $WorkRoot } +New-Item -ItemType Directory -Path $WorkRoot, $LogDir -Force | Out-Null + +# --- stage: serve.git with main parked at OLD -------------------------------- + +Step "staging serve.git (main -> OLD)" +# Tracked changes only (-uno): untracked files cannot leak into a bare clone. +$dirty = Invoke-Git @("-C", $RepoRoot, "status", "--porcelain", "-uno") +if ($dirty) { Fail "checkout has uncommitted tracked changes; the staged clone must be a reviewable commit" } + +if ($InstallRef -eq "auto") { + $tags = (Invoke-Git @("-C", $RepoRoot, "tag", "--list", "v[0-9]*", "--sort=-creatordate")) -split "`r?`n" + $InstallRef = $tags | Select-Object -First 1 + if (-not $InstallRef) { Fail "no release tags in the checkout to use as OLD" } +} +$OldSha = (Invoke-Git @("-C", $RepoRoot, "rev-parse", "$InstallRef^{commit}")).Trim() +$HeadSha = (Invoke-Git @("-C", $RepoRoot, "rev-parse", "HEAD")).Trim() +if ($OldSha -eq $HeadSha) { Fail "OLD ($InstallRef) IS HEAD; no update would be available" } + +Invoke-Git @("clone", "--bare", "--quiet", $RepoRoot, $ServeRepo) | Out-Null +Invoke-Git @("-C", $ServeRepo, "update-ref", "refs/heads/main", $OldSha) | Out-Null +Invoke-Git @("-C", $ServeRepo, "symbolic-ref", "HEAD", "refs/heads/main") | Out-Null +# The installer may pin a commit that is reachable but not at a ref tip. +Invoke-Git @("-C", $ServeRepo, "config", "uploadpack.allowAnySHA1InWant", "true") | Out-Null +Ok "serve.git main = $OldSha ($InstallRef), update target $HeadSha" + +# --- the git URL redirect ------------------------------------------------------ + +$gitCfg = Join-Path $WorkRoot "gitconfig" +$serveUrl = "file:///" + ($ServeRepo -replace "\\", "/") +@" +[url "$serveUrl"] + insteadOf = $RepoUrlHttps + insteadOf = $RepoUrlSsh +"@ | Set-Content -LiteralPath $gitCfg -Encoding Ascii +$env:GIT_CONFIG_GLOBAL = $gitCfg +Ok "git URL redirect via GIT_CONFIG_GLOBAL=$gitCfg" + +# Isolated install target: the runner may carry a preinstalled hermes. +# -HermesHome/-InstallDir are passed explicitly because the oldest sampled +# installers predate the HERMES_HOME env override. +$HermesHome = Join-Path $WorkRoot "hermes-home" +$InstallDir = Join-Path $HermesHome "hermes-agent" +New-Item -ItemType Directory -Path $HermesHome -Force | Out-Null +$env:HERMES_HOME = $HermesHome +# serve.git's file:// origin looks like a fork to the updater, whose "add +# the official repo as upstream?" prompt would hang a headless run. This +# marker is the product's own mechanism for suppressing it. +Set-Content -LiteralPath (Join-Path $HermesHome ".skip_upstream_prompt") -Value "" -Encoding Ascii + +function Invoke-Installer { + param([string]$Ref, [string]$Label) + $script = Join-Path $WorkRoot "install-$Label.ps1" + (Invoke-Git @("-C", $RepoRoot, "show", "$Ref`:scripts/install.ps1")) -join "`n" | + Set-Content -LiteralPath $script -Encoding UTF8 + # Flags must match the installer being run, not this checkout's: older + # releases reject parameters added later. -SkipSetup/-HermesHome/ + # -InstallDir go back further than any tag we sample; -NonInteractive + # is probed from the ref's own script text. + $flags = @("-SkipSetup", "-HermesHome", $HermesHome, "-InstallDir", $InstallDir) + $text = Get-Content -LiteralPath $script -Raw + if ($text -match '\$NonInteractive') { $flags += "-NonInteractive" } + $log = Join-Path $LogDir "install-$Label.log" + & powershell -NoProfile -ExecutionPolicy Bypass -File $script @flags *> $log + if ($LASTEXITCODE -ne 0) { + Get-Content -LiteralPath $log -Tail 50 | Write-Host + Fail "install.ps1 ($Label) exited $LASTEXITCODE; full log in $log" + } +} + +function Assert-Checkout { + param([string]$ExpectedSha, [string]$Label) + $got = (Invoke-Git @("-C", $InstallDir, "rev-parse", "HEAD")).Trim() + if ($got -ne $ExpectedSha) { Fail "installed checkout is $got, expected $Label ($ExpectedSha)" } + Ok "checkout is $Label ($ExpectedSha)" + $hermes = Join-Path $InstallDir "venv\Scripts\hermes.exe" + if (-not (Test-Path -LiteralPath $hermes)) { Fail "no hermes console script at $hermes" } + $verLog = Join-Path $LogDir "version-$Label.log" + & $hermes --version *> $verLog + if ($LASTEXITCODE -ne 0) { + Get-Content -LiteralPath $verLog | Write-Host + Fail "hermes --version failed after $Label; log in $verLog" + } + Ok "hermes --version works: $((Get-Content -LiteralPath $verLog -First 1))" +} + +# --- install OLD ------------------------------------------------------------------ + +Step "installing OLD ($InstallRef) via its own scripts/install.ps1" +Invoke-Installer $OldSha "old" +Assert-Checkout $OldSha "OLD" + +# --- update OLD -> HEAD -------------------------------------------------------------- + +Step "advancing served main to HEAD" +Invoke-Git @("-C", $ServeRepo, "update-ref", "refs/heads/main", $HeadSha) | Out-Null +Ok "serve.git main = $HeadSha" + +Step "updating via $UpdateMethod" +switch ($UpdateMethod) { + "hermes-update" { + $hermes = Join-Path $InstallDir "venv\Scripts\hermes.exe" + # `--yes` reaches the update subcommand only in later releases, and + # argparse rejects the whole invocation when it does not exist. + $updateArgs = @("update") + $helpText = & $hermes update --help 2>&1 | Out-String + if ($helpText -match '--yes') { $updateArgs += "--yes" } + $log = Join-Path $LogDir "update.log" + Push-Location $InstallDir + try { + & $hermes @updateArgs *> $log + $updateExit = $LASTEXITCODE + } finally { + Pop-Location + } + if ($updateExit -ne 0) { + Get-Content -LiteralPath $log -Tail 50 | Write-Host + Fail "hermes update exited $updateExit; full log in $log" + } + } + "installer-script" { + # A user re-running the one-liner today gets the CURRENT script. + Invoke-Installer $HeadSha "head" + } +} +Assert-Checkout $HeadSha "HEAD" + +Step "PASS: $InstallRef -> HEAD via $UpdateMethod" From ae3a822d4e3f736c102972dad9a83958004c29f1 Mon Sep 17 00:00:00 2001 From: ethernet Date: Tue, 11 Aug 2026 22:51:31 -0400 Subject: [PATCH 058/227] ci(windows-e2e): driver installs its own pinned @playwright/test - never the app tree's Run 31553293275: the desktop leg from v2026.6.19 failed at '@playwright/test resolvable from installed apps/desktop' - that release predates the dependency, and newer releases move it around via workspace hoisting, so resolving it from the installed tree made the leg's tooling a function of the version under test. Install a pinned @playwright/test (param, default 1.58.2 - the repo lockfile's current version) into a scratch driver dir with the managed node/npm every run and launch the driver from there. The driver talks to the app over Playwright's inspection pipe, so its Playwright is independent of the app; every OLD ref now runs the exact same driver stack. --- tests/install/windows-desktop-gui-e2e.ps1 | 43 +++++++++++++++-------- 1 file changed, 29 insertions(+), 14 deletions(-) diff --git a/tests/install/windows-desktop-gui-e2e.ps1 b/tests/install/windows-desktop-gui-e2e.ps1 index ea0aff0d9e7e..03eba70aa77a 100644 --- a/tests/install/windows-desktop-gui-e2e.ps1 +++ b/tests/install/windows-desktop-gui-e2e.ps1 @@ -95,7 +95,13 @@ param( [string]$WorkRoot = $(if ($env:HERMES_E2E_WORKROOT) { $env:HERMES_E2E_WORKROOT } else { Join-Path $env:TEMP "hermes-desktop-gui-e2e" }), - [string]$SetupExeUrl = "https://hermes-assets.nousresearch.com/Hermes-Setup.exe" + [string]$SetupExeUrl = "https://hermes-assets.nousresearch.com/Hermes-Setup.exe", + + # Pinned @playwright/test for the update-gui driver. Installed fresh + # into a scratch dir every run -- never resolved from the installed + # tree -- so the driver behaves identically for every OLD ref. Bump + # deliberately; keep roughly in step with the repo's own lockfile. + [string]$PlaywrightVersion = "1.58.2" ) $ErrorActionPreference = "Stop" @@ -521,29 +527,38 @@ function Invoke-GuiUpdateDesktopRoute([string]$TargetSha) { Remove-Item -LiteralPath $resultPath -Force -ErrorAction SilentlyContinue $node = Get-ManagedNode - $appsDesktop = Join-Path $InstallDir "apps\desktop" - # @playwright/test is a workspace devDependency; the root `npm ci` - # HOISTS it to the repo-root node_modules, not apps/desktop's. Resolve - # it the way Node will (walk up from apps/desktop) instead of asserting - # a hardcoded path that hoisting makes wrong. - $prevEap = $ErrorActionPreference; $ErrorActionPreference = "Continue" - & $node -e "require.resolve('@playwright/test', { paths: [process.argv[1]] })" $appsDesktop 2>&1 | Out-Null - $pwResolved = ($LASTEXITCODE -eq 0) - $ErrorActionPreference = $prevEap - Assert-True $pwResolved "@playwright/test resolvable from installed apps/desktop" + # The Playwright driver gets its OWN pinned @playwright/test in a + # scratch dir -- NEVER the installed tree's copy. The driver talks to + # the app over Playwright's inspection pipe, so its Playwright version + # is independent of the app under test; installing it ourselves makes + # the leg identical for every OLD ref (older releases predate the + # dependency entirely, and hoisting moves it around in newer ones). + $driverDir = Join-Path $WorkRoot "pw-driver" + New-Item -ItemType Directory -Path $driverDir -Force | Out-Null + $npmCli = Join-Path (Split-Path -Parent $node) "node_modules\npm\bin\npm-cli.js" + Assert-True (Test-Path -LiteralPath $npmCli) "managed npm exists beside the managed node" + Push-Location $driverDir + try { + & $node $npmCli install --no-save --no-audit --no-fund "@playwright/test@$PlaywrightVersion" 2>&1 | + Select-Object -Last 5 | ForEach-Object { Write-Host " npm| $_" } + $npmExit = $LASTEXITCODE + } finally { + Pop-Location + } + Assert-True ($npmExit -eq 0) "npm install @playwright/test@$PlaywrightVersion into the driver dir" $recorder = Start-DesktopRecorder (Join-Path $proof "desktop-frames") $recording = Start-ScreenRecording (Join-Path $proof "recording.mkv") try { # Launch the installed app and click through Settings -> About -> # Update now. Exit 0 = the app quit for the updater hand-off. - # Copy the driver INTO the installed apps/desktop first: Node resolves + # Copy the driver INTO $driverDir first: Node resolves # require('@playwright/test') from the SCRIPT's own directory upward, # so running it from the CI checkout would resolve the wrong (or no) # node_modules. - $driver = Join-Path $appsDesktop "e2e-drive-update.cjs" + $driver = Join-Path $driverDir "e2e-drive-update.cjs" Copy-Item (Join-Path $AssetsDir "drive-update.cjs") $driver -Force - Push-Location $appsDesktop + Push-Location $driverDir try { & $node $driver $desktopExe $proof 2>&1 | ForEach-Object { Write-Host " $_" } From dc23cbbb9b54584cd8e58095d04597bd6c1c9414 Mon Sep 17 00:00:00 2001 From: ethernet Date: Tue, 11 Aug 2026 22:52:38 -0400 Subject: [PATCH 059/227] ci(windows-e2e): stderr chatter is not failure - EAP=Continue around native calls First real run of windows-installer-script-e2e.ps1 died on 'Cloning into ...': git clone writes progress to stderr, and under EAP=Stop the outer PowerShell wraps a child's stderr into a terminating NativeCommandError. Drop to EAP=Continue around every native invocation that redirects with *> (install.ps1 run, hermes --version, hermes update) and judge by exit code alone - the same prevEap pattern the GUI driver already uses. --- tests/install/windows-installer-script-e2e.ps1 | 17 ++++++++++++++--- 1 file changed, 14 insertions(+), 3 deletions(-) diff --git a/tests/install/windows-installer-script-e2e.ps1 b/tests/install/windows-installer-script-e2e.ps1 index f011a5f16e98..c9cd906eb034 100644 --- a/tests/install/windows-installer-script-e2e.ps1 +++ b/tests/install/windows-installer-script-e2e.ps1 @@ -127,10 +127,16 @@ function Invoke-Installer { $text = Get-Content -LiteralPath $script -Raw if ($text -match '\$NonInteractive') { $flags += "-NonInteractive" } $log = Join-Path $LogDir "install-$Label.log" + # Native stderr (git clone progress, pip notices) must not become + # terminating NativeCommandErrors under EAP=Stop; the exit code is the + # verdict here, not stderr chatter. + $prevEap = $ErrorActionPreference; $ErrorActionPreference = "Continue" & powershell -NoProfile -ExecutionPolicy Bypass -File $script @flags *> $log - if ($LASTEXITCODE -ne 0) { + $installExit = $LASTEXITCODE + $ErrorActionPreference = $prevEap + if ($installExit -ne 0) { Get-Content -LiteralPath $log -Tail 50 | Write-Host - Fail "install.ps1 ($Label) exited $LASTEXITCODE; full log in $log" + Fail "install.ps1 ($Label) exited $installExit; full log in $log" } } @@ -142,8 +148,11 @@ function Assert-Checkout { $hermes = Join-Path $InstallDir "venv\Scripts\hermes.exe" if (-not (Test-Path -LiteralPath $hermes)) { Fail "no hermes console script at $hermes" } $verLog = Join-Path $LogDir "version-$Label.log" + $prevEap = $ErrorActionPreference; $ErrorActionPreference = "Continue" & $hermes --version *> $verLog - if ($LASTEXITCODE -ne 0) { + $verExit = $LASTEXITCODE + $ErrorActionPreference = $prevEap + if ($verExit -ne 0) { Get-Content -LiteralPath $verLog | Write-Host Fail "hermes --version failed after $Label; log in $verLog" } @@ -169,6 +178,7 @@ switch ($UpdateMethod) { # `--yes` reaches the update subcommand only in later releases, and # argparse rejects the whole invocation when it does not exist. $updateArgs = @("update") + $prevEap = $ErrorActionPreference; $ErrorActionPreference = "Continue" $helpText = & $hermes update --help 2>&1 | Out-String if ($helpText -match '--yes') { $updateArgs += "--yes" } $log = Join-Path $LogDir "update.log" @@ -178,6 +188,7 @@ switch ($UpdateMethod) { $updateExit = $LASTEXITCODE } finally { Pop-Location + $ErrorActionPreference = $prevEap } if ($updateExit -ne 0) { Get-Content -LiteralPath $log -Tail 50 | Write-Host From cde3b4ff093f9daf495a1292154e407486ddfc34 Mon Sep 17 00:00:00 2001 From: ethernet Date: Tue, 11 Aug 2026 23:00:42 -0400 Subject: [PATCH 060/227] fix some names --- .github/workflows/install-e2e-run.yml | 6 +----- .github/workflows/install-e2e-windows-run.yml | 7 ++----- 2 files changed, 3 insertions(+), 10 deletions(-) diff --git a/.github/workflows/install-e2e-run.yml b/.github/workflows/install-e2e-run.yml index 559601cdcd39..4a0cfcb3bf54 100644 --- a/.github/workflows/install-e2e-run.yml +++ b/.github/workflows/install-e2e-run.yml @@ -66,11 +66,7 @@ permissions: jobs: e2e: - # Static name on purpose: the caller's job name already carries the - # method pair, and GitHub renders name expressions UNEXPANDED (literal - # "${{ inputs... }}") on natively skipped jobs. Short because it is - # only a rendered tail (" / e2e"). - name: e2e + name: install & update # The pairs the driver can run today; anything else is a declared TODO # and natively skips. if: inputs.install-method == 'installer-script' && contains(fromJSON('["hermes-update", "installer-script"]'), inputs.update-method) diff --git a/.github/workflows/install-e2e-windows-run.yml b/.github/workflows/install-e2e-windows-run.yml index 40bf49fb3494..a7b1457df45f 100644 --- a/.github/workflows/install-e2e-windows-run.yml +++ b/.github/workflows/install-e2e-windows-run.yml @@ -88,10 +88,7 @@ permissions: jobs: # ---- arm 1: install via the website's Hermes-Setup.exe (GUI) ------------- e2e: - # Static name on purpose: the caller's job name already carries the - # method pair, and GitHub renders name expressions UNEXPANDED (literal - # "${{ inputs... }}") on natively skipped jobs. - name: e2e + name: Hermes-Setup.exe # The one pair the driver can run today, and only from a starting # version that ships the desktop app (the caller annotates # tag-has-desktop from the tag's own tree; releases before #20059 have @@ -185,7 +182,7 @@ jobs: # ---- arm 2: install via the irm | iex one-liner (headless) --------------- script-e2e: # Same static-name reasoning as e2e above. - name: e2e + name: install.ps1 # The pairs the script driver can run today. app-update from a script # install is a declared TODO (the desktop app also has to be BUILT by # that path first). From 80a0a198dd1882f91c6f99fc5e84ba4dc2487813 Mon Sep 17 00:00:00 2001 From: ethernet Date: Tue, 11 Aug 2026 23:06:42 -0400 Subject: [PATCH 061/227] ci(install-e2e): plan chart on the run summary - combination x tag markdown table generate-e2e-matrix.mjs grows --format markdown: one row per {os, install -> update} combination, one column per starting tag, appended to GITHUB_STEP_SUMMARY by the expand job. Cells mark dispatched legs; run-vs-grey stays the run workflows' call, so the only special cell is pre-desktop (the one annotation the plan owns). JSON mode unchanged. --- .github/workflows/install-e2e.yml | 5 +++ scripts/sandbox/generate-e2e-matrix.mjs | 53 ++++++++++++++++++++++++- 2 files changed, 57 insertions(+), 1 deletion(-) diff --git a/.github/workflows/install-e2e.yml b/.github/workflows/install-e2e.yml index 8f5fb3aa9ecd..ad7d901b914b 100644 --- a/.github/workflows/install-e2e.yml +++ b/.github/workflows/install-e2e.yml @@ -134,6 +134,11 @@ jobs: for key in linux windows macos; do echo "$key=$(echo "$matrices" | node -e 'let d="";process.stdin.on("data",c=>d+=c).on("end",()=>console.log(JSON.stringify(JSON.parse(d)[process.argv[1]])))' "$key")" >> "$GITHUB_OUTPUT" done + # The plan, human-readable: a combination x starting-tag chart on + # the run's summary page. + node scripts/sandbox/generate-e2e-matrix.mjs \ + --tags '${{ needs.pick-releases.outputs.tags }}' \ + --format markdown >> "$GITHUB_STEP_SUMMARY" linux: name: ${{ matrix.name }} diff --git a/scripts/sandbox/generate-e2e-matrix.mjs b/scripts/sandbox/generate-e2e-matrix.mjs index 6abbdabb3881..05301f22e113 100644 --- a/scripts/sandbox/generate-e2e-matrix.mjs +++ b/scripts/sandbox/generate-e2e-matrix.mjs @@ -165,14 +165,65 @@ export function buildMatrices(envs, tags) { return byOs; } +/** + * Render the plan as a markdown cross-table for $GITHUB_STEP_SUMMARY: + * one row per {os, install -> update} combination, one column per + * starting tag. Every cell is dispatched; whether it RUNS or greys out + * is the run workflow's call (capability lives there, not here), so the + * chart only distinguishes the one thing the plan itself knows: windows + * desktop-surface legs from tags that predate the desktop app. + * + * @param {{os: Os, install: string, update: string}[]} envs + * @param {TagAnnotation[]} tags + * @returns {string} + */ +export function renderMarkdownPlan(envs, tags) { + const needsDesktop = (/** @type {string} */ m) => + m.startsWith('desktop-installer') || m === 'app-update'; + const lines = [ + '### Install & Update E2E plan', + '', + `${envs.length} combinations x ${tags.length} starting tags = ${envs.length * tags.length} legs`, + '', + `| combination | ${tags.map((t) => t.ref).join(' | ')} |`, + `|---|${tags.map(() => '---').join('|')}|`, + ]; + for (const env of envs) { + const cells = tags.map((tag) => { + if ( + env.os === 'windows' && !tag.desktop && + (needsDesktop(env.install) || needsDesktop(env.update)) + ) { + return 'pre-desktop'; + } + return '✅'; + }); + lines.push(`| \`${env.os}: ${env.install} -> ${env.update}\` | ${cells.join(' | ')} |`); + } + lines.push( + '', + '✅ dispatched -- the OS run workflow decides run vs native skip', + '(unimplemented method pairs grey out there). `pre-desktop`: the tag', + "ships no desktop app, so desktop-surface legs grey out regardless.", + '', + ); + return lines.join('\n'); +} + function main() { const { values } = parseArgs({ options: { tags: { type: 'string', default: '[]' }, + format: { type: 'string', default: 'json' }, }, }); const tags = /** @type {TagAnnotation[]} */ (JSON.parse(values.tags)); - const matrices = buildMatrices(generateEnvironments(SPEC), tags); + const envs = generateEnvironments(SPEC); + if (values.format === 'markdown') { + process.stdout.write(renderMarkdownPlan(envs, tags)); + return; + } + const matrices = buildMatrices(envs, tags); process.stdout.write(`${JSON.stringify(matrices, null, 2)}\n`); } From db969ce696e4f9ab5c81fc9b88f2fb14c61fe766 Mon Sep 17 00:00:00 2001 From: ethernet Date: Tue, 11 Aug 2026 23:16:09 -0400 Subject: [PATCH 062/227] ci(install-e2e): result chart on the run summary - conclusions per combination x tag generate-e2e-matrix.mjs grows --format results: reads the run's own job list as NDJSON {name, conclusion} on stdin (per-leg conclusions are NOT reachable through needs - a matrix job collapses to one aggregate result) and re-renders the plan chart with each cell's outcome. Legs are recognized by the exact name shape buildMatrices mints, so unrelated jobs fall out; duplicate leg names (one windows job per driver arm, only one runs) merge by significance - real outcomes beat skips, failures beat successes. A final report job (if: always, needs all three OS jobs) appends the chart to its step summary via gh api with the default token. Verified against two real runs: 31536931863 renders 11 passed / 0 failed / 54 skipped all-green; 31557865241 (the pre-EAP-fix run) renders its 4 real failures + cancellations over the sibling arm's skips. --- .github/workflows/install-e2e.yml | 29 +++++++++ scripts/sandbox/generate-e2e-matrix.mjs | 83 ++++++++++++++++++++++++- 2 files changed, 110 insertions(+), 2 deletions(-) diff --git a/.github/workflows/install-e2e.yml b/.github/workflows/install-e2e.yml index ad7d901b914b..edef9d95e572 100644 --- a/.github/workflows/install-e2e.yml +++ b/.github/workflows/install-e2e.yml @@ -192,3 +192,32 @@ jobs: update-method: ${{ matrix.update_method }} install-ref: ${{ matrix.install_ref }} runner: macos-latest + + # The outcome, human-readable: the plan chart again, with each cell + # replaced by how that leg actually concluded. Per-leg conclusions are + # NOT reachable through `needs` (a matrix job's result collapses to one + # aggregate), so the table body comes from the run's own job list; the + # `needs` results only sequence this job after every leg and provide + # the per-OS aggregates. + report: + name: Result chart + if: always() + needs: [linux, windows, macos] + runs-on: ubuntu-latest + timeout-minutes: 5 + steps: + - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2 + with: + sparse-checkout: scripts/sandbox/generate-e2e-matrix.mjs + sparse-checkout-cone-mode: false + - env: + GH_TOKEN: ${{ github.token }} + run: | + set -euo pipefail + { + echo "OS jobs: linux ${{ needs.linux.result }}, windows ${{ needs.windows.result }}, macos ${{ needs.macos.result }}" + echo + gh api "repos/${{ github.repository }}/actions/runs/${{ github.run_id }}/jobs?per_page=100" \ + --paginate --jq '.jobs[] | {name, conclusion}' | + node scripts/sandbox/generate-e2e-matrix.mjs --format results + } >> "$GITHUB_STEP_SUMMARY" diff --git a/scripts/sandbox/generate-e2e-matrix.mjs b/scripts/sandbox/generate-e2e-matrix.mjs index 05301f22e113..36d8c7e9ffd5 100644 --- a/scripts/sandbox/generate-e2e-matrix.mjs +++ b/scripts/sandbox/generate-e2e-matrix.mjs @@ -210,13 +210,92 @@ export function renderMarkdownPlan(envs, tags) { return lines.join('\n'); } -function main() { +/** + * Render the run's OUTCOME as the same cross-table, from the run's own job + * list (GitHub Actions API): one row per combination, one column per tag, + * each cell the leg's conclusion. Input is NDJSON {name, conclusion} lines + * -- what `gh api --paginate --jq '.jobs[] | {name, conclusion}'` emits -- + * and legs are recognized by the exact name shape buildMatrices mints + * ("os: install -> update (tag -> HEAD) / ..."), so unrelated jobs + * (pick-releases, the report job itself) fall out naturally. + * + * @param {{name: string, conclusion: string | null}[]} jobs + * @returns {string} + */ +export function renderMarkdownResults(jobs) { + const LEG = /^(linux|windows|macos): (\S+) -> (\S+) \((\S+) -> HEAD\) \//; + // A combination can surface as SEVERAL jobs with the same leg name (the + // windows run workflow has one arm per driver; exactly one runs and the + // others natively skip), so cells merge by significance: a real outcome + // always beats a skip, and a bad outcome beats a good one. + const RANK = ['skip', '✅', 'running', 'cancelled', '❌']; + /** @type {Map>} */ + const rows = new Map(); + /** @type {string[]} */ + const tags = []; + for (const job of jobs) { + const m = job.name.match(LEG); + if (!m) continue; + const combo = `${m[1]}: ${m[2]} -> ${m[3]}`; + const tag = m[4]; + if (!tags.includes(tag)) tags.push(tag); + if (!rows.has(combo)) rows.set(combo, new Map()); + const cell = (() => { + switch (job.conclusion) { + case 'success': return '✅'; + case 'failure': return '❌'; + case 'skipped': return 'skip'; + case 'cancelled': return 'cancelled'; + default: return 'running'; + } + })(); + const byTag = /** @type {Map} */ (rows.get(combo)); + const prev = byTag.get(tag); + if (prev === undefined || RANK.indexOf(cell) > RANK.indexOf(prev)) { + byTag.set(tag, cell); + } + } + if (rows.size === 0) return '### Install & Update E2E results\n\n(no legs found in this run)\n'; + const cells = [...rows.values()].flatMap((r) => [...r.values()]); + const passed = cells.filter((c) => c === '✅').length; + const failed = cells.filter((c) => c === '❌').length; + const skipped = cells.filter((c) => c === 'skip').length; + const lines = [ + '### Install & Update E2E results', + '', + `${passed} passed, ${failed} failed, ${skipped} skipped (declared TODO / pre-desktop), ${cells.length} legs total`, + '', + `| combination | ${tags.join(' | ')} |`, + `|---|${tags.map(() => '---').join('|')}|`, + ]; + for (const [combo, byTag] of rows) { + lines.push(`| \`${combo}\` | ${tags.map((t) => byTag.get(t) || '-').join(' | ')} |`); + } + lines.push(''); + return lines.join('\n'); +} + +/** @returns {Promise} all of stdin */ +function readStdin() { + return new Promise((resolve) => { + let data = ''; + process.stdin.on('data', (c) => { data += c; }); + process.stdin.on('end', () => resolve(data)); + }); +} + +async function main() { const { values } = parseArgs({ options: { tags: { type: 'string', default: '[]' }, format: { type: 'string', default: 'json' }, }, }); + if (values.format === 'results') { + const jobs = (await readStdin()).split('\n').filter((l) => l.trim()).map((l) => JSON.parse(l)); + process.stdout.write(renderMarkdownResults(jobs)); + return; + } const tags = /** @type {TagAnnotation[]} */ (JSON.parse(values.tags)); const envs = generateEnvironments(SPEC); if (values.format === 'markdown') { @@ -228,5 +307,5 @@ function main() { } if (process.argv[1] && fileURLToPath(import.meta.url) === path.resolve(process.argv[1])) { - main(); + await main(); } From 8e6f6d863ce2d13e882c54fc782040b6c8132d95 Mon Sep 17 00:00:00 2001 From: ethernet Date: Tue, 11 Aug 2026 23:16:09 -0400 Subject: [PATCH 063/227] ci(install-e2e): result chart on the run summary - conclusions per combination x tag generate-e2e-matrix.mjs grows --format results: reads the run's own job list as NDJSON {name, conclusion} on stdin (per-leg conclusions are NOT reachable through needs - a matrix job collapses to one aggregate result) and re-renders the plan chart with each cell's outcome. Legs are recognized by the exact name shape buildMatrices mints, so unrelated jobs fall out; duplicate leg names (one windows job per driver arm, only one runs) merge by significance - real outcomes beat skips, failures beat successes. A final report job (if: always, needs all three OS jobs) appends the chart to its step summary via gh api with the default token. Verified against two real runs: 31536931863 renders 11 passed / 0 failed / 54 skipped all-green; 31557865241 (the pre-EAP-fix run) renders its 4 real failures + cancellations over the sibling arm's skips. --- scripts/sandbox/generate-e2e-matrix.mjs | 8 +------- 1 file changed, 1 insertion(+), 7 deletions(-) diff --git a/scripts/sandbox/generate-e2e-matrix.mjs b/scripts/sandbox/generate-e2e-matrix.mjs index 36d8c7e9ffd5..8b86572abb6f 100644 --- a/scripts/sandbox/generate-e2e-matrix.mjs +++ b/scripts/sandbox/generate-e2e-matrix.mjs @@ -196,16 +196,10 @@ export function renderMarkdownPlan(envs, tags) { ) { return 'pre-desktop'; } - return '✅'; + return '⏳ '; }); lines.push(`| \`${env.os}: ${env.install} -> ${env.update}\` | ${cells.join(' | ')} |`); } - lines.push( - '', - '✅ dispatched -- the OS run workflow decides run vs native skip', - '(unimplemented method pairs grey out there). `pre-desktop`: the tag', - "ships no desktop app, so desktop-surface legs grey out regardless.", - '', ); return lines.join('\n'); } From 31d958800117c2cd802eecbc1e61df969dd13904 Mon Sep 17 00:00:00 2001 From: ethernet Date: Tue, 11 Aug 2026 23:19:23 -0400 Subject: [PATCH 064/227] ci(install-e2e): full installer transcripts in the job log, folded in ::group:: All three drivers (installer-script-e2e.sh, windows-installer-script -e2e.ps1, windows-desktop-gui-e2e.ps1) now emit the complete install/update transcript into the job log wrapped in ::group::/ ::endgroup:: - collapsed by default, one click to expand, win or lose. Replaces the tail-50-only-on-failure pattern: a green install's transcript is how you diagnose the leg that fails next, and the artifact download was the only way to see it before. The GUI driver's bootstrap-installer.log / desktop-update-handoff.log tails become full folded dumps too. --- tests/install/installer-script-e2e.sh | 24 ++++++++++++------- tests/install/windows-desktop-gui-e2e.ps1 | 12 ++++++---- .../install/windows-installer-script-e2e.ps1 | 16 +++++++++---- 3 files changed, 35 insertions(+), 17 deletions(-) diff --git a/tests/install/installer-script-e2e.sh b/tests/install/installer-script-e2e.sh index 88c7327e88e9..3cce3287ca1e 100755 --- a/tests/install/installer-script-e2e.sh +++ b/tests/install/installer-script-e2e.sh @@ -69,6 +69,14 @@ SERVE_REPO="$WORK_ROOT/serve.git" step() { printf '\n=== %s ===\n' "$*"; } ok() { printf ' OK %s\n' "$*"; } fail() { printf 'E2E ASSERTION FAILED: %s\n' "$*" >&2; exit 1; } +# Full transcript in the job log, collapsed (GitHub renders ::group:: as a +# fold; plain text anywhere else). Win or lose -- a green install's log is +# how you diagnose the leg that fails next. +log_group() { + printf '::group::%s\n' "$1" + cat "$2" + printf '::endgroup::\n' +} rm -rf "$WORK_ROOT" mkdir -p "$WORK_ROOT" "$LOG_DIR" @@ -149,10 +157,10 @@ run_installer() { fi # "$LOG_DIR/install-$2.log" 2>&1; then - tail -50 "$LOG_DIR/install-$2.log" >&2 - fail "install.sh ($2) exited non-zero; full log in $LOG_DIR/install-$2.log" - fi + local rc=0 + bash "$script" "${flags[@]}" < /dev/null > "$LOG_DIR/install-$2.log" 2>&1 || rc=$? + log_group "install.sh ($2) transcript" "$LOG_DIR/install-$2.log" + [ "$rc" -eq 0 ] || fail "install.sh ($2) exited $rc; transcript above, log at $LOG_DIR/install-$2.log" } assert_checkout() { @@ -192,10 +200,10 @@ case "$UPDATE_METHOD" in else update_cmd=("$HERMES" update) fi - if ! (cd "$INSTALL_DIR" && "${update_cmd[@]}" < /dev/null > "$LOG_DIR/update.log" 2>&1); then - tail -50 "$LOG_DIR/update.log" >&2 - fail "hermes update exited non-zero; full log in $LOG_DIR/update.log" - fi + rc=0 + (cd "$INSTALL_DIR" && "${update_cmd[@]}" < /dev/null > "$LOG_DIR/update.log" 2>&1) || rc=$? + log_group "hermes update transcript" "$LOG_DIR/update.log" + [ "$rc" -eq 0 ] || fail "hermes update exited $rc; transcript above, log at $LOG_DIR/update.log" ;; installer-script) # A user re-running the one-liner today gets the CURRENT script. diff --git a/tests/install/windows-desktop-gui-e2e.ps1 b/tests/install/windows-desktop-gui-e2e.ps1 index 03eba70aa77a..9a4812babf57 100644 --- a/tests/install/windows-desktop-gui-e2e.ps1 +++ b/tests/install/windows-desktop-gui-e2e.ps1 @@ -460,11 +460,12 @@ function Invoke-PhaseInstallGui { finally { Stop-ScreenRecording $recording Stop-DesktopRecorder $recorder (Join-Path $proof "desktop-frames") - # Surface the installer's own log win or lose. + # Surface the installer's own log win or lose, full and folded. $bootLog = Join-Path $HermesHome "logs\bootstrap-installer.log" if (Test-Path -LiteralPath $bootLog) { - Write-Host " --- bootstrap-installer.log (tail) ---" - Get-Content -LiteralPath $bootLog -Tail 40 | ForEach-Object { Write-Host " | $_" } + Write-Host "::group::bootstrap-installer.log" + Get-Content -LiteralPath $bootLog | Write-Host + Write-Host "::endgroup::" Copy-Item $bootLog $proof -Force -ErrorAction SilentlyContinue } } @@ -661,8 +662,9 @@ function Invoke-GuiUpdateDesktopRoute([string]$TargetSha) { Stop-DesktopRecorder $recorder (Join-Path $proof "desktop-frames") $handoffLog = Join-Path $HermesHome "logs\desktop-update-handoff.log" if (Test-Path -LiteralPath $handoffLog) { - Write-Host " --- desktop-update-handoff.log (tail) ---" - Get-Content -LiteralPath $handoffLog -Tail 60 | ForEach-Object { Write-Host " | $_" } + Write-Host "::group::desktop-update-handoff.log" + Get-Content -LiteralPath $handoffLog | Write-Host + Write-Host "::endgroup::" Copy-Item $handoffLog (Join-Path $proof "desktop-update-handoff.log") -Force -ErrorAction SilentlyContinue } # Quit the relaunched app so job teardown is clean. diff --git a/tests/install/windows-installer-script-e2e.ps1 b/tests/install/windows-installer-script-e2e.ps1 index c9cd906eb034..58a8d8944af7 100644 --- a/tests/install/windows-installer-script-e2e.ps1 +++ b/tests/install/windows-installer-script-e2e.ps1 @@ -54,6 +54,14 @@ function Fail([string]$Message) { Write-Host "E2E ASSERTION FAILED: $Message" -ForegroundColor Red exit 1 } +# Full transcript in the job log, collapsed (GitHub renders ::group:: as a +# fold). Win or lose -- a green install's log is how you diagnose the leg +# that fails next. +function Write-LogGroup([string]$Title, [string]$LogPath) { + Write-Host "::group::$Title" + if (Test-Path -LiteralPath $LogPath) { Get-Content -LiteralPath $LogPath | Write-Host } + Write-Host "::endgroup::" +} function Invoke-Git { param([string[]]$GitArgs) @@ -134,9 +142,9 @@ function Invoke-Installer { & powershell -NoProfile -ExecutionPolicy Bypass -File $script @flags *> $log $installExit = $LASTEXITCODE $ErrorActionPreference = $prevEap + Write-LogGroup "install.ps1 ($Label) transcript" $log if ($installExit -ne 0) { - Get-Content -LiteralPath $log -Tail 50 | Write-Host - Fail "install.ps1 ($Label) exited $installExit; full log in $log" + Fail "install.ps1 ($Label) exited $installExit; transcript above, log at $log" } } @@ -190,9 +198,9 @@ switch ($UpdateMethod) { Pop-Location $ErrorActionPreference = $prevEap } + Write-LogGroup "hermes update transcript" $log if ($updateExit -ne 0) { - Get-Content -LiteralPath $log -Tail 50 | Write-Host - Fail "hermes update exited $updateExit; full log in $log" + Fail "hermes update exited $updateExit; transcript above, log at $log" } } "installer-script" { From 3cd663647172f11477b7b94fec029cfc03672422 Mon Sep 17 00:00:00 2001 From: ethernet Date: Tue, 11 Aug 2026 23:24:00 -0400 Subject: [PATCH 065/227] fix(install-e2e): stray paren broke the generator - every node invocation died --- scripts/sandbox/generate-e2e-matrix.mjs | 1 - 1 file changed, 1 deletion(-) diff --git a/scripts/sandbox/generate-e2e-matrix.mjs b/scripts/sandbox/generate-e2e-matrix.mjs index 8b86572abb6f..701b8ab77937 100644 --- a/scripts/sandbox/generate-e2e-matrix.mjs +++ b/scripts/sandbox/generate-e2e-matrix.mjs @@ -200,7 +200,6 @@ export function renderMarkdownPlan(envs, tags) { }); lines.push(`| \`${env.os}: ${env.install} -> ${env.update}\` | ${cells.join(' | ')} |`); } - ); return lines.join('\n'); } From a64b5f5cafe2b3e65bd874fe224a5ac51f0d5338 Mon Sep 17 00:00:00 2001 From: ethernet Date: Wed, 12 Aug 2026 03:30:54 -0400 Subject: [PATCH 066/227] test(install-e2e): smoke hermes desktop --build-only between install and update Each script-driver leg now proves the installed CLI can build the desktop app, after the install phase and again after the update. --build-only runs the full desktop pipeline and stops before the launch - the same call hermes update makes. Old releases that predate the flag skip the phase after a --help probe of the installed binary. Actually launching the app is a TODO: it needs the spawn-interception launcher and, on linux runners, a virtual display. Also adds tests/install/README.md describing how the test family works: the four layers, the git-redirect isolation, the phases, the probe-do-not-assume rule for old versions, skips, triggers, artifacts. --- tests/install/README.md | 69 +++++++++++++++++++ tests/install/installer-script-e2e.sh | 29 ++++++++ .../install/windows-installer-script-e2e.ps1 | 42 +++++++++++ 3 files changed, 140 insertions(+) create mode 100644 tests/install/README.md diff --git a/tests/install/README.md b/tests/install/README.md new file mode 100644 index 000000000000..25bbab5069b2 --- /dev/null +++ b/tests/install/README.md @@ -0,0 +1,69 @@ +# Install and Update E2E Tests + +These tests answer one question: can a user on a released version get to this commit? + +Each test leg installs an old released version, then updates it to HEAD. The install and the update run the real user surfaces. The legs do not use mocks and do not use headless proxies of GUI flows. + +## The layers + +The test family has four layers. Each layer has one job. + +1. `scripts/sandbox/generate-e2e-matrix.mjs` declares the support matrix. It lists every {os, install-method, update-method} pair. It expands the pairs against the sampled release tags. It knows nothing about which pairs CI can run. +2. `.github/workflows/install-e2e.yml` is the primary workflow. It picks the release tags, runs the generator, and fans out one matrix job per OS. It also writes the plan chart and the result chart on the run summary. +3. The run workflows own the capability knowledge. `install-e2e-run.yml` serves linux and macos with one OS-agnostic driver. `install-e2e-windows-run.yml` serves windows. A job-level `if:` gate in each run workflow lists the pairs its driver can run. All other pairs skip natively and show as grey. +4. The drivers do the work. `tests/install/installer-script-e2e.sh` is the POSIX driver. `tests/install/windows-installer-script-e2e.ps1` is the windows script driver. `tests/install/windows-desktop-gui-e2e.ps1` is the windows GUI driver. + +To declare a new method, edit the generator. To implement a method, flip the gate in the run workflow and extend a driver. + +## The isolation trick + +The drivers do not touch the network for git operations. Each driver makes a bare clone of the checkout at `serve.git`. Then it points every git process at this clone. The mechanism is a driver-owned `GIT_CONFIG_GLOBAL` file with `url..insteadOf` rewrites for both canonical repository URLs. + +The driver parks the `main` branch of `serve.git` at the old release. The installer runs and lands on the old release. Then the driver moves `main` to HEAD. An update becomes available in the same way that it does for a real user. + +The installer script is not downloaded. The install leg runs the copy from the old git ref. This is the copy that a user of that version executed. The update leg runs the copy from HEAD. + +## What one leg does + +Each leg with the script drivers has these phases: + +1. Stage: make the bare clone, park `main` at the old release. +2. Install: run the old release's own installer script. Make sure that the checkout is at the old commit and that `hermes --version` works. +3. Desktop smoke: run `hermes desktop --build-only` from the installed CLI. This proves that the installed version can build the desktop app. If the installed version does not have this flag, the phase reports a skip and continues. +4. Update: move `main` to HEAD. Apply one update method. Make sure that the checkout is at HEAD and that `hermes --version` works. +5. Desktop smoke again, at HEAD. + +The windows GUI driver replaces phases 2 and 4. It downloads the published `Hermes-Setup.exe`, clicks through the installer window with AutoHotkey, and clicks "Update now" in the running app with Playwright. + +## Old versions + +A leg can install a release from months back. The driver must not assume that the old version has today's CLI surface. The rule: probe, do not assume. + +- For the installer, read the flag from the old ref's own script text. +- For the installed CLI, ask the binary with `--help`. +- If a flag is not found, omit the flag. This is not an error. + +## Skips + +A grey leg is normal. There are two causes: + +- The method pair has no driver yet. The pair is a declared TODO. The gate in the run workflow lists the pairs that run. +- The starting release predates the surface under test. Example: a release without `apps/desktop` has no window to launch. The tag annotation `tag_has_desktop` from the primary workflow marks these releases. + +The result chart on the run summary shows each leg as passed, failed, or skipped. + +## Triggers + +The matrix does not run on pull requests. One leg installs real toolchains and takes more than 10 minutes. The triggers are: + +- A schedule, every 12 hours. This finds upstream drift. +- A release tag push. This is the moment the set of start versions changes. +- Manual dispatch. You can select the route and the tag count: + +``` +gh workflow run install-e2e.yml --ref -f route=both -f tag-count=2 +``` + +## Artifacts + +Each leg uploads its logs as an artifact. The windows GUI leg also uploads screenshots, a screen recording, and the update result file. Get them with `gh run download `. diff --git a/tests/install/installer-script-e2e.sh b/tests/install/installer-script-e2e.sh index 3cce3287ca1e..9c18cdcc77c1 100755 --- a/tests/install/installer-script-e2e.sh +++ b/tests/install/installer-script-e2e.sh @@ -176,11 +176,39 @@ assert_checkout() { ok "hermes --version works: $(head -c 120 "$LOG_DIR/version-$2.log" | tr -d '\n')" } +smoke_desktop() { + # $1: label (old|head). Prove the installed CLI can produce the desktop + # app: `hermes desktop --build-only` runs the full desktop pipeline + # (workspace install, renderer build, stamp write) and stops before the + # launch -- the same call `hermes update` itself makes. Probe the + # INSTALLED hermes for the flag rather than assuming this checkout's + # surface: sampled OLD releases may predate `hermes desktop` or + # --build-only entirely, and for them the phase skips, loudly. + local hermes="$INSTALL_DIR/venv/bin/hermes" + if ! "$hermes" desktop --help 2>/dev/null | grep -qF -- --build-only; then + ok "hermes desktop --build-only not supported at $1; skipping desktop smoke" + return 0 + fi + local rc=0 + (cd "$INSTALL_DIR" && "$hermes" desktop --build-only < /dev/null \ + > "$LOG_DIR/desktop-smoke-$1.log" 2>&1) || rc=$? + log_group "hermes desktop --build-only ($1) transcript" "$LOG_DIR/desktop-smoke-$1.log" + [ "$rc" -eq 0 ] || fail "hermes desktop --build-only ($1) exited $rc; transcript above" + ok "hermes desktop --build-only works at $1" + # TODO(launch): LAUNCH the built app and auto-close it. Mechanism when + # the pieces land: driver-side spawn interception (a sitecustomize.py on + # PYTHONPATH wraps subprocess.run under an env-var opt-in and captures + # the real argv/cwd/env at the spawn site) + Playwright _electron.launch + # on the captured spec; electronApp.close() is the auto-close. Blocked + # on that asset and, for linux runners, on a virtual display (Xvfb). +} + # --- install OLD --------------------------------------------------------------- step "installing OLD ($INSTALL_REF) via its own scripts/install.sh" run_installer "$OLD_SHA" old assert_checkout "$OLD_SHA" OLD +smoke_desktop old # --- update OLD -> HEAD ---------------------------------------------------------- @@ -211,5 +239,6 @@ case "$UPDATE_METHOD" in ;; esac assert_checkout "$HEAD_SHA" HEAD +smoke_desktop head step "PASS: $INSTALL_REF -> HEAD via $UPDATE_METHOD" diff --git a/tests/install/windows-installer-script-e2e.ps1 b/tests/install/windows-installer-script-e2e.ps1 index 58a8d8944af7..34157ccd81c8 100644 --- a/tests/install/windows-installer-script-e2e.ps1 +++ b/tests/install/windows-installer-script-e2e.ps1 @@ -167,11 +167,52 @@ function Assert-Checkout { Ok "hermes --version works: $((Get-Content -LiteralPath $verLog -First 1))" } +function Test-DesktopSmoke { + param([string]$Label) + # Prove the installed CLI can produce the desktop app: `hermes desktop + # --build-only` runs the full desktop pipeline (workspace install, + # renderer build, stamp write) and stops before the launch -- the same + # call `hermes update` itself makes. Probe the INSTALLED hermes for the + # flag rather than assuming this checkout's surface: sampled OLD + # releases may predate `hermes desktop` or --build-only entirely, and + # for them the phase skips, loudly. + $hermes = Join-Path $InstallDir "venv\Scripts\hermes.exe" + $prevEap = $ErrorActionPreference; $ErrorActionPreference = "Continue" + $helpText = & $hermes desktop --help 2>&1 | Out-String + $ErrorActionPreference = $prevEap + if ($helpText -notmatch '--build-only') { + Ok "hermes desktop --build-only not supported at $Label; skipping desktop smoke" + return + } + $log = Join-Path $LogDir "desktop-smoke-$Label.log" + $prevEap = $ErrorActionPreference; $ErrorActionPreference = "Continue" + Push-Location $InstallDir + try { + & $hermes desktop --build-only *> $log + $smokeExit = $LASTEXITCODE + } finally { + Pop-Location + $ErrorActionPreference = $prevEap + } + Write-LogGroup "hermes desktop --build-only ($Label) transcript" $log + if ($smokeExit -ne 0) { + Fail "hermes desktop --build-only ($Label) exited $smokeExit; transcript above" + } + Ok "hermes desktop --build-only works at $Label" + # TODO(launch): LAUNCH the built app and auto-close it. Mechanism when + # the pieces land: driver-side spawn interception (a sitecustomize.py + # on PYTHONPATH wraps subprocess.run under an env-var opt-in and + # captures the real argv/cwd/env at the spawn site) + Playwright + # _electron.launch on the captured spec; electronApp.close() is the + # auto-close. +} + # --- install OLD ------------------------------------------------------------------ Step "installing OLD ($InstallRef) via its own scripts/install.ps1" Invoke-Installer $OldSha "old" Assert-Checkout $OldSha "OLD" +Test-DesktopSmoke "old" # --- update OLD -> HEAD -------------------------------------------------------------- @@ -209,5 +250,6 @@ switch ($UpdateMethod) { } } Assert-Checkout $HeadSha "HEAD" +Test-DesktopSmoke "head" Step "PASS: $InstallRef -> HEAD via $UpdateMethod" From 1af20936638cfff1ea02ec9853d26c89f363aa07 Mon Sep 17 00:00:00 2001 From: ethernet Date: Wed, 12 Aug 2026 03:36:31 -0400 Subject: [PATCH 067/227] test(install-e2e): split app-update into open-app-update + hermes-desktop-app-update The desktop app has two launch paths, so app-update becomes two methods. open-app-update starts the app from the OS entry point the desktop installer created (the installed exe / the .app), so it exists only where a desktop installer does. hermes-desktop-app-update starts the app via hermes desktop, which every install method provides on every OS that ships the desktop app - on linux it is the only app surface, since no desktop installer or packaged artifact exists there. Both variants are desktop-surface methods on every OS, so the tag_has_desktop annotation moves from windows-only to every matrix entry, install-e2e-run.yml grows the input, and the plan chart marks pre-desktop cells on all OSes. The windows GUI arm's implemented pair renames to open-app-update; every other new combination is a declared TODO that natively skips. --- .github/workflows/install-e2e-run.yml | 7 ++- .github/workflows/install-e2e-windows-run.yml | 23 +++++---- .github/workflows/install-e2e.yml | 4 +- scripts/sandbox/generate-e2e-matrix.mjs | 49 ++++++++++++++----- tests/install/README.md | 7 +++ tests/install/windows-desktop-gui-e2e.ps1 | 18 ++++--- 6 files changed, 78 insertions(+), 30 deletions(-) diff --git a/.github/workflows/install-e2e-run.yml b/.github/workflows/install-e2e-run.yml index 4a0cfcb3bf54..e7501ec07f50 100644 --- a/.github/workflows/install-e2e-run.yml +++ b/.github/workflows/install-e2e-run.yml @@ -42,7 +42,7 @@ on: required: true type: string update-method: - description: 'How the install updates to HEAD. Supported: hermes-update (the updater) or installer-script (re-run the one-liner). Declared-but-TODO methods skip.' + description: 'How the install updates to HEAD. Supported: hermes-update (the updater) or installer-script (re-run the one-liner). Declared-but-TODO methods (open-app-update, hermes-desktop-app-update) skip.' required: true type: string install-ref: @@ -50,6 +50,11 @@ on: required: false type: string default: refs/heads/main + tag-has-desktop: + description: "Whether install-ref ships the desktop app (apps/desktop). The caller annotates this from the tag's own tree; desktop-method legs from pre-desktop releases natively skip." + required: false + type: boolean + default: true runner: description: 'Runner label.' required: false diff --git a/.github/workflows/install-e2e-windows-run.yml b/.github/workflows/install-e2e-windows-run.yml index a7b1457df45f..2975ad579d69 100644 --- a/.github/workflows/install-e2e-windows-run.yml +++ b/.github/workflows/install-e2e-windows-run.yml @@ -19,22 +19,26 @@ name: Install & Update E2E — Windows desktop (reusable) # headed, AutoHotkey clicks Install -> Launch, # the real Electron window must appear. Update # methods: -# app-update the app's own Update button: -# the installed Hermes.exe runs +# open-app-update the app's own Update +# button, app launched from the +# installed Hermes.exe: it runs # under Playwright's Electron # driver, which clicks Settings # -> About -> "Update now"; the # production hand-off chain runs # untouched. # (hermes-update, desktop-installer@latest, -# installer-script re-run: declared TODOs.) +# installer-script re-run, +# hermes-desktop-app-update: declared TODOs.) # installer-script the irm | iex one-liner # (tests/install/windows-installer-script-e2e # .ps1): runs the install.ps1 shipped AT the # OLD ref, headless. Update methods: # hermes-update venv hermes.exe update # installer-script re-run HEAD's install.ps1 -# (app-update from a script install: TODO.) +# (open-app-update and +# hermes-desktop-app-update from a script +# install: TODO.) # # Method pairs without a driver yet NATIVELY SKIP (grey check, no runner): # the capability knowledge lives here, next to the drivers, so the caller @@ -47,7 +51,7 @@ name: Install & Update E2E — Windows desktop (reusable) # uses: ./.github/workflows/install-e2e-windows-run.yml # with: # install-method: desktop-installer@latest -# update-method: app-update +# update-method: open-app-update # install-ref: v2026.8.3 on: @@ -58,7 +62,7 @@ on: required: true type: string update-method: - description: 'How the install updates to HEAD. Supported: app-update (Update button under Playwright, from a desktop install) and hermes-update / installer-script (from a script install). Declared-but-TODO pairs skip.' + description: 'How the install updates to HEAD. Supported: open-app-update (Update button under Playwright, app launched from the installed exe, from a desktop install) and hermes-update / installer-script (from a script install). Declared-but-TODO pairs (incl. hermes-desktop-app-update) skip.' required: true type: string install-ref: @@ -94,7 +98,7 @@ jobs: # tag-has-desktop from the tag's own tree; releases before #20059 have # no window to launch and no Update button to click). Anything else is # a native skip: a declared-TODO method pair, or a pre-desktop tag. - if: inputs.install-method == 'desktop-installer@latest' && inputs.update-method == 'app-update' && inputs.tag-has-desktop + if: inputs.install-method == 'desktop-installer@latest' && inputs.update-method == 'open-app-update' && inputs.tag-has-desktop runs-on: windows-latest timeout-minutes: ${{ inputs.timeout-minutes }} @@ -183,8 +187,9 @@ jobs: script-e2e: # Same static-name reasoning as e2e above. name: install.ps1 - # The pairs the script driver can run today. app-update from a script - # install is a declared TODO (the desktop app also has to be BUILT by + # The pairs the script driver can run today. open-app-update and + # hermes-desktop-app-update from a script install are declared TODOs + # (the desktop app also has to be BUILT by # that path first). if: inputs.install-method == 'installer-script' && contains(fromJSON('["hermes-update", "installer-script"]'), inputs.update-method) runs-on: windows-latest diff --git a/.github/workflows/install-e2e.yml b/.github/workflows/install-e2e.yml index edef9d95e572..0bfe470d5459 100644 --- a/.github/workflows/install-e2e.yml +++ b/.github/workflows/install-e2e.yml @@ -14,7 +14,7 @@ name: Install & Update E2E # clicked by AutoHotkey, update via the app, Playwright # clicking "Update now" (install-e2e-windows-run.yml) # Matrix: macos the same OS-agnostic driver as linux, on macos-latest -# (install-e2e-run.yml; app-update pairs are TODO) +# (install-e2e-run.yml; app-update variants are TODO) # # Every combination is dispatched to its OS's run workflow; the run # workflow natively skips (grey) what its driver cannot run yet -- an @@ -158,6 +158,7 @@ jobs: install-method: ${{ matrix.install_method }} update-method: ${{ matrix.update_method }} install-ref: ${{ matrix.install_ref }} + tag-has-desktop: ${{ matrix.tag_has_desktop }} windows: name: ${{ matrix.name }} @@ -191,6 +192,7 @@ jobs: install-method: ${{ matrix.install_method }} update-method: ${{ matrix.update_method }} install-ref: ${{ matrix.install_ref }} + tag-has-desktop: ${{ matrix.tag_has_desktop }} runner: macos-latest # The outcome, human-readable: the plan chart again, with each cell diff --git a/scripts/sandbox/generate-e2e-matrix.mjs b/scripts/sandbox/generate-e2e-matrix.mjs index 701b8ab77937..2011f531fb26 100644 --- a/scripts/sandbox/generate-e2e-matrix.mjs +++ b/scripts/sandbox/generate-e2e-matrix.mjs @@ -21,9 +21,9 @@ * * Prints JSON: { linux: {include:[...]}, windows: {include:[...]}, * macos: {include:[...]} } -- every entry is {name, install_method, - * update_method, install_ref}, and windows entries add tag_has_desktop - * (from the tag annotation) so the run workflow can natively skip - * desktop-surface legs from releases that predate the desktop app. + * update_method, install_ref, tag_has_desktop} (the tag annotation lets + * a run workflow natively skip desktop-surface legs from releases that + * predate the desktop app). */ import path from 'node:path'; @@ -41,10 +41,18 @@ import { fileURLToPath } from 'node:url'; * installer-script is the platform's one-liner (curl | bash on * linux/macos, irm | iex on windows); packaged-app is declared but not * used by any OS spec yet. - * @typedef {InstallMethod | 'hermes-update' | 'app-update'} UpdateMethod + * @typedef {InstallMethod | 'hermes-update' | 'open-app-update' | 'hermes-desktop-app-update'} UpdateMethod * Every install method doubles as an update method (re-run it over the - * existing install), plus the updater CLI and the running app's own - * Update button. + * existing install), plus the updater CLI and the two app-update + * variants. The variants differ by launch surface: open-app-update + * starts the app from the OS entry point the install registered + * (Start Menu / Desktop shortcuts) -- today only the windows desktop + * installer's stage registers one (install.sh --include-desktop + * builds the app but registers nothing), so these legs pair with a + * desktop-installer install; hermes-desktop-app-update starts the app + * via `hermes desktop`, which every install method provides on every + * OS that ships the desktop app. Both then update through the app's + * own Update button. * @typedef {'linux' | 'windows' | 'macos'} Os * * @typedef {{method: InstallMethod, versions?: InstallerVersion[]}} InstallEntry @@ -78,8 +86,11 @@ export const SPEC = { // Run the bootstrap exe again over an existing install (--update flow). { method: 'desktop-installer', versions: ['latest'] }, { method: 'hermes-update' }, - // Settings -> About -> "Update now" inside the running desktop app. - { method: 'app-update' }, + // Settings -> About -> "Update now", app launched from the installed + // exe (the entry point the desktop installer created). + { method: 'open-app-update' }, + // Same button, app launched via `hermes desktop`. + { method: 'hermes-desktop-app-update' }, ], }, macos: { @@ -89,7 +100,11 @@ export const SPEC = { update: [ { method: 'installer-script' }, { method: 'hermes-update' }, - { method: 'app-update' }, + // install.sh --include-desktop builds the .app inside the checkout + // but registers no OS entry point, so open-app-update legs pair + // with a desktop-installer install (the published dmg). + { method: 'open-app-update' }, + { method: 'hermes-desktop-app-update' }, ], }, linux: { @@ -99,6 +114,11 @@ export const SPEC = { update: [ { method: 'installer-script' }, { method: 'hermes-update' }, + // No desktop installer and no packaged desktop artifact exist for + // linux, so there is no open-app-update; `hermes desktop` is always + // the source-mode path (build apps/desktop from the checkout, launch + // electron) and is the one app surface a linux install has. + { method: 'hermes-desktop-app-update' }, ], }, }; @@ -157,8 +177,11 @@ export function buildMatrices(envs, tags) { install_method: env.install, update_method: env.update, install_ref: tag.ref, + // Every OS declares desktop-surface methods (both app-update + // variants at minimum), so every leg carries the annotation and + // its run workflow can natively skip pre-desktop tags. + tag_has_desktop: tag.desktop, }; - if (env.os === 'windows') entry.tag_has_desktop = tag.desktop; byOs[env.os].include.push(entry); } } @@ -170,7 +193,7 @@ export function buildMatrices(envs, tags) { * one row per {os, install -> update} combination, one column per * starting tag. Every cell is dispatched; whether it RUNS or greys out * is the run workflow's call (capability lives there, not here), so the - * chart only distinguishes the one thing the plan itself knows: windows + * chart only distinguishes the one thing the plan itself knows: * desktop-surface legs from tags that predate the desktop app. * * @param {{os: Os, install: string, update: string}[]} envs @@ -179,7 +202,7 @@ export function buildMatrices(envs, tags) { */ export function renderMarkdownPlan(envs, tags) { const needsDesktop = (/** @type {string} */ m) => - m.startsWith('desktop-installer') || m === 'app-update'; + m.startsWith('desktop-installer') || m === 'open-app-update' || m === 'hermes-desktop-app-update'; const lines = [ '### Install & Update E2E plan', '', @@ -191,7 +214,7 @@ export function renderMarkdownPlan(envs, tags) { for (const env of envs) { const cells = tags.map((tag) => { if ( - env.os === 'windows' && !tag.desktop && + !tag.desktop && (needsDesktop(env.install) || needsDesktop(env.update)) ) { return 'pre-desktop'; diff --git a/tests/install/README.md b/tests/install/README.md index 25bbab5069b2..be5b75b95683 100644 --- a/tests/install/README.md +++ b/tests/install/README.md @@ -43,6 +43,13 @@ A leg can install a release from months back. The driver must not assume that th - For the installed CLI, ask the binary with `--help`. - If a flag is not found, omit the flag. This is not an error. +## The two app-update variants + +The desktop app has two launch paths, so the matrix has two app-update methods. Both click "Update now" in the running app. They differ in how the app starts: + +- `open-app-update`: the app starts from the OS entry point that the install created. On windows these are the Start Menu and Desktop shortcuts to the installed `Hermes.exe`; the desktop installer always creates them. The installer scripts do not create entry points: their opt-in desktop stage (`--include-desktop` / `-IncludeDesktop`) builds the app inside the checkout but does not register it with the OS. So `open-app-update` legs pair with a `desktop-installer` install. +- `hermes-desktop-app-update`: the app starts with the `hermes desktop` command. Every install method provides this command, on each OS that ships the desktop app. On linux this is the only app surface: no desktop installer and no packaged desktop artifact exist for linux. + ## Skips A grey leg is normal. There are two causes: diff --git a/tests/install/windows-desktop-gui-e2e.ps1 b/tests/install/windows-desktop-gui-e2e.ps1 index 9a4812babf57..5bc5404b44b3 100644 --- a/tests/install/windows-desktop-gui-e2e.ps1 +++ b/tests/install/windows-desktop-gui-e2e.ps1 @@ -73,11 +73,11 @@ param( # Update method to exercise in the update-gui phase, named by the same # ids the combination generator (scripts/sandbox/generate-e2e-matrix - # .mjs) declares. Only "app-update" (the app's own Update button) is - # implemented; the others are declared arms so the surface is stable - # when they land. - [ValidateSet("app-update", "hermes-update", "desktop-installer@latest", "installer-script")] - [string]$Route = "app-update", + # .mjs) declares. Only "open-app-update" (the app's own Update button, + # app launched from the installed exe) is implemented; the others are + # declared arms so the surface is stable when they land. + [ValidateSet("open-app-update", "hermes-desktop-app-update", "hermes-update", "desktop-installer@latest", "installer-script")] + [string]$Route = "open-app-update", # The OLD version: the ref served as `main` while the installer runs, # i.e. what the user starts on. The published Hermes-Setup.exe carries @@ -675,9 +675,15 @@ function Invoke-GuiUpdateDesktopRoute([string]$TargetSha) { function Invoke-PhaseUpdateGui { $state = Read-State switch ($Route) { - "app-update" { + "open-app-update" { Invoke-GuiUpdateDesktopRoute $state.current } + "hermes-desktop-app-update" { + # TODO: launch the app via `hermes desktop` (spawn interception + # captures the real argv/cwd/env; Playwright launches from the + # captured spec), then the same Update-now click. + throw "update method 'hermes-desktop-app-update' is not implemented yet" + } "hermes-update" { # TODO: run `hermes update` from the installed venv -- the CLI # route. Needs the same completion/sha asserts minus the From 239523414e1b22c6140d8a92d17c992d08551398 Mon Sep 17 00:00:00 2001 From: ethernet Date: Wed, 12 Aug 2026 04:01:16 -0400 Subject: [PATCH 068/227] test(install-e2e): installer-script+desktop is its own install and update method The one-liner with its desktop stage opted in (--include-desktop / -IncludeDesktop) is a real install kind, distinct on both sides: on windows the stage builds Hermes.exe AND registers Start Menu / Desktop shortcuts - a second path to a hand-launchable app - while on linux/macos it builds into the checkout and registers no OS entry point. Declared on every OS and driven by both script drivers: the drivers pass the flag through (hard failure if the ref predates it - the tag-has-desktop gate already skips pre-desktop tags upstream) and assert the built app exists under apps/desktop/release afterwards. The run-workflow gates run +desktop pairs only on desktop-bearing tags; app-update pairs from +desktop installs stay declared TODOs. --- .github/workflows/install-e2e-run.yml | 14 ++-- .github/workflows/install-e2e-windows-run.yml | 13 ++-- scripts/sandbox/generate-e2e-matrix.mjs | 22 ++++-- tests/install/README.md | 6 ++ tests/install/installer-script-e2e.sh | 69 ++++++++++++++++--- .../install/windows-installer-script-e2e.ps1 | 44 ++++++++++-- 6 files changed, 143 insertions(+), 25 deletions(-) diff --git a/.github/workflows/install-e2e-run.yml b/.github/workflows/install-e2e-run.yml index e7501ec07f50..e61f1db39134 100644 --- a/.github/workflows/install-e2e-run.yml +++ b/.github/workflows/install-e2e-run.yml @@ -38,11 +38,11 @@ on: workflow_call: inputs: install-method: - description: 'How the starting version gets installed. Supported: installer-script (the real curl | install.sh one-liner).' + description: 'How the starting version gets installed. Supported: installer-script (the real curl | install.sh one-liner) and installer-script+desktop (the same one-liner with --include-desktop).' required: true type: string update-method: - description: 'How the install updates to HEAD. Supported: hermes-update (the updater) or installer-script (re-run the one-liner). Declared-but-TODO methods (open-app-update, hermes-desktop-app-update) skip.' + description: 'How the install updates to HEAD. Supported: hermes-update (the updater), installer-script (re-run the one-liner), installer-script+desktop (re-run with --include-desktop). Declared-but-TODO methods (open-app-update, hermes-desktop-app-update) skip.' required: true type: string install-ref: @@ -73,8 +73,13 @@ jobs: e2e: name: install & update # The pairs the driver can run today; anything else is a declared TODO - # and natively skips. - if: inputs.install-method == 'installer-script' && contains(fromJSON('["hermes-update", "installer-script"]'), inputs.update-method) + # and natively skips. The +desktop variants also need the starting tag + # to ship apps/desktop (the --include-desktop flag shipped with it). + if: >- + (inputs.install-method == 'installer-script' + || (inputs.install-method == 'installer-script+desktop' && inputs.tag-has-desktop)) + && (contains(fromJSON('["hermes-update", "installer-script"]'), inputs.update-method) + || (inputs.update-method == 'installer-script+desktop' && inputs.tag-has-desktop)) runs-on: ${{ inputs.runner }} timeout-minutes: ${{ inputs.timeout-minutes }} @@ -90,6 +95,7 @@ jobs: run: | set -euo pipefail tests/install/installer-script-e2e.sh \ + --install-method '${{ inputs.install-method }}' \ --update-method '${{ inputs.update-method }}' \ --install-ref '${{ inputs.install-ref }}' env: diff --git a/.github/workflows/install-e2e-windows-run.yml b/.github/workflows/install-e2e-windows-run.yml index 2975ad579d69..09303dc0a61a 100644 --- a/.github/workflows/install-e2e-windows-run.yml +++ b/.github/workflows/install-e2e-windows-run.yml @@ -58,7 +58,7 @@ on: workflow_call: inputs: install-method: - description: 'How OLD gets installed. Supported: desktop-installer@latest (website exe, AHK-clicked) and installer-script (irm | iex install.ps1). Declared-but-TODO methods skip.' + description: 'How OLD gets installed. Supported: desktop-installer@latest (website exe, AHK-clicked), installer-script (irm | iex install.ps1) and installer-script+desktop (the same with -IncludeDesktop). Declared-but-TODO methods skip.' required: true type: string update-method: @@ -190,8 +190,13 @@ jobs: # The pairs the script driver can run today. open-app-update and # hermes-desktop-app-update from a script install are declared TODOs # (the desktop app also has to be BUILT by - # that path first). - if: inputs.install-method == 'installer-script' && contains(fromJSON('["hermes-update", "installer-script"]'), inputs.update-method) + # that path first). The +desktop variants also need the starting tag + # to ship apps/desktop (the -IncludeDesktop parameter shipped with it). + if: >- + (inputs.install-method == 'installer-script' + || (inputs.install-method == 'installer-script+desktop' && inputs.tag-has-desktop)) + && (contains(fromJSON('["hermes-update", "installer-script"]'), inputs.update-method) + || (inputs.update-method == 'installer-script+desktop' && inputs.tag-has-desktop)) runs-on: windows-latest timeout-minutes: ${{ inputs.timeout-minutes }} @@ -205,7 +210,7 @@ jobs: - name: Run install + update E2E shell: powershell - run: powershell -NoProfile -ExecutionPolicy Bypass -File tests\install\windows-installer-script-e2e.ps1 -UpdateMethod "${{ inputs.update-method }}" -InstallRef "${{ inputs.install-ref }}" + run: powershell -NoProfile -ExecutionPolicy Bypass -File tests\install\windows-installer-script-e2e.ps1 -InstallMethod "${{ inputs.install-method }}" -UpdateMethod "${{ inputs.update-method }}" -InstallRef "${{ inputs.install-ref }}" env: HERMES_E2E_LOG_DIR: ${{ runner.temp }}\e2e-logs diff --git a/scripts/sandbox/generate-e2e-matrix.mjs b/scripts/sandbox/generate-e2e-matrix.mjs index 2011f531fb26..16cb8e176f12 100644 --- a/scripts/sandbox/generate-e2e-matrix.mjs +++ b/scripts/sandbox/generate-e2e-matrix.mjs @@ -37,10 +37,14 @@ import { fileURLToPath } from 'node:url'; * @typedef {'latest'} InstallerVersion * The artifact published on the website right now -- Hermes-Setup.exe has * no versioned archive yet. Widen this union when one exists. - * @typedef {'installer-script' | 'desktop-installer' | 'packaged-app'} InstallMethod + * @typedef {'installer-script' | 'installer-script+desktop' | 'desktop-installer' | 'packaged-app'} InstallMethod * installer-script is the platform's one-liner (curl | bash on - * linux/macos, irm | iex on windows); packaged-app is declared but not - * used by any OS spec yet. + * linux/macos, irm | iex on windows); installer-script+desktop is the + * same one-liner with its desktop stage opted in (--include-desktop / + * -IncludeDesktop), which also builds the desktop app -- on windows it + * registers Start Menu / Desktop shortcuts too, on linux/macos it + * builds into the checkout without registering an OS entry point; + * packaged-app is declared but not used by any OS spec yet. * @typedef {InstallMethod | 'hermes-update' | 'open-app-update' | 'hermes-desktop-app-update'} UpdateMethod * Every install method doubles as an update method (re-run it over the * existing install), plus the updater CLI and the two app-update @@ -78,11 +82,16 @@ export const SPEC = { install: [ // irm https://hermes.nousresearch.com/install.ps1 | iex { method: 'installer-script' }, + // The same one-liner with -IncludeDesktop: builds Hermes.exe AND + // registers Start Menu / Desktop shortcuts, so it is a second real + // path to a hand-launchable app. + { method: 'installer-script+desktop' }, // Website Hermes-Setup.exe, clicked through the GUI. { method: 'desktop-installer', versions: ['latest'] }, ], update: [ { method: 'installer-script' }, + { method: 'installer-script+desktop' }, // Run the bootstrap exe again over an existing install (--update flow). { method: 'desktop-installer', versions: ['latest'] }, { method: 'hermes-update' }, @@ -96,9 +105,11 @@ export const SPEC = { macos: { install: [ { method: 'installer-script' }, + { method: 'installer-script+desktop' }, ], update: [ { method: 'installer-script' }, + { method: 'installer-script+desktop' }, { method: 'hermes-update' }, // install.sh --include-desktop builds the .app inside the checkout // but registers no OS entry point, so open-app-update legs pair @@ -110,9 +121,11 @@ export const SPEC = { linux: { install: [ { method: 'installer-script' }, + { method: 'installer-script+desktop' }, ], update: [ { method: 'installer-script' }, + { method: 'installer-script+desktop' }, { method: 'hermes-update' }, // No desktop installer and no packaged desktop artifact exist for // linux, so there is no open-app-update; `hermes desktop` is always @@ -202,7 +215,8 @@ export function buildMatrices(envs, tags) { */ export function renderMarkdownPlan(envs, tags) { const needsDesktop = (/** @type {string} */ m) => - m.startsWith('desktop-installer') || m === 'open-app-update' || m === 'hermes-desktop-app-update'; + m.startsWith('desktop-installer') || m === 'installer-script+desktop' || + m === 'open-app-update' || m === 'hermes-desktop-app-update'; const lines = [ '### Install & Update E2E plan', '', diff --git a/tests/install/README.md b/tests/install/README.md index be5b75b95683..0e36ebb6d9b4 100644 --- a/tests/install/README.md +++ b/tests/install/README.md @@ -43,6 +43,12 @@ A leg can install a release from months back. The driver must not assume that th - For the installed CLI, ask the binary with `--help`. - If a flag is not found, omit the flag. This is not an error. +## The install methods + +- `installer-script`: the platform's one-liner (`curl | bash` on linux and macos, `irm | iex` on windows). +- `installer-script+desktop`: the same one-liner with its desktop stage opted in (`--include-desktop` / `-IncludeDesktop`). The stage builds the desktop app during the install. On windows it also registers Start Menu and Desktop shortcuts. On linux and macos it builds the app inside the checkout and registers no OS entry point. +- `desktop-installer@latest`: the published GUI installer (`Hermes-Setup.exe` on windows), clicked through the real window. + ## The two app-update variants The desktop app has two launch paths, so the matrix has two app-update methods. Both click "Update now" in the running app. They differ in how the app starts: diff --git a/tests/install/installer-script-e2e.sh b/tests/install/installer-script-e2e.sh index 9c18cdcc77c1..d2e1a0a009c2 100755 --- a/tests/install/installer-script-e2e.sh +++ b/tests/install/installer-script-e2e.sh @@ -25,11 +25,16 @@ # the checkout landed on HEAD with a working `hermes` # # Usage: -# tests/install/installer-script-e2e.sh --update-method hermes-update|installer-script +# tests/install/installer-script-e2e.sh --update-method hermes-update|installer-script|installer-script+desktop +# [--install-method installer-script|installer-script+desktop] # [--install-ref REF] # +# --install-method installer-script the plain one-liner (default) +# installer-script+desktop the one-liner with its desktop +# stage opted in (--include-desktop) # --update-method hermes-update `hermes update` # installer-script re-run install.sh (HEAD's copy) +# installer-script+desktop re-run with --include-desktop # --install-ref what to install first; anything git resolves. Default: # the newest release tag in the checkout. # @@ -37,23 +42,31 @@ set -euo pipefail +INSTALL_METHOD="installer-script" UPDATE_METHOD="" INSTALL_REF="" while [ "$#" -gt 0 ]; do case "$1" in + --install-method) + [ "$#" -ge 2 ] || { echo 'error: --install-method needs a value' >&2; exit 1; } + INSTALL_METHOD="$2"; shift 2 ;; --update-method) [ "$#" -ge 2 ] || { echo 'error: --update-method needs a value' >&2; exit 1; } UPDATE_METHOD="$2"; shift 2 ;; --install-ref) [ "$#" -ge 2 ] || { echo 'error: --install-ref needs a value' >&2; exit 1; } INSTALL_REF="$2"; shift 2 ;; - -h|--help) sed -n '2,37p' "$0"; exit 0 ;; + -h|--help) sed -n '2,45p' "$0"; exit 0 ;; *) echo "error: unknown argument: $1" >&2; exit 1 ;; esac done +case "$INSTALL_METHOD" in + installer-script|installer-script+desktop) ;; + *) echo "error: --install-method must be installer-script or installer-script+desktop, got '$INSTALL_METHOD'" >&2; exit 1 ;; +esac case "$UPDATE_METHOD" in - hermes-update|installer-script) ;; - *) echo "error: --update-method must be hermes-update or installer-script, got '$UPDATE_METHOD'" >&2; exit 1 ;; + hermes-update|installer-script|installer-script+desktop) ;; + *) echo "error: --update-method must be hermes-update, installer-script or installer-script+desktop, got '$UPDATE_METHOD'" >&2; exit 1 ;; esac REPO_ROOT="$(cd "$(dirname "$0")/../.." && pwd)" @@ -144,7 +157,8 @@ installer_supports() { } run_installer() { - # $1: ref whose scripts/install.sh to run; $2: log name + # $1: ref whose scripts/install.sh to run; $2: log name; $3: "desktop" to + # opt the desktop stage in (--include-desktop) local script="$WORK_ROOT/install-$2.sh" git -C "$REPO_ROOT" show "$1:scripts/install.sh" > "$script" chmod +x "$script" @@ -155,6 +169,15 @@ run_installer() { if installer_supports "$1" "--skip-browser"; then flags+=(--skip-browser) fi + if [ "${3:-}" = "desktop" ]; then + # The desktop stage is the point of this leg, so a ref without the + # flag is a hard failure, not a silent downgrade to a plain install. + # (Releases that predate apps/desktop are already skipped upstream by + # the tag-has-desktop gate; the flag shipped with the app.) + installer_supports "$1" "--include-desktop" \ + || fail "ref $1 does not support --include-desktop; this leg cannot mean what it claims" + flags+=(--include-desktop) + fi # HEAD ---------------------------------------------------------- @@ -237,6 +286,10 @@ case "$UPDATE_METHOD" in # A user re-running the one-liner today gets the CURRENT script. run_installer "$HEAD_SHA" head ;; + installer-script+desktop) + run_installer "$HEAD_SHA" head desktop + assert_desktop_artifact HEAD + ;; esac assert_checkout "$HEAD_SHA" HEAD smoke_desktop head diff --git a/tests/install/windows-installer-script-e2e.ps1 b/tests/install/windows-installer-script-e2e.ps1 index 34157ccd81c8..53fdf4cca892 100644 --- a/tests/install/windows-installer-script-e2e.ps1 +++ b/tests/install/windows-installer-script-e2e.ps1 @@ -31,7 +31,9 @@ #Requires -Version 5.1 param( - [ValidateSet("hermes-update", "installer-script")] + [ValidateSet("installer-script", "installer-script+desktop")] + [string]$InstallMethod = "installer-script", + [ValidateSet("hermes-update", "installer-script", "installer-script+desktop")] [string]$UpdateMethod = "hermes-update", [string]$InstallRef = "auto" ) @@ -123,7 +125,7 @@ $env:HERMES_HOME = $HermesHome Set-Content -LiteralPath (Join-Path $HermesHome ".skip_upstream_prompt") -Value "" -Encoding Ascii function Invoke-Installer { - param([string]$Ref, [string]$Label) + param([string]$Ref, [string]$Label, [switch]$IncludeDesktop) $script = Join-Path $WorkRoot "install-$Label.ps1" (Invoke-Git @("-C", $RepoRoot, "show", "$Ref`:scripts/install.ps1")) -join "`n" | Set-Content -LiteralPath $script -Encoding UTF8 @@ -134,6 +136,16 @@ function Invoke-Installer { $flags = @("-SkipSetup", "-HermesHome", $HermesHome, "-InstallDir", $InstallDir) $text = Get-Content -LiteralPath $script -Raw if ($text -match '\$NonInteractive') { $flags += "-NonInteractive" } + if ($IncludeDesktop) { + # The desktop stage is the point of this leg, so a ref without the + # parameter is a hard failure, not a silent downgrade to a plain + # install. (Pre-desktop releases are already skipped upstream by + # the tag-has-desktop gate; the parameter shipped with the app.) + if ($text -notmatch '\$IncludeDesktop') { + Fail "ref $Ref does not support -IncludeDesktop; this leg cannot mean what it claims" + } + $flags += "-IncludeDesktop" + } $log = Join-Path $LogDir "install-$Label.log" # Native stderr (git clone progress, pip notices) must not become # terminating NativeCommandErrors under EAP=Stop; the exit code is the @@ -167,6 +179,18 @@ function Assert-Checkout { Ok "hermes --version works: $((Get-Content -LiteralPath $verLog -First 1))" } +function Assert-DesktopArtifact { + param([string]$Label) + # After a +desktop install the built app must exist under the checkout; + # the installer also registers Start Menu / Desktop shortcuts, but the + # artifact is the ground truth a headless job can check. + $exe = Join-Path $InstallDir "apps\desktop\release\win-unpacked\Hermes.exe" + if (-not (Test-Path -LiteralPath $exe)) { + Fail "no desktop app at $exe after $Label (+desktop install)" + } + Ok "desktop app built by installer at ${Label}: $exe" +} + function Test-DesktopSmoke { param([string]$Label) # Prove the installed CLI can produce the desktop app: `hermes desktop @@ -209,9 +233,15 @@ function Test-DesktopSmoke { # --- install OLD ------------------------------------------------------------------ -Step "installing OLD ($InstallRef) via its own scripts/install.ps1" -Invoke-Installer $OldSha "old" -Assert-Checkout $OldSha "OLD" +Step "installing OLD ($InstallRef) via its own scripts/install.ps1 ($InstallMethod)" +if ($InstallMethod -eq "installer-script+desktop") { + Invoke-Installer $OldSha "old" -IncludeDesktop + Assert-Checkout $OldSha "OLD" + Assert-DesktopArtifact "OLD" +} else { + Invoke-Installer $OldSha "old" + Assert-Checkout $OldSha "OLD" +} Test-DesktopSmoke "old" # --- update OLD -> HEAD -------------------------------------------------------------- @@ -248,6 +278,10 @@ switch ($UpdateMethod) { # A user re-running the one-liner today gets the CURRENT script. Invoke-Installer $HeadSha "head" } + "installer-script+desktop" { + Invoke-Installer $HeadSha "head" -IncludeDesktop + Assert-DesktopArtifact "HEAD" + } } Assert-Checkout $HeadSha "HEAD" Test-DesktopSmoke "head" From cdaf0cf0910baf1daaf46b6f72223c7f471ce15e Mon Sep 17 00:00:00 2001 From: ethernet Date: Wed, 12 Aug 2026 04:15:24 -0400 Subject: [PATCH 069/227] ci(install-e2e): one screen-recording mechanism on every runner, Xvfb for headless linux The composite action .github/actions/e2e-screen-record owns setup and lifecycle on all three OSes: ffmpeg via apt/brew-verify/winget+cache, capture via x11grab/gdigrab/avfoundation, mkv at 15fps stopped by 'q' on live stdin with kill fallback. Linux runners have no display, so start brings up a dedicated Xvfb :99 and exports DISPLAY - one display serves both the recorder and any app a later step launches. Recording moves out of the GUI driver into workflow infrastructure - that is what makes it uniform - and a missing ffmpeg or a zero-frame file now FAILS the leg instead of skipping silently: the graceful-skip path is how the windows leg shipped no recording.mkv while green. Lifecycle proven locally: start against lavfi testsrc, q-stop, ffprobe duration check (record-start.sh/record-stop.sh under nix ffmpeg). --- .github/actions/e2e-screen-record/action.yml | 102 ++++++++++++++++++ .github/workflows/install-e2e-run.yml | 15 +++ .github/workflows/install-e2e-windows-run.yml | 52 ++++----- tests/install/README.md | 2 +- tests/install/e2e-assets/record-start.ps1 | 62 +++++++++++ tests/install/e2e-assets/record-start.sh | 71 ++++++++++++ tests/install/e2e-assets/record-stop.ps1 | 53 +++++++++ tests/install/e2e-assets/record-stop.sh | 49 +++++++++ tests/install/windows-desktop-gui-e2e.ps1 | 39 ------- 9 files changed, 381 insertions(+), 64 deletions(-) create mode 100644 .github/actions/e2e-screen-record/action.yml create mode 100644 tests/install/e2e-assets/record-start.ps1 create mode 100755 tests/install/e2e-assets/record-start.sh create mode 100644 tests/install/e2e-assets/record-stop.ps1 create mode 100755 tests/install/e2e-assets/record-stop.sh diff --git a/.github/actions/e2e-screen-record/action.yml b/.github/actions/e2e-screen-record/action.yml new file mode 100644 index 000000000000..d4923d27afdc --- /dev/null +++ b/.github/actions/e2e-screen-record/action.yml @@ -0,0 +1,102 @@ +name: E2E screen recording +description: > + One screen-recording mechanism for every install-e2e runner. mode=start + installs what the OS needs (ffmpeg everywhere; Xvfb on headless linux), + starts the recorder, and exports DISPLAY for later steps. mode=stop + finalizes the recording and fails on a zero-frame file. A missing ffmpeg + is a hard error - a graceful skip makes the missing tool invisible and + the artifact silently loses its recording. + +inputs: + mode: + description: start | stop + required: true + output: + description: Path of the recording (mkv). + required: true + +runs: + using: composite + steps: + # ---- setup + start ------------------------------------------------------ + - name: Install ffmpeg + Xvfb (linux) + if: inputs.mode == 'start' && runner.os == 'Linux' + shell: bash + run: | + set -euo pipefail + if ! command -v ffmpeg >/dev/null 2>&1 || ! command -v Xvfb >/dev/null 2>&1; then + sudo apt-get update -qq + sudo apt-get install -y -qq --no-install-recommends ffmpeg xvfb + fi + + - name: Start Xvfb (linux, headless) + if: inputs.mode == 'start' && runner.os == 'Linux' + shell: bash + run: | + set -euo pipefail + # A dedicated display rather than xvfb-run-wrapping each command, so + # ONE display serves both the app under test and the recorder. + if [ -z "${DISPLAY:-}" ]; then + Xvfb :99 -screen 0 1920x1080x24 & + echo "$!" > "$RUNNER_TEMP/xvfb.pid" + echo "DISPLAY=:99" >> "$GITHUB_ENV" + export DISPLAY=:99 + fi + # Wait until the display accepts connections; xdpyinfo may not be + # installed, so probe with the X socket. + for _ in $(seq 1 50); do + [ -S "/tmp/.X11-unix/X99" ] && break + sleep 0.2 + done + [ -S "/tmp/.X11-unix/X99" ] || { echo "Xvfb :99 did not come up" >&2; exit 1; } + + - name: Verify ffmpeg (macos) + if: inputs.mode == 'start' && runner.os == 'macOS' + shell: bash + run: | + set -euo pipefail + # Do not trust "preinstalled" claims - verify, install on miss. + command -v ffmpeg >/dev/null 2>&1 || brew install --quiet ffmpeg + + - name: Restore cached ffmpeg (windows) + if: inputs.mode == 'start' && runner.os == 'Windows' + id: ffmpeg-cache + uses: actions/cache@27d5ce7f107fe9357f9df03efb73ab90386fccae # v5 + with: + path: ${{ runner.temp }}\test-bins\ffmpeg + key: e2e-ffmpeg-${{ runner.os }}-v1 + + - name: Install ffmpeg (windows) + if: inputs.mode == 'start' && runner.os == 'Windows' && steps.ffmpeg-cache.outputs.cache-hit != 'true' + shell: pwsh + run: | + $bins = "$env:RUNNER_TEMP\test-bins\ffmpeg" + New-Item -ItemType Directory -Path $bins -Force | Out-Null + winget install -e --id Gyan.FFmpeg --silent --accept-source-agreements --accept-package-agreements --disable-interactivity --location "$env:RUNNER_TEMP\ffmpeg_dir" + Copy-Item -Path "$env:RUNNER_TEMP\ffmpeg_dir\*\*" -Destination $bins -Recurse -Force + + - name: Add ffmpeg to PATH (windows) + if: inputs.mode == 'start' && runner.os == 'Windows' + shell: pwsh + run: Add-Content -Path $env:GITHUB_PATH -Value "$env:RUNNER_TEMP\test-bins\ffmpeg\bin" + + - name: Start recording (posix) + if: inputs.mode == 'start' && runner.os != 'Windows' + shell: bash + run: bash "$GITHUB_ACTION_PATH/../../../tests/install/e2e-assets/record-start.sh" '${{ inputs.output }}' + + - name: Start recording (windows) + if: inputs.mode == 'start' && runner.os == 'Windows' + shell: powershell + run: powershell -NoProfile -ExecutionPolicy Bypass -File "$env:GITHUB_ACTION_PATH\..\..\..\tests\install\e2e-assets\record-start.ps1" -OutFile "${{ inputs.output }}" + + # ---- stop --------------------------------------------------------------- + - name: Stop recording (posix) + if: inputs.mode == 'stop' && runner.os != 'Windows' + shell: bash + run: bash "$GITHUB_ACTION_PATH/../../../tests/install/e2e-assets/record-stop.sh" '${{ inputs.output }}' + + - name: Stop recording (windows) + if: inputs.mode == 'stop' && runner.os == 'Windows' + shell: powershell + run: powershell -NoProfile -ExecutionPolicy Bypass -File "$env:GITHUB_ACTION_PATH\..\..\..\tests\install\e2e-assets\record-stop.ps1" -OutFile "${{ inputs.output }}" diff --git a/.github/workflows/install-e2e-run.yml b/.github/workflows/install-e2e-run.yml index e61f1db39134..36553831c073 100644 --- a/.github/workflows/install-e2e-run.yml +++ b/.github/workflows/install-e2e-run.yml @@ -91,6 +91,14 @@ jobs: with: fetch-depth: 0 + # One recording mechanism on every OS (Xvfb gives headless linux a + # display; the same display serves any app the driver launches). + - name: Start screen recording + uses: ./.github/actions/e2e-screen-record + with: + mode: start + output: ${{ runner.temp }}/e2e-logs/recording.mkv + - name: Run install + update E2E run: | set -euo pipefail @@ -103,6 +111,13 @@ jobs: # would trip the driver's own dirty-tree guard. HERMES_E2E_LOG_DIR: ${{ runner.temp }}/e2e-logs + - name: Stop screen recording + if: always() + uses: ./.github/actions/e2e-screen-record + with: + mode: stop + output: ${{ runner.temp }}/e2e-logs/recording.mkv + # Artifact names cannot contain '/', and install-ref may be a full ref # like refs/heads/main. GitHub Actions expressions have no string-replace # function, so build the safe name here. Runs even on failure -- that is diff --git a/.github/workflows/install-e2e-windows-run.yml b/.github/workflows/install-e2e-windows-run.yml index 09303dc0a61a..f1e99531044f 100644 --- a/.github/workflows/install-e2e-windows-run.yml +++ b/.github/workflows/install-e2e-windows-run.yml @@ -119,31 +119,15 @@ jobs: with: fetch-depth: 0 - # ffmpeg records the screen for the whole run so a failed click is - # diagnosable from the artifact instead of by guesswork. NOT - # preinstalled on windows-latest; the driver skips recording - # gracefully when it's absent, so this step is what makes the - # recording actually happen. Cached — winget's ffmpeg download is - # the slow part. - - name: Restore cached ffmpeg - id: ffmpeg-cache - uses: actions/cache@27d5ce7f107fe9357f9df03efb73ab90386fccae # v5 + # One recording mechanism on every OS: the composite action installs + # ffmpeg (cached - winget's download is the slow part), starts the + # capture, and record-stop fails on a zero-frame file so a silently + # missing recording cannot go green. + - name: Start screen recording + uses: ./.github/actions/e2e-screen-record with: - path: ${{ runner.temp }}\test-bins\ffmpeg - key: e2e-ffmpeg-${{ runner.os }}-v1 - - - name: Install ffmpeg - if: steps.ffmpeg-cache.outputs.cache-hit != 'true' - shell: pwsh - run: | - $bins = "$env:RUNNER_TEMP\test-bins\ffmpeg" - New-Item -ItemType Directory -Path $bins -Force | Out-Null - winget install -e --id Gyan.FFmpeg --silent --accept-source-agreements --accept-package-agreements --disable-interactivity --location "$env:RUNNER_TEMP\ffmpeg_dir" - Copy-Item -Path "$env:RUNNER_TEMP\ffmpeg_dir\*\*" -Destination $bins -Recurse -Force - - - name: Add ffmpeg to PATH - shell: pwsh - run: Add-Content -Path $env:GITHUB_PATH -Value "$env:RUNNER_TEMP\test-bins\ffmpeg\bin" + mode: start + output: ${{ github.workspace }}\gui-e2e-proof\recording.mkv - name: Stage serve repo (main -> ${{ inputs.install-ref }}) shell: powershell @@ -157,6 +141,13 @@ jobs: shell: powershell run: powershell -NoProfile -ExecutionPolicy Bypass -File tests\install\windows-desktop-gui-e2e.ps1 -Phase update-gui -Route "${{ inputs.update-method }}" -InstallRef "${{ inputs.install-ref }}" -SetupExeUrl ${{ inputs.setup-exe-url }} + - name: Stop screen recording + if: always() + uses: ./.github/actions/e2e-screen-record + with: + mode: stop + output: ${{ github.workspace }}\gui-e2e-proof\recording.mkv + - name: Collect proof + logs if: always() shell: powershell @@ -208,12 +199,25 @@ jobs: with: fetch-depth: 0 + - name: Start screen recording + uses: ./.github/actions/e2e-screen-record + with: + mode: start + output: ${{ runner.temp }}\e2e-logs\recording.mkv + - name: Run install + update E2E shell: powershell run: powershell -NoProfile -ExecutionPolicy Bypass -File tests\install\windows-installer-script-e2e.ps1 -InstallMethod "${{ inputs.install-method }}" -UpdateMethod "${{ inputs.update-method }}" -InstallRef "${{ inputs.install-ref }}" env: HERMES_E2E_LOG_DIR: ${{ runner.temp }}\e2e-logs + - name: Stop screen recording + if: always() + uses: ./.github/actions/e2e-screen-record + with: + mode: stop + output: ${{ runner.temp }}\e2e-logs\recording.mkv + - name: Upload installer logs if: always() uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1 diff --git a/tests/install/README.md b/tests/install/README.md index 0e36ebb6d9b4..e9de9fb35bc0 100644 --- a/tests/install/README.md +++ b/tests/install/README.md @@ -79,4 +79,4 @@ gh workflow run install-e2e.yml --ref -f route=both -f tag-count=2 ## Artifacts -Each leg uploads its logs as an artifact. The windows GUI leg also uploads screenshots, a screen recording, and the update result file. Get them with `gh run download `. +Each leg uploads its logs as an artifact. Every leg also records the screen for its whole run: the composite action `.github/actions/e2e-screen-record` installs ffmpeg, records with the OS's capture backend (x11grab on linux, gdigrab on windows, avfoundation on macos), and fails the leg if the recording is missing or has zero frames. Linux runners have no display, so the action starts `Xvfb :99` first and exports `DISPLAY` for every later step — the app under test and the recorder share that display. The windows GUI leg also uploads screenshots and the update result file. Get them with `gh run download `. diff --git a/tests/install/e2e-assets/record-start.ps1 b/tests/install/e2e-assets/record-start.ps1 new file mode 100644 index 000000000000..a49e9ac4a505 --- /dev/null +++ b/tests/install/e2e-assets/record-start.ps1 @@ -0,0 +1,62 @@ +# Start a continuous ffmpeg screen recording in the background (windows). +# +# Usage: powershell -File record-start.ps1 -OutFile recording.mkv +# +# The graceful stop is the character 'q' on ffmpeg's LIVE stdin, which only +# System.Diagnostics.Process exposes (Start-Process -RedirectStandardInput +# hands ffmpeg a file handle already at EOF). This script therefore spawns a +# detached HOLDER powershell that owns the ffmpeg process and its stdin pipe, +# and stops it when a STOP marker file appears; record-stop.ps1 writes the +# marker. mkv on purpose: it stays playable even unfinalized. A missing +# ffmpeg is a HARD error - a graceful skip makes the missing tool invisible +# and the artifact silently loses its recording. + +#Requires -Version 5.1 +param( + [Parameter(Mandatory = $true)][string]$OutFile +) +$ErrorActionPreference = "Stop" + +if (-not (Get-Command ffmpeg -ErrorAction SilentlyContinue)) { + Write-Host "record-start: ffmpeg not on PATH (the workflow must install it)" + exit 1 +} + +$outDir = Split-Path -Parent $OutFile +if ($outDir -and -not (Test-Path -LiteralPath $outDir)) { + New-Item -ItemType Directory -Path $outDir -Force | Out-Null +} +$stopMarker = "$OutFile.stop" +$stateFile = "$OutFile.state" +Remove-Item -LiteralPath $stopMarker, $stateFile -Force -ErrorAction SilentlyContinue + +$holder = "$OutFile.holder.ps1" +@' +param([string]$OutFile, [string]$StopMarker) +$psi = New-Object System.Diagnostics.ProcessStartInfo +$psi.FileName = "ffmpeg" +$psi.Arguments = "-y -f gdigrab -framerate 15 -i desktop " + + "-hide_banner -loglevel error " + + "-c:v libx264 -preset ultrafast -pix_fmt yuv420p `"$OutFile`"" +$psi.RedirectStandardInput = $true +$psi.UseShellExecute = $false +$proc = [System.Diagnostics.Process]::Start($psi) +Set-Content -LiteralPath "$OutFile.ffpid" -Value $proc.Id +while (-not $proc.HasExited) { + if (Test-Path -LiteralPath $StopMarker) { + try { + $proc.StandardInput.Write("q") + $proc.StandardInput.Close() + } catch {} + if (-not $proc.WaitForExit(15000)) { try { $proc.Kill() } catch {} } + break + } + Start-Sleep -Milliseconds 500 +} +'@ | Set-Content -LiteralPath $holder -Encoding UTF8 + +$holderProc = Start-Process -FilePath "powershell.exe" ` + -ArgumentList "-NoProfile", "-ExecutionPolicy", "Bypass", "-File", $holder, "-OutFile", $OutFile, "-StopMarker", $stopMarker ` + -WindowStyle Hidden -PassThru +Set-Content -LiteralPath $stateFile -Value "$($holderProc.Id) $stopMarker" +Write-Host "record-start: recording to $OutFile (holder pid $($holderProc.Id))" diff --git a/tests/install/e2e-assets/record-start.sh b/tests/install/e2e-assets/record-start.sh new file mode 100755 index 000000000000..6d6b8f97c409 --- /dev/null +++ b/tests/install/e2e-assets/record-start.sh @@ -0,0 +1,71 @@ +#!/usr/bin/env bash +# Start a continuous ffmpeg screen recording in the background. +# +# Usage: record-start.sh OUTPUT.mkv [INPUT_ARGS...] +# OUTPUT.mkv where to record (mkv: stays playable even unfinalized) +# INPUT_ARGS optional ffmpeg input override; default picks per-OS: +# linux -f x11grab -i "$DISPLAY" (Xvfb or real) +# macos -f avfoundation -i +# +# Writes state (pid + control fifo path) next to OUTPUT as OUTPUT.state so +# record-stop.sh can finalize gracefully: ffmpeg stops cleanly on the +# character 'q' on live stdin, so stdin is a fifo we hold open. A missing +# ffmpeg is a HARD error - a graceful skip makes the missing tool invisible +# and the artifact silently loses its recording. + +set -euo pipefail + +OUT="${1:?usage: record-start.sh OUTPUT.mkv [input args...]}" +shift || true + +command -v ffmpeg >/dev/null 2>&1 || { + echo "record-start: ffmpeg not on PATH (the workflow must install it)" >&2 + exit 1 +} + +INPUT=("$@") +if [ "${#INPUT[@]}" -eq 0 ]; then + case "$(uname -s)" in + Linux) + : "${DISPLAY:?record-start: DISPLAY not set (start Xvfb first on headless runners)}" + INPUT=(-f x11grab -framerate 15 -i "$DISPLAY") + ;; + Darwin) + # avfoundation lists devices on stderr; the first "Capture screen" + # index is the whole display. Parse it rather than hardcoding: the + # index shifts with attached cameras. + screen_idx="$(ffmpeg -f avfoundation -list_devices true -i "" 2>&1 \ + | sed -n 's/^\[AVFoundation[^]]*\] \[\([0-9]*\)\] Capture screen.*/\1/p' | head -1)" + [ -n "$screen_idx" ] || { echo "record-start: no capture screen device found" >&2; exit 1; } + INPUT=(-f avfoundation -framerate 15 -capture_cursor 1 -i "${screen_idx}:none") + ;; + *) + echo "record-start: unsupported OS $(uname -s) (windows uses record-start.ps1)" >&2 + exit 1 + ;; + esac +fi + +mkdir -p "$(dirname "$OUT")" +FIFO="$OUT.ctl" +STATE="$OUT.state" +rm -f "$FIFO" "$STATE" +mkfifo "$FIFO" + +# Hold the fifo's write end open in a shepherd process; ffmpeg reads its +# stdin from the fifo. record-stop.sh writes 'q' into the fifo. +ffmpeg -hide_banner -loglevel error "${INPUT[@]}" \ + -pix_fmt yuv420p -c:v libx264 -preset ultrafast "$OUT" < "$FIFO" & +FFMPEG_PID=$! +# Open a persistent write fd so the fifo doesn't EOF before stop. +exec 9> "$FIFO" +# Hand the fd to a shepherd that outlives this script. +( + exec 9>&9 + while kill -0 "$FFMPEG_PID" 2>/dev/null; do sleep 1; done +) & +SHEPHERD_PID=$! +disown "$SHEPHERD_PID" 2>/dev/null || true + +printf '%s %s %s\n' "$FFMPEG_PID" "$FIFO" "$SHEPHERD_PID" > "$STATE" +echo "record-start: recording to $OUT (ffmpeg pid $FFMPEG_PID)" diff --git a/tests/install/e2e-assets/record-stop.ps1 b/tests/install/e2e-assets/record-stop.ps1 new file mode 100644 index 000000000000..594b3a417bf1 --- /dev/null +++ b/tests/install/e2e-assets/record-stop.ps1 @@ -0,0 +1,53 @@ +# Stop a recording started by record-start.ps1 and verify the file is real. +# +# Usage: powershell -File record-stop.ps1 -OutFile recording.mkv +# +# Drops the STOP marker the holder watches for (it writes 'q' to ffmpeg's +# live stdin), waits for the holder to exit, then asserts the output exists +# and has a decodable duration - a zero-frame recording is the classic +# silent failure. + +#Requires -Version 5.1 +param( + [Parameter(Mandatory = $true)][string]$OutFile +) +$ErrorActionPreference = "Stop" + +$stateFile = "$OutFile.state" +if (-not (Test-Path -LiteralPath $stateFile)) { + Write-Host "record-stop: no state at $stateFile (was record-start run?)" + exit 1 +} +$state = (Get-Content -LiteralPath $stateFile -Raw).Trim() -split " ", 2 +$holderPid = [int]$state[0] +$stopMarker = $state[1] + +Set-Content -LiteralPath $stopMarker -Value "stop" +try { + $holder = Get-Process -Id $holderPid -ErrorAction SilentlyContinue + if ($holder) { $holder.WaitForExit(20000) | Out-Null } +} catch {} +# Belt and braces: if ffmpeg outlived the holder, kill it directly. +$ffpidFile = "$OutFile.ffpid" +if (Test-Path -LiteralPath $ffpidFile) { + $ffpid = [int](Get-Content -LiteralPath $ffpidFile -Raw).Trim() + try { Stop-Process -Id $ffpid -Force -ErrorAction SilentlyContinue } catch {} +} +Remove-Item -LiteralPath $stateFile, $stopMarker, $ffpidFile, "$OutFile.holder.ps1" -Force -ErrorAction SilentlyContinue + +if (-not (Test-Path -LiteralPath $OutFile) -or (Get-Item -LiteralPath $OutFile).Length -eq 0) { + Write-Host "record-stop: $OutFile missing or empty" + exit 1 +} +if (Get-Command ffprobe -ErrorAction SilentlyContinue) { + $prevEap = $ErrorActionPreference; $ErrorActionPreference = "Continue" + $dur = (& ffprobe -v error -show_entries format=duration -of csv=p=0 $OutFile 2>&1 | Out-String).Trim() + $ErrorActionPreference = $prevEap + if (-not $dur -or $dur -eq "0" -or $dur -like "0.0*") { + Write-Host "record-stop: $OutFile has no duration (zero-frame recording)" + exit 1 + } + Write-Host "record-stop: $OutFile finalized (${dur}s)" +} else { + Write-Host "record-stop: $OutFile finalized (ffprobe absent; size $((Get-Item -LiteralPath $OutFile).Length) bytes)" +} diff --git a/tests/install/e2e-assets/record-stop.sh b/tests/install/e2e-assets/record-stop.sh new file mode 100755 index 000000000000..2d06eae8dfd7 --- /dev/null +++ b/tests/install/e2e-assets/record-stop.sh @@ -0,0 +1,49 @@ +#!/usr/bin/env bash +# Stop a recording started by record-start.sh and verify the file is real. +# +# Usage: record-stop.sh OUTPUT.mkv +# +# Graceful stop: the character 'q' on ffmpeg's live stdin (the control +# fifo). Falls back to SIGINT (also a clean finalize for ffmpeg), then +# SIGKILL. Fails if the output is missing or has no decodable duration - +# a zero-frame recording is the classic silent failure. + +set -euo pipefail + +OUT="${1:?usage: record-stop.sh OUTPUT.mkv}" +STATE="$OUT.state" + +[ -f "$STATE" ] || { echo "record-stop: no state at $STATE (was record-start run?)" >&2; exit 1; } +read -r FFMPEG_PID FIFO _SHEPHERD < "$STATE" + +if kill -0 "$FFMPEG_PID" 2>/dev/null; then + # Write the quit key; don't hang if the reader is already gone. + { printf 'q' > "$FIFO"; } 2>/dev/null & + WRITER=$! + for _ in $(seq 1 50); do + kill -0 "$FFMPEG_PID" 2>/dev/null || break + sleep 0.2 + done + kill "$WRITER" 2>/dev/null || true + if kill -0 "$FFMPEG_PID" 2>/dev/null; then + echo "record-stop: q did not stop ffmpeg; SIGINT" >&2 + kill -INT "$FFMPEG_PID" 2>/dev/null || true + for _ in $(seq 1 25); do + kill -0 "$FFMPEG_PID" 2>/dev/null || break + sleep 0.2 + done + kill -9 "$FFMPEG_PID" 2>/dev/null || true + fi +fi +rm -f "$FIFO" "$STATE" + +[ -s "$OUT" ] || { echo "record-stop: $OUT missing or empty" >&2; exit 1; } +if command -v ffprobe >/dev/null 2>&1; then + dur="$(ffprobe -v error -show_entries format=duration -of csv=p=0 "$OUT" || echo 0)" + case "$dur" in + ''|0|0.*) echo "record-stop: $OUT has no duration (zero-frame recording)" >&2; exit 1 ;; + esac + echo "record-stop: $OUT finalized (${dur}s)" +else + echo "record-stop: $OUT finalized (ffprobe absent; size $(wc -c < "$OUT") bytes)" +fi diff --git a/tests/install/windows-desktop-gui-e2e.ps1 b/tests/install/windows-desktop-gui-e2e.ps1 index 5bc5404b44b3..a77feeba09d7 100644 --- a/tests/install/windows-desktop-gui-e2e.ps1 +++ b/tests/install/windows-desktop-gui-e2e.ps1 @@ -225,41 +225,6 @@ function Save-DesktopScreenshot([string]$OutFile) { } } -function Start-ScreenRecording([string]$OutFile) { - # Continuous ffmpeg screen capture (gdigrab, 15fps). ffmpeg ships on the - # windows-latest runner image; skip gracefully elsewhere. mkv on purpose: - # it stays playable even if the process dies without finalizing. - # - # ffmpeg must be started, fed, and stopped from THIS process: the - # graceful stop is the character 'q' on its LIVE stdin pipe, which only - # System.Diagnostics.Process exposes (Start-Process - # -RedirectStandardInput hands it a file handle already at EOF). - if (-not (Get-Command ffmpeg -ErrorAction SilentlyContinue)) { - Write-Host " (ffmpeg not on PATH; skipping screen recording)" - return $null - } - $psi = New-Object System.Diagnostics.ProcessStartInfo - $psi.FileName = "ffmpeg" - $psi.Arguments = "-y -f gdigrab -framerate 15 -i desktop " + - "-hide_banner -loglevel error " + - "-c:v libx264 -preset ultrafast -pix_fmt yuv420p `"$OutFile`"" - $psi.RedirectStandardInput = $true - $psi.UseShellExecute = $false - $proc = [System.Diagnostics.Process]::Start($psi) - Write-Host " screen recording started (pid $($proc.Id)) -> $OutFile" - return $proc -} - -function Stop-ScreenRecording($proc) { - if ($proc -and -not $proc.HasExited) { - try { - $proc.StandardInput.Write("q") - $proc.StandardInput.Close() - } catch {} - if (-not $proc.WaitForExit(15000)) { try { $proc.Kill() } catch {} } - } -} - function Start-DesktopRecorder([string]$OutDir) { # Rolling desktop capture: one PNG every 3s from a detached PowerShell, # capped at 800 frames (~40 min). Proof that survives any step failure. @@ -419,7 +384,6 @@ function Invoke-PhaseInstallGui { New-Item -ItemType Directory -Path $HermesHome -Force | Out-Null $recorder = Start-DesktopRecorder (Join-Path $proof "desktop-frames") - $recording = Start-ScreenRecording (Join-Path $proof "recording.mkv") $ahkLog = Join-Path $proof "ahk.log" try { Save-DesktopScreenshot (Join-Path $proof "00-before-installer.png") @@ -458,7 +422,6 @@ function Invoke-PhaseInstallGui { Assert-True $installer.HasExited "Hermes-Setup.exe exited after Launch" } finally { - Stop-ScreenRecording $recording Stop-DesktopRecorder $recorder (Join-Path $proof "desktop-frames") # Surface the installer's own log win or lose, full and folded. $bootLog = Join-Path $HermesHome "logs\bootstrap-installer.log" @@ -549,7 +512,6 @@ function Invoke-GuiUpdateDesktopRoute([string]$TargetSha) { Assert-True ($npmExit -eq 0) "npm install @playwright/test@$PlaywrightVersion into the driver dir" $recorder = Start-DesktopRecorder (Join-Path $proof "desktop-frames") - $recording = Start-ScreenRecording (Join-Path $proof "recording.mkv") try { # Launch the installed app and click through Settings -> About -> # Update now. Exit 0 = the app quit for the updater hand-off. @@ -658,7 +620,6 @@ function Invoke-GuiUpdateDesktopRoute([string]$TargetSha) { Save-DesktopScreenshot (Join-Path $proof "99-relaunched-desktop.png") } finally { - Stop-ScreenRecording $recording Stop-DesktopRecorder $recorder (Join-Path $proof "desktop-frames") $handoffLog = Join-Path $HermesHome "logs\desktop-update-handoff.log" if (Test-Path -LiteralPath $handoffLog) { From 0f903e14a32d039c6e9ff67f1b281920ee6d1367 Mon Sep 17 00:00:00 2001 From: ethernet Date: Wed, 12 Aug 2026 04:20:56 -0400 Subject: [PATCH 070/227] test(install-e2e): hermes-desktop-app-update goes live on the script driver Playwright must own the spawn (it needs the inspection pipe), but hermes desktop is not just build+launch - stamp checks, integrity gates, sandbox fixups, and a constructed child environment. So the driver intercepts the product's own launch: a sitecustomize.py on PYTHONPATH (opt-in via HERMES_E2E_CAPTURE_LAUNCH) wraps subprocess.run, captures argv/cwd/env at the spawn site, and fakes success instead of spawning; launch-from-spec.mjs then _electron.launch-es exactly that spec and clicks Settings -> About -> Update now. Completion is product state, not a Playwright event: the handoff result file or the checkout reaching the expected sha (source installs write no result file). Ships with the driver, so it works unchanged on every sampled OLD ref - no product flag, no pre-flag fallback split. Both launch shapes are matched (npm exec electron / packaged exe under apps/desktop/release); npm BUILD calls pass through untouched. Exit 0 without a capture fails the leg: a version that never reached its launch must not pass. Probe-the-probe: scripts/launch_capture_probe.sh runs control rows (no opt-in, non-launch argv) and both treatment shapes - all green locally. Gate flips on the shared run workflow for linux/macos; windows adopts the same path with the driver restructuring. --- .github/workflows/install-e2e-run.yml | 13 +- scripts/launch_capture_probe.sh | 80 ++++++++ tests/install/README.md | 2 +- .../launch-capture/sitecustomize.py | 102 ++++++++++ tests/install/e2e-assets/launch-from-spec.mjs | 175 ++++++++++++++++++ tests/install/installer-script-e2e.sh | 52 +++++- 6 files changed, 415 insertions(+), 9 deletions(-) create mode 100755 scripts/launch_capture_probe.sh create mode 100644 tests/install/e2e-assets/launch-capture/sitecustomize.py create mode 100644 tests/install/e2e-assets/launch-from-spec.mjs diff --git a/.github/workflows/install-e2e-run.yml b/.github/workflows/install-e2e-run.yml index 36553831c073..f45848a5e67b 100644 --- a/.github/workflows/install-e2e-run.yml +++ b/.github/workflows/install-e2e-run.yml @@ -42,7 +42,7 @@ on: required: true type: string update-method: - description: 'How the install updates to HEAD. Supported: hermes-update (the updater), installer-script (re-run the one-liner), installer-script+desktop (re-run with --include-desktop). Declared-but-TODO methods (open-app-update, hermes-desktop-app-update) skip.' + description: 'How the install updates to HEAD. Supported: hermes-update (the updater), installer-script (re-run the one-liner), installer-script+desktop (re-run with --include-desktop), hermes-desktop-app-update (launch via hermes desktop under Playwright, click Update now). Declared-but-TODO methods (open-app-update) skip.' required: true type: string install-ref: @@ -61,10 +61,10 @@ on: type: string default: ubuntu-latest timeout-minutes: - description: 'Job timeout. A cold run installs real toolchains twice.' + description: 'Job timeout. A cold run installs real toolchains twice, and app-update legs add a full Electron build + launch.' required: false type: number - default: 45 + default: 75 permissions: contents: read @@ -73,13 +73,14 @@ jobs: e2e: name: install & update # The pairs the driver can run today; anything else is a declared TODO - # and natively skips. The +desktop variants also need the starting tag - # to ship apps/desktop (the --include-desktop flag shipped with it). + # and natively skips. Desktop-surface methods (+desktop installs, + # hermes-desktop-app-update) also need the starting tag to ship + # apps/desktop (their flags shipped with it). if: >- (inputs.install-method == 'installer-script' || (inputs.install-method == 'installer-script+desktop' && inputs.tag-has-desktop)) && (contains(fromJSON('["hermes-update", "installer-script"]'), inputs.update-method) - || (inputs.update-method == 'installer-script+desktop' && inputs.tag-has-desktop)) + || (contains(fromJSON('["installer-script+desktop", "hermes-desktop-app-update"]'), inputs.update-method) && inputs.tag-has-desktop)) runs-on: ${{ inputs.runner }} timeout-minutes: ${{ inputs.timeout-minutes }} diff --git a/scripts/launch_capture_probe.sh b/scripts/launch_capture_probe.sh new file mode 100755 index 000000000000..45dc5ee06fbb --- /dev/null +++ b/scripts/launch_capture_probe.sh @@ -0,0 +1,80 @@ +#!/usr/bin/env bash +# Probe-the-probe for launch-capture/sitecustomize.py, no real install needed. +# Control rows FIRST: without the opt-in env var, and for non-launch argv +# shapes, subprocess.run must behave untouched. Then treatment rows: both +# launch shapes must be captured without spawning. +set -euo pipefail +CAP_DIR="$(cd "$(dirname "$0")" && pwd)/../tests/install/e2e-assets/launch-capture" +WORK="$(mktemp -d)" +trap 'rm -rf "$WORK"' EXIT + +run_py() { + # $1: with_var (yes/no); rest: python -c payload + local with_var="$1"; shift + if [ "$with_var" = yes ]; then + PYTHONPATH="$CAP_DIR${PYTHONPATH:+:$PYTHONPATH}" \ + HERMES_E2E_CAPTURE_LAUNCH="$WORK/spec.json" python3 "$@" + else + PYTHONPATH="$CAP_DIR${PYTHONPATH:+:$PYTHONPATH}" python3 "$@" + fi +} + +fail() { echo "PROBE FAILED: $*" >&2; exit 1; } + +echo "--- control 1: no env var -> run() untouched, real spawn happens" +out="$(run_py no -c 'import subprocess; print(subprocess.run(["echo","real-spawn"],capture_output=True,text=True).stdout.strip())')" +[ "$out" = "real-spawn" ] || fail "control 1: expected real-spawn, got '$out'" +[ ! -e "$WORK/spec.json" ] || fail "control 1: spec written without opt-in" +echo "OK" + +echo "--- control 2: env var set, NON-launch argv (npm run pack) -> passthrough" +rm -f "$WORK"/spec.json* +out="$(run_py yes -c 'import subprocess; r=subprocess.run(["echo","npm-build-ran"],capture_output=True,text=True); print(r.stdout.strip())')" +[ "$out" = "npm-build-ran" ] || fail "control 2: echo did not run" +# and an actual npm-shaped BUILD argv (argv[0]=npm but no electron token): +run_py yes -c 'import subprocess,sys; r=subprocess.run(["npm","run","pack"],capture_output=True); sys.exit(0)' 2>/dev/null || true +[ ! -e "$WORK/spec.json" ] || fail "control 2: build argv was wrongly captured" +echo "OK" + +echo "--- treatment 1: source shape (npm exec -- electron .) captured, not spawned" +rm -f "$WORK"/spec.json* +run_py yes -c ' +import subprocess +r = subprocess.run(["npm", "exec", "--", "electron", "."], cwd="/tmp", env={"HERMES_DESKTOP_CWD": "/tmp", "PATH": "/usr/bin"}) +assert r.returncode == 0, r +' +[ -e "$WORK/spec.json" ] || fail "treatment 1: no spec written" +[ "$(cat "$WORK/spec.json.captured")" = "source" ] || fail "treatment 1: wrong shape" +python3 - "$WORK/spec.json" <<'EOF' +import json, sys +spec = json.load(open(sys.argv[1])) +assert spec["argv"] == ["npm", "exec", "--", "electron", "."], spec["argv"] +assert spec["cwd"] == "/tmp", spec["cwd"] +assert spec["env"]["HERMES_DESKTOP_CWD"] == "/tmp", "env= kwarg not captured" +assert spec["matchedShape"] == "source" +print("spec contents OK") +EOF +echo "OK" + +echo "--- treatment 2: packaged shape captured, not spawned" +rm -f "$WORK"/spec.json* +run_py yes -c ' +import subprocess +exe = "/x/apps/desktop/release/linux-unpacked/Hermes" +r = subprocess.run([exe, "--no-sandbox"], cwd="/tmp", env={"PATH": "/usr/bin"}) +assert r.returncode == 0, r # a real spawn of this path would ENOENT +' +[ "$(cat "$WORK/spec.json.captured")" = "packaged" ] || fail "treatment 2: wrong shape" +echo "OK" + +echo "--- treatment 3: windows-style packaged argv matches too" +rm -f "$WORK"/spec.json* +run_py yes -c ' +import subprocess +r = subprocess.run(["C:\\x\\apps\\desktop\\release\\win-unpacked\\Hermes.exe"], env={}) +assert r.returncode == 0 +' +[ "$(cat "$WORK/spec.json.captured")" = "packaged" ] || fail "treatment 3: wrong shape" +echo "OK" + +echo "ALL PROBES PASSED" diff --git a/tests/install/README.md b/tests/install/README.md index e9de9fb35bc0..54cb7f895754 100644 --- a/tests/install/README.md +++ b/tests/install/README.md @@ -54,7 +54,7 @@ A leg can install a release from months back. The driver must not assume that th The desktop app has two launch paths, so the matrix has two app-update methods. Both click "Update now" in the running app. They differ in how the app starts: - `open-app-update`: the app starts from the OS entry point that the install created. On windows these are the Start Menu and Desktop shortcuts to the installed `Hermes.exe`; the desktop installer always creates them. The installer scripts do not create entry points: their opt-in desktop stage (`--include-desktop` / `-IncludeDesktop`) builds the app inside the checkout but does not register it with the OS. So `open-app-update` legs pair with a `desktop-installer` install. -- `hermes-desktop-app-update`: the app starts with the `hermes desktop` command. Every install method provides this command, on each OS that ships the desktop app. On linux this is the only app surface: no desktop installer and no packaged desktop artifact exist for linux. +- `hermes-desktop-app-update`: the app starts with the `hermes desktop` command. Every install method provides this command, on each OS that ships the desktop app. On linux this is the only app surface: no desktop installer and no packaged desktop artifact exist for linux. The driver captures the product's own launch call (argv, cwd, environment) with `e2e-assets/launch-capture/sitecustomize.py` and re-executes it under Playwright, which owns the app and clicks the update flow. ## Skips diff --git a/tests/install/e2e-assets/launch-capture/sitecustomize.py b/tests/install/e2e-assets/launch-capture/sitecustomize.py new file mode 100644 index 000000000000..bf7b39116967 --- /dev/null +++ b/tests/install/e2e-assets/launch-capture/sitecustomize.py @@ -0,0 +1,102 @@ +"""Driver-side spawn interception for hermes desktop E2E legs. + +The installed ``hermes`` is a venv console script, so its interpreter +imports ``sitecustomize`` at startup when this directory is on +``PYTHONPATH``. Behind an explicit env-var opt-in the module wraps +``subprocess.run`` so the FINAL electron launch call of ``hermes +desktop`` is captured -- argv, cwd, and the fully-constructed ``env`` +kwarg written to a JSON spec -- and replaced with a fake success instead +of spawning. Everything before the spawn (build, stamps, integrity gate, +sandbox fixup) runs for real, in the REAL installed code of whatever +version is under test; Playwright's ``_electron.launch`` then owns the +app from spawn using exactly the spec the product would have used. + +This ships with the DRIVER, never with the product, so it works +identically on every sampled OLD ref -- version drift in the launch +shapes is the matcher's problem, which lives here, next to the driver +(the same maintenance model as ``installer_supports()``). + +Opt-in: ``HERMES_E2E_CAPTURE_LAUNCH=`` -- the spec is written +there, and the marker file ``.captured`` distinguishes "hermes +desktop exited 0 and we captured" from "exited 0 without reaching a +launch" (a version that errors out earlier must FAIL the leg, loudly). + +Launch shapes across sampled desktop-era tags (verified against each +tag's own hermes_cli/main.py): + + v2026.6.5 subprocess.run([npm, "exec", "--", "electron", "."], ...) + v0.20.0+ the npm-exec form AND subprocess.run(launch_command, ...) + where launch_command[0] is the packaged app executable + under apps/desktop/release/ + +Both go through ``subprocess.run`` with an explicit ``env=`` kwarg. npm +BUILD calls (``npm run build`` / ``npm run pack``) carry no ``electron`` +token in argv and pass through untouched -- they must run for real. +""" + +import os + +_SPEC_PATH = os.environ.get("HERMES_E2E_CAPTURE_LAUNCH") + +if _SPEC_PATH: + import json + import subprocess + + _SPEC: str = _SPEC_PATH + _real_run = subprocess.run + + def _basename_noext(token: str) -> str: + base = os.path.basename(str(token)) + for ext in (".exe", ".cmd", ".bat"): + if base.lower().endswith(ext): + base = base[: -len(ext)] + return base.lower() + + def _match_shape(argv: "list[str]") -> str: + """Return the launch shape for argv, or '' when it is not a launch.""" + if not argv: + return "" + tokens = [str(t) for t in argv] + head = _basename_noext(tokens[0]) + # Source shape: npm/npx invoking electron ("npm exec -- electron ."). + # Membership, not position: absorb argv drift across versions. Build + # calls ("npm run pack") carry no bare "electron" token. + if head in ("npm", "npx"): + if any(_basename_noext(t) == "electron" for t in tokens[1:]): + return "source" + return "" + # Packaged shape: argv[0] is the packaged app executable under + # apps/desktop/release/ (win-unpacked/Hermes.exe, linux-unpacked/..., + # mac*/Hermes.app/Contents/MacOS/...). + first = tokens[0].replace("\\", "/") + if "apps/desktop/release/" in first: + return "packaged" + return "" + + def _capturing_run(*args, **kwargs): + argv = args[0] if args else kwargs.get("args") + if not isinstance(argv, (list, tuple)): + return _real_run(*args, **kwargs) + tokens = [str(t) for t in argv] + shape = _match_shape(tokens) + if not shape: + return _real_run(*args, **kwargs) + env = kwargs.get("env") + spec = { + "argv": tokens, + "cwd": str(kwargs.get("cwd") or os.getcwd()), + # Capture what the child would ACTUALLY get: the constructed + # env= when present, the ambient environment when not. + "env": dict(env) if env is not None else dict(os.environ), + "matchedShape": shape, + } + tmp = _SPEC + ".tmp" + with open(tmp, "w", encoding="utf-8") as fh: + json.dump(spec, fh, indent=2) + os.replace(tmp, _SPEC) + with open(_SPEC + ".captured", "w", encoding="utf-8") as fh: + fh.write(shape) + print(f"[e2e launch-capture] captured {shape} launch -> {_SPEC} (not spawning)") + return subprocess.CompletedProcess(tokens, 0, stdout=None, stderr=None) + + subprocess.run = _capturing_run diff --git a/tests/install/e2e-assets/launch-from-spec.mjs b/tests/install/e2e-assets/launch-from-spec.mjs new file mode 100644 index 000000000000..e203ba5714bb --- /dev/null +++ b/tests/install/e2e-assets/launch-from-spec.mjs @@ -0,0 +1,175 @@ +// @ts-check +/** + * Launch the Hermes desktop app from a captured launch spec and click the + * real update flow: Settings -> About -> "Update now". + * + * The spec is written by launch-capture/sitecustomize.py at `hermes + * desktop`'s own spawn site, so argv, cwd, and the fully-constructed env + * are the product's own -- this launcher only translates the npm-exec + * source shape into a direct electron binary path (Playwright needs a + * real executable, and the electron npm shim would re-spawn out of our + * control). + * + * Usage (from the scratch dir where the driver installed @playwright/test): + * node launch-from-spec.mjs --spec /path/launch-spec.json \ + * [--result $HERMES_HOME/.hermes-update-result.json] \ + * [--expect-sha --repo-dir ] [--no-update] + * + * --no-update: launch + wait for the window + close. The smoke arm. + * Otherwise: click Update now, then poll for completion. Two signals, + * either satisfies (poll whichever are given, first hit wins): + * --result the windows hand-off's result file + * (HERMES_HOME/.hermes-update-result.json) + * --expect-sha the installed checkout reaching the expected commit - + * the source-install signal, where the About pane's update + * runs `hermes update` and no result file exists. + * The Playwright close event is unreliable across the update handoff, so + * neither signal is an app event. + */ + +import fs from 'node:fs'; +import path from 'node:path'; +import { execFileSync } from 'node:child_process'; +import { parseArgs } from 'node:util'; +import { _electron } from '@playwright/test'; + +/** + * @typedef {{argv: string[], cwd: string, env: Record, + * matchedShape: 'source' | 'packaged'}} LaunchSpec + */ + +/** + * Resolve what _electron.launch needs from a captured spec. + * @param {LaunchSpec} spec + * @returns {{executablePath: string, args: string[], cwd: string, + * env: Record}} + */ +export function resolveLaunch(spec) { + if (spec.matchedShape === 'packaged') { + return { + executablePath: spec.argv[0], + args: spec.argv.slice(1), + cwd: spec.cwd, + env: spec.env, + }; + } + // Source shape: ["npm", "exec", "--", "electron", ".", ...extra] running + // in apps/desktop. Electron's real binary lives in the workspace-hoisted + // node_modules; `electron/index.js` exports its path but requires the + // module -- cheaper here to read the path file it derives from. + const desktopDir = spec.cwd; + const idx = spec.argv.findIndex((t) => t === 'electron'); + const extra = idx >= 0 ? spec.argv.slice(idx + 1).filter((t) => t !== '.') : []; + const candidates = [ + path.join(desktopDir, 'node_modules', 'electron'), + path.join(desktopDir, '..', '..', 'node_modules', 'electron'), + ]; + for (const moduleDir of candidates) { + const pathTxt = path.join(moduleDir, 'path.txt'); + if (!fs.existsSync(pathTxt)) continue; + const rel = fs.readFileSync(pathTxt, 'utf8').trim(); + const exe = path.join(moduleDir, 'dist', rel); + if (fs.existsSync(exe)) { + return { executablePath: exe, args: ['.', ...extra], cwd: desktopDir, env: spec.env }; + } + } + throw new Error(`no electron binary found under ${candidates.join(' or ')}`); +} + +/** @param {string} msg */ +function log(msg) { + console.log(`[launch-from-spec] ${msg}`); +} + +async function main() { + const { values } = parseArgs({ + options: { + spec: { type: 'string' }, + result: { type: 'string' }, + 'expect-sha': { type: 'string' }, + 'repo-dir': { type: 'string' }, + 'no-update': { type: 'boolean', default: false }, + 'timeout-ms': { type: 'string', default: '600000' }, + }, + }); + if (!values.spec) throw new Error('--spec is required'); + /** @type {LaunchSpec} */ + const spec = JSON.parse(fs.readFileSync(values.spec, 'utf8')); + const launch = resolveLaunch(spec); + log(`launching ${launch.executablePath} (shape: ${spec.matchedShape})`); + + const app = await _electron.launch({ + executablePath: launch.executablePath, + args: launch.args, + cwd: launch.cwd, + env: launch.env, + }); + const window = await app.firstWindow({ timeout: 120_000 }); + await window.waitForLoadState('domcontentloaded'); + log(`window up: ${await window.title()}`); + await window.screenshot({ path: `${values.spec}.window.png` }).catch(() => {}); + + if (values['no-update']) { + log('smoke mode: window proven, closing'); + await app.close().catch(() => {}); + return; + } + + if (!values.result && !(values['expect-sha'] && values['repo-dir'])) { + throw new Error('need --result and/or --expect-sha + --repo-dir unless --no-update'); + } + const deadline = Date.now() + Number(values['timeout-ms']); + + // Dismiss the onboarding overlay when present (fresh HERMES_HOME). + const skip = window.getByRole('button', { name: /skip|get started|continue/i }).first(); + if (await skip.isVisible({ timeout: 5_000 }).catch(() => false)) { + await skip.click().catch(() => {}); + } + + // Settings -> About -> Update now. Selectors favor accessible names over + // DOM structure so renderer refactors don't break the leg. + await window.getByRole('button', { name: /settings/i }).first().click(); + await window.getByRole('tab', { name: /about/i }).or( + window.getByRole('button', { name: /about/i })).first().click(); + const updateNow = window.getByRole('button', { name: /update now/i }).first(); + await updateNow.waitFor({ state: 'visible', timeout: 60_000 }); + await updateNow.click(); + log('clicked Update now; polling for result file'); + + // The app may relaunch/exit during the update; completion signals are + // product state, not Playwright events. + const resultPath = values.result; + const expectSha = values['expect-sha']; + const repoDir = values['repo-dir']; + /** @returns {string} */ + const headSha = () => { + try { + return execFileSync('git', ['-C', /** @type {string} */ (repoDir), 'rev-parse', 'HEAD'], { + encoding: 'utf8', + }).trim(); + } catch { + return ''; + } + }; + for (;;) { + if (resultPath && fs.existsSync(resultPath)) { + log(`update result present: ${fs.readFileSync(resultPath, 'utf8').slice(0, 200)}`); + break; + } + if (expectSha && repoDir && headSha() === expectSha) { + log(`checkout reached expected sha ${expectSha}`); + break; + } + if (Date.now() > deadline) { + await window.screenshot({ path: `${values.spec}.timeout.png` }).catch(() => {}); + throw new Error('update completion signal never appeared (result file / expected sha)'); + } + await new Promise((r) => setTimeout(r, 2_000)); + } + await app.close().catch(() => {}); +} + +const invoked = process.argv[1] && path.resolve(process.argv[1]) === (await import('node:url')).fileURLToPath(import.meta.url); +if (invoked) { + await main(); +} diff --git a/tests/install/installer-script-e2e.sh b/tests/install/installer-script-e2e.sh index d2e1a0a009c2..bc6078f3bef7 100755 --- a/tests/install/installer-script-e2e.sh +++ b/tests/install/installer-script-e2e.sh @@ -35,6 +35,9 @@ # --update-method hermes-update `hermes update` # installer-script re-run install.sh (HEAD's copy) # installer-script+desktop re-run with --include-desktop +# hermes-desktop-app-update launch the app via `hermes +# desktop` (spawn captured, Playwright +# drives it) and click Update now # --install-ref what to install first; anything git resolves. Default: # the newest release tag in the checkout. # @@ -65,8 +68,8 @@ case "$INSTALL_METHOD" in *) echo "error: --install-method must be installer-script or installer-script+desktop, got '$INSTALL_METHOD'" >&2; exit 1 ;; esac case "$UPDATE_METHOD" in - hermes-update|installer-script|installer-script+desktop) ;; - *) echo "error: --update-method must be hermes-update, installer-script or installer-script+desktop, got '$UPDATE_METHOD'" >&2; exit 1 ;; + hermes-update|installer-script|installer-script+desktop|hermes-desktop-app-update) ;; + *) echo "error: --update-method must be hermes-update, installer-script, installer-script+desktop or hermes-desktop-app-update, got '$UPDATE_METHOD'" >&2; exit 1 ;; esac REPO_ROOT="$(cd "$(dirname "$0")/../.." && pwd)" @@ -290,6 +293,51 @@ case "$UPDATE_METHOD" in run_installer "$HEAD_SHA" head desktop assert_desktop_artifact HEAD ;; + hermes-desktop-app-update) + # The real user surface: `hermes desktop` launches the app, the user + # clicks Settings -> About -> Update now. Playwright must OWN the spawn + # (it needs the inspection pipe), so the driver intercepts the product's + # own launch call - argv/cwd/env captured at the spawn site by + # e2e-assets/launch-capture/sitecustomize.py - and re-executes it under + # _electron.launch. Everything before the spawn (build, stamps, sandbox + # fixup) runs for real in the installed code. + HERMES="$INSTALL_DIR/venv/bin/hermes" + ASSETS="$REPO_ROOT/tests/install/e2e-assets" + SPEC="$WORK_ROOT/launch-spec.json" + + step "capturing the hermes desktop launch spec (build runs for real)" + rc=0 + (cd "$INSTALL_DIR" && \ + PYTHONPATH="$ASSETS/launch-capture${PYTHONPATH:+:$PYTHONPATH}" \ + HERMES_E2E_CAPTURE_LAUNCH="$SPEC" \ + "$HERMES" desktop < /dev/null > "$LOG_DIR/desktop-launch-capture.log" 2>&1) || rc=$? + log_group "hermes desktop (launch capture) transcript" "$LOG_DIR/desktop-launch-capture.log" + [ "$rc" -eq 0 ] || fail "hermes desktop exited $rc during launch capture; transcript above" + # Exit 0 without a capture means a version that never reached its + # launch - that must fail loudly, not pass as a no-op. + [ -f "$SPEC.captured" ] || fail "hermes desktop exited 0 but no launch was captured at $SPEC" + ok "captured $(cat "$SPEC.captured") launch spec" + + step "driving the app under Playwright: Settings -> About -> Update now" + # Driver tooling comes from the driver: a scratch dir with our own + # pinned @playwright/test, never resolved from the installed tree + # (older OLD refs predate the dependency; hoisting moves it around). + PW_DIR="$WORK_ROOT/playwright" + mkdir -p "$PW_DIR" + (cd "$PW_DIR" && npm install --no-save --no-audit --no-fund \ + "@playwright/test@1.58.2" > "$LOG_DIR/playwright-install.log" 2>&1) \ + || { log_group "playwright install transcript" "$LOG_DIR/playwright-install.log"; fail "playwright install failed"; } + cp "$ASSETS/launch-from-spec.mjs" "$PW_DIR/" + rc=0 + (cd "$PW_DIR" && node launch-from-spec.mjs \ + --spec "$SPEC" \ + --result "$HERMES_HOME/.hermes-update-result.json" \ + --expect-sha "$HEAD_SHA" \ + --repo-dir "$INSTALL_DIR" \ + > "$LOG_DIR/app-update.log" 2>&1) || rc=$? + log_group "app update (Playwright) transcript" "$LOG_DIR/app-update.log" + [ "$rc" -eq 0 ] || fail "app-driven update exited $rc; transcript above" + ;; esac assert_checkout "$HEAD_SHA" HEAD smoke_desktop head From 11f00b823ca3c6209fc4e180bd1324711317cd49 Mon Sep 17 00:00:00 2001 From: ethernet Date: Wed, 12 Aug 2026 04:23:43 -0400 Subject: [PATCH 071/227] fix(install-e2e): installer_supports lied under pipefail - buffer the probe git show | grep -qF exits at grep's first match; install.sh is ~140KB with the flag strings in the first few KB, so git show takes SIGPIPE on its next write and the pipeline reports 141 under set -o pipefail. The probe answered NO for flags the ref HAS - timing-dependent, green without pipefail (every local check), red on the runner. It hid while a probe miss just meant omitting --skip-browser; the first probe where NO is a hard failure (--include-desktop) exposed it on its first CI leg. Buffer git show into a variable and grep the string: git always completes, grep judges bytes. --- tests/install/installer-script-e2e.sh | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-) diff --git a/tests/install/installer-script-e2e.sh b/tests/install/installer-script-e2e.sh index bc6078f3bef7..b07e36a4dc2e 100755 --- a/tests/install/installer-script-e2e.sh +++ b/tests/install/installer-script-e2e.sh @@ -155,8 +155,15 @@ INSTALL_DIR="$HERMES_HOME/hermes-agent" # install.sh rather than assuming this checkout's flag set: the point of the # matrix is to install releases from months back, whose installers predate # options we take for granted. +# +# Buffered through a variable, NOT `git show | grep -q`: under pipefail, +# grep -q exits at the first match (install.sh is ~140KB, the flags appear +# in the first few KB), git show takes SIGPIPE on its next write, and the +# pipeline reports 141 -- the probe answers NO for a flag the ref HAS. installer_supports() { - git -C "$REPO_ROOT" show "$1:scripts/install.sh" | grep -qF -- "$2" + local text + text="$(git -C "$REPO_ROOT" show "$1:scripts/install.sh")" + grep -qF -- "$2" <<< "$text" } run_installer() { From adf7d55f4b63412dfde2c0b6fefff03238c08317 Mon Sep 17 00:00:00 2001 From: ethernet Date: Wed, 12 Aug 2026 04:30:12 -0400 Subject: [PATCH 072/227] ci(install-e2e): windows composes install x update - one driver, one job windows-desktop-gui-e2e.ps1 and windows-installer-script-e2e.ps1 fold into tests/install/windows-e2e.ps1 with orthogonal -InstallMethod and -Route axes: the install phase dispatches on one, the update phase on the other, and shared workroot state carries how OLD landed - so any implemented update method can follow any implemented install method. Implementing a new pair is now a driver function plus a gate edit, never a new job. The run workflow collapses to ONE inner job whose if: is the implemented-pairs table. Newly cheap pairs go live with the merge: desktop-installer@latest -> hermes-update / installer-script / installer-script+desktop / hermes-desktop-app-update installer-script(+desktop) -> hermes-desktop-app-update installer-script+desktop -> open-app-update (the -IncludeDesktop install registers real Start Menu / Desktop shortcuts) Only desktop-installer@latest as an UPDATE method stays a declared TODO. scripts/windows_e2e_harness.ps1 executes the parse/parameter/ dispatch checks under pwsh before any Windows runner spins up. --- .github/workflows/install-e2e-windows-run.yml | 168 ++++------ scripts/sandbox/generate-e2e-matrix.mjs | 8 +- scripts/windows_e2e_harness.ps1 | 48 +++ tests/install/README.md | 4 +- tests/install/installer-script-e2e.sh | 2 +- ...ws-desktop-gui-e2e.ps1 => windows-e2e.ps1} | 239 +++++++++++++-- .../install/windows-installer-script-e2e.ps1 | 289 ------------------ 7 files changed, 320 insertions(+), 438 deletions(-) create mode 100644 scripts/windows_e2e_harness.ps1 rename tests/install/{windows-desktop-gui-e2e.ps1 => windows-e2e.ps1} (76%) delete mode 100644 tests/install/windows-installer-script-e2e.ps1 diff --git a/.github/workflows/install-e2e-windows-run.yml b/.github/workflows/install-e2e-windows-run.yml index f1e99531044f..cd043151d2be 100644 --- a/.github/workflows/install-e2e-windows-run.yml +++ b/.github/workflows/install-e2e-windows-run.yml @@ -1,48 +1,34 @@ -name: Install & Update E2E — Windows desktop (reusable) - -# Runs ONE Windows {install-method, update-method} combination: install -# OLD, then update OLD -> HEAD through the route a user would take. +# Reusable runner for ONE Windows install/update combination. # -# The Windows sibling of install-e2e-run.yml. No bubblewrap anywhere -- the -# drivers fake GitHub with git's own transport rewrite: a driver-owned -# GIT_CONFIG_GLOBAL carrying url..insteadOf for both -# canonical repo URLs, so the installer and updater run byte-for-byte -# against their real URLs and land on a local bare repo. Its `main` serves -# OLD (install-ref, default: the newest release tag) during the install, -# then advances to HEAD for the update leg -- an update becomes available -# exactly the way it does for a real user. +# One job, two orthogonal axes: tests/install/windows-e2e.ps1 dispatches its +# install phase on install-method and its update phase on update-method, so +# implementing a new pair is a driver function + a gate edit here - never a +# new job. The driver's phases share state via the workroot, and every leg +# runs the REAL user surface for its methods: # -# Two install arms, two drivers: -# desktop-installer@latest the REAL desktop user flow -# (tests/install/windows-desktop-gui-e2e.ps1): -# the website's published Hermes-Setup.exe runs -# headed, AutoHotkey clicks Install -> Launch, -# the real Electron window must appear. Update -# methods: -# open-app-update the app's own Update -# button, app launched from the -# installed Hermes.exe: it runs -# under Playwright's Electron -# driver, which clicks Settings -# -> About -> "Update now"; the -# production hand-off chain runs -# untouched. -# (hermes-update, desktop-installer@latest, -# installer-script re-run, -# hermes-desktop-app-update: declared TODOs.) -# installer-script the irm | iex one-liner -# (tests/install/windows-installer-script-e2e -# .ps1): runs the install.ps1 shipped AT the -# OLD ref, headless. Update methods: -# hermes-update venv hermes.exe update -# installer-script re-run HEAD's install.ps1 -# (open-app-update and -# hermes-desktop-app-update from a script -# install: TODO.) +# desktop-installer@latest the website's Hermes-Setup.exe, downloaded and +# run headed, AutoHotkey clicks Install -> +# Launch, the real Electron window must appear. +# installer-script the irm | iex one-liner: the install.ps1 +# shipped AT the OLD ref, headless. +# installer-script+desktop the same one-liner with -IncludeDesktop: +# builds Hermes.exe AND registers Start Menu / +# Desktop shortcuts. +# hermes-update venv hermes.exe update. +# open-app-update the app's own Update button, app launched +# from the installed exe under Playwright's +# Electron driver (Settings -> About -> +# "Update now"); the production hand-off chain +# runs untouched. +# hermes-desktop-app-update the same button, app launched via `hermes +# desktop`: the driver captures the product's +# own spawn (argv/cwd/env) and re-executes it +# under Playwright. # -# Method pairs without a driver yet NATIVELY SKIP (grey check, no runner): -# the capability knowledge lives here, next to the drivers, so the caller -# can dispatch every declared combination without knowing which ones work. +# Method pairs without a driver arm yet NATIVELY SKIP (grey check, no +# runner): the capability knowledge lives here, next to the driver, so the +# caller can dispatch every declared combination without knowing which ones +# work. # # Call it: # @@ -54,6 +40,8 @@ name: Install & Update E2E — Windows desktop (reusable) # update-method: open-app-update # install-ref: v2026.8.3 +name: install-e2e windows leg + on: workflow_call: inputs: @@ -62,7 +50,7 @@ on: required: true type: string update-method: - description: 'How the install updates to HEAD. Supported: open-app-update (Update button under Playwright, app launched from the installed exe, from a desktop install) and hermes-update / installer-script (from a script install). Declared-but-TODO pairs (incl. hermes-desktop-app-update) skip.' + description: 'How the install updates to HEAD. Supported: open-app-update (Update button under Playwright, from a desktop-bearing install), hermes-desktop-app-update (same button, app launched via hermes desktop), hermes-update, installer-script, installer-script+desktop. desktop-installer@latest re-run is a declared TODO and skips.' required: true type: string install-ref: @@ -90,15 +78,24 @@ permissions: contents: read jobs: - # ---- arm 1: install via the website's Hermes-Setup.exe (GUI) ------------- e2e: - name: Hermes-Setup.exe - # The one pair the driver can run today, and only from a starting - # version that ships the desktop app (the caller annotates - # tag-has-desktop from the tag's own tree; releases before #20059 have - # no window to launch and no Update button to click). Anything else is - # a native skip: a declared-TODO method pair, or a pre-desktop tag. - if: inputs.install-method == 'desktop-installer@latest' && inputs.update-method == 'open-app-update' && inputs.tag-has-desktop + # Short static name on purpose: name expressions render UNEXPANDED on + # skipped jobs. + name: e2e + # The implemented {install x update} pairs. Two rules feed the table: + # * every desktop-surface method needs the starting tag to ship + # apps/desktop (pre-desktop releases have no window to launch, no + # Update button to click, no -IncludeDesktop to pass); + # * open-app-update needs an OS entry point, which only the + # desktop-bearing installs create. + # desktop-installer@latest as an UPDATE method is the one declared TODO. + if: >- + (inputs.install-method == 'installer-script' + || (contains(fromJSON('["installer-script+desktop", "desktop-installer@latest"]'), inputs.install-method) && inputs.tag-has-desktop)) + && (contains(fromJSON('["hermes-update", "installer-script"]'), inputs.update-method) + || (contains(fromJSON('["installer-script+desktop", "hermes-desktop-app-update"]'), inputs.update-method) && inputs.tag-has-desktop) + || (inputs.update-method == 'open-app-update' && inputs.tag-has-desktop + && contains(fromJSON('["desktop-installer@latest", "installer-script+desktop"]'), inputs.install-method))) runs-on: windows-latest timeout-minutes: ${{ inputs.timeout-minutes }} @@ -112,9 +109,8 @@ jobs: steps: # Full history: the driver bare-clones this checkout as the repo the - # installer/updater talk to, and the installer's baked release pin - # must be reachable in that clone. A shallow checkout cannot serve - # either need. + # installer/updater talk to, and both OLD and HEAD must be reachable + # in that clone. A shallow checkout cannot serve either need. - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2 with: fetch-depth: 0 @@ -131,15 +127,15 @@ jobs: - name: Stage serve repo (main -> ${{ inputs.install-ref }}) shell: powershell - run: powershell -NoProfile -ExecutionPolicy Bypass -File tests\install\windows-desktop-gui-e2e.ps1 -Phase stage -Route "${{ inputs.update-method }}" -InstallRef "${{ inputs.install-ref }}" -SetupExeUrl ${{ inputs.setup-exe-url }} + run: powershell -NoProfile -ExecutionPolicy Bypass -File tests\install\windows-e2e.ps1 -Phase stage -InstallMethod "${{ inputs.install-method }}" -Route "${{ inputs.update-method }}" -InstallRef "${{ inputs.install-ref }}" -SetupExeUrl ${{ inputs.setup-exe-url }} - - name: Install ${{ inputs.install-ref }} via website Hermes-Setup.exe (headed, AHK-clicked) + - name: Install ${{ inputs.install-ref }} (${{ inputs.install-method }}) shell: powershell - run: powershell -NoProfile -ExecutionPolicy Bypass -File tests\install\windows-desktop-gui-e2e.ps1 -Phase install-gui -Route "${{ inputs.update-method }}" -InstallRef "${{ inputs.install-ref }}" -SetupExeUrl ${{ inputs.setup-exe-url }} + run: powershell -NoProfile -ExecutionPolicy Bypass -File tests\install\windows-e2e.ps1 -Phase install -InstallMethod "${{ inputs.install-method }}" -Route "${{ inputs.update-method }}" -InstallRef "${{ inputs.install-ref }}" -SetupExeUrl ${{ inputs.setup-exe-url }} - name: Update ${{ inputs.install-ref }} -> HEAD (${{ inputs.update-method }}) shell: powershell - run: powershell -NoProfile -ExecutionPolicy Bypass -File tests\install\windows-desktop-gui-e2e.ps1 -Phase update-gui -Route "${{ inputs.update-method }}" -InstallRef "${{ inputs.install-ref }}" -SetupExeUrl ${{ inputs.setup-exe-url }} + run: powershell -NoProfile -ExecutionPolicy Bypass -File tests\install\windows-e2e.ps1 -Phase update -InstallMethod "${{ inputs.install-method }}" -Route "${{ inputs.update-method }}" -InstallRef "${{ inputs.install-ref }}" -SetupExeUrl ${{ inputs.setup-exe-url }} - name: Stop screen recording if: always() @@ -158,6 +154,7 @@ jobs: $home_ = Join-Path $work "hermes-home" foreach ($pair in @( @{ src = (Join-Path $work "proof"); dst = "proof" }, + @{ src = (Join-Path $work "logs"); dst = "driver-logs" }, @{ src = (Join-Path $work "shas.json"); dst = "shas.json" }, @{ src = (Join-Path $home_ "logs"); dst = "logs" }, @{ src = (Join-Path $home_ ".hermes-update-result.json"); dst = ".hermes-update-result.json" } @@ -169,60 +166,7 @@ jobs: if: always() uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1 with: - name: install-e2e-windows-${{ inputs.update-method }}-${{ inputs.install-ref }}-${{ github.sha }} + name: install-e2e-windows-${{ inputs.install-method }}-${{ inputs.update-method }}-${{ inputs.install-ref }}-${{ github.sha }} path: gui-e2e-proof retention-days: 14 if-no-files-found: ignore - - # ---- arm 2: install via the irm | iex one-liner (headless) --------------- - script-e2e: - # Same static-name reasoning as e2e above. - name: install.ps1 - # The pairs the script driver can run today. open-app-update and - # hermes-desktop-app-update from a script install are declared TODOs - # (the desktop app also has to be BUILT by - # that path first). The +desktop variants also need the starting tag - # to ship apps/desktop (the -IncludeDesktop parameter shipped with it). - if: >- - (inputs.install-method == 'installer-script' - || (inputs.install-method == 'installer-script+desktop' && inputs.tag-has-desktop)) - && (contains(fromJSON('["hermes-update", "installer-script"]'), inputs.update-method) - || (inputs.update-method == 'installer-script+desktop' && inputs.tag-has-desktop)) - runs-on: windows-latest - timeout-minutes: ${{ inputs.timeout-minutes }} - - steps: - # Full history: the driver bare-clones this checkout as the repo the - # installer/updater talk to, and both OLD and HEAD must be reachable - # in that clone. - - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2 - with: - fetch-depth: 0 - - - name: Start screen recording - uses: ./.github/actions/e2e-screen-record - with: - mode: start - output: ${{ runner.temp }}\e2e-logs\recording.mkv - - - name: Run install + update E2E - shell: powershell - run: powershell -NoProfile -ExecutionPolicy Bypass -File tests\install\windows-installer-script-e2e.ps1 -InstallMethod "${{ inputs.install-method }}" -UpdateMethod "${{ inputs.update-method }}" -InstallRef "${{ inputs.install-ref }}" - env: - HERMES_E2E_LOG_DIR: ${{ runner.temp }}\e2e-logs - - - name: Stop screen recording - if: always() - uses: ./.github/actions/e2e-screen-record - with: - mode: stop - output: ${{ runner.temp }}\e2e-logs\recording.mkv - - - name: Upload installer logs - if: always() - uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1 - with: - name: install-e2e-windows-script-${{ inputs.update-method }}-${{ inputs.install-ref }}-${{ github.sha }} - path: ${{ runner.temp }}\e2e-logs - retention-days: 14 - if-no-files-found: ignore diff --git a/scripts/sandbox/generate-e2e-matrix.mjs b/scripts/sandbox/generate-e2e-matrix.mjs index 16cb8e176f12..c50a4385460a 100644 --- a/scripts/sandbox/generate-e2e-matrix.mjs +++ b/scripts/sandbox/generate-e2e-matrix.mjs @@ -254,10 +254,10 @@ export function renderMarkdownPlan(envs, tags) { */ export function renderMarkdownResults(jobs) { const LEG = /^(linux|windows|macos): (\S+) -> (\S+) \((\S+) -> HEAD\) \//; - // A combination can surface as SEVERAL jobs with the same leg name (the - // windows run workflow has one arm per driver; exactly one runs and the - // others natively skip), so cells merge by significance: a real outcome - // always beats a skip, and a bad outcome beats a good one. + // A combination can surface as SEVERAL jobs with the same leg name (a + // run workflow may have one inner job per driver arm; exactly one runs + // and the others natively skip), so cells merge by significance: a real + // outcome always beats a skip, and a bad outcome beats a good one. const RANK = ['skip', '✅', 'running', 'cancelled', '❌']; /** @type {Map>} */ const rows = new Map(); diff --git a/scripts/windows_e2e_harness.ps1 b/scripts/windows_e2e_harness.ps1 new file mode 100644 index 000000000000..4ad2d0e2ecff --- /dev/null +++ b/scripts/windows_e2e_harness.ps1 @@ -0,0 +1,48 @@ +# Local harness for windows-e2e.ps1's non-GUI logic, runnable under linux +# pwsh (nix run nixpkgs#powershell). Exercises the pieces that cost CI +# round-trips when wrong: parameter surface, ref-flag probing, and the +# install/update dispatch reaching the right arm names. +param() +$ErrorActionPreference = "Stop" +$driver = Join-Path $PSScriptRoot "..\tests\install\windows-e2e.ps1" + +# 1. Parse cleanly. +$tok = $null; $err = $null +[System.Management.Automation.Language.Parser]::ParseFile((Resolve-Path $driver), [ref]$tok, [ref]$err) | Out-Null +if ($err -and $err.Count) { $err | ForEach-Object { Write-Host $_.Message }; exit 1 } +Write-Host "parse: OK" + +# 2. Parameter surface: both axes present with the generator's ids. +$ast = [System.Management.Automation.Language.Parser]::ParseFile((Resolve-Path $driver), [ref]$tok, [ref]$err) +$params = $ast.ParamBlock.Parameters +$byName = @{} +foreach ($p in $params) { $byName[$p.Name.VariablePath.UserPath] = $p } +foreach ($required in @("Phase", "InstallMethod", "Route", "InstallRef")) { + if (-not $byName.ContainsKey($required)) { Write-Host "missing param: $required"; exit 1 } +} +$imSet = ($byName["InstallMethod"].Attributes | Where-Object { $_.TypeName.Name -eq "ValidateSet" }).PositionalArguments.Value +foreach ($m in @("desktop-installer@latest", "installer-script", "installer-script+desktop")) { + if ($imSet -notcontains $m) { Write-Host "InstallMethod ValidateSet missing $m"; exit 1 } +} +$rSet = ($byName["Route"].Attributes | Where-Object { $_.TypeName.Name -eq "ValidateSet" }).PositionalArguments.Value +foreach ($m in @("open-app-update", "hermes-desktop-app-update", "hermes-update", "installer-script", "installer-script+desktop", "desktop-installer@latest")) { + if ($rSet -notcontains $m) { Write-Host "Route ValidateSet missing $m"; exit 1 } +} +Write-Host "parameter surface: OK" + +# 3. Dispatch bodies reference the right arms (AST-level: the switch on +# InstallMethod contains the three arms; the switch on Route contains all +# six, with desktop-installer@latest throwing). +$text = Get-Content -LiteralPath $driver -Raw +foreach ($needle in @( + 'function Invoke-PhaseInstall', + 'function Invoke-PhaseUpdate', + 'Invoke-RefInstaller $state.old "old" -IncludeDesktop', + 'Invoke-HermesDesktopAppUpdate $state.current', + 'Invoke-HermesUpdate', + "update method 'desktop-installer@latest' is not implemented yet" +)) { + if ($text.IndexOf($needle) -lt 0) { Write-Host "dispatch missing: $needle"; exit 1 } +} +Write-Host "dispatch arms: OK" +Write-Host "ALL HARNESS CHECKS PASSED" diff --git a/tests/install/README.md b/tests/install/README.md index 54cb7f895754..c50c32444e64 100644 --- a/tests/install/README.md +++ b/tests/install/README.md @@ -11,7 +11,7 @@ The test family has four layers. Each layer has one job. 1. `scripts/sandbox/generate-e2e-matrix.mjs` declares the support matrix. It lists every {os, install-method, update-method} pair. It expands the pairs against the sampled release tags. It knows nothing about which pairs CI can run. 2. `.github/workflows/install-e2e.yml` is the primary workflow. It picks the release tags, runs the generator, and fans out one matrix job per OS. It also writes the plan chart and the result chart on the run summary. 3. The run workflows own the capability knowledge. `install-e2e-run.yml` serves linux and macos with one OS-agnostic driver. `install-e2e-windows-run.yml` serves windows. A job-level `if:` gate in each run workflow lists the pairs its driver can run. All other pairs skip natively and show as grey. -4. The drivers do the work. `tests/install/installer-script-e2e.sh` is the POSIX driver. `tests/install/windows-installer-script-e2e.ps1` is the windows script driver. `tests/install/windows-desktop-gui-e2e.ps1` is the windows GUI driver. +4. The drivers do the work. `tests/install/installer-script-e2e.sh` is the POSIX driver. `tests/install/windows-e2e.ps1` is the windows driver; its install phase and update phase dispatch on separate method parameters, so any implemented update method can follow any implemented install method. To declare a new method, edit the generator. To implement a method, flip the gate in the run workflow and extend a driver. @@ -33,7 +33,7 @@ Each leg with the script drivers has these phases: 4. Update: move `main` to HEAD. Apply one update method. Make sure that the checkout is at HEAD and that `hermes --version` works. 5. Desktop smoke again, at HEAD. -The windows GUI driver replaces phases 2 and 4. It downloads the published `Hermes-Setup.exe`, clicks through the installer window with AutoHotkey, and clicks "Update now" in the running app with Playwright. +The windows GUI driver replaces phases 2 and 4 when the install method is `desktop-installer@latest`. It downloads the published `Hermes-Setup.exe`, clicks through the installer window with AutoHotkey, and clicks "Update now" in the running app with Playwright. ## Old versions diff --git a/tests/install/installer-script-e2e.sh b/tests/install/installer-script-e2e.sh index b07e36a4dc2e..77be0c0f6e09 100755 --- a/tests/install/installer-script-e2e.sh +++ b/tests/install/installer-script-e2e.sh @@ -1,7 +1,7 @@ #!/usr/bin/env bash # Prove a user who installed OLD via the installer script can reach HEAD. # -# The POSIX sibling of tests/install/windows-desktop-gui-e2e.ps1, sharing its +# The POSIX sibling of tests/install/windows-e2e.ps1, sharing its # staging trick and replacing the old bubblewrap sandbox: instead of a fake # Internet (MITM proxy + upload-pack shim), every git process is pointed at a # local bare clone with url..insteadOf rewrites for both diff --git a/tests/install/windows-desktop-gui-e2e.ps1 b/tests/install/windows-e2e.ps1 similarity index 76% rename from tests/install/windows-desktop-gui-e2e.ps1 rename to tests/install/windows-e2e.ps1 index a77feeba09d7..60739e1e7f8b 100644 --- a/tests/install/windows-desktop-gui-e2e.ps1 +++ b/tests/install/windows-e2e.ps1 @@ -61,22 +61,31 @@ # Real GUI users are on the official origin and never see it. # # USAGE (local Windows box or CI): -# powershell -File tests\install\windows-desktop-gui-e2e.ps1 -Phase all -# ... -Phase stage / install-gui / update-gui +# powershell -File tests\install\windows-e2e.ps1 -Phase all +# ... -Phase stage / install / update # Phases share state via \shas.json, so CI can run them as -# separate steps for readable logs. +# separate steps for readable logs. -InstallMethod and -Route are +# orthogonal axes: the install phase dispatches on -InstallMethod, the +# update phase on -Route, and install writes what update needs (paths, +# how OLD landed) into the shared state - so any implemented update can +# follow any implemented install. # ============================================================================ param( - [ValidateSet("stage", "install-gui", "update-gui", "all")] + [ValidateSet("stage", "install", "update", "all")] [string]$Phase = "all", - # Update method to exercise in the update-gui phase, named by the same - # ids the combination generator (scripts/sandbox/generate-e2e-matrix - # .mjs) declares. Only "open-app-update" (the app's own Update button, - # app launched from the installed exe) is implemented; the others are - # declared arms so the surface is stable when they land. - [ValidateSet("open-app-update", "hermes-desktop-app-update", "hermes-update", "desktop-installer@latest", "installer-script")] + # How OLD gets installed, named by the same ids the combination + # generator (scripts/sandbox/generate-e2e-matrix.mjs) declares. + [ValidateSet("desktop-installer@latest", "installer-script", "installer-script+desktop")] + [string]$InstallMethod = "desktop-installer@latest", + + # Update method to exercise in the update phase, same id namespace. + # open-app-update (from a desktop-installer install) and hermes-update / + # installer-script / installer-script+desktop (from script installs) are + # implemented; the rest are declared arms so the surface is stable when + # they land. + [ValidateSet("open-app-update", "hermes-desktop-app-update", "hermes-update", "desktop-installer@latest", "installer-script", "installer-script+desktop")] [string]$Route = "open-app-update", # The OLD version: the ref served as `main` while the installer runs, @@ -209,6 +218,130 @@ function Test-HermesRuns([string]$Label) { Assert-True ($LASTEXITCODE -eq 0) "$Label -- hermes --version exits 0" } +# ---------------------------------------------------------------------------- +# Script-install arm: the irm | iex one-liner, headless (the install.ps1 +# shipped AT the ref under test, run with flags probed from that ref's own +# script text - older releases reject parameters added later). +# ---------------------------------------------------------------------------- +function Write-LogGroup([string]$Title, [string]$LogPath) { + Write-Host "::group::$Title" + if (Test-Path -LiteralPath $LogPath) { Get-Content -LiteralPath $LogPath | Write-Host } + Write-Host "::endgroup::" +} + +function Invoke-RefInstaller { + param([string]$Ref, [string]$Label, [switch]$IncludeDesktop) + $script = Join-Path $WorkRoot "install-$Label.ps1" + (Invoke-Git @("-C", $RepoRoot, "show", "$Ref`:scripts/install.ps1")) -join "`n" | + Set-Content -LiteralPath $script -Encoding UTF8 + $flags = @("-SkipSetup", "-HermesHome", $HermesHome, "-InstallDir", $InstallDir) + $text = Get-Content -LiteralPath $script -Raw + if ($text -match '\$NonInteractive') { $flags += "-NonInteractive" } + if ($IncludeDesktop) { + # The desktop stage is the point of this leg: a ref without the + # parameter is a hard failure, not a silent plain install. + if ($text -notmatch '\$IncludeDesktop') { + throw "E2E ASSERTION FAILED: ref $Ref does not support -IncludeDesktop; this leg cannot mean what it claims" + } + $flags += "-IncludeDesktop" + } + New-Item -ItemType Directory -Path (Join-Path $WorkRoot "logs") -Force | Out-Null + $log = Join-Path $WorkRoot "logs\install-$Label.log" + $prevEap = $ErrorActionPreference; $ErrorActionPreference = "Continue" + & powershell -NoProfile -ExecutionPolicy Bypass -File $script @flags *> $log + $installExit = $LASTEXITCODE + $ErrorActionPreference = $prevEap + Write-LogGroup "install.ps1 ($Label) transcript" $log + Assert-True ($installExit -eq 0) "install.ps1 ($Label) exited 0" +} + +function Assert-DesktopArtifact([string]$Label) { + Assert-True ($null -ne (Get-DesktopExe)) "$Label -- desktop app built by installer under apps\desktop\release" +} + +function Invoke-HermesUpdate { + # The venv updater. --yes reaches the update subcommand only in later + # releases; ask the installed binary, never parse its source. + $hermesExe = Join-Path $InstallDir "venv\Scripts\hermes.exe" + $updateArgs = @("update") + $prevEap = $ErrorActionPreference; $ErrorActionPreference = "Continue" + $helpText = & $hermesExe update --help 2>&1 | Out-String + if ($helpText -match '--yes') { $updateArgs += "--yes" } + New-Item -ItemType Directory -Path (Join-Path $WorkRoot "logs") -Force | Out-Null + $log = Join-Path $WorkRoot "logs\update.log" + Push-Location $InstallDir + try { + & $hermesExe @updateArgs *> $log + $updateExit = $LASTEXITCODE + } finally { + Pop-Location + $ErrorActionPreference = $prevEap + } + Write-LogGroup "hermes update transcript" $log + Assert-True ($updateExit -eq 0) "hermes update exited 0" +} + +function Invoke-HermesDesktopAppUpdate([string]$TargetSha) { + # The hermes-desktop launch surface: `hermes desktop` runs its whole + # real pipeline; the driver intercepts the product's final spawn + # (argv/cwd/env captured by e2e-assets/launch-capture/sitecustomize.py) + # and re-executes it under Playwright, which clicks Update now. + $hermesExe = Join-Path $InstallDir "venv\Scripts\hermes.exe" + $spec = Join-Path $WorkRoot "launch-spec.json" + New-Item -ItemType Directory -Path (Join-Path $WorkRoot "logs") -Force | Out-Null + $log = Join-Path $WorkRoot "logs\desktop-launch-capture.log" + + $capDir = Join-Path $AssetsDir "launch-capture" + $prevPy = $env:PYTHONPATH + $prevCap = $env:HERMES_E2E_CAPTURE_LAUNCH + $env:PYTHONPATH = if ($prevPy) { "$capDir;$prevPy" } else { $capDir } + $env:HERMES_E2E_CAPTURE_LAUNCH = $spec + $prevEap = $ErrorActionPreference; $ErrorActionPreference = "Continue" + Push-Location $InstallDir + try { + & $hermesExe desktop *> $log + $capExit = $LASTEXITCODE + } finally { + Pop-Location + $ErrorActionPreference = $prevEap + $env:PYTHONPATH = $prevPy + $env:HERMES_E2E_CAPTURE_LAUNCH = $prevCap + } + Write-LogGroup "hermes desktop (launch capture) transcript" $log + Assert-True ($capExit -eq 0) "hermes desktop exited 0 during launch capture" + Assert-True (Test-Path -LiteralPath "$spec.captured") "a launch was actually captured (exit 0 without a launch must not pass)" + + $node = Get-ManagedNode + $driverDir = Join-Path $WorkRoot "pw-driver" + New-Item -ItemType Directory -Path $driverDir -Force | Out-Null + $npmCli = Join-Path (Split-Path -Parent $node) "node_modules\npm\bin\npm-cli.js" + Assert-True (Test-Path -LiteralPath $npmCli) "managed npm exists beside the managed node" + Push-Location $driverDir + try { + & $node $npmCli install --no-save --no-audit --no-fund "@playwright/test@$PlaywrightVersion" 2>&1 | + Select-Object -Last 5 | ForEach-Object { Write-Host " npm| $_" } + $npmExit = $LASTEXITCODE + } finally { + Pop-Location + } + Assert-True ($npmExit -eq 0) "npm install @playwright/test@$PlaywrightVersion into the driver dir" + + Copy-Item (Join-Path $AssetsDir "launch-from-spec.mjs") (Join-Path $driverDir "launch-from-spec.mjs") -Force + $prevEap = $ErrorActionPreference; $ErrorActionPreference = "Continue" + Push-Location $driverDir + try { + & $node "launch-from-spec.mjs" --spec $spec ` + --result (Join-Path $HermesHome ".hermes-update-result.json") ` + --expect-sha $TargetSha --repo-dir $InstallDir 2>&1 | + ForEach-Object { Write-Host " pw| $_" } + $driveExit = $LASTEXITCODE + } finally { + Pop-Location + $ErrorActionPreference = $prevEap + } + Assert-True ($driveExit -eq 0) "app driven via captured hermes desktop spec; update completed" +} + function Save-DesktopScreenshot([string]$OutFile) { # Single full-desktop screenshot (primary screen). try { @@ -633,23 +766,70 @@ function Invoke-GuiUpdateDesktopRoute([string]$TargetSha) { } } -function Invoke-PhaseUpdateGui { +function Invoke-PhaseInstall { + # Dispatch on the install axis. Each arm ends with the same contract: + # checkout at OLD, hermes runs, and state carries how OLD landed so any + # update arm can follow any install arm. $state = Read-State + # Isolated install target for every arm; serve.git's file:// origin + # looks like a fork to the updater, whose "add the official repo as + # upstream?" prompt would hang a headless run - the marker is the + # product's own suppression mechanism. + $env:HERMES_HOME = $HermesHome + New-Item -ItemType Directory -Path $HermesHome -Force | Out-Null + New-Item -ItemType File -Path (Join-Path $HermesHome ".skip_upstream_prompt") -Force | Out-Null + switch ($InstallMethod) { + "desktop-installer@latest" { + Invoke-PhaseInstallGui + } + "installer-script" { + Write-Step "INSTALL (script): OLD's own install.ps1, headless" + Invoke-RefInstaller $state.old "old" + Assert-True ((Get-InstalledHead) -eq $state.old) "installed checkout is at OLD" + Test-HermesRuns "post-install-script" + } + "installer-script+desktop" { + Write-Step "INSTALL (script+desktop): OLD's own install.ps1 -IncludeDesktop, headless" + Invoke-RefInstaller $state.old "old" -IncludeDesktop + Assert-True ((Get-InstalledHead) -eq $state.old) "installed checkout is at OLD" + Test-HermesRuns "post-install-script-desktop" + Assert-DesktopArtifact "OLD" + } + } +} + +function Invoke-PhaseUpdate { + $state = Read-State + $env:HERMES_HOME = $HermesHome + + # The update becomes available the way it does for a real user: the + # remote's main moves forward. The GUI route re-advances harmlessly + # (same sha); script routes need it here because only the GUI arm's + # helper used to own this step. + Invoke-Git @("-C", $ServeRepo, "update-ref", "refs/heads/main", $state.current) | Out-Null + Write-Host " serve.git main advanced to $($state.current)" + switch ($Route) { "open-app-update" { + # Meaningful only where an OS entry point exists - install.ps1 + # -IncludeDesktop registers shortcuts too, so both desktop- + # bearing installs qualify; the workflow gate enforces which + # pairs are dispatched. Invoke-GuiUpdateDesktopRoute $state.current } "hermes-desktop-app-update" { - # TODO: launch the app via `hermes desktop` (spawn interception - # captures the real argv/cwd/env; Playwright launches from the - # captured spec), then the same Update-now click. - throw "update method 'hermes-desktop-app-update' is not implemented yet" + Invoke-HermesDesktopAppUpdate $state.current } "hermes-update" { - # TODO: run `hermes update` from the installed venv -- the CLI - # route. Needs the same completion/sha asserts minus the - # app-quit dance. - throw "update method 'hermes-update' is not implemented yet" + Invoke-HermesUpdate + } + "installer-script" { + # A user re-running the one-liner today gets the CURRENT script. + Invoke-RefInstaller $state.current "head" + } + "installer-script+desktop" { + Invoke-RefInstaller $state.current "head" -IncludeDesktop + Assert-DesktopArtifact "HEAD" } "desktop-installer@latest" { # TODO: re-run the bootstrap Hermes-Setup.exe over the existing @@ -657,19 +837,18 @@ function Invoke-PhaseUpdateGui { # runs unattended). throw "update method 'desktop-installer@latest' is not implemented yet" } - "installer-script" { - # TODO: re-run the irm | iex one-liner over the existing - # install. - throw "update method 'installer-script' is not implemented yet" - } } + + Assert-True ((Get-InstalledHead) -eq $state.current) "checkout landed on HEAD" + Test-HermesRuns "post-update" } # ---------------------------------------------------------------------------- # Dispatch # ---------------------------------------------------------------------------- -Write-Host "Windows Desktop GUI E2E driver (real user flow)" +Write-Host "Windows install/update E2E driver (real user flows)" Write-Host " phase: $Phase" +Write-Host " install: $InstallMethod" Write-Host " route: $Route" Write-Host " repo: $RepoRoot" Write-Host " workroot: $WorkRoot" @@ -677,13 +856,13 @@ Write-Host " workroot: $WorkRoot" Set-GitRedirect switch ($Phase) { - "stage" { Invoke-PhaseStage } - "install-gui" { Invoke-PhaseInstallGui } - "update-gui" { Invoke-PhaseUpdateGui } + "stage" { Invoke-PhaseStage } + "install" { Invoke-PhaseInstall } + "update" { Invoke-PhaseUpdate } "all" { Invoke-PhaseStage - Invoke-PhaseInstallGui - Invoke-PhaseUpdateGui + Invoke-PhaseInstall + Invoke-PhaseUpdate } } diff --git a/tests/install/windows-installer-script-e2e.ps1 b/tests/install/windows-installer-script-e2e.ps1 deleted file mode 100644 index 53fdf4cca892..000000000000 --- a/tests/install/windows-installer-script-e2e.ps1 +++ /dev/null @@ -1,289 +0,0 @@ -# Prove a Windows user who installed OLD via the installer script (irm | -# iex) can reach HEAD. -# -# The install.ps1 sibling of tests/install/installer-script-e2e.sh, sharing -# its staging trick with the GUI driver (windows-desktop-gui-e2e.ps1): -# every git process is pointed at a local bare clone with -# url..insteadOf rewrites for both canonical repo URLs in -# a driver-owned GIT_CONFIG_GLOBAL. The installer and updater run -# byte-for-byte against their real URLs and land on serve.git; `main` -# serves OLD during the install, then advances to HEAD for the update leg. -# (NOT GIT_CONFIG_COUNT/KEY_n/VALUE_n env config -- install.ps1 SETS those -# itself and would clobber ours.) -# -# install.ps1 itself is not downloaded: the install leg runs the copy -# shipped AT the OLD ref (what a user who installed then actually -# executed), and the installer-script update leg runs HEAD's copy (what -# the website serves at update time). -# -# Usage: -# powershell -File tests\install\windows-installer-script-e2e.ps1 ` -# -UpdateMethod hermes-update -InstallRef v0.20.2 -# -# -UpdateMethod hermes-update venv\Scripts\hermes.exe update -# installer-script re-run install.ps1 (HEAD's copy) -# -InstallRef what to install first; anything git resolves. auto = -# the newest release tag in the checkout. -# -# Requires a clean full-history checkout with release tags fetched. -# PowerShell 5.1-safe, pure ASCII (OEM codepages explode on fancy dashes). - -#Requires -Version 5.1 - -param( - [ValidateSet("installer-script", "installer-script+desktop")] - [string]$InstallMethod = "installer-script", - [ValidateSet("hermes-update", "installer-script", "installer-script+desktop")] - [string]$UpdateMethod = "hermes-update", - [string]$InstallRef = "auto" -) - -$ErrorActionPreference = "Stop" - -$RepoRoot = (Resolve-Path (Join-Path $PSScriptRoot "..\..")).Path -$RepoUrlSsh = "git@github.com:NousResearch/hermes-agent.git" -$RepoUrlHttps = "https://github.com/NousResearch/hermes-agent.git" - -# Everything lives OUTSIDE the checkout; an untracked dir inside the repo -# would trip the dirty-tree guard below on the next run. -$WorkRoot = Join-Path $(if ($env:RUNNER_TEMP) { $env:RUNNER_TEMP } else { $env:TEMP }) "hermes-installer-script-e2e" -$LogDir = if ($env:HERMES_E2E_LOG_DIR) { $env:HERMES_E2E_LOG_DIR } else { Join-Path $WorkRoot "logs" } -$ServeRepo = Join-Path $WorkRoot "serve.git" - -function Step([string]$Message) { Write-Host "`n=== $Message ===" } -function Ok([string]$Message) { Write-Host " OK $Message" } -function Fail([string]$Message) { - Write-Host "E2E ASSERTION FAILED: $Message" -ForegroundColor Red - exit 1 -} -# Full transcript in the job log, collapsed (GitHub renders ::group:: as a -# fold). Win or lose -- a green install's log is how you diagnose the leg -# that fails next. -function Write-LogGroup([string]$Title, [string]$LogPath) { - Write-Host "::group::$Title" - if (Test-Path -LiteralPath $LogPath) { Get-Content -LiteralPath $LogPath | Write-Host } - Write-Host "::endgroup::" -} - -function Invoke-Git { - param([string[]]$GitArgs) - $out = & git @GitArgs 2>&1 - if ($LASTEXITCODE -ne 0) { - Fail "git $($GitArgs -join ' ') exited $LASTEXITCODE`: $out" - } - return $out -} - -if (Test-Path -LiteralPath $WorkRoot) { Remove-Item -Recurse -Force -LiteralPath $WorkRoot } -New-Item -ItemType Directory -Path $WorkRoot, $LogDir -Force | Out-Null - -# --- stage: serve.git with main parked at OLD -------------------------------- - -Step "staging serve.git (main -> OLD)" -# Tracked changes only (-uno): untracked files cannot leak into a bare clone. -$dirty = Invoke-Git @("-C", $RepoRoot, "status", "--porcelain", "-uno") -if ($dirty) { Fail "checkout has uncommitted tracked changes; the staged clone must be a reviewable commit" } - -if ($InstallRef -eq "auto") { - $tags = (Invoke-Git @("-C", $RepoRoot, "tag", "--list", "v[0-9]*", "--sort=-creatordate")) -split "`r?`n" - $InstallRef = $tags | Select-Object -First 1 - if (-not $InstallRef) { Fail "no release tags in the checkout to use as OLD" } -} -$OldSha = (Invoke-Git @("-C", $RepoRoot, "rev-parse", "$InstallRef^{commit}")).Trim() -$HeadSha = (Invoke-Git @("-C", $RepoRoot, "rev-parse", "HEAD")).Trim() -if ($OldSha -eq $HeadSha) { Fail "OLD ($InstallRef) IS HEAD; no update would be available" } - -Invoke-Git @("clone", "--bare", "--quiet", $RepoRoot, $ServeRepo) | Out-Null -Invoke-Git @("-C", $ServeRepo, "update-ref", "refs/heads/main", $OldSha) | Out-Null -Invoke-Git @("-C", $ServeRepo, "symbolic-ref", "HEAD", "refs/heads/main") | Out-Null -# The installer may pin a commit that is reachable but not at a ref tip. -Invoke-Git @("-C", $ServeRepo, "config", "uploadpack.allowAnySHA1InWant", "true") | Out-Null -Ok "serve.git main = $OldSha ($InstallRef), update target $HeadSha" - -# --- the git URL redirect ------------------------------------------------------ - -$gitCfg = Join-Path $WorkRoot "gitconfig" -$serveUrl = "file:///" + ($ServeRepo -replace "\\", "/") -@" -[url "$serveUrl"] - insteadOf = $RepoUrlHttps - insteadOf = $RepoUrlSsh -"@ | Set-Content -LiteralPath $gitCfg -Encoding Ascii -$env:GIT_CONFIG_GLOBAL = $gitCfg -Ok "git URL redirect via GIT_CONFIG_GLOBAL=$gitCfg" - -# Isolated install target: the runner may carry a preinstalled hermes. -# -HermesHome/-InstallDir are passed explicitly because the oldest sampled -# installers predate the HERMES_HOME env override. -$HermesHome = Join-Path $WorkRoot "hermes-home" -$InstallDir = Join-Path $HermesHome "hermes-agent" -New-Item -ItemType Directory -Path $HermesHome -Force | Out-Null -$env:HERMES_HOME = $HermesHome -# serve.git's file:// origin looks like a fork to the updater, whose "add -# the official repo as upstream?" prompt would hang a headless run. This -# marker is the product's own mechanism for suppressing it. -Set-Content -LiteralPath (Join-Path $HermesHome ".skip_upstream_prompt") -Value "" -Encoding Ascii - -function Invoke-Installer { - param([string]$Ref, [string]$Label, [switch]$IncludeDesktop) - $script = Join-Path $WorkRoot "install-$Label.ps1" - (Invoke-Git @("-C", $RepoRoot, "show", "$Ref`:scripts/install.ps1")) -join "`n" | - Set-Content -LiteralPath $script -Encoding UTF8 - # Flags must match the installer being run, not this checkout's: older - # releases reject parameters added later. -SkipSetup/-HermesHome/ - # -InstallDir go back further than any tag we sample; -NonInteractive - # is probed from the ref's own script text. - $flags = @("-SkipSetup", "-HermesHome", $HermesHome, "-InstallDir", $InstallDir) - $text = Get-Content -LiteralPath $script -Raw - if ($text -match '\$NonInteractive') { $flags += "-NonInteractive" } - if ($IncludeDesktop) { - # The desktop stage is the point of this leg, so a ref without the - # parameter is a hard failure, not a silent downgrade to a plain - # install. (Pre-desktop releases are already skipped upstream by - # the tag-has-desktop gate; the parameter shipped with the app.) - if ($text -notmatch '\$IncludeDesktop') { - Fail "ref $Ref does not support -IncludeDesktop; this leg cannot mean what it claims" - } - $flags += "-IncludeDesktop" - } - $log = Join-Path $LogDir "install-$Label.log" - # Native stderr (git clone progress, pip notices) must not become - # terminating NativeCommandErrors under EAP=Stop; the exit code is the - # verdict here, not stderr chatter. - $prevEap = $ErrorActionPreference; $ErrorActionPreference = "Continue" - & powershell -NoProfile -ExecutionPolicy Bypass -File $script @flags *> $log - $installExit = $LASTEXITCODE - $ErrorActionPreference = $prevEap - Write-LogGroup "install.ps1 ($Label) transcript" $log - if ($installExit -ne 0) { - Fail "install.ps1 ($Label) exited $installExit; transcript above, log at $log" - } -} - -function Assert-Checkout { - param([string]$ExpectedSha, [string]$Label) - $got = (Invoke-Git @("-C", $InstallDir, "rev-parse", "HEAD")).Trim() - if ($got -ne $ExpectedSha) { Fail "installed checkout is $got, expected $Label ($ExpectedSha)" } - Ok "checkout is $Label ($ExpectedSha)" - $hermes = Join-Path $InstallDir "venv\Scripts\hermes.exe" - if (-not (Test-Path -LiteralPath $hermes)) { Fail "no hermes console script at $hermes" } - $verLog = Join-Path $LogDir "version-$Label.log" - $prevEap = $ErrorActionPreference; $ErrorActionPreference = "Continue" - & $hermes --version *> $verLog - $verExit = $LASTEXITCODE - $ErrorActionPreference = $prevEap - if ($verExit -ne 0) { - Get-Content -LiteralPath $verLog | Write-Host - Fail "hermes --version failed after $Label; log in $verLog" - } - Ok "hermes --version works: $((Get-Content -LiteralPath $verLog -First 1))" -} - -function Assert-DesktopArtifact { - param([string]$Label) - # After a +desktop install the built app must exist under the checkout; - # the installer also registers Start Menu / Desktop shortcuts, but the - # artifact is the ground truth a headless job can check. - $exe = Join-Path $InstallDir "apps\desktop\release\win-unpacked\Hermes.exe" - if (-not (Test-Path -LiteralPath $exe)) { - Fail "no desktop app at $exe after $Label (+desktop install)" - } - Ok "desktop app built by installer at ${Label}: $exe" -} - -function Test-DesktopSmoke { - param([string]$Label) - # Prove the installed CLI can produce the desktop app: `hermes desktop - # --build-only` runs the full desktop pipeline (workspace install, - # renderer build, stamp write) and stops before the launch -- the same - # call `hermes update` itself makes. Probe the INSTALLED hermes for the - # flag rather than assuming this checkout's surface: sampled OLD - # releases may predate `hermes desktop` or --build-only entirely, and - # for them the phase skips, loudly. - $hermes = Join-Path $InstallDir "venv\Scripts\hermes.exe" - $prevEap = $ErrorActionPreference; $ErrorActionPreference = "Continue" - $helpText = & $hermes desktop --help 2>&1 | Out-String - $ErrorActionPreference = $prevEap - if ($helpText -notmatch '--build-only') { - Ok "hermes desktop --build-only not supported at $Label; skipping desktop smoke" - return - } - $log = Join-Path $LogDir "desktop-smoke-$Label.log" - $prevEap = $ErrorActionPreference; $ErrorActionPreference = "Continue" - Push-Location $InstallDir - try { - & $hermes desktop --build-only *> $log - $smokeExit = $LASTEXITCODE - } finally { - Pop-Location - $ErrorActionPreference = $prevEap - } - Write-LogGroup "hermes desktop --build-only ($Label) transcript" $log - if ($smokeExit -ne 0) { - Fail "hermes desktop --build-only ($Label) exited $smokeExit; transcript above" - } - Ok "hermes desktop --build-only works at $Label" - # TODO(launch): LAUNCH the built app and auto-close it. Mechanism when - # the pieces land: driver-side spawn interception (a sitecustomize.py - # on PYTHONPATH wraps subprocess.run under an env-var opt-in and - # captures the real argv/cwd/env at the spawn site) + Playwright - # _electron.launch on the captured spec; electronApp.close() is the - # auto-close. -} - -# --- install OLD ------------------------------------------------------------------ - -Step "installing OLD ($InstallRef) via its own scripts/install.ps1 ($InstallMethod)" -if ($InstallMethod -eq "installer-script+desktop") { - Invoke-Installer $OldSha "old" -IncludeDesktop - Assert-Checkout $OldSha "OLD" - Assert-DesktopArtifact "OLD" -} else { - Invoke-Installer $OldSha "old" - Assert-Checkout $OldSha "OLD" -} -Test-DesktopSmoke "old" - -# --- update OLD -> HEAD -------------------------------------------------------------- - -Step "advancing served main to HEAD" -Invoke-Git @("-C", $ServeRepo, "update-ref", "refs/heads/main", $HeadSha) | Out-Null -Ok "serve.git main = $HeadSha" - -Step "updating via $UpdateMethod" -switch ($UpdateMethod) { - "hermes-update" { - $hermes = Join-Path $InstallDir "venv\Scripts\hermes.exe" - # `--yes` reaches the update subcommand only in later releases, and - # argparse rejects the whole invocation when it does not exist. - $updateArgs = @("update") - $prevEap = $ErrorActionPreference; $ErrorActionPreference = "Continue" - $helpText = & $hermes update --help 2>&1 | Out-String - if ($helpText -match '--yes') { $updateArgs += "--yes" } - $log = Join-Path $LogDir "update.log" - Push-Location $InstallDir - try { - & $hermes @updateArgs *> $log - $updateExit = $LASTEXITCODE - } finally { - Pop-Location - $ErrorActionPreference = $prevEap - } - Write-LogGroup "hermes update transcript" $log - if ($updateExit -ne 0) { - Fail "hermes update exited $updateExit; transcript above, log at $log" - } - } - "installer-script" { - # A user re-running the one-liner today gets the CURRENT script. - Invoke-Installer $HeadSha "head" - } - "installer-script+desktop" { - Invoke-Installer $HeadSha "head" -IncludeDesktop - Assert-DesktopArtifact "HEAD" - } -} -Assert-Checkout $HeadSha "HEAD" -Test-DesktopSmoke "head" - -Step "PASS: $InstallRef -> HEAD via $UpdateMethod" From 2b39b885d6ca048fb9b18ec9edd27d28b9a32e51 Mon Sep 17 00:00:00 2001 From: ethernet Date: Wed, 12 Aug 2026 04:36:03 -0400 Subject: [PATCH 073/227] test(install-e2e): macos desktop-installer arm - the published dmg, driven for real macos gains the desktop-installer@latest install method: the website's Hermes-Setup.dmg (verified live), mounted with hdiutil and its app binary run DIRECTLY - an open-launched app inherits none of the git redirect env, so direct exec is what keeps the isolation honest while staying the same binary and first-launch flow. install-e2e-macos-run.yml takes the windows shape: one workflow, one inner job per driver arm, native skips. Arm 1 delegates script installs to the shared OS-agnostic run workflow; arm 2 stages, installs from the dmg, and drives both app-update methods through launch-from-spec.mjs - open-app-update launches the installed .app (the double-click surface, env via Playwright), hermes-desktop-app-update captures the product's own hermes desktop spawn. Both end on sha asserts, never version strings. --- .github/workflows/install-e2e-macos-run.yml | 149 +++++++++++ .github/workflows/install-e2e.yml | 12 +- scripts/sandbox/generate-e2e-matrix.mjs | 2 + tests/install/README.md | 2 +- tests/install/macos-desktop-e2e.sh | 275 ++++++++++++++++++++ 5 files changed, 434 insertions(+), 6 deletions(-) create mode 100644 .github/workflows/install-e2e-macos-run.yml create mode 100755 tests/install/macos-desktop-e2e.sh diff --git a/.github/workflows/install-e2e-macos-run.yml b/.github/workflows/install-e2e-macos-run.yml new file mode 100644 index 000000000000..3e03740e2497 --- /dev/null +++ b/.github/workflows/install-e2e-macos-run.yml @@ -0,0 +1,149 @@ +# Reusable runner for ONE macOS install/update combination. +# +# Two driver arms, one runs per dispatch (the other natively skips): +# +# e2e (script arms) tests/install/installer-script-e2e.sh - the +# OS-agnostic git-redirect driver shared with +# linux. installer-script(+desktop) installs, +# script/updater/hermes-desktop-app-update +# updates. +# gui-e2e (desktop arm) tests/install/macos-desktop-e2e.sh - the +# published Hermes-Setup.dmg, mounted and run, +# then the app driven by Playwright for the +# app-update methods. +# +# Method pairs without a driver arm yet NATIVELY SKIP (grey check, no +# runner): the capability knowledge lives here, next to the drivers. + +name: install-e2e macos leg + +on: + workflow_call: + inputs: + install-method: + description: 'How OLD gets installed. Supported: installer-script, installer-script+desktop (curl | bash one-liner, optionally with --include-desktop) and desktop-installer@latest (the published Hermes-Setup.dmg).' + required: true + type: string + update-method: + description: 'How the install updates to HEAD. Script installs support hermes-update / installer-script / installer-script+desktop / hermes-desktop-app-update; dmg installs support open-app-update / hermes-desktop-app-update.' + required: true + type: string + install-ref: + description: 'What to install before updating: a branch, a tag, or a SHA reachable from main.' + required: false + type: string + default: refs/heads/main + tag-has-desktop: + description: "Whether install-ref ships the desktop app (apps/desktop). Desktop-method legs from pre-desktop releases natively skip." + required: false + type: boolean + default: true + dmg-url: + description: 'Bootstrap dmg to install OLD with. Default: the latest published one — what a user downloads today.' + required: false + type: string + default: https://hermes-assets.nousresearch.com/Hermes-Setup.dmg + timeout-minutes: + description: 'Job timeout. App-update legs do a full Electron build.' + required: false + type: number + default: 120 + +permissions: + contents: read + +jobs: + # ---- arm 1: script installs (the shared OS-agnostic driver) -------------- + e2e: + name: install & update + if: >- + (inputs.install-method == 'installer-script' + || (inputs.install-method == 'installer-script+desktop' && inputs.tag-has-desktop)) + && (contains(fromJSON('["hermes-update", "installer-script"]'), inputs.update-method) + || (contains(fromJSON('["installer-script+desktop", "hermes-desktop-app-update"]'), inputs.update-method) && inputs.tag-has-desktop)) + uses: ./.github/workflows/install-e2e-run.yml + with: + install-method: ${{ inputs.install-method }} + update-method: ${{ inputs.update-method }} + install-ref: ${{ inputs.install-ref }} + tag-has-desktop: ${{ inputs.tag-has-desktop }} + runner: macos-latest + timeout-minutes: ${{ inputs.timeout-minutes }} + + # ---- arm 2: the published dmg, then Playwright drives the app ------------ + gui-e2e: + # Short static name on purpose: name expressions render UNEXPANDED on + # skipped jobs. + name: Hermes-Setup.dmg + if: >- + inputs.install-method == 'desktop-installer@latest' && inputs.tag-has-desktop + && contains(fromJSON('["open-app-update", "hermes-desktop-app-update"]'), inputs.update-method) + runs-on: macos-latest + timeout-minutes: ${{ inputs.timeout-minutes }} + + steps: + # Full history: the driver bare-clones this checkout as the repo the + # installer/updater talk to. + - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2 + with: + fetch-depth: 0 + + - name: Start screen recording + uses: ./.github/actions/e2e-screen-record + with: + mode: start + output: ${{ runner.temp }}/e2e-logs/recording.mkv + + - name: Stage serve repo (main -> ${{ inputs.install-ref }}) + run: | + set -euo pipefail + tests/install/macos-desktop-e2e.sh --phase stage \ + --update-method '${{ inputs.update-method }}' \ + --install-ref '${{ inputs.install-ref }}' \ + --dmg-url '${{ inputs.dmg-url }}' + env: + HERMES_E2E_LOG_DIR: ${{ runner.temp }}/e2e-logs + + - name: Install ${{ inputs.install-ref }} via Hermes-Setup.dmg + run: | + set -euo pipefail + tests/install/macos-desktop-e2e.sh --phase install \ + --update-method '${{ inputs.update-method }}' \ + --install-ref '${{ inputs.install-ref }}' \ + --dmg-url '${{ inputs.dmg-url }}' + env: + HERMES_E2E_LOG_DIR: ${{ runner.temp }}/e2e-logs + + - name: Update ${{ inputs.install-ref }} -> HEAD (${{ inputs.update-method }}) + run: | + set -euo pipefail + tests/install/macos-desktop-e2e.sh --phase update \ + --update-method '${{ inputs.update-method }}' \ + --install-ref '${{ inputs.install-ref }}' \ + --dmg-url '${{ inputs.dmg-url }}' + env: + HERMES_E2E_LOG_DIR: ${{ runner.temp }}/e2e-logs + + - name: Stop screen recording + if: always() + uses: ./.github/actions/e2e-screen-record + with: + mode: stop + output: ${{ runner.temp }}/e2e-logs/recording.mkv + + # Artifact names cannot contain '/'; install-ref may be a full ref. + - name: Build artifact name + id: artifact + if: always() + run: | + safe_ref="$(printf '%s' '${{ inputs.install-ref }}' | tr '/:' '--')" + echo "name=install-e2e-macos-dmg-${{ inputs.update-method }}-${safe_ref}-${{ github.sha }}" >> "$GITHUB_OUTPUT" + + - name: Upload logs + if: always() + uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1 + with: + name: ${{ steps.artifact.outputs.name }} + path: ${{ runner.temp }}/e2e-logs + retention-days: 14 + if-no-files-found: ignore diff --git a/.github/workflows/install-e2e.yml b/.github/workflows/install-e2e.yml index 0bfe470d5459..1641fc95932b 100644 --- a/.github/workflows/install-e2e.yml +++ b/.github/workflows/install-e2e.yml @@ -13,8 +13,10 @@ name: Install & Update E2E # Matrix: windows the real desktop user flow: website Hermes-Setup.exe # clicked by AutoHotkey, update via the app, Playwright # clicking "Update now" (install-e2e-windows-run.yml) -# Matrix: macos the same OS-agnostic driver as linux, on macos-latest -# (install-e2e-run.yml; app-update variants are TODO) +# Matrix: macos script installs on the shared OS-agnostic driver, +# plus the real desktop user flow: website +# Hermes-Setup.dmg mounted and run, updates via the +# app under Playwright (install-e2e-macos-run.yml) # # Every combination is dispatched to its OS's run workflow; the run # workflow natively skips (grey) what its driver cannot run yet -- an @@ -186,14 +188,14 @@ jobs: # 10x-cost runners: keep concurrency low. max-parallel: 2 matrix: ${{ fromJSON(needs.generate-matrix.outputs.macos) }} - # The same OS-agnostic driver as linux -- only the runner differs. - uses: ./.github/workflows/install-e2e-run.yml + # Two driver arms: the OS-agnostic script driver (shared with linux) + # and the published-dmg GUI driver; the run workflow routes. + uses: ./.github/workflows/install-e2e-macos-run.yml with: install-method: ${{ matrix.install_method }} update-method: ${{ matrix.update_method }} install-ref: ${{ matrix.install_ref }} tag-has-desktop: ${{ matrix.tag_has_desktop }} - runner: macos-latest # The outcome, human-readable: the plan chart again, with each cell # replaced by how that leg actually concluded. Per-leg conclusions are diff --git a/scripts/sandbox/generate-e2e-matrix.mjs b/scripts/sandbox/generate-e2e-matrix.mjs index c50a4385460a..93be11ca1952 100644 --- a/scripts/sandbox/generate-e2e-matrix.mjs +++ b/scripts/sandbox/generate-e2e-matrix.mjs @@ -106,6 +106,8 @@ export const SPEC = { install: [ { method: 'installer-script' }, { method: 'installer-script+desktop' }, + // The published Hermes-Setup.dmg from the website, mounted and run. + { method: 'desktop-installer', versions: ['latest'] }, ], update: [ { method: 'installer-script' }, diff --git a/tests/install/README.md b/tests/install/README.md index c50c32444e64..fdc1c33e8bea 100644 --- a/tests/install/README.md +++ b/tests/install/README.md @@ -47,7 +47,7 @@ A leg can install a release from months back. The driver must not assume that th - `installer-script`: the platform's one-liner (`curl | bash` on linux and macos, `irm | iex` on windows). - `installer-script+desktop`: the same one-liner with its desktop stage opted in (`--include-desktop` / `-IncludeDesktop`). The stage builds the desktop app during the install. On windows it also registers Start Menu and Desktop shortcuts. On linux and macos it builds the app inside the checkout and registers no OS entry point. -- `desktop-installer@latest`: the published GUI installer (`Hermes-Setup.exe` on windows), clicked through the real window. +- `desktop-installer@latest`: the published GUI installer (`Hermes-Setup.exe` on windows, `Hermes-Setup.dmg` on macos), driven through the real user flow. ## The two app-update variants diff --git a/tests/install/macos-desktop-e2e.sh b/tests/install/macos-desktop-e2e.sh new file mode 100755 index 000000000000..51418a8f5d51 --- /dev/null +++ b/tests/install/macos-desktop-e2e.sh @@ -0,0 +1,275 @@ +#!/usr/bin/env bash +# Prove a macOS user who installed OLD via the published desktop installer +# (Hermes-Setup.dmg from the website) can reach HEAD. +# +# The macOS sibling of tests/install/windows-e2e.ps1's desktop-installer +# arm, sharing the staging trick: every git process is pointed at a local +# bare clone via url..insteadOf in a driver-owned +# GIT_CONFIG_GLOBAL. The published dmg carries no commit pin - it installs +# whatever `main` serves - so parking serve.git's main at OLD stages the +# "user on the current release" start, and advancing it to HEAD makes an +# update available exactly the way it does for a real user. +# +# Phases (state shared via the workroot, mirroring the windows driver): +# stage bare-clone this checkout to serve.git, park main at OLD +# install download the dmg, hdiutil attach, run the installer app's +# binary DIRECTLY (env inheritance: an `open`-launched app sees +# none of our redirect env), wait for the install to land +# update advance served main to HEAD, apply ONE update method: +# open-app-update launch the installed app binary +# under Playwright, click Update now +# hermes-desktop-app-update capture `hermes desktop`'s spawn, +# launch the spec under Playwright, +# click Update now +# +# Usage: +# tests/install/macos-desktop-e2e.sh --phase stage|install|update|all +# --update-method open-app-update|hermes-desktop-app-update +# [--install-ref REF] [--dmg-url URL] +# +# Requires a clean full-history checkout with release tags fetched, on a +# macOS host with a window server (the GitHub macos runners qualify). + +set -euo pipefail + +PHASE="all" +UPDATE_METHOD="" +INSTALL_REF="" +DMG_URL="https://hermes-assets.nousresearch.com/Hermes-Setup.dmg" +PLAYWRIGHT_VERSION="1.58.2" +while [ "$#" -gt 0 ]; do + case "$1" in + --phase) + [ "$#" -ge 2 ] || { echo 'error: --phase needs a value' >&2; exit 1; } + PHASE="$2"; shift 2 ;; + --update-method) + [ "$#" -ge 2 ] || { echo 'error: --update-method needs a value' >&2; exit 1; } + UPDATE_METHOD="$2"; shift 2 ;; + --install-ref) + [ "$#" -ge 2 ] || { echo 'error: --install-ref needs a value' >&2; exit 1; } + INSTALL_REF="$2"; shift 2 ;; + --dmg-url) + [ "$#" -ge 2 ] || { echo 'error: --dmg-url needs a value' >&2; exit 1; } + DMG_URL="$2"; shift 2 ;; + -h|--help) sed -n '2,32p' "$0"; exit 0 ;; + *) echo "error: unknown argument: $1" >&2; exit 1 ;; + esac +done +case "$UPDATE_METHOD" in + open-app-update|hermes-desktop-app-update) ;; + *) echo "error: --update-method must be open-app-update or hermes-desktop-app-update, got '$UPDATE_METHOD'" >&2; exit 1 ;; +esac +[ "$(uname -s)" = "Darwin" ] || { echo "error: this driver runs on macOS only" >&2; exit 1; } + +REPO_ROOT="$(cd "$(dirname "$0")/../.." && pwd)" +REPO_URL_SSH="git@github.com:NousResearch/hermes-agent.git" +REPO_URL_HTTPS="https://github.com/NousResearch/hermes-agent.git" +ASSETS="$REPO_ROOT/tests/install/e2e-assets" + +WORK_ROOT="${HERMES_E2E_WORKROOT:-${RUNNER_TEMP:-${TMPDIR:-/tmp}}/hermes-macos-desktop-e2e}" +LOG_DIR="${HERMES_E2E_LOG_DIR:-$WORK_ROOT/logs}" +SERVE_REPO="$WORK_ROOT/serve.git" +STATE="$WORK_ROOT/shas.env" +export HOME_SANDBOX="$WORK_ROOT/home" + +step() { printf '\n=== %s ===\n' "$*"; } +ok() { printf ' OK %s\n' "$*"; } +fail() { printf 'E2E ASSERTION FAILED: %s\n' "$*" >&2; exit 1; } +log_group() { + printf '::group::%s\n' "$1" + cat "$2" + printf '::endgroup::\n' +} + +# Every phase runs in its own process (separate CI steps), so the redirect +# env is re-established here, not inherited. +arm_redirect() { + GIT_CFG="$WORK_ROOT/gitconfig" + export GIT_CONFIG_GLOBAL="$GIT_CFG" + export HOME="$HOME_SANDBOX" + export PATH="$HOME/.local/bin:$PATH" + export HERMES_HOME="$HOME/.hermes" + export INSTALL_DIR="$HERMES_HOME/hermes-agent" +} + +phase_stage() { + step "staging serve.git (main -> OLD)" + [ -z "$(git -C "$REPO_ROOT" status --porcelain -uno)" ] \ + || fail "checkout has uncommitted tracked changes; the staged clone must be a reviewable commit" + + rm -rf "$WORK_ROOT" + mkdir -p "$WORK_ROOT" "$LOG_DIR" "$HOME_SANDBOX/.local/bin" + + local old_ref="$INSTALL_REF" + if [ -z "$old_ref" ] || [ "$old_ref" = "auto" ]; then + old_ref="$(git -C "$REPO_ROOT" tag --list 'v[0-9]*' --sort=-creatordate | head -1)" + [ -n "$old_ref" ] || fail "no release tags in the checkout to use as OLD" + fi + local old_sha head_sha + old_sha="$(git -C "$REPO_ROOT" rev-parse "${old_ref}^{commit}")" + head_sha="$(git -C "$REPO_ROOT" rev-parse HEAD)" + [ "$old_sha" != "$head_sha" ] || fail "OLD ($old_ref) IS HEAD; no update would be available" + + git clone --bare --quiet "$REPO_ROOT" "$SERVE_REPO" + git -C "$SERVE_REPO" update-ref refs/heads/main "$old_sha" + git -C "$SERVE_REPO" symbolic-ref HEAD refs/heads/main + git -C "$SERVE_REPO" config uploadpack.allowAnySHA1InWant true + + cat > "$WORK_ROOT/gitconfig" < "$STATE" + ok "serve.git main = $old_sha ($old_ref), update target $head_sha" +} + +find_installed_app() { + # The bootstrap installs the packaged app; look where the product puts it + # (the checkout's release dir), plus /Applications for a copied bundle. + local cand + for cand in \ + "$INSTALL_DIR/apps/desktop/release/mac-arm64/Hermes.app" \ + "$INSTALL_DIR/apps/desktop/release/mac/Hermes.app" \ + "/Applications/Hermes.app"; do + [ -d "$cand" ] && { printf '%s' "$cand"; return 0; } + done + return 1 +} + +phase_install() { + # shellcheck disable=SC1090 + . "$STATE" + arm_redirect + step "installing OLD ($OLD_REF) via the published Hermes-Setup.dmg" + + local dmg="$WORK_ROOT/Hermes-Setup.dmg" + [ -f "$dmg" ] || curl -fsSL -o "$dmg" "$DMG_URL" + [ "$(stat -f%z "$dmg")" -gt 1000000 ] || fail "dmg download too small: $(stat -f%z "$dmg") bytes" + # curl'd files carry no quarantine attr, but belt and braces on a runner. + xattr -dr com.apple.quarantine "$dmg" 2>/dev/null || true + + local mount + mount="$(hdiutil attach -nobrowse -readonly "$dmg" | awk -F'\t' '/\/Volumes\//{print $NF; exit}')" + [ -n "$mount" ] || fail "hdiutil attach produced no mount point" + ok "dmg mounted at $mount" + + local app_bin="" + local app + app="$(find "$mount" -maxdepth 1 -name '*.app' | head -1)" + [ -n "$app" ] || { hdiutil detach "$mount" >/dev/null 2>&1 || true; fail "no .app inside the dmg"; } + app_bin="$(find "$app/Contents/MacOS" -type f -perm +111 | head -1)" + [ -n "$app_bin" ] || fail "no executable inside $app/Contents/MacOS" + + # Run the installer binary DIRECTLY: `open` launches via launchd, which + # inherits NONE of the redirect env (GIT_CONFIG_GLOBAL, HOME) - the whole + # isolation would silently evaporate. Direct exec is the same binary and + # the same first-launch flow. + local rc=0 + "$app_bin" > "$LOG_DIR/bootstrap-install.log" 2>&1 || rc=$? + log_group "Hermes-Setup (dmg bootstrap) transcript" "$LOG_DIR/bootstrap-install.log" + hdiutil detach "$mount" >/dev/null 2>&1 || true + [ "$rc" -eq 0 ] || fail "dmg bootstrap exited $rc; transcript above" + + [ -d "$INSTALL_DIR/.git" ] || fail "no checkout landed at $INSTALL_DIR" + local got + got="$(git -C "$INSTALL_DIR" rev-parse HEAD)" + [ "$got" = "$OLD_SHA" ] || fail "installed checkout is $got, expected OLD ($OLD_SHA)" + ok "checkout is OLD ($OLD_SHA)" + local hermes="$INSTALL_DIR/venv/bin/hermes" + [ -x "$hermes" ] || fail "no hermes console script at $hermes" + "$hermes" --version > "$LOG_DIR/version-old.log" 2>&1 || fail "hermes --version failed after install" + ok "hermes --version works: $(head -c 120 "$LOG_DIR/version-old.log" | tr -d '\n')" + find_installed_app >/dev/null || fail "no installed Hermes.app after the dmg bootstrap" + ok "installed app: $(find_installed_app)" +} + +run_playwright_update() { + # $1: spec file to launch from. Installs the driver's OWN pinned + # @playwright/test into a scratch dir (never the installed tree's copy). + local spec="$1" + local pw_dir="$WORK_ROOT/playwright" + mkdir -p "$pw_dir" + (cd "$pw_dir" && npm install --no-save --no-audit --no-fund \ + "@playwright/test@$PLAYWRIGHT_VERSION" > "$LOG_DIR/playwright-install.log" 2>&1) \ + || { log_group "playwright install transcript" "$LOG_DIR/playwright-install.log"; fail "playwright install failed"; } + cp "$ASSETS/launch-from-spec.mjs" "$pw_dir/" + local rc=0 + (cd "$pw_dir" && node launch-from-spec.mjs \ + --spec "$spec" \ + --result "$HERMES_HOME/.hermes-update-result.json" \ + --expect-sha "$HEAD_SHA" \ + --repo-dir "$INSTALL_DIR" \ + > "$LOG_DIR/app-update.log" 2>&1) || rc=$? + log_group "app update (Playwright) transcript" "$LOG_DIR/app-update.log" + [ "$rc" -eq 0 ] || fail "app-driven update exited $rc; transcript above" +} + +phase_update() { + # shellcheck disable=SC1090 + . "$STATE" + arm_redirect + step "advancing served main to HEAD" + git -C "$SERVE_REPO" update-ref refs/heads/main "$HEAD_SHA" + ok "serve.git main = $HEAD_SHA" + + step "updating via $UPDATE_METHOD" + case "$UPDATE_METHOD" in + open-app-update) + # The installed app IS the user surface here (double-click the .app); + # hand-build the spec Playwright launches from. Env: the redirect set, + # which is exactly what the app's children (git, hermes update) need. + local app app_bin + app="$(find_installed_app)" || fail "no installed app to launch" + app_bin="$(find "$app/Contents/MacOS" -type f -perm +111 | head -1)" + python3 - "$app_bin" "$WORK_ROOT/launch-spec.json" <<'PYEOF' +import json, os, sys +spec = { + "argv": [sys.argv[1]], + "cwd": os.path.dirname(sys.argv[1]), + "env": dict(os.environ), + "matchedShape": "packaged", +} +with open(sys.argv[2], "w") as fh: + json.dump(spec, fh, indent=2) +PYEOF + run_playwright_update "$WORK_ROOT/launch-spec.json" + ;; + hermes-desktop-app-update) + # The product's own launch, captured at its spawn site. + local hermes="$INSTALL_DIR/venv/bin/hermes" + local spec="$WORK_ROOT/launch-spec.json" + local rc=0 + (cd "$INSTALL_DIR" && \ + PYTHONPATH="$ASSETS/launch-capture${PYTHONPATH:+:$PYTHONPATH}" \ + HERMES_E2E_CAPTURE_LAUNCH="$spec" \ + "$hermes" desktop < /dev/null > "$LOG_DIR/desktop-launch-capture.log" 2>&1) || rc=$? + log_group "hermes desktop (launch capture) transcript" "$LOG_DIR/desktop-launch-capture.log" + [ "$rc" -eq 0 ] || fail "hermes desktop exited $rc during launch capture" + [ -f "$spec.captured" ] || fail "hermes desktop exited 0 but no launch was captured" + ok "captured $(cat "$spec.captured") launch spec" + run_playwright_update "$spec" + ;; + esac + + local got + got="$(git -C "$INSTALL_DIR" rev-parse HEAD)" + [ "$got" = "$HEAD_SHA" ] || fail "checkout is $got, expected HEAD ($HEAD_SHA)" + ok "checkout landed on HEAD ($HEAD_SHA)" + "$INSTALL_DIR/venv/bin/hermes" --version > "$LOG_DIR/version-head.log" 2>&1 \ + || fail "hermes --version failed after update" + ok "hermes --version works post-update" + step "PASS: $OLD_REF -> HEAD via $UPDATE_METHOD" +} + +case "$PHASE" in + stage) phase_stage ;; + install) phase_install ;; + update) phase_update ;; + all) phase_stage; phase_install; phase_update ;; + *) echo "error: --phase must be stage, install, update or all" >&2; exit 1 ;; +esac From f4c4c17dc678c35fd6468854e53b659fca40561c Mon Sep 17 00:00:00 2001 From: ethernet Date: Wed, 12 Aug 2026 06:02:44 -0400 Subject: [PATCH 074/227] maybe make macos screencap work? --- tests/install/e2e-assets/record-start.sh | 19 +++++++++++++++++-- 1 file changed, 17 insertions(+), 2 deletions(-) diff --git a/tests/install/e2e-assets/record-start.sh b/tests/install/e2e-assets/record-start.sh index 6d6b8f97c409..a9bf9cd021da 100755 --- a/tests/install/e2e-assets/record-start.sh +++ b/tests/install/e2e-assets/record-start.sh @@ -34,9 +34,24 @@ if [ "${#INPUT[@]}" -eq 0 ]; then # avfoundation lists devices on stderr; the first "Capture screen" # index is the whole display. Parse it rather than hardcoding: the # index shifts with attached cameras. - screen_idx="$(ffmpeg -f avfoundation -list_devices true -i "" 2>&1 \ + # + # `-list_devices true -i ""` always exits non-zero. + # Capture output/status explicitly (with `|| true`) instead of letting a bare assignment + # trip `set -e`, which would abort before we ever get to report + # anything useful. + probe_out="$(ffmpeg -f avfoundation -list_devices true -i "" 2>&1 || true)" + screen_idx="$(printf '%s\n' "$probe_out" \ | sed -n 's/^\[AVFoundation[^]]*\] \[\([0-9]*\)\] Capture screen.*/\1/p' | head -1)" - [ -n "$screen_idx" ] || { echo "record-start: no capture screen device found" >&2; exit 1; } + + if [ -z "$screen_idx" ]; then + if printf '%s\n' "$probe_out" | grep -qiE 'Input/output error|Unknown input format|errno 5'; then + echo "record-start: avfoundation could not enumerate devices (I/O error)" >&2 + else + echo "record-start: no capture screen device found in avfoundation device list:" >&2 + fi + printf '%s\n' "$probe_out" >&2 + exit 1 + fi INPUT=(-f avfoundation -framerate 15 -capture_cursor 1 -i "${screen_idx}:none") ;; *) From 368164d7dcbd98527a37c89d54bdd490c058e440 Mon Sep 17 00:00:00 2001 From: ethernet Date: Wed, 12 Aug 2026 07:24:32 -0400 Subject: [PATCH 075/227] git shims for fork detection disablement --- tests/install/installer-script-e2e.sh | 73 ++++++++++++++++++++----- tests/install/macos-desktop-e2e.sh | 62 +++++++++++++++++++-- tests/install/windows-e2e.ps1 | 77 +++++++++++++++++++-------- 3 files changed, 170 insertions(+), 42 deletions(-) diff --git a/tests/install/installer-script-e2e.sh b/tests/install/installer-script-e2e.sh index 77be0c0f6e09..ec5e553c705c 100755 --- a/tests/install/installer-script-e2e.sh +++ b/tests/install/installer-script-e2e.sh @@ -122,18 +122,67 @@ git -C "$SERVE_REPO" symbolic-ref HEAD refs/heads/main git -C "$SERVE_REPO" config uploadpack.allowAnySHA1InWant true ok "serve.git main = $OLD_SHA ($INSTALL_REF), update target $HEAD_SHA" -# --- the git URL redirect ----------------------------------------------------- - -# A driver-owned global gitconfig, NOT GIT_CONFIG_COUNT/KEY_n/VALUE_n env -# config: install.sh sets those itself and would clobber ours. -GIT_CFG="$WORK_ROOT/gitconfig" -cat > "$GIT_CFG" < "$GIT_CFG" < "$SHIM_DIR/git" < $REAL_GIT (origin reports $REPO_URL_HTTPS)" +} + +# later, we might factor this out into a separate step like the macos desktop one. +arm_redirect # Isolated HOME: the runner's real one may carry a preinstalled hermes or a # developer config, and old installer scripts hardcode $HOME/.hermes (the @@ -144,10 +193,6 @@ mkdir -p "$HOME/.local/bin" export PATH="$HOME/.local/bin:$PATH" export HERMES_HOME="$HOME/.hermes" mkdir -p "$HERMES_HOME" -# serve.git's file:// origin looks like a fork to the updater, whose "add the -# official repo as upstream?" prompt would hang a headless run. This marker is -# the product's own mechanism for suppressing it. -touch "$HERMES_HOME/.skip_upstream_prompt" INSTALL_DIR="$HERMES_HOME/hermes-agent" diff --git a/tests/install/macos-desktop-e2e.sh b/tests/install/macos-desktop-e2e.sh index 51418a8f5d51..f52867bc6164 100755 --- a/tests/install/macos-desktop-e2e.sh +++ b/tests/install/macos-desktop-e2e.sh @@ -84,8 +84,65 @@ log_group() { # Every phase runs in its own process (separate CI steps), so the redirect # env is re-established here, not inherited. arm_redirect() { + # --- the git URL redirect ----------------------------------------------------- + + # we redirect to our own repo so we can play around with what commit hermes thinks we're on. + # A driver-owned global gitconfig, NOT GIT_CONFIG_COUNT/KEY_n/VALUE_n env + # config: install.sh sets those itself and would clobber ours. + actual_git_url="$(git -C "$REPO_ROOT" remote get-url origin)" GIT_CFG="$WORK_ROOT/gitconfig" + cat > "$GIT_CFG" < "$SHIM_DIR/git" < $REAL_GIT (origin reports $REPO_URL_HTTPS)" + + # ------- export HOME="$HOME_SANDBOX" export PATH="$HOME/.local/bin:$PATH" export HERMES_HOME="$HOME/.hermes" @@ -115,11 +172,6 @@ phase_stage() { git -C "$SERVE_REPO" symbolic-ref HEAD refs/heads/main git -C "$SERVE_REPO" config uploadpack.allowAnySHA1InWant true - cat > "$WORK_ROOT/gitconfig" < $fileUrl" + + + # shim git and make 'git remote get-url origin' report the actual HA upstream + + # insteadOf is transparent for transport but `git remote get-url origin` gives you the + # replacement, so _get_origin_url() sees file://$SERVE_REPO and _is_fork() would return true. + # we check for the arguments "remote get-url origin" in order in any position + # to allow for e.g. -c with some config being passed. + # if we didn't do this, we'd need the .skip_upstream_prompt file to prevent a hang in headless,"add the + # official repo as upstream?" prompt would hang a headless run. But we don't anymore :D + + $realGit = (Get-Command git.exe -ErrorAction Stop).Source + $shimDir = Join-Path $WorkRoot "shim" + New-Item -ItemType Directory -Path $shimDir -Force | Out-Null + $shimPath = Join-Path $shimDir "git.bat" + @" +@echo off +setlocal enabledelayedexpansion +set prev2= +set prev1= +:loop +if "%~1"=="" goto passthrough +if /I "!prev2!"=="remote" if /I "!prev1!"=="get-url" if /I "%~1"=="origin" ( + echo $RepoUrlHttps + exit /b 0 +) +set prev2=!prev1! +set prev1=%~1 +shift +goto loop +:passthrough +`"$realGit`" %* +exit /b %ERRORLEVEL% +"@ | Set-Content -LiteralPath $shimPath -Encoding ASCII + + $env:PATH = "$shimDir;$env:PATH" + + # check it worked + $observedGitUrl = Invoke-Git @("-C", $RepoRoot, "remote", "get-url", "origin") + Assert-True ($observedGitUrl -eq $RepoUrlHttps) "git remote get-url shim: origin resolves to '$observedGitUrl', expected '$RepoUrlHttps'" + Write-Host " git remote get-url shim: $shimPath -> $realGit" + Write-Host " 'remote get-url origin' now reports $RepoUrlHttps" } function Read-State { @@ -586,19 +631,6 @@ function Invoke-PhaseInstallGui { Add-Content -LiteralPath $envFile -Value "OPENROUTER_API_KEY=sk-or-...-key" } Write-Host " seeded placeholder provider key for the update leg" - - # Suppress the interactive "add upstream remote?" prompt during the GUI - # update leg. Our serve.git origin (file://) looks like a fork to - # `hermes update`, so `_sync_with_upstream_if_needed` would call bare - # input() -- which HANGS FOREVER when the Desktop spawns the hand-off via - # `cmd start /min` (a real but empty console; input() blocks waiting for a - # keystroke that never comes). Real GUI users on the official github - # origin never hit this path (_is_fork is false). The skip marker - # (.skip_upstream_prompt in HERMES_HOME) is the product's own mechanism - # for "don't ask about upstream", so setting it keeps the update flow - # faithful while avoiding the fork-only prompt. - New-Item -ItemType File -Path (Join-Path $HermesHome ".skip_upstream_prompt") -Force | Out-Null - Write-Host " set .skip_upstream_prompt (serve.git origin looks like a fork; avoids the fork-only input() hang)" } # ---------------------------------------------------------------------------- @@ -777,7 +809,6 @@ function Invoke-PhaseInstall { # product's own suppression mechanism. $env:HERMES_HOME = $HermesHome New-Item -ItemType Directory -Path $HermesHome -Force | Out-Null - New-Item -ItemType File -Path (Join-Path $HermesHome ".skip_upstream_prompt") -Force | Out-Null switch ($InstallMethod) { "desktop-installer@latest" { Invoke-PhaseInstallGui From 6fd7f05de48e9163f64f97a7facca4bdb4eee821 Mon Sep 17 00:00:00 2001 From: ethernet Date: Wed, 12 Aug 2026 07:40:20 -0400 Subject: [PATCH 076/227] fix macos screen record --- .github/actions/e2e-screen-record/action.yml | 8 ++++++++ 1 file changed, 8 insertions(+) diff --git a/.github/actions/e2e-screen-record/action.yml b/.github/actions/e2e-screen-record/action.yml index d4923d27afdc..2e947217533a 100644 --- a/.github/actions/e2e-screen-record/action.yml +++ b/.github/actions/e2e-screen-record/action.yml @@ -58,6 +58,14 @@ runs: # Do not trust "preinstalled" claims - verify, install on miss. command -v ffmpeg >/dev/null 2>&1 || brew install --quiet ffmpeg + # until new runner image is published by github, we have to hack on screen record approvals + # see hhttps://github.com/actions/runner-images/issues/14474 - as of this hermes agent commit, + # it's merged but the image isn't updated. + approvalsPlist="$HOME/Library/Group Containers/group.com.apple.replayd/ScreenCaptureApprovals.plist" + mkdir -p "$(dirname "$approvalsPlist")" + defaults write "$approvalsPlist" "/opt/hca/hosted-compute-agent" -date "3024-01-01 00:00:00 +0000" + killall cfprefsd 2>/dev/null || true + - name: Restore cached ffmpeg (windows) if: inputs.mode == 'start' && runner.os == 'Windows' id: ffmpeg-cache From 4578b5d0a65c7fed24336fe07352c14ed79ab902 Mon Sep 17 00:00:00 2001 From: ethernet Date: Wed, 12 Aug 2026 07:48:35 -0400 Subject: [PATCH 077/227] ci(install-e2e): result chart says WHY a cell skipped Skipped cells split into their reason: pre-desktop (a desktop-surface method against a tag that predates apps/desktop) vs TODO (declared, no driver arm yet). The report job passes pick-releases' annotated tags into --format results; the shared methodNeedsDesktop() is the same predicate the plan chart uses, so plan and results agree about what pre-desktop means. Without --tags the renderer keeps the flat skip label (backward compatible). Verified against run 31579084845 real job list: 82 legs, 44 skips labeled correctly. --- .github/workflows/install-e2e.yml | 7 +++- scripts/sandbox/generate-e2e-matrix.mjs | 51 ++++++++++++++++++++----- 2 files changed, 46 insertions(+), 12 deletions(-) diff --git a/.github/workflows/install-e2e.yml b/.github/workflows/install-e2e.yml index 1641fc95932b..714d10d3799f 100644 --- a/.github/workflows/install-e2e.yml +++ b/.github/workflows/install-e2e.yml @@ -206,7 +206,7 @@ jobs: report: name: Result chart if: always() - needs: [linux, windows, macos] + needs: [pick-releases, linux, windows, macos] runs-on: ubuntu-latest timeout-minutes: 5 steps: @@ -221,7 +221,10 @@ jobs: { echo "OS jobs: linux ${{ needs.linux.result }}, windows ${{ needs.windows.result }}, macos ${{ needs.macos.result }}" echo + # The tag annotations let the chart say WHY a cell skipped + # (pre-desktop vs declared TODO) instead of a flat "skip". gh api "repos/${{ github.repository }}/actions/runs/${{ github.run_id }}/jobs?per_page=100" \ --paginate --jq '.jobs[] | {name, conclusion}' | - node scripts/sandbox/generate-e2e-matrix.mjs --format results + node scripts/sandbox/generate-e2e-matrix.mjs --format results \ + --tags '${{ needs.pick-releases.outputs.tags }}' } >> "$GITHUB_STEP_SUMMARY" diff --git a/scripts/sandbox/generate-e2e-matrix.mjs b/scripts/sandbox/generate-e2e-matrix.mjs index 93be11ca1952..bba6a435df42 100644 --- a/scripts/sandbox/generate-e2e-matrix.mjs +++ b/scripts/sandbox/generate-e2e-matrix.mjs @@ -203,6 +203,18 @@ export function buildMatrices(envs, tags) { return byOs; } +/** + * Does this method id need the starting tag to ship the desktop app? + * Shared by the plan chart (pre-desktop cells) and the results chart + * (labeling WHY a skipped leg skipped). + * @param {string} m + * @returns {boolean} + */ +export function methodNeedsDesktop(m) { + return m.startsWith('desktop-installer') || m === 'installer-script+desktop' || + m === 'open-app-update' || m === 'hermes-desktop-app-update'; +} + /** * Render the plan as a markdown cross-table for $GITHUB_STEP_SUMMARY: * one row per {os, install -> update} combination, one column per @@ -216,9 +228,6 @@ export function buildMatrices(envs, tags) { * @returns {string} */ export function renderMarkdownPlan(envs, tags) { - const needsDesktop = (/** @type {string} */ m) => - m.startsWith('desktop-installer') || m === 'installer-script+desktop' || - m === 'open-app-update' || m === 'hermes-desktop-app-update'; const lines = [ '### Install & Update E2E plan', '', @@ -231,7 +240,7 @@ export function renderMarkdownPlan(envs, tags) { const cells = tags.map((tag) => { if ( !tag.desktop && - (needsDesktop(env.install) || needsDesktop(env.update)) + (methodNeedsDesktop(env.install) || methodNeedsDesktop(env.update)) ) { return 'pre-desktop'; } @@ -252,15 +261,36 @@ export function renderMarkdownPlan(envs, tags) { * (pick-releases, the report job itself) fall out naturally. * * @param {{name: string, conclusion: string | null}[]} jobs + * @param {TagAnnotation[]} [tagAnnotations] When given (the same --tags the + * plan got), skipped cells carry their REASON: `pre-desktop` when a + * desktop-surface method meets a tag that predates apps/desktop, `TODO` + * when the pair is declared but no driver arm runs it yet. * @returns {string} */ -export function renderMarkdownResults(jobs) { +export function renderMarkdownResults(jobs, tagAnnotations = []) { const LEG = /^(linux|windows|macos): (\S+) -> (\S+) \((\S+) -> HEAD\) \//; + /** @type {Map} */ + const desktopByTag = new Map(tagAnnotations.map((t) => [t.ref, t.desktop])); + /** + * @param {string} install @param {string} update @param {string} tag + * @returns {string} + */ + const skipLabel = (install, update, tag) => { + if (desktopByTag.size === 0) return 'skip'; + if ( + desktopByTag.get(tag) === false && + (methodNeedsDesktop(install) || methodNeedsDesktop(update)) + ) { + return 'pre-desktop'; + } + return 'TODO'; + }; // A combination can surface as SEVERAL jobs with the same leg name (a // run workflow may have one inner job per driver arm; exactly one runs // and the others natively skip), so cells merge by significance: a real // outcome always beats a skip, and a bad outcome beats a good one. - const RANK = ['skip', '✅', 'running', 'cancelled', '❌']; + const RANK = ['skip', 'TODO', 'pre-desktop', '✅', 'running', 'cancelled', '❌']; + const SKIPS = ['skip', 'TODO', 'pre-desktop']; /** @type {Map>} */ const rows = new Map(); /** @type {string[]} */ @@ -276,7 +306,7 @@ export function renderMarkdownResults(jobs) { switch (job.conclusion) { case 'success': return '✅'; case 'failure': return '❌'; - case 'skipped': return 'skip'; + case 'skipped': return skipLabel(m[2], m[3], tag); case 'cancelled': return 'cancelled'; default: return 'running'; } @@ -291,11 +321,11 @@ export function renderMarkdownResults(jobs) { const cells = [...rows.values()].flatMap((r) => [...r.values()]); const passed = cells.filter((c) => c === '✅').length; const failed = cells.filter((c) => c === '❌').length; - const skipped = cells.filter((c) => c === 'skip').length; + const skipped = cells.filter((c) => SKIPS.includes(c)).length; const lines = [ '### Install & Update E2E results', '', - `${passed} passed, ${failed} failed, ${skipped} skipped (declared TODO / pre-desktop), ${cells.length} legs total`, + `${passed} passed, ${failed} failed, ${skipped} skipped (TODO = declared, no driver arm yet; pre-desktop = the starting release predates apps/desktop), ${cells.length} legs total`, '', `| combination | ${tags.join(' | ')} |`, `|---|${tags.map(() => '---').join('|')}|`, @@ -325,7 +355,8 @@ async function main() { }); if (values.format === 'results') { const jobs = (await readStdin()).split('\n').filter((l) => l.trim()).map((l) => JSON.parse(l)); - process.stdout.write(renderMarkdownResults(jobs)); + const annotations = /** @type {TagAnnotation[]} */ (JSON.parse(values.tags)); + process.stdout.write(renderMarkdownResults(jobs, annotations)); return; } const tags = /** @type {TagAnnotation[]} */ (JSON.parse(values.tags)); From 6e139b2c22f609210015d740cca64ff753edabc6 Mon Sep 17 00:00:00 2001 From: ethernet Date: Wed, 12 Aug 2026 09:46:09 -0400 Subject: [PATCH 078/227] fix(windows-e2e): pin the driver's git to git.exe - never through its own shim The stage shim (git.bat, lying to the product about origin's URL) sits on PATH, so Invoke-Git's bare 'git' routed the driver's own plumbing through cmd - whose parser eats unquoted carets. PowerShell only quotes args containing whitespace, so rev-parse 'v2026.8.3^{commit}' reached the bat as v2026.8.3{commit}: bad revision. The shim must stay a .bat: its audience is the product's python callers (fork detection's remote get-url), which resolve via PATHEXT and never see a .ps1. So the split is by audience - Invoke-Git pins the resolved git.exe for every driver call; the shim serves the product, whose shimmed flows use no caret revs (documented as the accepted hole, with a loud bad-revision failure if that ever changes). The shim self-check now probes through PATH, since Invoke-Git deliberately bypasses it. --- tests/install/windows-e2e.ps1 | 33 ++++++++++++++++++++++++++++----- 1 file changed, 28 insertions(+), 5 deletions(-) diff --git a/tests/install/windows-e2e.ps1 b/tests/install/windows-e2e.ps1 index 84e11f1a3354..140538a2a742 100644 --- a/tests/install/windows-e2e.ps1 +++ b/tests/install/windows-e2e.ps1 @@ -149,10 +149,14 @@ function Invoke-Git([string[]]$GitArgs) { # NativeCommandError even when it exits 0 (git loves stderr for # progress/notices). Relax EAP around the native call only; exit-code # checking below is the real error gate. + # + # ALWAYS the real git.exe, never the shim we ship. + # annoying bug where .bat files eat ^ args. + # if hermes ever adds a git command that calls something with ^ this will break, lol. $prevEap = $ErrorActionPreference $ErrorActionPreference = "Continue" try { - $output = & git @GitArgs 2>&1 + $output = & $script:RealGitExe @GitArgs 2>&1 if ($LASTEXITCODE -ne 0) { throw "git $($GitArgs -join ' ') failed (exit $LASTEXITCODE): $output" } @@ -164,7 +168,6 @@ function Invoke-Git([string[]]$GitArgs) { function Set-GitRedirect { # we redirect to our own repo so we can play around with what commit hermes thinks we're on. - # # MECHANISM: a driver-owned global gitconfig selected via # GIT_CONFIG_GLOBAL. Do NOT use GIT_CONFIG_COUNT/KEY_n/VALUE_n env # config here -- install.ps1 SETS those itself (GIT_CONFIG_COUNT=1, @@ -192,7 +195,6 @@ function Set-GitRedirect { Write-Host " git URL redirect via GIT_CONFIG_GLOBAL=$gitCfg" Write-Host " $RepoUrlHttps -> $fileUrl" - # shim git and make 'git remote get-url origin' report the actual HA upstream # insteadOf is transparent for transport but `git remote get-url origin` gives you the @@ -228,9 +230,28 @@ exit /b %ERRORLEVEL% $env:PATH = "$shimDir;$env:PATH" - # check it worked - $observedGitUrl = Invoke-Git @("-C", $RepoRoot, "remote", "get-url", "origin") + # Check it worked THROUGH the shim - deliberately not Invoke-Git, which + # pins the real git.exe. `git` via PATH here is exactly how the + # product's callers resolve it. Probe the intercepted verb AND plain + # passthrough. + # + # KNOWN HOLE, accepted: cmd parses the command line before the bat sees + # %*, and callers only quote args containing whitespace (PowerShell + # native binding and python's list2cmdline alike) - so a caret arg like + # rev-parse HEAD^{commit} loses its caret THROUGH ANY .bat, unfixably. + # A .ps1 shim would dodge cmd but PATHEXT-resolving callers (python - + # the shim's entire audience) never see .ps1 files, so .bat it stays. + # The driver's own git plumbing therefore pins git.exe (Invoke-Git), + # and the product's shimmed flows (fork detection: remote get-url) use + # no caret revs. If a product path ever sends carets through the shim, + # the leg fails loudly on a bad-revision error naming the mangled arg. + $prevEap = $ErrorActionPreference; $ErrorActionPreference = "Continue" + $observedGitUrl = (& git -C $RepoRoot remote get-url origin 2>&1 | Out-String).Trim() + $passthroughProbe = (& git -C $RepoRoot rev-parse HEAD 2>&1 | Out-String).Trim() + $passthroughExit = $LASTEXITCODE + $ErrorActionPreference = $prevEap Assert-True ($observedGitUrl -eq $RepoUrlHttps) "git remote get-url shim: origin resolves to '$observedGitUrl', expected '$RepoUrlHttps'" + Assert-True ($passthroughExit -eq 0 -and $passthroughProbe -match '^[0-9a-f]{40}$') "shim passthrough works: rev-parse HEAD -> '$passthroughProbe'" Write-Host " git remote get-url shim: $shimPath -> $realGit" Write-Host " 'remote get-url origin' now reports $RepoUrlHttps" } @@ -884,6 +905,8 @@ Write-Host " route: $Route" Write-Host " repo: $RepoRoot" Write-Host " workroot: $WorkRoot" +$script:RealGitExe = (Get-Command git.exe -ErrorAction Stop).Source + Set-GitRedirect switch ($Phase) { From fd02bd86218bc58ff56e5d336bc06e26784c0c79 Mon Sep 17 00:00:00 2001 From: ethernet Date: Wed, 12 Aug 2026 10:15:25 -0400 Subject: [PATCH 079/227] refactor(desktop-e2e): one mock inference server in tests-js/scripts The dev:mock script duplicated the e2e mock server. The copy had only the plain chat reply; every scripted path lived only in the e2e version. A single mock server now lives in tests-js/scripts/mock-server.ts. The e2e suite imports it as a library. Running the file directly starts the server, writes a mock config, and launches the desktop app. The dev:mock script now runs that file. The e2e tsconfig lists tests-js/scripts in its include, because the composite project rule requires every imported file to be listed. --- apps/desktop/e2e/chat.spec.ts | 2 +- .../e2e/correction-session-switch.spec.ts | 2 +- apps/desktop/e2e/fixtures.ts | 2 +- .../e2e/hidden-history-messages.spec.ts | 2 +- .../e2e/image-attachment-resume.spec.ts | 2 +- apps/desktop/e2e/interim-messages.spec.ts | 2 +- apps/desktop/e2e/large-session-resume.spec.ts | 2 +- apps/desktop/e2e/queue-turn-boundary.spec.ts | 2 +- ...session-compression-and-queue-stop.spec.ts | 2 +- apps/desktop/e2e/sidebar-states.spec.ts | 2 +- apps/desktop/e2e/tile-unread-bug.spec.ts | 2 +- apps/desktop/e2e/warm-resume-jitter.spec.ts | 2 +- .../e2e/worktree-branch-status.spec.ts | 2 +- apps/desktop/package.json | 2 +- apps/desktop/scripts/dev-mock.mjs | 237 ------------------ apps/desktop/tsconfig.e2e.json | 2 +- .../e2e => tests-js/scripts}/mock-server.ts | 182 +++++++++++++- 17 files changed, 196 insertions(+), 253 deletions(-) delete mode 100644 apps/desktop/scripts/dev-mock.mjs rename {apps/desktop/e2e => tests-js/scripts}/mock-server.ts (86%) diff --git a/apps/desktop/e2e/chat.spec.ts b/apps/desktop/e2e/chat.spec.ts index 9a55d9fc8eeb..3d79c315e4aa 100644 --- a/apps/desktop/e2e/chat.spec.ts +++ b/apps/desktop/e2e/chat.spec.ts @@ -11,7 +11,7 @@ import { expect, test } from './test' import { type MockBackendFixture, setupMockBackend, waitForAppReady } from './fixtures' -import { BLOCKING_CLARIFY_QUESTION, BLOCKING_CLARIFY_TRIGGER } from './mock-server' +import { BLOCKING_CLARIFY_QUESTION, BLOCKING_CLARIFY_TRIGGER } from '../../../tests-js/scripts/mock-server' import { expectVisualSnapshot } from './visual-snapshot' let fixture: MockBackendFixture | null = null diff --git a/apps/desktop/e2e/correction-session-switch.spec.ts b/apps/desktop/e2e/correction-session-switch.spec.ts index dc435b1d5384..fc7e0359bca8 100644 --- a/apps/desktop/e2e/correction-session-switch.spec.ts +++ b/apps/desktop/e2e/correction-session-switch.spec.ts @@ -10,7 +10,7 @@ import { type TestInfo } from '@playwright/test' import { expect, test, type Page } from './test' import { type MockBackendFixture, setupMockBackend, waitForAppReady } from './fixtures' -import { CORRECTION_SWITCH_TRIGGER, MOCK_REPLY } from './mock-server' +import { CORRECTION_SWITCH_TRIGGER, MOCK_REPLY } from '../../../tests-js/scripts/mock-server' const OTHER_SESSION_PROMPT = 'E2E persisted session used for a warm resume.' const ORIGINAL_PROMPT = `${CORRECTION_SWITCH_TRIGGER}: original prompt must remain singular after a correction.` diff --git a/apps/desktop/e2e/fixtures.ts b/apps/desktop/e2e/fixtures.ts index 787be421886b..69ba0914e5ee 100644 --- a/apps/desktop/e2e/fixtures.ts +++ b/apps/desktop/e2e/fixtures.ts @@ -27,7 +27,7 @@ import * as path from 'node:path' import { _electron, type ElectronApplication, type Page } from '@playwright/test' -import { startMockServer, type MockServerOptions } from './mock-server' +import { startMockServer, type MockServerOptions } from '../../../tests-js/scripts/mock-server' import { installErrorBannerGuard } from './test' const DESKTOP_ROOT = path.resolve(import.meta.dirname, '..') diff --git a/apps/desktop/e2e/hidden-history-messages.spec.ts b/apps/desktop/e2e/hidden-history-messages.spec.ts index 23076f766e7f..77241d336753 100644 --- a/apps/desktop/e2e/hidden-history-messages.spec.ts +++ b/apps/desktop/e2e/hidden-history-messages.spec.ts @@ -25,7 +25,7 @@ import { startMockServer, VERIFICATION_STOP_TEXT, VERIFICATION_STOP_TRIGGER, -} from './mock-server' +} from '../../../tests-js/scripts/mock-server' import { RealSessionBuilder } from './real-session-builder' const SESSION_TITLE = 'E2E Hidden History Messages' diff --git a/apps/desktop/e2e/image-attachment-resume.spec.ts b/apps/desktop/e2e/image-attachment-resume.spec.ts index a4f8da68e39d..4382e76035a9 100644 --- a/apps/desktop/e2e/image-attachment-resume.spec.ts +++ b/apps/desktop/e2e/image-attachment-resume.spec.ts @@ -23,7 +23,7 @@ import { writeEnvFile, writeMockProviderConfig, } from './fixtures' -import { type MockServer, startMockServer } from './mock-server' +import { type MockServer, startMockServer } from '../../../tests-js/scripts/mock-server' import { RealSessionBuilder } from './real-session-builder' import { type ElectronApplication, expect, type Page, test } from './test' diff --git a/apps/desktop/e2e/interim-messages.spec.ts b/apps/desktop/e2e/interim-messages.spec.ts index 2f6da013912b..3837084985c2 100644 --- a/apps/desktop/e2e/interim-messages.spec.ts +++ b/apps/desktop/e2e/interim-messages.spec.ts @@ -36,7 +36,7 @@ import { setupMockBackend, waitForAppReady, } from './fixtures' -import { INTERIM_TEXTS, restartMockServer } from './mock-server' +import { INTERIM_TEXTS, restartMockServer } from '../../../tests-js/scripts/mock-server' // ─── Helpers ────────────────────────────────────────────────────────── diff --git a/apps/desktop/e2e/large-session-resume.spec.ts b/apps/desktop/e2e/large-session-resume.spec.ts index 02c5f16937d1..4dfaaa7d922d 100644 --- a/apps/desktop/e2e/large-session-resume.spec.ts +++ b/apps/desktop/e2e/large-session-resume.spec.ts @@ -13,7 +13,7 @@ import { writeEnvFile, writeMockProviderConfig, } from './fixtures' -import { MOCK_REPLY, startMockServer, type MockServer, type MockServerOptions } from './mock-server' +import { MOCK_REPLY, startMockServer, type MockServer, type MockServerOptions } from '../../../tests-js/scripts/mock-server' import { RealSessionBuilder } from './real-session-builder' const DESKTOP_ROOT = path.resolve(import.meta.dirname, '..') diff --git a/apps/desktop/e2e/queue-turn-boundary.spec.ts b/apps/desktop/e2e/queue-turn-boundary.spec.ts index 2308d321a553..7c97115753de 100644 --- a/apps/desktop/e2e/queue-turn-boundary.spec.ts +++ b/apps/desktop/e2e/queue-turn-boundary.spec.ts @@ -10,7 +10,7 @@ import { expect, test, type Page } from './test' import { type MockBackendFixture, setupMockBackend, waitForAppReady } from './fixtures' -import { MOCK_REPLY } from './mock-server' +import { MOCK_REPLY } from '../../../tests-js/scripts/mock-server' const ACTIVE_PROMPT = 'E2E_QUEUE_TURN_BOUNDARY_ACTIVE' const QUEUED_PROMPT = 'E2E_QUEUE_TURN_BOUNDARY_QUEUED' diff --git a/apps/desktop/e2e/session-compression-and-queue-stop.spec.ts b/apps/desktop/e2e/session-compression-and-queue-stop.spec.ts index e48c52cda2b0..192fc8a45a9e 100644 --- a/apps/desktop/e2e/session-compression-and-queue-stop.spec.ts +++ b/apps/desktop/e2e/session-compression-and-queue-stop.spec.ts @@ -5,7 +5,7 @@ import { expect, test, type Page } from '@playwright/test' import { type MockBackendFixture, setupMockBackend, waitForAppReady } from './fixtures' -import { MOCK_REPLY, receivedUserTexts, restartMockServer } from './mock-server' +import { MOCK_REPLY, receivedUserTexts, restartMockServer } from '../../../tests-js/scripts/mock-server' async function send(page: Page, text: string, delay = 15): Promise { const composer = page.locator('[contenteditable="true"]').first() diff --git a/apps/desktop/e2e/sidebar-states.spec.ts b/apps/desktop/e2e/sidebar-states.spec.ts index 6d8c0c2c9c9c..e760835caa7d 100644 --- a/apps/desktop/e2e/sidebar-states.spec.ts +++ b/apps/desktop/e2e/sidebar-states.spec.ts @@ -21,7 +21,7 @@ import { restartMockServer, SIDEBAR_CROSS_TEXTS, SIDEBAR_TEXTS, -} from './mock-server' +} from '../../../tests-js/scripts/mock-server' /** Background-running dot aria-label (from i18n en.ts). */ const BG_DOT_LABEL = 'Background task running' diff --git a/apps/desktop/e2e/tile-unread-bug.spec.ts b/apps/desktop/e2e/tile-unread-bug.spec.ts index fa614a21d2f0..e11782036f0d 100644 --- a/apps/desktop/e2e/tile-unread-bug.spec.ts +++ b/apps/desktop/e2e/tile-unread-bug.spec.ts @@ -27,7 +27,7 @@ import { createBackgroundReleaseHandle, restartMockServer, SIDEBAR_CROSS_TEXTS, -} from './mock-server' +} from '../../../tests-js/scripts/mock-server' /** Finished-unread dot aria-label. */ const UNREAD_DOT_LABEL = 'Finished — unread' diff --git a/apps/desktop/e2e/warm-resume-jitter.spec.ts b/apps/desktop/e2e/warm-resume-jitter.spec.ts index dbc0fe5dd113..b8e95c199e95 100644 --- a/apps/desktop/e2e/warm-resume-jitter.spec.ts +++ b/apps/desktop/e2e/warm-resume-jitter.spec.ts @@ -41,7 +41,7 @@ import { buildAppEnv, launchDesktop, } from './fixtures' -import { startMockServer } from './mock-server' +import { startMockServer } from '../../../tests-js/scripts/mock-server' import { RealSessionBuilder } from './real-session-builder' const SESSION_TITLE = 'E2E Warm Resume Jitter Test' diff --git a/apps/desktop/e2e/worktree-branch-status.spec.ts b/apps/desktop/e2e/worktree-branch-status.spec.ts index 971a1f6b4126..66c821a7bc8d 100644 --- a/apps/desktop/e2e/worktree-branch-status.spec.ts +++ b/apps/desktop/e2e/worktree-branch-status.spec.ts @@ -11,7 +11,7 @@ import { writeEnvFile, writeMockProviderConfig, } from './fixtures' -import { startMockServer } from './mock-server' +import { startMockServer } from '../../../tests-js/scripts/mock-server' import { expect, test } from './test' import { expectVisualSnapshot } from './visual-snapshot' diff --git a/apps/desktop/package.json b/apps/desktop/package.json index 7e21884540e3..624afff1be8b 100644 --- a/apps/desktop/package.json +++ b/apps/desktop/package.json @@ -17,7 +17,7 @@ "clean:electron": "tsc --build tsconfig.electron.json --clean", "dev": "concurrently -k \"npm:dev:renderer\" \"npm:dev:electron\"", "dev:fake-boot": "cross-env HERMES_DESKTOP_BOOT_FAKE=1 HERMES_DESKTOP_BOOT_FAKE_STEP_MS=650 npm run dev", - "dev:mock": "node scripts/dev-mock.mjs", + "dev:mock": "node ../../tests-js/scripts/mock-server.ts", "dev:renderer": "node scripts/assert-root-install.mjs && npm run clean:renderer && vite --host 127.0.0.1 --port 5174", "dev:electron": "tsc --build tsconfig.electron.json && wait-on http://127.0.0.1:5174 && node scripts/bundle-electron-main.mjs --dev && cross-env XCURSOR_SIZE=24 HERMES_DESKTOP_DEV_SERVER=http://127.0.0.1:5174 electron .", "profile:main": "tsc --build tsconfig.electron.json && wait-on http://127.0.0.1:5174 && node scripts/bundle-electron-main.mjs --dev && cross-env XCURSOR_SIZE=24 HERMES_DESKTOP_DEV_SERVER=http://127.0.0.1:5174 electron --inspect=9229 .", diff --git a/apps/desktop/scripts/dev-mock.mjs b/apps/desktop/scripts/dev-mock.mjs deleted file mode 100644 index 7b523d88b3c3..000000000000 --- a/apps/desktop/scripts/dev-mock.mjs +++ /dev/null @@ -1,237 +0,0 @@ -#!/usr/bin/env node -/** - * Launch the desktop app with a mock inference provider — no real API - * keys needed. Starts a local OpenAI-compatible server that returns a - * canned reply, writes an isolated config.yaml + .env, and launches the - * built Electron app against them. - * - * This reuses the same mock-server and config format as the E2E fixtures - * (apps/desktop/e2e/mock-server.ts + fixtures.ts), so local dev and CI - * test the same chain. - * - * Prerequisite: `npm run build` must have been run so dist/ exists. - * - * Usage: - * node scripts/dev-mock.mjs - * npm run dev:mock - * - * The mock server listens on an ephemeral port and replies to every - * chat completion with: - * "Hello from the mock inference server! The full boot chain is working." - */ - -import http from 'node:http' -import fs from 'node:fs' -import os from 'node:os' -import path from 'node:path' -import { spawn, spawnSync } from 'node:child_process' - -const DESKTOP_ROOT = path.resolve(import.meta.dirname, '..') -const REPO_ROOT = path.resolve(DESKTOP_ROOT, '..', '..') - -// ── Canned reply ─────────────────────────────────────────────────────── - -const CANNED_REPLY = - 'Hello from the mock inference server! The full boot chain is working.' - -// ── Mock server (mirrors e2e/mock-server.ts) ─────────────────────────── - -function startMockServer() { - return new Promise((resolve, reject) => { - const server = http.createServer((req, res) => { - res.setHeader('Access-Control-Allow-Origin', '*') - res.setHeader('Access-Control-Allow-Headers', '*') - res.setHeader('Access-Control-Allow-Methods', 'GET, POST, OPTIONS') - - if (req.method === 'OPTIONS') { - res.writeHead(204) - res.end() - return - } - - if (req.method === 'GET' && req.url === '/v1/models') { - res.writeHead(200, { 'Content-Type': 'application/json' }) - res.end( - JSON.stringify({ - object: 'list', - data: [{ id: 'mock-model', object: 'model', created: 0, owned_by: 'mock' }], - }), - ) - return - } - - if (req.method === 'POST' && req.url?.startsWith('/v1/chat/completions')) { - let body = '' - req.on('data', (chunk) => { body += chunk.toString() }) - req.on('end', () => { - let parsed = {} - try { parsed = JSON.parse(body) } catch { /* non-streaming */ } - - const stream = parsed.stream === true - const model = parsed.model || 'mock-model' - - if (stream) { - res.writeHead(200, { - 'Content-Type': 'text/event-stream', - 'Cache-Control': 'no-cache', - Connection: 'keep-alive', - }) - const words = CANNED_REPLY.split(' ') - let i = 0 - const sendChunk = () => { - if (i >= words.length) { - res.write( - `data: ${JSON.stringify({ - id: 'mock-completion', object: 'chat.completion.chunk', - created: 0, model, - choices: [{ index: 0, delta: {}, finish_reason: 'stop' }], - })}\n\n`, - ) - res.write('data: [DONE]\n\n') - res.end() - return - } - const word = i === 0 ? words[i] : ' ' + words[i] - res.write( - `data: ${JSON.stringify({ - id: 'mock-completion', object: 'chat.completion.chunk', - created: 0, model, - choices: [{ index: 0, delta: { content: word }, finish_reason: null }], - })}\n\n`, - ) - i++ - setTimeout(sendChunk, 20) - } - sendChunk() - } else { - res.writeHead(200, { 'Content-Type': 'application/json' }) - res.end( - JSON.stringify({ - id: 'mock-completion', object: 'chat.completion', - created: 0, model, - choices: [{ - index: 0, - message: { role: 'assistant', content: CANNED_REPLY }, - finish_reason: 'stop', - }], - usage: { prompt_tokens: 10, completion_tokens: 20, total_tokens: 30 }, - }), - ) - } - }) - req.on('error', () => { res.writeHead(400); res.end('Bad request') }) - return - } - - res.writeHead(404, { 'Content-Type': 'application/json' }) - res.end(JSON.stringify({ error: 'Not found' })) - }) - - server.on('error', reject) - server.listen(0, '127.0.0.1', () => { - const addr = server.address() - if (addr === null || typeof addr === 'string') { - reject(new Error('Failed to get server address')) - return - } - resolve({ port: addr.port, url: `http://127.0.0.1:${addr.port}`, close: () => server.close() }) - }) - }) -} - -// ── Config + env writing (mirrors e2e/fixtures.ts) ───────────────────── - -function createSandbox() { - const root = fs.mkdtempSync(path.join(os.tmpdir(), `hermes-dev-mock-${Date.now()}`)) - const hermesHome = path.join(root, 'hermes-home') - const userDataDir = path.join(root, 'electron-user-data') - fs.mkdirSync(hermesHome, { recursive: true }) - fs.mkdirSync(userDataDir, { recursive: true }) - return { root, hermesHome, userDataDir, cleanup: () => fs.rmSync(root, { recursive: true, force: true }) } -} - -function writeMockConfig(hermesHome, mockUrl) { - fs.writeFileSync( - path.join(hermesHome, 'config.yaml'), - `# Auto-generated by dev-mock.mjs -model: - default: mock-model - provider: mock -providers: - mock: - api: ${mockUrl}/v1 - name: Mock - api_mode: chat_completions - key_env: MOCK_API_KEY - models: - mock-model: {} - context_length: 4096 -`, - 'utf8', - ) - fs.writeFileSync(path.join(hermesHome, '.env'), 'MOCK_API_KEY=e2e-mock-key\n', 'utf8') -} - -// ── Electron launch ──────────────────────────────────────────────────── - -function findElectron() { - const local = path.join(REPO_ROOT, 'node_modules', 'electron', 'dist', 'electron') - if (fs.existsSync(local)) return local - const r = spawnSync('which', ['electron'], { encoding: 'utf8' }) - if (r.status === 0 && r.stdout.trim()) return r.stdout.trim() - throw new Error('Electron binary not found. Run "npm install" from the repo root.') -} - -function assertDistBuilt() { - const electronMain = path.join(DESKTOP_ROOT, 'dist', 'electron-main.mjs') - const indexHtml = path.join(DESKTOP_ROOT, 'dist', 'index.html') - if (!fs.existsSync(electronMain) || !fs.existsSync(indexHtml)) { - throw new Error( - `Desktop dist not built. Run 'cd apps/desktop && npm run build' first.\n` + - `Missing: ${electronMain}`, - ) - } -} - -// ── Main ─────────────────────────────────────────────────────────────── - -async function main() { - assertDistBuilt() - - console.log('Starting mock inference server...') - const mock = await startMockServer() - console.log(` Mock server: ${mock.url}`) - - const sandbox = createSandbox() - writeMockConfig(sandbox.hermesHome, mock.url) - console.log(` HERMES_HOME: ${sandbox.hermesHome}`) - - const electronBin = findElectron() - - const env = { - ...process.env, - HERMES_HOME: sandbox.hermesHome, - HERMES_DESKTOP_USER_DATA_DIR: sandbox.userDataDir, - HERMES_DESKTOP_IGNORE_EXISTING: '1', - HERMES_DESKTOP_HERMES_ROOT: REPO_ROOT, - HERMES_DESKTOP_APP_NAME: `HermesDevMock-${Date.now()}`, - } - - console.log('Launching Electron...') - const child = spawn(electronBin, [DESKTOP_ROOT, '--disable-gpu', '--no-sandbox'], { - env, - cwd: DESKTOP_ROOT, - stdio: 'inherit', - }) - - child.on('exit', (code) => { - mock.close() - sandbox.cleanup() - process.exit(code ?? 0) - }) -} - -main().catch((err) => { - console.error(err) - process.exit(1) -}) diff --git a/apps/desktop/tsconfig.e2e.json b/apps/desktop/tsconfig.e2e.json index e0beb8961941..5485ca4cc43d 100644 --- a/apps/desktop/tsconfig.e2e.json +++ b/apps/desktop/tsconfig.e2e.json @@ -4,6 +4,6 @@ "types": ["node", "@playwright/test"], "composite": true }, - "include": ["e2e", "playwright.config.ts"], + "include": ["e2e", "playwright.config.ts", "../../tests-js/scripts"], "exclude": ["src", "electron"] } diff --git a/apps/desktop/e2e/mock-server.ts b/tests-js/scripts/mock-server.ts similarity index 86% rename from apps/desktop/e2e/mock-server.ts rename to tests-js/scripts/mock-server.ts index ce4665d1775c..61f40f403ad5 100644 --- a/apps/desktop/e2e/mock-server.ts +++ b/tests-js/scripts/mock-server.ts @@ -1,5 +1,6 @@ /** - * Minimal OpenAI-compatible mock inference server for E2E tests. + * Minimal OpenAI-compatible mock inference server for E2E tests and the + * dev:mock dev flow. * * Implements just enough of the /v1/* surface for `hermes serve` to resolve a * provider, list models, and stream a canned chat completion back to the @@ -12,13 +13,19 @@ * The canned response is a short, deterministic assistant message. Tool-call * requests are not simulated — the E2E tests only need the chat surface to * prove the full boot → gateway → inference → renderer chain works. + * + * Import the module to get the server as a library (the Playwright E2E + * suite). Run the file directly to also write an isolated mock config and + * launch the built Electron app against it (`npm run dev:mock`). */ +import { spawn, spawnSync } from 'node:child_process' import fs from 'node:fs' import http from 'node:http' import type { ServerResponse } from 'node:http' import os from 'node:os' import nodePath from 'node:path' +import { pathToFileURL } from 'node:url' /** A canned assistant reply used for every chat completion request. */ export const MOCK_REPLY = 'Hello from the mock inference server! The full boot chain is working.' @@ -200,10 +207,12 @@ function sidebarCrossBgCommand(releasePath?: string): string { if (!releasePath) { return 'echo "long bg output" && sleep 5 && echo "finished"' } + // Bounded wait (60s): if a test forgets to release (or crashes mid-way), // the process still exits instead of hanging the worker until the suite // times out. const quoted = JSON.stringify(releasePath) + return [ 'echo "long bg output"', `for _ in $(seq 1 600); do [ -e ${quoted} ] && break; sleep 0.1; done`, @@ -327,12 +336,15 @@ export function startMockServer(options: MockServerOptions = {}): Promise void) | null = null let releaseHeldStream: (() => void) | null = null let heldCompletionCount = 0 + const heldStreamStarted = new Promise(resolveHeld => { resolveHeldStreamStarted = resolveHeld }) + const heldStreamReleased = new Promise(resolveRelease => { releaseHeldStream = resolveRelease }) + const server = http.createServer((req, res) => { // CORS headers — the Electron renderer doesn't need them, but they // don't hurt and make the server usable from a browser context too. @@ -343,6 +355,7 @@ export function startMockServer(options: MockServerOptions = {}): Promise m?.role === 'user') const userText = typeof lastUserMsg?.content === 'string' ? lastUserMsg.content : '' + if (userText) { _receivedUserTexts.push(userText) } + const isInterimTrigger = userText.includes('E2E_INTERIM_TRIGGER') const isSidebarTrigger = userText.includes('E2E_SIDEBAR_TRIGGER') const isSidebarCrossTrigger = userText.includes('E2E_SIDEBAR_CROSS') const isQueueStopTrigger = userText.includes('E2E_QUEUE_STOP_TRIGGER') + const isVerificationStopTrigger = messages.some( message => typeof message?.content === 'string' && message.content.includes(VERIFICATION_STOP_TRIGGER), ) + const isCorrectionSwitchTrigger = messages.some( message => typeof message?.content === 'string' && message.content.includes(CORRECTION_SWITCH_TRIGGER), ) @@ -427,17 +446,20 @@ export function startMockServer(options: MockServerOptions = {}): Promise { if (holdThisCompletion) { heldCompletionCount++ } + resolveHeldStreamStarted?.() + return heldStreamReleased } : undefined) } else { @@ -527,6 +560,7 @@ export function startMockServer(options: MockServerOptions = {}): Promise { const addr = server.address() + if (addr === null || typeof addr === 'string') { reject(new Error('Failed to get server address')) + return } @@ -607,16 +643,20 @@ function streamTextResponse( res.write(sseChunk(model, {}, 'stop')) res.write('data: [DONE]\n\n') res.end() + return } const word = i === 0 ? words[i] : ' ' + words[i] res.write(sseChunk(model, { content: word })) i++ + if (waitForRelease && i === 1) { waitForRelease().then(() => setTimeout(sendChunk, 20)) + return } + setTimeout(sendChunk, 20) } @@ -683,8 +723,10 @@ function streamScriptedTurn( } else { res.write(sseChunk(model, {}, finishReason)) } + res.write('data: [DONE]\n\n') res.end() + return } @@ -709,8 +751,10 @@ function streamScriptedTurn( } else { res.write(sseChunk(model, {}, finishReason)) } + res.write('data: [DONE]\n\n') res.end() + return } @@ -733,9 +777,11 @@ function nonStreamingScriptedTurn( const finishReason = hasToolCalls ? 'tool_calls' : 'stop' const message: Record = { role: 'assistant' } + if (turn.text) { message.content = turn.text } + if (hasToolCalls) { message.tool_calls = turn.toolCalls!.map((tc, idx) => ({ id: `call_e2e_${_scriptIndex}_${idx}`, @@ -804,6 +850,7 @@ export function createBackgroundReleaseHandle(): BackgroundReleaseHandle { os.tmpdir(), `hermes-e2e-bg-release-${process.pid}-${Date.now()}-${Math.random().toString(36).slice(2, 8)}`, ) + return { path, release: () => { @@ -867,3 +914,136 @@ export const SIDEBAR_CROSS_TEXTS = { /** The subagent's goal. */ subagentGoal: 'Analyze cross-session state', } as const + +// ─── Dev launcher ────────────────────────────────────────────────────── +// +// Running this file directly (`node tests-js/scripts/mock-server.ts`) +// starts the server, writes an isolated config.yaml + .env that point at +// it, and launches the built Electron app against them — the `dev:mock` +// flow. Importing the module never runs this block: the Playwright E2E +// suite imports the server as a library instead. + +interface DevSandbox { + root: string + hermesHome: string + userDataDir: string + cleanup: () => void +} + +/** Create an isolated HERMES_HOME + Electron user-data dir in the OS temp dir. */ +function createDevSandbox(): DevSandbox { + const root = fs.mkdtempSync(nodePath.join(os.tmpdir(), `hermes-dev-mock-${Date.now()}`)) + const hermesHome = nodePath.join(root, 'hermes-home') + const userDataDir = nodePath.join(root, 'electron-user-data') + fs.mkdirSync(hermesHome, { recursive: true }) + fs.mkdirSync(userDataDir, { recursive: true }) + + return { + root, + hermesHome, + userDataDir, + cleanup: () => { + try { + fs.rmSync(root, { recursive: true, force: true }) + } catch { + // best-effort + } + }, + } +} + +/** Write a config.yaml + .env that pre-configure the mock provider. */ +function writeMockConfig(hermesHome: string, mockUrl: string): void { + fs.writeFileSync( + nodePath.join(hermesHome, 'config.yaml'), + `# Auto-generated by dev-mock +model: + default: mock-model + provider: mock +providers: + mock: + api: ${mockUrl}/v1 + name: Mock + api_mode: chat_completions + key_env: MOCK_API_KEY + models: + mock-model: {} + context_length: 4096 +`, + 'utf8', + ) + fs.writeFileSync(nodePath.join(hermesHome, '.env'), 'MOCK_API_KEY=e2e-mock-key\n', 'utf8') +} + +/** Resolve the Electron binary: the repo's own install, then PATH. */ +function findElectron(repoRoot: string): string { + const local = nodePath.join(repoRoot, 'node_modules', 'electron', 'dist', 'electron') + + if (fs.existsSync(local)) {return local} + const r = spawnSync('which', ['electron'], { encoding: 'utf8' }) + + if (r.status === 0 && r.stdout.trim()) {return r.stdout.trim()} + throw new Error('Electron binary not found. Run "npm install" from the repo root.') +} + +/** Fail fast with a clear message when the desktop dist/ is missing. */ +function assertDistBuilt(desktopRoot: string): void { + const electronMain = nodePath.join(desktopRoot, 'dist', 'electron-main.mjs') + const indexHtml = nodePath.join(desktopRoot, 'dist', 'index.html') + + if (!fs.existsSync(electronMain) || !fs.existsSync(indexHtml)) { + throw new Error( + `Desktop dist not built. Run 'cd apps/desktop && npm run build' first.\n` + + `Missing: ${electronMain}`, + ) + } +} + +/** Start the mock, write the sandbox, and launch the built Electron app. */ +async function runDevLaunch(): Promise { + const desktopRoot = nodePath.resolve(import.meta.dirname, '..', '..', 'apps', 'desktop') + const repoRoot = nodePath.resolve(desktopRoot, '..', '..') + + assertDistBuilt(desktopRoot) + + console.log('Starting mock inference server...') + const mock = await startMockServer() + console.log(` Mock server: ${mock.url}`) + + const sandbox = createDevSandbox() + writeMockConfig(sandbox.hermesHome, mock.url) + console.log(` HERMES_HOME: ${sandbox.hermesHome}`) + + const electronBin = findElectron(repoRoot) + + const env: Record = { + ...process.env, + HERMES_HOME: sandbox.hermesHome, + HERMES_DESKTOP_USER_DATA_DIR: sandbox.userDataDir, + HERMES_DESKTOP_IGNORE_EXISTING: '1', + HERMES_DESKTOP_HERMES_ROOT: repoRoot, + HERMES_DESKTOP_APP_NAME: `HermesDevMock-${Date.now()}`, + } + + console.log('Launching Electron...') + + const child = spawn(electronBin, [desktopRoot, '--disable-gpu', '--no-sandbox'], { + env, + cwd: desktopRoot, + stdio: 'inherit', + }) + + child.on('exit', (code: number | null) => { + void mock.close() + sandbox.cleanup() + process.exit(code ?? 0) + }) +} + +// Only run the dev launcher when this file is executed directly. +if (process.argv[1] !== undefined && import.meta.url === pathToFileURL(process.argv[1]).href) { + runDevLaunch().catch((err: unknown) => { + console.error(err) + process.exit(1) + }) +} From 7721b26a9aead0603505b1af4ab5732c34851ce5 Mon Sep 17 00:00:00 2001 From: ethernet Date: Wed, 12 Aug 2026 10:36:09 -0400 Subject: [PATCH 080/227] fix(install-e2e): boot the app-update legs against a REAL configured provider The app-update legs died on the onboarding overlay - a fullscreen div that intercepts every click (the Settings click timed out under it). The old plan seeded a fake provider key, which lies: the overlay vanishes but the app is broken. Instead the driver now runs the desktop E2E suite's own mock inference server (tests-js/scripts/mock-server.ts, zero deps, bare-node type stripping >=22.18) and configures it into HERMES_HOME byte-for-byte like the dev:mock flow: config.yaml provider + MOCK_API_KEY env. The app boots genuinely configured - no overlay, real chat surface. e2e-assets/mock-provider.{sh,mjs} own start/stop (pid + url files; the wrapper lives until SIGTERM - gating on stdin-close made the server die instantly, a background process's stdin is already EOF) and the config write. Wired into the posix script driver's hermes-desktop-app-update arm and the macos driver's update phase (both app-update methods). The Playwright flow keeps its defense-in- depth: the real escape hatch ('I'll choose a provider later') and the verified 'Open settings' selector. Probed locally: models + streamed/non-streamed completions answer, server stops cleanly on kill. --- tests/install/e2e-assets/launch-from-spec.mjs | 23 ++++-- tests/install/e2e-assets/mock-provider.mjs | 31 ++++++++ tests/install/e2e-assets/mock-provider.sh | 71 +++++++++++++++++++ tests/install/installer-script-e2e.sh | 9 +++ tests/install/macos-desktop-e2e.sh | 7 ++ 5 files changed, 134 insertions(+), 7 deletions(-) create mode 100644 tests/install/e2e-assets/mock-provider.mjs create mode 100755 tests/install/e2e-assets/mock-provider.sh diff --git a/tests/install/e2e-assets/launch-from-spec.mjs b/tests/install/e2e-assets/launch-from-spec.mjs index e203ba5714bb..0fb3a85b941e 100644 --- a/tests/install/e2e-assets/launch-from-spec.mjs +++ b/tests/install/e2e-assets/launch-from-spec.mjs @@ -120,15 +120,24 @@ async function main() { } const deadline = Date.now() + Number(values['timeout-ms']); - // Dismiss the onboarding overlay when present (fresh HERMES_HOME). - const skip = window.getByRole('button', { name: /skip|get started|continue/i }).first(); - if (await skip.isVisible({ timeout: 5_000 }).catch(() => false)) { - await skip.click().catch(() => {}); + // Dismiss the onboarding overlay when present. Two layers of defense: + // the drivers seed a provider key so the app's runtime check reports + // configured=true and the overlay never mounts; if it shows anyway + // (fresh HERMES_HOME, slow readiness check), click the real escape + // hatch - "I'll choose a provider later" (i18n en: chooseLater). The + // overlay is a fullscreen div that intercepts ALL clicks, so this must + // resolve before any Settings navigation. + const later = window + .getByRole('button', { name: /choose a provider later|skip/i }) + .first(); + if (await later.isVisible({ timeout: 10_000 }).catch(() => false)) { + await later.click().catch(() => {}); + await later.waitFor({ state: 'hidden', timeout: 15_000 }).catch(() => {}); } - // Settings -> About -> Update now. Selectors favor accessible names over - // DOM structure so renderer refactors don't break the leg. - await window.getByRole('button', { name: /settings/i }).first().click(); + // Settings -> About -> Update now. The settings trigger is an icon + // button whose accessible name is "Open settings". + await window.getByRole('button', { name: /open settings|settings/i }).first().click(); await window.getByRole('tab', { name: /about/i }).or( window.getByRole('button', { name: /about/i })).first().click(); const updateNow = window.getByRole('button', { name: /update now/i }).first(); diff --git a/tests/install/e2e-assets/mock-provider.mjs b/tests/install/e2e-assets/mock-provider.mjs new file mode 100644 index 000000000000..d56ca9a9c656 --- /dev/null +++ b/tests/install/e2e-assets/mock-provider.mjs @@ -0,0 +1,31 @@ +// Driver-side wrapper around tests-js/scripts/mock-server.ts (the desktop +// E2E suite's OpenAI-compatible mock): starts the server as a LIBRARY +// (importing, not executing, so the dev-launcher block never runs) and +// publishes its URL to a file for the shell driver to consume. +// +// Usage: node mock-provider.mjs +// Writes "" with the base URL (http://127.0.0.1:) once +// the server is listening, then stays alive until stdin closes. + +// @ts-check +import fs from 'node:fs'; +import process from 'node:process'; +import { fileURLToPath } from 'node:url'; + +const mockUrl = fileURLToPath(new URL('../../../tests-js/scripts/mock-server.ts', import.meta.url)); + +const { startMockServer } = await import(mockUrl); +const mock = await startMockServer(); + +const urlFile = process.argv[2]; +if (!urlFile) { + console.error('usage: node mock-provider.mjs '); + process.exit(1); +} +fs.writeFileSync(urlFile, mock.url); +console.log(`[mock-provider] listening at ${mock.url}`); + +// Live until killed. NOT gated on stdin closing: a backgrounded process's +// stdin is already at EOF, so an end-event exit would fire immediately +// and the server would die right after writing its URL. +await new Promise(() => {}); diff --git a/tests/install/e2e-assets/mock-provider.sh b/tests/install/e2e-assets/mock-provider.sh new file mode 100755 index 000000000000..ba6e46b83a48 --- /dev/null +++ b/tests/install/e2e-assets/mock-provider.sh @@ -0,0 +1,71 @@ +#!/usr/bin/env bash +# Start/stop the mock inference server and point the app at it as a REAL +# configured provider. +# +# Why: the desktop app boots to the onboarding overlay when no provider is +# configured - a fullscreen div that intercepts every click, which killed +# the app-update legs. Seeding a fake key makes the overlay vanish but +# leaves a lying app; this makes the app GENUINELY configured (config.yaml +# provider + key env, exactly what tests-js/scripts/mock-server.ts's +# dev:mock flow writes) with a real, chat-capable backend. +# +# Usage (sourced from a driver): +# mock_start start, write config into $HERMES_HOME +# mock_stop kill the background server +# +# Requires: ASSETS (e2e-assets dir), LOG_DIR, HERMES_HOME, ok/fail helpers. + +MOCK_PIDFILE="" +MOCK_URLFILE="" + +mock_start() { + local workroot="${1:?mock_start needs a workroot}" + MOCK_PIDFILE="$workroot/mock.pid" + MOCK_URLFILE="$workroot/mock.url" + rm -f "$MOCK_PIDFILE" "$MOCK_URLFILE" + + # Bare `node file.ts` type-stripping works on node >=22.18 (the images + # ship 22.22+; the installed managed node is >=26) - same contract as + # the repo's own `dev:mock` script. + node "$ASSETS/mock-provider.mjs" "$MOCK_URLFILE" > "$LOG_DIR/mock.log" 2>&1 & + echo $! > "$MOCK_PIDFILE" + + local _i + for _i in 1 2 3 4 5 6 7 8 9 10; do + [ -s "$MOCK_URLFILE" ] && break + sleep 0.2 + done + if [ ! -s "$MOCK_URLFILE" ]; then + log_group "mock server transcript" "$LOG_DIR/mock.log" + fail "mock inference server did not come up; transcript above" + fi + local url + url="$(cat "$MOCK_URLFILE")" + ok "mock inference server: $url" + + # The provider config, byte-compatible with writeMockConfig() in + # tests-js/scripts/mock-server.ts. + cat > "$HERMES_HOME/config.yaml" <> "$HERMES_HOME/.env" + ok "provider 'mock' configured in $HERMES_HOME (api $url/v1)" +} + +mock_stop() { + if [ -n "$MOCK_PIDFILE" ] && [ -f "$MOCK_PIDFILE" ]; then + kill "$(cat "$MOCK_PIDFILE")" 2>/dev/null || true + rm -f "$MOCK_PIDFILE" + fi +} diff --git a/tests/install/installer-script-e2e.sh b/tests/install/installer-script-e2e.sh index ec5e553c705c..ec553fc0c257 100755 --- a/tests/install/installer-script-e2e.sh +++ b/tests/install/installer-script-e2e.sh @@ -357,6 +357,15 @@ case "$UPDATE_METHOD" in ASSETS="$REPO_ROOT/tests/install/e2e-assets" SPEC="$WORK_ROOT/launch-spec.json" + # A REAL configured provider: the mock inference server (the desktop E2E + # suite's own) is configured into HERMES_HOME exactly like the dev:mock + # flow does. The app then boots genuinely configured - no onboarding + # overlay (a fullscreen div that intercepts every click) - and the chat + # surface is real too. + source "$ASSETS/mock-provider.sh" + mock_start "$WORK_ROOT" + trap mock_stop EXIT + step "capturing the hermes desktop launch spec (build runs for real)" rc=0 (cd "$INSTALL_DIR" && \ diff --git a/tests/install/macos-desktop-e2e.sh b/tests/install/macos-desktop-e2e.sh index f52867bc6164..da86c117f89f 100755 --- a/tests/install/macos-desktop-e2e.sh +++ b/tests/install/macos-desktop-e2e.sh @@ -270,6 +270,13 @@ phase_update() { ok "serve.git main = $HEAD_SHA" step "updating via $UPDATE_METHOD" + # The app must boot configured or the onboarding overlay (a fullscreen + # div) eats every click: configure the mock inference server exactly like + # the dev:mock flow does, so the app is genuinely configured. + # shellcheck source=../install/e2e-assets/mock-provider.sh + source "$ASSETS/mock-provider.sh" + mock_start "$WORK_ROOT" + trap mock_stop EXIT case "$UPDATE_METHOD" in open-app-update) # The installed app IS the user surface here (double-click the .app); From 1ae3961d3adb31499f33b3eb13a3cfa17d854ebc Mon Sep 17 00:00:00 2001 From: ethernet Date: Wed, 12 Aug 2026 12:10:57 -0400 Subject: [PATCH 081/227] fix ssh redirgithub --- tests/install/installer-script-e2e.sh | 2 ++ tests/install/macos-desktop-e2e.sh | 2 ++ tests/install/windows-e2e.ps1 | 2 ++ 3 files changed, 6 insertions(+) diff --git a/tests/install/installer-script-e2e.sh b/tests/install/installer-script-e2e.sh index ec553fc0c257..ae732843de91 100755 --- a/tests/install/installer-script-e2e.sh +++ b/tests/install/installer-script-e2e.sh @@ -132,6 +132,8 @@ arm_redirect() { cat > "$GIT_CFG" < "$GIT_CFG" < Date: Wed, 12 Aug 2026 15:23:39 -0400 Subject: [PATCH 082/227] hog the pool --- .github/workflows/install-e2e.yml | 6 ------ 1 file changed, 6 deletions(-) diff --git a/.github/workflows/install-e2e.yml b/.github/workflows/install-e2e.yml index 714d10d3799f..5ee1b119a92c 100644 --- a/.github/workflows/install-e2e.yml +++ b/.github/workflows/install-e2e.yml @@ -153,7 +153,6 @@ jobs: # One leg breaking is worth knowing about even if another already # failed, so let every leg report. fail-fast: false - max-parallel: 4 matrix: ${{ fromJSON(needs.generate-matrix.outputs.linux) }} uses: ./.github/workflows/install-e2e-run.yml with: @@ -168,9 +167,6 @@ jobs: needs: generate-matrix strategy: fail-fast: false - # 2x-cost runners at ~14 min a leg: two at a time is throughput - # enough without hogging the Windows pool. - max-parallel: 2 matrix: ${{ fromJSON(needs.generate-matrix.outputs.windows) }} uses: ./.github/workflows/install-e2e-windows-run.yml with: @@ -185,8 +181,6 @@ jobs: needs: generate-matrix strategy: fail-fast: false - # 10x-cost runners: keep concurrency low. - max-parallel: 2 matrix: ${{ fromJSON(needs.generate-matrix.outputs.macos) }} # Two driver arms: the OS-agnostic script driver (shared with linux) # and the published-dmg GUI driver; the run workflow routes. From 05ffab9d1857044e03f591abd54ab74c18c6af8a Mon Sep 17 00:00:00 2001 From: ethernet Date: Wed, 12 Aug 2026 15:28:08 -0400 Subject: [PATCH 083/227] feat(install-e2e): playback.html leg player - zip in, video + time-synced logs A static single-file player (tests/install/e2e-assets/playback.html): ?zip= unzips in-browser (JSZip), plays the screen recording with a timer pinned top-left, and renders every *.log with video<->log sync: the video follows the driver's transcript, clicking a log line seeks the video. A sync-offset slider aligns the recording start (ffmpeg comes up first) with the driver's relative clock. Sync axis: drivers now prefix every transcript line with [+MM:SS] relative to driver start (ts-prefix.sh / ts-prefix.ps1, pipe-safe under pipefail / relaxed EAP). Browsers cannot play Matroska, so each leg remuxes recording.mkv -> recording.mp4 (-c copy, no re-encode) before the artifact upload, on all three OSes. Verified end-to-end in a real browser against a generated artifact zip: zip load, mp4 playback, timer, tab switching, follow-sync at t=6/t=12, click-to-seek, autoplay policy (expected NotAllowedError on synthetic play; real clicks fine). Also fixes the shim fail message's dead variable ( -> observed_git_url) in both posix drivers. --- .github/workflows/install-e2e-macos-run.yml | 9 + .github/workflows/install-e2e-run.yml | 11 + .github/workflows/install-e2e-windows-run.yml | 9 + tests/install/e2e-assets/playback.html | 308 ++++++++++++++++++ tests/install/e2e-assets/ts-prefix.ps1 | 18 + tests/install/e2e-assets/ts-prefix.sh | 21 ++ tests/install/installer-script-e2e.sh | 22 +- tests/install/macos-desktop-e2e.sh | 18 +- tests/install/windows-e2e.ps1 | 9 +- 9 files changed, 404 insertions(+), 21 deletions(-) create mode 100644 tests/install/e2e-assets/playback.html create mode 100644 tests/install/e2e-assets/ts-prefix.ps1 create mode 100644 tests/install/e2e-assets/ts-prefix.sh diff --git a/.github/workflows/install-e2e-macos-run.yml b/.github/workflows/install-e2e-macos-run.yml index 3e03740e2497..61b5335c8115 100644 --- a/.github/workflows/install-e2e-macos-run.yml +++ b/.github/workflows/install-e2e-macos-run.yml @@ -131,6 +131,15 @@ jobs: mode: stop output: ${{ runner.temp }}/e2e-logs/recording.mkv + - name: Remux recording for browser playback + if: always() + run: | + set -euo pipefail + if [ -f "${{ runner.temp }}/e2e-logs/recording.mkv" ]; then + ffmpeg -y -hide_banner -loglevel error -i "${{ runner.temp }}/e2e-logs/recording.mkv" \ + -c copy "${{ runner.temp }}/e2e-logs/recording.mp4" + fi + # Artifact names cannot contain '/'; install-ref may be a full ref. - name: Build artifact name id: artifact diff --git a/.github/workflows/install-e2e-run.yml b/.github/workflows/install-e2e-run.yml index f45848a5e67b..226e7aeb8141 100644 --- a/.github/workflows/install-e2e-run.yml +++ b/.github/workflows/install-e2e-run.yml @@ -119,6 +119,17 @@ jobs: mode: stop output: ${{ runner.temp }}/e2e-logs/recording.mkv + # Browsers cannot play Matroska: remux (copy codec, no re-encode) so + # the artifact zip feeds the static playback.html leg player directly. + - name: Remux recording for browser playback + if: always() + run: | + set -euo pipefail + if [ -f "${{ runner.temp }}/e2e-logs/recording.mkv" ]; then + ffmpeg -y -hide_banner -loglevel error -i "${{ runner.temp }}/e2e-logs/recording.mkv" \ + -c copy "${{ runner.temp }}/e2e-logs/recording.mp4" + fi + # Artifact names cannot contain '/', and install-ref may be a full ref # like refs/heads/main. GitHub Actions expressions have no string-replace # function, so build the safe name here. Runs even on failure -- that is diff --git a/.github/workflows/install-e2e-windows-run.yml b/.github/workflows/install-e2e-windows-run.yml index cd043151d2be..13fed4d1fdaf 100644 --- a/.github/workflows/install-e2e-windows-run.yml +++ b/.github/workflows/install-e2e-windows-run.yml @@ -144,6 +144,15 @@ jobs: mode: stop output: ${{ github.workspace }}\gui-e2e-proof\recording.mkv + - name: Remux recording for browser playback + if: always() + shell: pwsh + run: | + $mkv = "$env:GITHUB_WORKSPACE\gui-e2e-proof\recording.mkv" + if (Test-Path -LiteralPath $mkv) { + & ffmpeg -y -hide_banner -loglevel error -i $mkv -c copy "$env:GITHUB_WORKSPACE\gui-e2e-proof\recording.mp4" + } + - name: Collect proof + logs if: always() shell: powershell diff --git a/tests/install/e2e-assets/playback.html b/tests/install/e2e-assets/playback.html new file mode 100644 index 000000000000..405bfb221caa --- /dev/null +++ b/tests/install/e2e-assets/playback.html @@ -0,0 +1,308 @@ + + + + + +hermes install-e2e leg player + + + + +
loading…
+
+
+ install-e2e leg player + +
+
+
+ +
0:00 / 0:00
+
+
+ + + + + + +
+
+
+
+
— following —
+
+
+
+
+
+ + + diff --git a/tests/install/e2e-assets/ts-prefix.ps1 b/tests/install/e2e-assets/ts-prefix.ps1 new file mode 100644 index 000000000000..86398ca32800 --- /dev/null +++ b/tests/install/e2e-assets/ts-prefix.ps1 @@ -0,0 +1,18 @@ +# Add-TsPrefix: prefix each pipeline line with [+MM:SS] relative to the +# moment the pipeline started (the playback.html leg player's sync axis). +# +# Usage (dot-sourced from a driver): +# & cmd 2>&1 | Add-TsPrefix | Out-File -Encoding UTF8 $log +# +# Call under the relaxed-EAP dance the driver already uses around native +# invocations (merging stderr through a pipe under EAP=Stop turns native +# stderr chatter into a terminating NativeCommandError). + +$script:TsPrefixStart = Get-Date + +function Add-TsPrefix { + process { + $t = (Get-Date) - $script:TsPrefixStart + "[+{0:D2}:{1:D2}] {2}" -f [math]::Floor($t.TotalMinutes), [math]::Floor($t.TotalSeconds % 60), $_ + } +} diff --git a/tests/install/e2e-assets/ts-prefix.sh b/tests/install/e2e-assets/ts-prefix.sh new file mode 100644 index 000000000000..b340517dc14f --- /dev/null +++ b/tests/install/e2e-assets/ts-prefix.sh @@ -0,0 +1,21 @@ +#!/usr/bin/env bash +# ts_prefix: prefix each stdin line with [+MM:SS] relative to the moment +# the pipe started. The playback.html leg player uses these prefixes to +# sync a log file to the screen recording's timeline (the video starts a +# few seconds before the driver, hence the player's offset slider). +# +# Usage (sourced from a driver): +# cmd 2>&1 | ts_prefix > "$LOG_DIR/x.log" +# +# Must be the LAST consumer in the pipe: with `set -o pipefail` the exit +# status stays the command's, while ts_prefix's own status is always 0. +# SECONDS is bash-specific, so this helper is bash-only. + +ts_prefix() { + local _start=$SECONDS + local _line _t + while IFS= read -r _line; do + _t=$((SECONDS - _start)) + printf '[+%02d:%02d] %s\n' $((_t / 60)) $((_t % 60)) "$_line" + done +} diff --git a/tests/install/installer-script-e2e.sh b/tests/install/installer-script-e2e.sh index ae732843de91..d7f3e77b37f4 100755 --- a/tests/install/installer-script-e2e.sh +++ b/tests/install/installer-script-e2e.sh @@ -85,6 +85,8 @@ SERVE_REPO="$WORK_ROOT/serve.git" step() { printf '\n=== %s ===\n' "$*"; } ok() { printf ' OK %s\n' "$*"; } fail() { printf 'E2E ASSERTION FAILED: %s\n' "$*" >&2; exit 1; } +# shellcheck source=../e2e-assets/ts-prefix.sh +source "$(dirname "$0")/e2e-assets/ts-prefix.sh" 2>/dev/null || ts_prefix() { cat; } # Full transcript in the job log, collapsed (GitHub renders ::group:: as a # fold; plain text anywhere else). Win or lose -- a green install's log is # how you diagnose the leg that fails next. @@ -178,7 +180,7 @@ EOF # check it worked observed_git_url="$(git -C "$REPO_ROOT" remote get-url origin)" if [[ "$observed_git_url" != "$REPO_URL_HTTPS" ]]; then - fail "failed git remote get-url shim: origin resolves to '$actual', expected '$REPO_URL_HTTPS'" + fail "failed git remote get-url shim: origin resolves to '$observed_git_url', expected '$REPO_URL_HTTPS'" fi ok "git remote get-url shim: $SHIM_DIR/git -> $REAL_GIT (origin reports $REPO_URL_HTTPS)" } @@ -238,7 +240,7 @@ run_installer() { # "$LOG_DIR/install-$2.log" 2>&1 || rc=$? + bash "$script" "${flags[@]}" < /dev/null 2>&1 | ts_prefix > "$LOG_DIR/install-$2.log" || rc=$? log_group "install.sh ($2) transcript" "$LOG_DIR/install-$2.log" [ "$rc" -eq 0 ] || fail "install.sh ($2) exited $rc; transcript above, log at $LOG_DIR/install-$2.log" } @@ -271,7 +273,7 @@ assert_checkout() { ok "checkout is $2 ($1)" local hermes="$INSTALL_DIR/venv/bin/hermes" [ -x "$hermes" ] || fail "no hermes console script at $hermes" - "$hermes" --version > "$LOG_DIR/version-$2.log" 2>&1 \ + "$hermes" --version 2>&1 | ts_prefix > "$LOG_DIR/version-$2.log" \ || fail "hermes --version failed after $2; log in $LOG_DIR/version-$2.log" ok "hermes --version works: $(head -c 120 "$LOG_DIR/version-$2.log" | tr -d '\n')" } @@ -290,8 +292,8 @@ smoke_desktop() { return 0 fi local rc=0 - (cd "$INSTALL_DIR" && "$hermes" desktop --build-only < /dev/null \ - > "$LOG_DIR/desktop-smoke-$1.log" 2>&1) || rc=$? + (cd "$INSTALL_DIR" && "$hermes" desktop --build-only < /dev/null 2>&1 \ + | ts_prefix > "$LOG_DIR/desktop-smoke-$1.log") || rc=$? log_group "hermes desktop --build-only ($1) transcript" "$LOG_DIR/desktop-smoke-$1.log" [ "$rc" -eq 0 ] || fail "hermes desktop --build-only ($1) exited $rc; transcript above" ok "hermes desktop --build-only works at $1" @@ -335,7 +337,7 @@ case "$UPDATE_METHOD" in update_cmd=("$HERMES" update) fi rc=0 - (cd "$INSTALL_DIR" && "${update_cmd[@]}" < /dev/null > "$LOG_DIR/update.log" 2>&1) || rc=$? + (cd "$INSTALL_DIR" && "${update_cmd[@]}" < /dev/null 2>&1 | ts_prefix > "$LOG_DIR/update.log") || rc=$? log_group "hermes update transcript" "$LOG_DIR/update.log" [ "$rc" -eq 0 ] || fail "hermes update exited $rc; transcript above, log at $LOG_DIR/update.log" ;; @@ -373,7 +375,7 @@ case "$UPDATE_METHOD" in (cd "$INSTALL_DIR" && \ PYTHONPATH="$ASSETS/launch-capture${PYTHONPATH:+:$PYTHONPATH}" \ HERMES_E2E_CAPTURE_LAUNCH="$SPEC" \ - "$HERMES" desktop < /dev/null > "$LOG_DIR/desktop-launch-capture.log" 2>&1) || rc=$? + "$HERMES" desktop < /dev/null 2>&1 | ts_prefix > "$LOG_DIR/desktop-launch-capture.log") || rc=$? log_group "hermes desktop (launch capture) transcript" "$LOG_DIR/desktop-launch-capture.log" [ "$rc" -eq 0 ] || fail "hermes desktop exited $rc during launch capture; transcript above" # Exit 0 without a capture means a version that never reached its @@ -388,7 +390,7 @@ case "$UPDATE_METHOD" in PW_DIR="$WORK_ROOT/playwright" mkdir -p "$PW_DIR" (cd "$PW_DIR" && npm install --no-save --no-audit --no-fund \ - "@playwright/test@1.58.2" > "$LOG_DIR/playwright-install.log" 2>&1) \ + "@playwright/test@1.58.2" 2>&1 | ts_prefix > "$LOG_DIR/playwright-install.log") \ || { log_group "playwright install transcript" "$LOG_DIR/playwright-install.log"; fail "playwright install failed"; } cp "$ASSETS/launch-from-spec.mjs" "$PW_DIR/" rc=0 @@ -396,8 +398,8 @@ case "$UPDATE_METHOD" in --spec "$SPEC" \ --result "$HERMES_HOME/.hermes-update-result.json" \ --expect-sha "$HEAD_SHA" \ - --repo-dir "$INSTALL_DIR" \ - > "$LOG_DIR/app-update.log" 2>&1) || rc=$? + --repo-dir "$INSTALL_DIR" 2>&1 \ + | ts_prefix > "$LOG_DIR/app-update.log") || rc=$? log_group "app update (Playwright) transcript" "$LOG_DIR/app-update.log" [ "$rc" -eq 0 ] || fail "app-driven update exited $rc; transcript above" ;; diff --git a/tests/install/macos-desktop-e2e.sh b/tests/install/macos-desktop-e2e.sh index 7eb374fc8aeb..cc644e9dafb5 100755 --- a/tests/install/macos-desktop-e2e.sh +++ b/tests/install/macos-desktop-e2e.sh @@ -75,6 +75,8 @@ export HOME_SANDBOX="$WORK_ROOT/home" step() { printf '\n=== %s ===\n' "$*"; } ok() { printf ' OK %s\n' "$*"; } fail() { printf 'E2E ASSERTION FAILED: %s\n' "$*" >&2; exit 1; } +# shellcheck source=../e2e-assets/ts-prefix.sh +source "$(dirname "$0")/e2e-assets/ts-prefix.sh" 2>/dev/null || ts_prefix() { cat; } log_group() { printf '::group::%s\n' "$1" cat "$2" @@ -140,7 +142,7 @@ EOF # check it worked observed_git_url="$(git -C "$REPO_DIR" remote get-url origin)" if [[ "$observed_git_url" != "$REPO_URL_HTTPS" ]]; then - fail "failed git remote get-url shim: origin resolves to '$actual', expected '$REPO_URL_HTTPS'" + fail "failed git remote get-url shim: origin resolves to '$observed_git_url', expected '$REPO_URL_HTTPS'" fi ok "git remote get-url shim: $SHIM_DIR/git -> $REAL_GIT (origin reports $REPO_URL_HTTPS)" @@ -224,7 +226,7 @@ phase_install() { # isolation would silently evaporate. Direct exec is the same binary and # the same first-launch flow. local rc=0 - "$app_bin" > "$LOG_DIR/bootstrap-install.log" 2>&1 || rc=$? + "$app_bin" 2>&1 | ts_prefix > "$LOG_DIR/bootstrap-install.log" || rc=$? log_group "Hermes-Setup (dmg bootstrap) transcript" "$LOG_DIR/bootstrap-install.log" hdiutil detach "$mount" >/dev/null 2>&1 || true [ "$rc" -eq 0 ] || fail "dmg bootstrap exited $rc; transcript above" @@ -236,7 +238,7 @@ phase_install() { ok "checkout is OLD ($OLD_SHA)" local hermes="$INSTALL_DIR/venv/bin/hermes" [ -x "$hermes" ] || fail "no hermes console script at $hermes" - "$hermes" --version > "$LOG_DIR/version-old.log" 2>&1 || fail "hermes --version failed after install" + "$hermes" --version 2>&1 | ts_prefix > "$LOG_DIR/version-old.log" || fail "hermes --version failed after install" ok "hermes --version works: $(head -c 120 "$LOG_DIR/version-old.log" | tr -d '\n')" find_installed_app >/dev/null || fail "no installed Hermes.app after the dmg bootstrap" ok "installed app: $(find_installed_app)" @@ -249,7 +251,7 @@ run_playwright_update() { local pw_dir="$WORK_ROOT/playwright" mkdir -p "$pw_dir" (cd "$pw_dir" && npm install --no-save --no-audit --no-fund \ - "@playwright/test@$PLAYWRIGHT_VERSION" > "$LOG_DIR/playwright-install.log" 2>&1) \ + "@playwright/test@$PLAYWRIGHT_VERSION" 2>&1 | ts_prefix > "$LOG_DIR/playwright-install.log") \ || { log_group "playwright install transcript" "$LOG_DIR/playwright-install.log"; fail "playwright install failed"; } cp "$ASSETS/launch-from-spec.mjs" "$pw_dir/" local rc=0 @@ -257,8 +259,8 @@ run_playwright_update() { --spec "$spec" \ --result "$HERMES_HOME/.hermes-update-result.json" \ --expect-sha "$HEAD_SHA" \ - --repo-dir "$INSTALL_DIR" \ - > "$LOG_DIR/app-update.log" 2>&1) || rc=$? + --repo-dir "$INSTALL_DIR" 2>&1 \ + | ts_prefix > "$LOG_DIR/app-update.log") || rc=$? log_group "app update (Playwright) transcript" "$LOG_DIR/app-update.log" [ "$rc" -eq 0 ] || fail "app-driven update exited $rc; transcript above" } @@ -308,7 +310,7 @@ PYEOF (cd "$INSTALL_DIR" && \ PYTHONPATH="$ASSETS/launch-capture${PYTHONPATH:+:$PYTHONPATH}" \ HERMES_E2E_CAPTURE_LAUNCH="$spec" \ - "$hermes" desktop < /dev/null > "$LOG_DIR/desktop-launch-capture.log" 2>&1) || rc=$? + "$hermes" desktop < /dev/null 2>&1 | ts_prefix > "$LOG_DIR/desktop-launch-capture.log") || rc=$? log_group "hermes desktop (launch capture) transcript" "$LOG_DIR/desktop-launch-capture.log" [ "$rc" -eq 0 ] || fail "hermes desktop exited $rc during launch capture" [ -f "$spec.captured" ] || fail "hermes desktop exited 0 but no launch was captured" @@ -321,7 +323,7 @@ PYEOF got="$(git -C "$INSTALL_DIR" rev-parse HEAD)" [ "$got" = "$HEAD_SHA" ] || fail "checkout is $got, expected HEAD ($HEAD_SHA)" ok "checkout landed on HEAD ($HEAD_SHA)" - "$INSTALL_DIR/venv/bin/hermes" --version > "$LOG_DIR/version-head.log" 2>&1 \ + "$INSTALL_DIR/venv/bin/hermes" --version 2>&1 | ts_prefix > "$LOG_DIR/version-head.log" \ || fail "hermes --version failed after update" ok "hermes --version works post-update" step "PASS: $OLD_REF -> HEAD via $UPDATE_METHOD" diff --git a/tests/install/windows-e2e.ps1 b/tests/install/windows-e2e.ps1 index 3480b628a9bd..5e2e77c6d98b 100644 --- a/tests/install/windows-e2e.ps1 +++ b/tests/install/windows-e2e.ps1 @@ -291,6 +291,9 @@ function Test-HermesRuns([string]$Label) { # shipped AT the ref under test, run with flags probed from that ref's own # script text - older releases reject parameters added later). # ---------------------------------------------------------------------------- +# shellcheck source=../e2e-assets/ts-prefix.ps1 +. (Join-Path $PSScriptRoot "e2e-assets\ts-prefix.ps1") + function Write-LogGroup([string]$Title, [string]$LogPath) { Write-Host "::group::$Title" if (Test-Path -LiteralPath $LogPath) { Get-Content -LiteralPath $LogPath | Write-Host } @@ -316,7 +319,7 @@ function Invoke-RefInstaller { New-Item -ItemType Directory -Path (Join-Path $WorkRoot "logs") -Force | Out-Null $log = Join-Path $WorkRoot "logs\install-$Label.log" $prevEap = $ErrorActionPreference; $ErrorActionPreference = "Continue" - & powershell -NoProfile -ExecutionPolicy Bypass -File $script @flags *> $log + & powershell -NoProfile -ExecutionPolicy Bypass -File $script @flags 2>&1 | Add-TsPrefix | Out-File -Encoding UTF8 $log $installExit = $LASTEXITCODE $ErrorActionPreference = $prevEap Write-LogGroup "install.ps1 ($Label) transcript" $log @@ -339,7 +342,7 @@ function Invoke-HermesUpdate { $log = Join-Path $WorkRoot "logs\update.log" Push-Location $InstallDir try { - & $hermesExe @updateArgs *> $log + & $hermesExe @updateArgs 2>&1 | Add-TsPrefix | Out-File -Encoding UTF8 $log $updateExit = $LASTEXITCODE } finally { Pop-Location @@ -367,7 +370,7 @@ function Invoke-HermesDesktopAppUpdate([string]$TargetSha) { $prevEap = $ErrorActionPreference; $ErrorActionPreference = "Continue" Push-Location $InstallDir try { - & $hermesExe desktop *> $log + & $hermesExe desktop 2>&1 | Add-TsPrefix | Out-File -Encoding UTF8 $log $capExit = $LASTEXITCODE } finally { Pop-Location From e138cb555dce08c01ba54e5f456e44ae18f443c3 Mon Sep 17 00:00:00 2001 From: ethernet Date: Wed, 12 Aug 2026 15:50:46 -0400 Subject: [PATCH 084/227] feat(install-e2e): hook the leg player into the results table MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Each leg uploads playback.html as a single-file artifact (archive: false) before the driver runs, so it exists even on failure. The results chart now links every leg that RAN (pass or fail, not skip) to its player with ?zip= pointing at that leg's logs artifact. Leg<->artifact mapping: the generator mints a leg_id per matrix entry (sanitized matrix name, exported legId()), every run workflow names its artifacts install-e2e-{player,logs}-, and the report job feeds the run's artifact name->id list to the results renderer, which rebuilds the leg id from the parsed job name. GitHub does not link jobs to artifacts, so the deterministic name is the join key. Empirical finding: GitHub artifact downloads are auth-gated (the download URL 307s to /suites/... which is 404 anonymous), so a locally-opened player page cannot fetch the zip cross-origin. The player now degrades gracefully: ?zip= fetch failure renders a real download link for the zip (a normal click carries the user's session) plus a drag-and-drop / file-picker path, and no-param opens as a pure drop target. Verified in a real browser against a real artifact URL. Verified: generator emits leg_id, results renderer emits ✅/❌ [📼](...?zip=...) only on ran cells, npm run check PASS, install tests 36/36, strict tsc PASS, actionlint x4 PASS. --- .github/workflows/install-e2e-macos-run.yml | 20 ++++- .github/workflows/install-e2e-run.yml | 21 ++++- .github/workflows/install-e2e-windows-run.yml | 16 +++- .github/workflows/install-e2e.yml | 14 +++- scripts/sandbox/generate-e2e-matrix.mjs | 62 +++++++++++--- tests/install/e2e-assets/playback.html | 82 +++++++++++++++---- 6 files changed, 177 insertions(+), 38 deletions(-) diff --git a/.github/workflows/install-e2e-macos-run.yml b/.github/workflows/install-e2e-macos-run.yml index 61b5335c8115..e2d18951059f 100644 --- a/.github/workflows/install-e2e-macos-run.yml +++ b/.github/workflows/install-e2e-macos-run.yml @@ -34,10 +34,14 @@ on: type: string default: refs/heads/main tag-has-desktop: - description: "Whether install-ref ships the desktop app (apps/desktop). Desktop-method legs from pre-desktop releases natively skip." + description: "Whether install-ref ships the desktop app (apps/desktop). The caller annotates this from the tag's own tree; desktop-method legs from pre-desktop releases natively skip." required: false type: boolean default: true + leg-id: + description: 'Artifact-safe matrix leg id (from generate-e2e-matrix.mjs legId). Names this leg''s logs + player artifacts so the report job can link a row to its zip.' + required: true + type: string dmg-url: description: 'Bootstrap dmg to install OLD with. Default: the latest published one — what a user downloads today.' required: false @@ -67,6 +71,7 @@ jobs: update-method: ${{ inputs.update-method }} install-ref: ${{ inputs.install-ref }} tag-has-desktop: ${{ inputs.tag-has-desktop }} + leg-id: ${{ inputs.leg-id }} runner: macos-latest timeout-minutes: ${{ inputs.timeout-minutes }} @@ -88,6 +93,16 @@ jobs: with: fetch-depth: 0 + - name: Upload leg player + if: always() + uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1 + with: + name: install-e2e-player-${{ inputs.leg-id }} + path: tests/install/e2e-assets/playback.html + archive: false + retention-days: 14 + if-no-files-found: error + - name: Start screen recording uses: ./.github/actions/e2e-screen-record with: @@ -145,8 +160,7 @@ jobs: id: artifact if: always() run: | - safe_ref="$(printf '%s' '${{ inputs.install-ref }}' | tr '/:' '--')" - echo "name=install-e2e-macos-dmg-${{ inputs.update-method }}-${safe_ref}-${{ github.sha }}" >> "$GITHUB_OUTPUT" + echo "name=install-e2e-logs-${{ inputs.leg-id }}" >> "$GITHUB_OUTPUT" - name: Upload logs if: always() diff --git a/.github/workflows/install-e2e-run.yml b/.github/workflows/install-e2e-run.yml index 226e7aeb8141..f47aea2c8d86 100644 --- a/.github/workflows/install-e2e-run.yml +++ b/.github/workflows/install-e2e-run.yml @@ -50,6 +50,10 @@ on: required: false type: string default: refs/heads/main + leg-id: + description: 'Artifact-safe matrix leg id (from generate-e2e-matrix.mjs legId). Names this leg''s logs + player artifacts so the report job can link a row to its zip.' + required: true + type: string tag-has-desktop: description: "Whether install-ref ships the desktop app (apps/desktop). The caller annotates this from the tag's own tree; desktop-method legs from pre-desktop releases natively skip." required: false @@ -130,6 +134,19 @@ jobs: -c copy "${{ runner.temp }}/e2e-logs/recording.mp4" fi + # The leg player: one static HTML, uploaded BEFORE the driver runs + # (if: always() covers failures too). Single file, no zip, so the + # artifacts URL serves playback.html directly. + - name: Upload leg player + if: always() + uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1 + with: + name: install-e2e-player-${{ inputs.leg-id }} + path: tests/install/e2e-assets/playback.html + archive: false + retention-days: 14 + if-no-files-found: error + # Artifact names cannot contain '/', and install-ref may be a full ref # like refs/heads/main. GitHub Actions expressions have no string-replace # function, so build the safe name here. Runs even on failure -- that is @@ -139,9 +156,7 @@ jobs: id: artifact run: | set -euo pipefail - safe_ref='${{ inputs.install-ref }}' - safe_ref="${safe_ref//\//-}" - echo "name=install-e2e-${{ runner.os }}-${{ inputs.update-method }}-${safe_ref}" >> "$GITHUB_OUTPUT" + echo "name=install-e2e-logs-${{ inputs.leg-id }}" >> "$GITHUB_OUTPUT" # The installer's own transcripts say far more than the assertion that # tripped when a real install breaks. diff --git a/.github/workflows/install-e2e-windows-run.yml b/.github/workflows/install-e2e-windows-run.yml index 13fed4d1fdaf..ea67fafb7aa1 100644 --- a/.github/workflows/install-e2e-windows-run.yml +++ b/.github/workflows/install-e2e-windows-run.yml @@ -63,6 +63,10 @@ on: required: false type: boolean default: true + leg-id: + description: 'Artifact-safe matrix leg id (from generate-e2e-matrix.mjs legId). Names this leg''s logs + player artifacts so the report job can link a row to its zip.' + required: true + type: string setup-exe-url: description: 'Bootstrap installer to install OLD with. Default: the latest published one — what a user downloads today.' required: false @@ -115,6 +119,16 @@ jobs: with: fetch-depth: 0 + - name: Upload leg player + if: always() + uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1 + with: + name: install-e2e-player-${{ inputs.leg-id }} + path: tests/install/e2e-assets/playback.html + archive: false + retention-days: 14 + if-no-files-found: error + # One recording mechanism on every OS: the composite action installs # ffmpeg (cached - winget's download is the slow part), starts the # capture, and record-stop fails on a zero-frame file so a silently @@ -175,7 +189,7 @@ jobs: if: always() uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1 with: - name: install-e2e-windows-${{ inputs.install-method }}-${{ inputs.update-method }}-${{ inputs.install-ref }}-${{ github.sha }} + name: install-e2e-logs-${{ inputs.leg-id }} path: gui-e2e-proof retention-days: 14 if-no-files-found: ignore diff --git a/.github/workflows/install-e2e.yml b/.github/workflows/install-e2e.yml index 5ee1b119a92c..eddeec299c04 100644 --- a/.github/workflows/install-e2e.yml +++ b/.github/workflows/install-e2e.yml @@ -160,6 +160,7 @@ jobs: update-method: ${{ matrix.update_method }} install-ref: ${{ matrix.install_ref }} tag-has-desktop: ${{ matrix.tag_has_desktop }} + leg-id: ${{ matrix.leg_id }} windows: name: ${{ matrix.name }} @@ -174,6 +175,7 @@ jobs: update-method: ${{ matrix.update_method }} install-ref: ${{ matrix.install_ref }} tag-has-desktop: ${{ matrix.tag_has_desktop }} + leg-id: ${{ matrix.leg_id }} macos: name: ${{ matrix.name }} @@ -190,6 +192,7 @@ jobs: update-method: ${{ matrix.update_method }} install-ref: ${{ matrix.install_ref }} tag-has-desktop: ${{ matrix.tag_has_desktop }} + leg-id: ${{ matrix.leg_id }} # The outcome, human-readable: the plan chart again, with each cell # replaced by how that leg actually concluded. Per-leg conclusions are @@ -218,7 +221,12 @@ jobs: # The tag annotations let the chart say WHY a cell skipped # (pre-desktop vs declared TODO) instead of a flat "skip". gh api "repos/${{ github.repository }}/actions/runs/${{ github.run_id }}/jobs?per_page=100" \ - --paginate --jq '.jobs[] | {name, conclusion}' | - node scripts/sandbox/generate-e2e-matrix.mjs --format results \ - --tags '${{ needs.pick-releases.outputs.tags }}' + --paginate --jq '.jobs[] | {name, conclusion}' > /tmp/e2e-jobs.ndjson + gh api "repos/${{ github.repository }}/actions/runs/${{ github.run_id }}/artifacts?per_page=100" \ + --paginate --jq '.artifacts[] | {name, id}' > /tmp/e2e-artifacts.ndjson + echo 'Legend: ✅ ran green · ❌ ran red · pre-desktop / TODO = why a leg skipped · 📼 opens the leg player (recording + synced logs)' + echo + node scripts/sandbox/generate-e2e-matrix.mjs --format results \ + --tags '${{ needs.pick-releases.outputs.tags }}' \ + --artifacts /tmp/e2e-artifacts.ndjson < /tmp/e2e-jobs.ndjson } >> "$GITHUB_STEP_SUMMARY" diff --git a/scripts/sandbox/generate-e2e-matrix.mjs b/scripts/sandbox/generate-e2e-matrix.mjs index bba6a435df42..00fa1b110a8c 100644 --- a/scripts/sandbox/generate-e2e-matrix.mjs +++ b/scripts/sandbox/generate-e2e-matrix.mjs @@ -26,6 +26,7 @@ * predate the desktop app). */ +import fs from 'node:fs'; import path from 'node:path'; import { parseArgs } from 'node:util'; import { fileURLToPath } from 'node:url'; @@ -72,10 +73,24 @@ import { fileURLToPath } from 'node:url'; * A picked release tag plus what its own tree ships (annotated by * pick-releases in install-e2e.yml). * - * @typedef {{name: string, install_method: string, update_method: string, + * @typedef {{name: string, leg_id: string, install_method: string, update_method: string, * install_ref: string, tag_has_desktop?: boolean}} MatrixEntry */ +/** + * Artifact-safe key for one leg: the matrix name with every character + * outside [A-Za-z0-9._-] collapsed to '-'. Reconstructed by the results + * renderer from the parsed job name (same formula, same bytes), and used + * by every run workflow to name its logs/player artifacts, so the report + * job can map a concluded leg back to its artifacts without GitHub + * linking jobs to artifacts. + * @param {string} name + * @returns {string} + */ +export function legId(name) { + return name.replace(/[^A-Za-z0-9._-]+/g, '-'); +} + /** @type {Record} */ export const SPEC = { windows: { @@ -189,6 +204,7 @@ export function buildMatrices(envs, tags) { /** @type {MatrixEntry} */ const entry = { name: `${env.os}: ${env.install} -> ${env.update} (${tag.ref} -> HEAD)`, + leg_id: legId(`${env.os}: ${env.install} -> ${env.update} (${tag.ref} -> HEAD)`), install_method: env.install, update_method: env.update, install_ref: tag.ref, @@ -265,10 +281,16 @@ export function renderMarkdownPlan(envs, tags) { * plan got), skipped cells carry their REASON: `pre-desktop` when a * desktop-surface method meets a tag that predates apps/desktop, `TODO` * when the pair is declared but no driver arm runs it yet. + * @param {Map} [artifactById] Artifact name -> id for this + * run (the report job's --artifacts). Legs that RAN (success or failure) + * get a 📼 link: the leg's player artifact (playback.html, archive:false) + * with ?zip= pointing at the leg's logs artifact. Both artifacts are + * named install-e2e-{player,logs}-; leg_id is rebuilt from the + * parsed job name with the same formula the generator mints (legId). * @returns {string} */ -export function renderMarkdownResults(jobs, tagAnnotations = []) { - const LEG = /^(linux|windows|macos): (\S+) -> (\S+) \((\S+) -> HEAD\) \//; +export function renderMarkdownResults(jobs, tagAnnotations = [], artifactById = new Map()) { + const LEG = /^(linux|windows|macos): (\S+) -> (\S+) \(([^)]+) -> HEAD\) \//; /** @type {Map} */ const desktopByTag = new Map(tagAnnotations.map((t) => [t.ref, t.desktop])); /** @@ -303,14 +325,20 @@ export function renderMarkdownResults(jobs, tagAnnotations = []) { if (!tags.includes(tag)) tags.push(tag); if (!rows.has(combo)) rows.set(combo, new Map()); const cell = (() => { - switch (job.conclusion) { - case 'success': return '✅'; - case 'failure': return '❌'; - case 'skipped': return skipLabel(m[2], m[3], tag); - case 'cancelled': return 'cancelled'; - default: return 'running'; - } - })(); + const legId2 = legId(`${m[1]}: ${m[2]} -> ${m[3]} (${m[4]} -> HEAD)`); + const playerId = artifactById.get(`install-e2e-player-${legId2}`); + const logsId = artifactById.get(`install-e2e-logs-${legId2}`); + const reel = (playerId !== undefined && logsId !== undefined) + ? ` [📼](https://github.com/${process.env.GITHUB_REPOSITORY}/actions/runs/${process.env.GITHUB_RUN_ID}/artifacts/${playerId}?zip=${encodeURIComponent(`https://github.com/${process.env.GITHUB_REPOSITORY}/actions/runs/${process.env.GITHUB_RUN_ID}/artifacts/${logsId}`)})` + : ''; + switch (job.conclusion) { + case 'success': return `✅${reel}`; + case 'failure': return `❌${reel}`; + case 'skipped': return skipLabel(m[2], m[3], tag); + case 'cancelled': return 'cancelled'; + default: return 'running'; + } + })(); const byTag = /** @type {Map} */ (rows.get(combo)); const prev = byTag.get(tag); if (prev === undefined || RANK.indexOf(cell) > RANK.indexOf(prev)) { @@ -351,12 +379,22 @@ async function main() { options: { tags: { type: 'string', default: '[]' }, format: { type: 'string', default: 'json' }, + artifacts: { type: 'string' }, }, }); if (values.format === 'results') { const jobs = (await readStdin()).split('\n').filter((l) => l.trim()).map((l) => JSON.parse(l)); const annotations = /** @type {TagAnnotation[]} */ (JSON.parse(values.tags)); - process.stdout.write(renderMarkdownResults(jobs, annotations)); + /** @type {Map} */ + const artifactById = new Map(); + if (values.artifacts) { + for (const line of (await fs.promises.readFile(values.artifacts, 'utf8')).split('\n')) { + if (!line.trim()) continue; + const a = JSON.parse(line); + if (typeof a.name === 'string' && typeof a.id === 'number') artifactById.set(a.name, a.id); + } + } + process.stdout.write(renderMarkdownResults(jobs, annotations, artifactById)); return; } const tags = /** @type {TagAnnotation[]} */ (JSON.parse(values.tags)); diff --git a/tests/install/e2e-assets/playback.html b/tests/install/e2e-assets/playback.html index 405bfb221caa..4bece90d7b08 100644 --- a/tests/install/e2e-assets/playback.html +++ b/tests/install/e2e-assets/playback.html @@ -227,11 +227,8 @@ } /* ── zip loading ───────────────────────────────────────────────────────── */ -async function loadZip(url) { - statusEl.textContent = `fetching ${url} …`; - const resp = await fetch(url, { credentials: 'same-origin' }); - if (!resp.ok) throw new Error(`zip fetch failed: ${resp.status} ${resp.statusText}`); - const zip = await JSZip.loadAsync(await resp.arrayBuffer()); +async function loadZipFromBlob(blob, label) { + const zip = await JSZip.loadAsync(await blob.arrayBuffer()); const names = Object.keys(zip.files).filter((n) => !zip.files[n].dir); const videoName = names.find((n) => n === 'recording.mp4') || @@ -245,26 +242,64 @@ state.logs.set(n, { lines, hasTs: lines.some((l) => l.ts !== null) }); } if (state.logs.size === 0) throw new Error('no *.log files in the zip'); - state.zipName = url.split('/').pop() || url; + state.zipName = label || 'leg artifact'; state.activeTab = [...state.logs.keys()][0]; } -/* ── boot ──────────────────────────────────────────────────────────────── */ -async function main() { - const params = new URLSearchParams(location.search); - const zipUrl = params.get('zip'); - if (!zipUrl) { - fail('no ?zip= param — pass the artifact zip URL, e.g.\n' + - 'playback.html?zip=https://github.com//actions/runs//artifacts/'); - return; +async function loadZip(url) { + statusEl.textContent = `fetching ${url} …`; + try { + const resp = await fetch(url, { credentials: 'same-origin' }); + if (!resp.ok) throw new Error(`zip fetch failed: ${resp.status} ${resp.statusText}`); + await loadZipFromBlob(await resp.blob(), url.split('/').pop()); + } catch (e) { + // GitHub artifact downloads are auth-gated: a cross-origin page (this + // file opened locally) cannot fetch them. The page then becomes the + // download-and-drop path — a normal link click DOES carry the user's + // GitHub session. + statusEl.className = 'error'; + statusEl.innerHTML = + `Could not fetch the zip directly (GitHub artifact downloads need your session — ` + + `a locally-opened page cannot send it). ` + + `Download the leg zip: ${url.split('/').pop()} ` + + `— then drop it anywhere on this page (or click here to pick the file).`; + statusEl.onclick = () => fileInput.click(); + return false; } + return true; +} + +/* drag-drop + file picker: the universal transport for auth-gated zips */ +const fileInput = document.createElement('input'); +fileInput.type = 'file'; +fileInput.accept = '.zip,application/zip'; +fileInput.style.display = 'none'; +fileInput.addEventListener('change', async () => { + const f = fileInput.files && fileInput.files[0]; + if (!f) return; try { - await loadZip(zipUrl); + await loadZipFromBlob(f, f.name); + boot(); } catch (e) { fail(`could not load zip: ${e.message}`); - return; } +}); +document.body.appendChild(fileInput); +document.addEventListener('dragover', (e) => e.preventDefault()); +document.addEventListener('drop', async (e) => { + e.preventDefault(); + const f = [...(e.dataTransfer?.files ?? [])][0]; + if (!f || !/\.zip$/i.test(f.name)) { fail('drop a .zip artifact file'); return; } + try { + await loadZipFromBlob(f, f.name); + boot(); + } catch (err) { + fail(`could not load zip: ${err.message}`); + } +}); +/* ── boot ──────────────────────────────────────────────────────────────── */ +function boot() { statusEl.style.display = 'none'; $('app').style.display = 'flex'; $('zipName').textContent = state.zipName; @@ -302,6 +337,21 @@ renderTabs(); renderPane(true); } + +async function main() { + const params = new URLSearchParams(location.search); + const zipUrl = params.get('zip'); + if (zipUrl) { + if (await loadZip(zipUrl)) boot(); + } else { + // No ?zip=: this is the manual path — the leg player artifact page or + // a local copy. Drop or pick a leg zip. + statusEl.className = 'error'; + statusEl.innerHTML = + 'Drop a leg artifact zip anywhere on this page (or click here to pick the file).'; + statusEl.onclick = () => fileInput.click(); + } +} main(); From 41e9fee5b145117ec237288f6728c63c1cfec346 Mon Sep 17 00:00:00 2001 From: ethernet Date: Wed, 12 Aug 2026 15:54:42 -0400 Subject: [PATCH 085/227] default to 2 tags --- .github/workflows/install-e2e.yml | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/.github/workflows/install-e2e.yml b/.github/workflows/install-e2e.yml index eddeec299c04..746d540c91f5 100644 --- a/.github/workflows/install-e2e.yml +++ b/.github/workflows/install-e2e.yml @@ -97,7 +97,7 @@ jobs: - id: pick run: | set -euo pipefail - tags="$(scripts/sandbox/pick-release-tags.sh --count '${{ inputs.tag-count || 5 }}')" + tags="$(scripts/sandbox/pick-release-tags.sh --count '${{ inputs.tag-count || 2 }}')" echo "Testing updates from: $tags" # Annotate each tag with what its own tree supports, so run # workflows can natively skip surfaces the starting version does From 08d9f27773c4077f634ce3a54d71489241ab054a Mon Sep 17 00:00:00 2001 From: ethernet Date: Thu, 13 Aug 2026 14:41:04 -0400 Subject: [PATCH 086/227] =?UTF-8?q?fix(install-e2e):=20results=20chart=20?= =?UTF-8?q?=F0=9F=93=BC=20links=20against=20real=20artifact=20names?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Debugged against run 31635036702 (real jobs + artifacts replayed through the renderer). Two naming realities the renderer ignored: 1. upload-artifact with archive:false IGNORES the name: input and registers the artifact under the FILE's basename - every leg's player artifact is called 'playback.html' (all the same blob). The renderer now uses any 'playback.html' artifact for the player half of the 📼 link instead of install-e2e-player-. 2. The posix arms append - to the logs artifact name at upload (install-e2e-logs--), while windows does not. Match by prefix instead of exact name. Also fixed the pass/fail summary counters, which compared cells with === against '✅' - the appended reel link made every ran cell count as neither passed nor failed (the summary said 0 passed while the table was full of checks). Replayed against the real run: 31 passed, 25 failed, 149 skipped, 24 📼 links on ran cells (both outcomes), none on skips. --- scripts/sandbox/generate-e2e-matrix.mjs | 28 ++++++++++++++++++------- 1 file changed, 20 insertions(+), 8 deletions(-) diff --git a/scripts/sandbox/generate-e2e-matrix.mjs b/scripts/sandbox/generate-e2e-matrix.mjs index 00fa1b110a8c..70a9f28acce1 100644 --- a/scripts/sandbox/generate-e2e-matrix.mjs +++ b/scripts/sandbox/generate-e2e-matrix.mjs @@ -283,10 +283,14 @@ export function renderMarkdownPlan(envs, tags) { * when the pair is declared but no driver arm runs it yet. * @param {Map} [artifactById] Artifact name -> id for this * run (the report job's --artifacts). Legs that RAN (success or failure) - * get a 📼 link: the leg's player artifact (playback.html, archive:false) - * with ?zip= pointing at the leg's logs artifact. Both artifacts are - * named install-e2e-{player,logs}-; leg_id is rebuilt from the - * parsed job name with the same formula the generator mints (legId). + * get a 📼 link: the leg's player artifact (playback.html, archive:false + * — GitHub names those after the FILE, so they all collide on + * `playback.html` and any one of them works, they are the same blob) + * with ?zip= pointing at the leg's logs artifact. Logs artifacts are + * `install-e2e-logs-` (windows) or `install-e2e-logs--` + * (posix arms append the sha at upload) — matched by prefix. leg_id is + * rebuilt from the parsed job name with the same formula the generator + * mints (legId). * @returns {string} */ export function renderMarkdownResults(jobs, tagAnnotations = [], artifactById = new Map()) { @@ -326,8 +330,16 @@ export function renderMarkdownResults(jobs, tagAnnotations = [], artifactById = if (!rows.has(combo)) rows.set(combo, new Map()); const cell = (() => { const legId2 = legId(`${m[1]}: ${m[2]} -> ${m[3]} (${m[4]} -> HEAD)`); - const playerId = artifactById.get(`install-e2e-player-${legId2}`); - const logsId = artifactById.get(`install-e2e-logs-${legId2}`); + // Logs artifacts: `install-e2e-logs-` (windows) or with a + // trailing `-` (posix arms append it at upload). Match by prefix. + const logsName = [...artifactById.keys()].find((n) => n.startsWith(`install-e2e-logs-${legId2}`)); + // Player artifacts: upload-artifact `archive: false` names the + // artifact after the FILE (playback.html), ignoring `name:` — so + // every leg's player artifact is called `playback.html`, and they + // are all the same blob. Any one of them works. + const playerName = [...artifactById.keys()].find((n) => n === 'playback.html'); + const playerId = playerName !== undefined ? artifactById.get(playerName) : undefined; + const logsId = logsName !== undefined ? artifactById.get(logsName) : undefined; const reel = (playerId !== undefined && logsId !== undefined) ? ` [📼](https://github.com/${process.env.GITHUB_REPOSITORY}/actions/runs/${process.env.GITHUB_RUN_ID}/artifacts/${playerId}?zip=${encodeURIComponent(`https://github.com/${process.env.GITHUB_REPOSITORY}/actions/runs/${process.env.GITHUB_RUN_ID}/artifacts/${logsId}`)})` : ''; @@ -347,8 +359,8 @@ export function renderMarkdownResults(jobs, tagAnnotations = [], artifactById = } if (rows.size === 0) return '### Install & Update E2E results\n\n(no legs found in this run)\n'; const cells = [...rows.values()].flatMap((r) => [...r.values()]); - const passed = cells.filter((c) => c === '✅').length; - const failed = cells.filter((c) => c === '❌').length; + const passed = cells.filter((c) => c.startsWith('✅')).length; + const failed = cells.filter((c) => c.startsWith('❌')).length; const skipped = cells.filter((c) => SKIPS.includes(c)).length; const lines = [ '### Install & Update E2E results', From cd667debfa09b70e4e1287bde430c7756f5e6490 Mon Sep 17 00:00:00 2001 From: ethernet Date: Thu, 13 Aug 2026 22:51:44 -0400 Subject: [PATCH 087/227] fix(install-e2e): windows transcripts were empty; player gets #zip= hash + one player per run MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Windows transcripts were ZERO bytes: ts-prefix.ps1 formatted with {0:D2}, but Floor() returns a double and the D specifier is integer-only - it threw per line, and under the driver's relaxed EAP every line errored into the void. {0:00} fixes it (custom numeric format works on doubles). Reproduced the exact pipeline locally (empty file + Format specifier invalid), verified the fix produces prefixed merged stdout+stderr with exit code intact. That is also why the log timeline never auto-synced: there was nothing in the files to sync. The GitHub artifact URL 307s to /suites/... server-side and strips the ?zip= query param. The player now reads the zip URL from a #zip= HASH param (client-side, survives the redirect) with ?zip= as fallback; the hash path was verified in a real browser against a real leg zip (auto-fetch + boot). Per ethie's design, one player artifact for the whole run: new leg-player job uploads playback.html (archive:false) before the matrix legs, the report job needs it, and each ran cell gets TWO links - 📼 to the player with #zip= and ⬇️ to the raw zip. Per-leg player uploads removed from all three run workflows. --- .github/workflows/install-e2e-macos-run.yml | 10 ------- .github/workflows/install-e2e-run.yml | 21 +++----------- .github/workflows/install-e2e-windows-run.yml | 10 ------- .github/workflows/install-e2e.yml | 26 ++++++++++++++++- scripts/sandbox/generate-e2e-matrix.mjs | 28 ++++++++++--------- tests/install/e2e-assets/playback.html | 23 ++++++++++++--- tests/install/e2e-assets/ts-prefix.ps1 | 5 +++- 7 files changed, 67 insertions(+), 56 deletions(-) diff --git a/.github/workflows/install-e2e-macos-run.yml b/.github/workflows/install-e2e-macos-run.yml index e2d18951059f..de90b863b242 100644 --- a/.github/workflows/install-e2e-macos-run.yml +++ b/.github/workflows/install-e2e-macos-run.yml @@ -93,16 +93,6 @@ jobs: with: fetch-depth: 0 - - name: Upload leg player - if: always() - uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1 - with: - name: install-e2e-player-${{ inputs.leg-id }} - path: tests/install/e2e-assets/playback.html - archive: false - retention-days: 14 - if-no-files-found: error - - name: Start screen recording uses: ./.github/actions/e2e-screen-record with: diff --git a/.github/workflows/install-e2e-run.yml b/.github/workflows/install-e2e-run.yml index f47aea2c8d86..29cc5a8c51eb 100644 --- a/.github/workflows/install-e2e-run.yml +++ b/.github/workflows/install-e2e-run.yml @@ -134,23 +134,10 @@ jobs: -c copy "${{ runner.temp }}/e2e-logs/recording.mp4" fi - # The leg player: one static HTML, uploaded BEFORE the driver runs - # (if: always() covers failures too). Single file, no zip, so the - # artifacts URL serves playback.html directly. - - name: Upload leg player - if: always() - uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1 - with: - name: install-e2e-player-${{ inputs.leg-id }} - path: tests/install/e2e-assets/playback.html - archive: false - retention-days: 14 - if-no-files-found: error - - # Artifact names cannot contain '/', and install-ref may be a full ref - # like refs/heads/main. GitHub Actions expressions have no string-replace - # function, so build the safe name here. Runs even on failure -- that is - # exactly when the logs are wanted. + # The leg player: ONE static HTML per run, uploaded up front by the + # leg-player job in install-e2e.yml (archive: false, so GitHub names + # the artifact after the file: playback.html). The report job links + # every ran leg to it with the leg's zip as a #zip= hash param. - name: Build artifact name if: always() id: artifact diff --git a/.github/workflows/install-e2e-windows-run.yml b/.github/workflows/install-e2e-windows-run.yml index ea67fafb7aa1..c90831a77231 100644 --- a/.github/workflows/install-e2e-windows-run.yml +++ b/.github/workflows/install-e2e-windows-run.yml @@ -119,16 +119,6 @@ jobs: with: fetch-depth: 0 - - name: Upload leg player - if: always() - uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1 - with: - name: install-e2e-player-${{ inputs.leg-id }} - path: tests/install/e2e-assets/playback.html - archive: false - retention-days: 14 - if-no-files-found: error - # One recording mechanism on every OS: the composite action installs # ffmpeg (cached - winget's download is the slow part), starts the # capture, and record-stop fails on a zero-frame file so a silently diff --git a/.github/workflows/install-e2e.yml b/.github/workflows/install-e2e.yml index 746d540c91f5..93d3b3ae5d59 100644 --- a/.github/workflows/install-e2e.yml +++ b/.github/workflows/install-e2e.yml @@ -194,6 +194,30 @@ jobs: tag-has-desktop: ${{ matrix.tag_has_desktop }} leg-id: ${{ matrix.leg_id }} + # The leg player: one static HTML for the whole run. Uploaded BEFORE the + # matrix legs so it exists even when every leg dies; the report job links + # every ran leg to it with that leg's logs zip as a #zip= hash param + # (hash survives the artifact URL's server-side redirect, the query does + # not). archive: false makes GitHub name the artifact after the FILE + # (playback.html), ignoring the name: input -- harmless, the renderer + # looks it up by that name. + leg-player: + name: Upload leg player + runs-on: ubuntu-latest + timeout-minutes: 5 + steps: + - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2 + with: + sparse-checkout: tests/install/e2e-assets/playback.html + sparse-checkout-cone-mode: false + - uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1 + with: + name: install-e2e-player + path: tests/install/e2e-assets/playback.html + archive: false + retention-days: 14 + if-no-files-found: error + # The outcome, human-readable: the plan chart again, with each cell # replaced by how that leg actually concluded. Per-leg conclusions are # NOT reachable through `needs` (a matrix job's result collapses to one @@ -203,7 +227,7 @@ jobs: report: name: Result chart if: always() - needs: [pick-releases, linux, windows, macos] + needs: [leg-player, pick-releases, linux, windows, macos] runs-on: ubuntu-latest timeout-minutes: 5 steps: diff --git a/scripts/sandbox/generate-e2e-matrix.mjs b/scripts/sandbox/generate-e2e-matrix.mjs index 70a9f28acce1..35053d8a653e 100644 --- a/scripts/sandbox/generate-e2e-matrix.mjs +++ b/scripts/sandbox/generate-e2e-matrix.mjs @@ -283,14 +283,14 @@ export function renderMarkdownPlan(envs, tags) { * when the pair is declared but no driver arm runs it yet. * @param {Map} [artifactById] Artifact name -> id for this * run (the report job's --artifacts). Legs that RAN (success or failure) - * get a 📼 link: the leg's player artifact (playback.html, archive:false - * — GitHub names those after the FILE, so they all collide on - * `playback.html` and any one of them works, they are the same blob) - * with ?zip= pointing at the leg's logs artifact. Logs artifacts are - * `install-e2e-logs-` (windows) or `install-e2e-logs--` - * (posix arms append the sha at upload) — matched by prefix. leg_id is - * rebuilt from the parsed job name with the same formula the generator - * mints (legId). + * get TWO links: 📼 to the run's single player artifact (playback.html, + * archive:false names artifacts after the FILE) with the leg's logs zip + * as a #zip= HASH param — the artifact URL 307s server-side and strips + * the query, the hash survives — and ⬇️ to the raw logs zip. Logs + * artifacts are `install-e2e-logs-` (windows) or + * `install-e2e-logs--` (posix arms append the sha at + * upload) — matched by prefix. leg_id is rebuilt from the parsed job + * name with the same formula the generator mints (legId). * @returns {string} */ export function renderMarkdownResults(jobs, tagAnnotations = [], artifactById = new Map()) { @@ -333,15 +333,17 @@ export function renderMarkdownResults(jobs, tagAnnotations = [], artifactById = // Logs artifacts: `install-e2e-logs-` (windows) or with a // trailing `-` (posix arms append it at upload). Match by prefix. const logsName = [...artifactById.keys()].find((n) => n.startsWith(`install-e2e-logs-${legId2}`)); - // Player artifacts: upload-artifact `archive: false` names the - // artifact after the FILE (playback.html), ignoring `name:` — so - // every leg's player artifact is called `playback.html`, and they - // are all the same blob. Any one of them works. + // Player artifacts: the single per-run leg-player upload; archive: + // false names it after the FILE (playback.html), ignoring `name:`. const playerName = [...artifactById.keys()].find((n) => n === 'playback.html'); const playerId = playerName !== undefined ? artifactById.get(playerName) : undefined; const logsId = logsName !== undefined ? artifactById.get(logsName) : undefined; + // Two links per ran leg: the player with the zip as a HASH param + // (GitHub's artifact URL 307s to /suites/... and strips the query — + // the hash survives client-side), and the raw zip download. + const runBase = `https://github.com/${process.env.GITHUB_REPOSITORY}/actions/runs/${process.env.GITHUB_RUN_ID}`; const reel = (playerId !== undefined && logsId !== undefined) - ? ` [📼](https://github.com/${process.env.GITHUB_REPOSITORY}/actions/runs/${process.env.GITHUB_RUN_ID}/artifacts/${playerId}?zip=${encodeURIComponent(`https://github.com/${process.env.GITHUB_REPOSITORY}/actions/runs/${process.env.GITHUB_RUN_ID}/artifacts/${logsId}`)})` + ? ` [📼](${runBase}/artifacts/${playerId}#zip=${encodeURIComponent(`${runBase}/artifacts/${logsId}`)}) [⬇️](${runBase}/artifacts/${logsId})` : ''; switch (job.conclusion) { case 'success': return `✅${reel}`; diff --git a/tests/install/e2e-assets/playback.html b/tests/install/e2e-assets/playback.html index 4bece90d7b08..315d002b64e4 100644 --- a/tests/install/e2e-assets/playback.html +++ b/tests/install/e2e-assets/playback.html @@ -269,6 +269,19 @@ return true; } +/** + * Resolve the zip URL: hash param wins (?zip= gets stripped by GitHub's + * artifact redirect, the hash survives), then the query param. + * @returns {string | null} + */ +function resolveZipUrl() { + const m = location.hash.match(/(?:^|[#&])zip=([^&]+)/); + if (m) { + try { return decodeURIComponent(m[1]); } catch { return m[1]; } + } + return new URLSearchParams(location.search).get('zip'); +} + /* drag-drop + file picker: the universal transport for auth-gated zips */ const fileInput = document.createElement('input'); fileInput.type = 'file'; @@ -339,13 +352,15 @@ } async function main() { - const params = new URLSearchParams(location.search); - const zipUrl = params.get('zip'); + const zipUrl = resolveZipUrl(); if (zipUrl) { + // Attempt the direct fetch (works when the page is served on + // github.com itself — same-origin, session cookies flow). On failure + // the status line becomes a download link + drop target. if (await loadZip(zipUrl)) boot(); } else { - // No ?zip=: this is the manual path — the leg player artifact page or - // a local copy. Drop or pick a leg zip. + // No zip param: this is the manual path — the leg player artifact + // page or a local copy. Drop or pick a leg zip. statusEl.className = 'error'; statusEl.innerHTML = 'Drop a leg artifact zip anywhere on this page (or click here to pick the file).'; diff --git a/tests/install/e2e-assets/ts-prefix.ps1 b/tests/install/e2e-assets/ts-prefix.ps1 index 86398ca32800..cc9acc878055 100644 --- a/tests/install/e2e-assets/ts-prefix.ps1 +++ b/tests/install/e2e-assets/ts-prefix.ps1 @@ -13,6 +13,9 @@ $script:TsPrefixStart = Get-Date function Add-TsPrefix { process { $t = (Get-Date) - $script:TsPrefixStart - "[+{0:D2}:{1:D2}] {2}" -f [math]::Floor($t.TotalMinutes), [math]::Floor($t.TotalSeconds % 60), $_ + # {0:00} not {0:D2}: Floor() returns a double and the D specifier + # is integer-only - it throws per line, and under the driver's + # relaxed EAP every line errors into the void (empty transcripts). + "[+{0:00}:{1:00}] {2}" -f [math]::Floor($t.TotalMinutes), [math]::Floor($t.TotalSeconds % 60), $_ } } From 91ba10835a54dc28e97ab69d9d6dce7cc83d2ae7 Mon Sep 17 00:00:00 2001 From: ethernet Date: Fri, 14 Aug 2026 03:51:27 -0400 Subject: [PATCH 088/227] fix(install-e2e): wait longer for init --- tests/install/e2e-assets/launch-from-spec.mjs | 22 +++++++++++++------ 1 file changed, 15 insertions(+), 7 deletions(-) diff --git a/tests/install/e2e-assets/launch-from-spec.mjs b/tests/install/e2e-assets/launch-from-spec.mjs index 0fb3a85b941e..3dbec88f7d76 100644 --- a/tests/install/e2e-assets/launch-from-spec.mjs +++ b/tests/install/e2e-assets/launch-from-spec.mjs @@ -120,6 +120,7 @@ async function main() { } const deadline = Date.now() + Number(values['timeout-ms']); + // Dismiss the onboarding overlay when present. Two layers of defense: // the drivers seed a provider key so the app's runtime check reports // configured=true and the overlay never mounts; if it shows anyway @@ -127,13 +128,20 @@ async function main() { // hatch - "I'll choose a provider later" (i18n en: chooseLater). The // overlay is a fullscreen div that intercepts ALL clicks, so this must // resolve before any Settings navigation. - const later = window - .getByRole('button', { name: /choose a provider later|skip/i }) - .first(); - if (await later.isVisible({ timeout: 10_000 }).catch(() => false)) { - await later.click().catch(() => {}); - await later.waitFor({ state: 'hidden', timeout: 15_000 }).catch(() => {}); - } + const later = window.getByRole('button', { name: /choose a provider later|skip/i }).first() + const settingsButton = window.getByRole('button', { name: /open settings|settings/i }).first() + + await Promise.race([ + settingsButton.isVisible({ timeout: 60_000 }).catch(() => false), + later + .isVisible({ timeout: 90_000 }) + .catch(() => false) + .then(async () => { + await later.click().catch(() => {}) + await later.waitFor({ state: 'hidden', timeout: 15_000 }).catch(() => {}) + }) + ]) + // Settings -> About -> Update now. The settings trigger is an icon // button whose accessible name is "Open settings". From 4bc6bfede1ef0ad92ab6a60c559cf357a2a5f1ab Mon Sep 17 00:00:00 2001 From: ethernet Date: Fri, 14 Aug 2026 15:10:45 -0400 Subject: [PATCH 089/227] =?UTF-8?q?feat(playback):=20ALL=C2=B7merged=20tab?= =?UTF-8?q?=20-=20every=20log=20on=20one=20video-synced=20timeline?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The merged view interleaves every *.log in time order across the whole leg, filename-prefixed per line (green .file span), and syncs like any other tab: the video highlights the right file's line at the right moment, clicking a line seeks the video. Untimed lines inherit their file's previous timestamp so intra-file order survives the sort; the merged tab is first and the default. Verified in a real browser with an interleaved synthetic zip (two logs, overlapping timelines): time-order merge, prefixes, follow-sync at t=22 highlighting the app-update.log line, click-to-seek to t=30. --- tests/install/e2e-assets/playback.html | 31 ++++++++++++++++++++++++++ 1 file changed, 31 insertions(+) diff --git a/tests/install/e2e-assets/playback.html b/tests/install/e2e-assets/playback.html index 315d002b64e4..fb309d0d0393 100644 --- a/tests/install/e2e-assets/playback.html +++ b/tests/install/e2e-assets/playback.html @@ -60,6 +60,7 @@ border-radius: 3px; } .line:hover { background: rgba(88,166,255,.08); } .line .ts { color: var(--dim); margin-right: 8px; user-select: none; } + .line .file { color: #7ee787; margin-right: 8px; user-select: none; } .line.synced { color: var(--accent); } #noSync { color: var(--dim); padding: 4px 10px; font-size: 12px; } #syncing { position: sticky; top: 0; z-index: 3; background: rgba(13,17,23,.9); @@ -191,6 +192,12 @@ $('video').play().catch(() => {}); }; } + if (ln.file) { + const file = document.createElement('span'); + file.className = 'file'; + file.textContent = `${ln.file}: `; + div.appendChild(file); + } div.appendChild(document.createTextNode(ln.text)); frag.appendChild(div); } @@ -227,6 +234,27 @@ } /* ── zip loading ───────────────────────────────────────────────────────── */ +/** + * Merge every log into one timeline, in time order, with each line + * filename-prefixed. Untimed lines inherit the previous timed line of + * their file so ordering within a file survives the sort. + * @param {Map} logs + * @returns {{lines: {ts: number, text: string, file: string}[], hasTs: boolean}} + */ +function buildMergedLog(logs) { + /** @type {{ts: number, text: string, file: string}[]} */ + const merged = []; + for (const [name, entry] of logs) { + let lastTs = 0; + for (const ln of entry.lines) { + if (ln.ts !== null) lastTs = ln.ts; + merged.push({ ts: lastTs, text: ln.text, file: name }); + } + } + merged.sort((a, b) => a.ts - b.ts); + return { lines: merged, hasTs: true }; +} + async function loadZipFromBlob(blob, label) { const zip = await JSZip.loadAsync(await blob.arrayBuffer()); const names = Object.keys(zip.files).filter((n) => !zip.files[n].dir); @@ -242,6 +270,9 @@ state.logs.set(n, { lines, hasTs: lines.some((l) => l.ts !== null) }); } if (state.logs.size === 0) throw new Error('no *.log files in the zip'); + // The merged view comes first: one timeline across every log, synced + // to the video like any other tab. + state.logs = new Map([['ALL · merged', buildMergedLog(state.logs)], ...state.logs]); state.zipName = label || 'leg artifact'; state.activeTab = [...state.logs.keys()][0]; } From 4c3ee593c0f2d8f4a845a38fc811d09d2fbf44ed Mon Sep 17 00:00:00 2001 From: ethernet Date: Fri, 14 Aug 2026 15:58:51 -0400 Subject: [PATCH 090/227] feat(playback): ansi n styles --- tests/install/e2e-assets/playback.html | 87 +++++++++++++++++--------- 1 file changed, 58 insertions(+), 29 deletions(-) diff --git a/tests/install/e2e-assets/playback.html b/tests/install/e2e-assets/playback.html index fb309d0d0393..df1026d77c1f 100644 --- a/tests/install/e2e-assets/playback.html +++ b/tests/install/e2e-assets/playback.html @@ -13,8 +13,8 @@ * { box-sizing: border-box; } body { margin: 0; background: var(--bg); color: var(--text); font: 13px/1.45 ui-monospace, SFMono-Regular, Menlo, Consolas, monospace; } - #status { position: fixed; inset: 0; display: flex; align-items: center; - justify-content: center; color: var(--dim); font-size: 15px; z-index: 5; } + #status {text-align: center; position: fixed; inset: 0; display: flex-col; align-items: center; + justify-content: center; color: var(--dim); font-size: 15px; z-index: 5; padding: 24px;} #status.error { color: #f85149; } #app { display: none; height: 100vh; flex-direction: column; } @@ -107,7 +107,7 @@ /* Hermes install-e2e leg player. * * One static file. Point it at a run's artifact zip: - * playback.html?zip= + * playback.html#zip= * It unzips in-browser (JSZip from CDN), plays recording.mp4 (preferred) or * recording.mkv, and renders every *.log with a top-left timer and * video↔log sync. Sync axis: driver log lines carry a [+MM:SS] prefix, @@ -171,6 +171,43 @@ renderPane(true); } +function ansiToHtml(text) { + const colors = { + 30: 'black', 31: 'red', 32: 'green', 33: 'yellow', + 34: 'blue', 35: 'magenta', 36: 'cyan', 37: 'white', + 90: 'gray', 91: 'lightred', 92: 'lightgreen', 93: 'lightyellow', + 94: 'lightblue', 95: 'lightmagenta', 96: 'lightcyan', 97: 'white' + }; + + const escapeHtml = s => s.replace(/[&<>]/g, c => ({'&':'&','<':'<','>':'>'}[c])); + + let html = ''; + let openSpan = false; + const parts = text.split(/\x1b\[([0-9;]*)m/); + + // parts alternates: [text, codes, text, codes, ...] + for (let i = 0; i < parts.length; i++) { + if (i % 2 === 0) { + html += escapeHtml(parts[i]); + } else { + if (openSpan) { html += ''; openSpan = false; } + const codes = parts[i].split(';').map(Number); + const styles = []; + for (const code of codes) { + if (code === 0) continue; // reset + if (code === 1) styles.push('font-weight:bold'); + else if (colors[code]) styles.push(`color:${colors[code]}`); + } + if (styles.length) { + html += ``; + openSpan = true; + } + } + } + if (openSpan) html += ''; + return html; +} + function renderPane(scrollToEnd) { const pane = $('pane'); const entry = state.logs.get(state.activeTab); @@ -198,7 +235,9 @@ file.textContent = `${ln.file}: `; div.appendChild(file); } - div.appendChild(document.createTextNode(ln.text)); + const text = document.createElement('span') + text.innerHTML = ansiToHtml(ln.text); + div.appendChild(text); frag.appendChild(div); } pane.appendChild(frag); @@ -278,26 +317,20 @@ } async function loadZip(url) { - statusEl.textContent = `fetching ${url} …`; - try { - const resp = await fetch(url, { credentials: 'same-origin' }); - if (!resp.ok) throw new Error(`zip fetch failed: ${resp.status} ${resp.statusText}`); - await loadZipFromBlob(await resp.blob(), url.split('/').pop()); - } catch (e) { - // GitHub artifact downloads are auth-gated: a cross-origin page (this - // file opened locally) cannot fetch them. The page then becomes the - // download-and-drop path — a normal link click DOES carry the user's - // GitHub session. - statusEl.className = 'error'; - statusEl.innerHTML = - `Could not fetch the zip directly (GitHub artifact downloads need your session — ` + - `a locally-opened page cannot send it). ` + - `Download the leg zip: ${url.split('/').pop()} ` + - `— then drop it anywhere on this page (or click here to pick the file).`; - statusEl.onclick = () => fileInput.click(); - return false; + // GitHub artifact downloads are auth-gated: a cross-origin page (this + // file opened locally) cannot fetch them. The page then becomes the + // download-and-drop path — a normal link click DOES carry the user's + // GitHub session. + statusEl.innerHTML = + `
Download the zip from here: ${url.split('/').pop()}
` + + `
then drop it anywhere on this page
(or click anywhere on the page to pick the file).
`; + statusEl.onclick = (e) => { + fileInput.click() + }; + egg.onclick = (e) => { + e.stopPropagation(); } - return true; + return false; } /** @@ -385,16 +418,12 @@ async function main() { const zipUrl = resolveZipUrl(); if (zipUrl) { - // Attempt the direct fetch (works when the page is served on - // github.com itself — same-origin, session cookies flow). On failure - // the status line becomes a download link + drop target. - if (await loadZip(zipUrl)) boot(); + loadZip(zipUrl) } else { // No zip param: this is the manual path — the leg player artifact // page or a local copy. Drop or pick a leg zip. - statusEl.className = 'error'; statusEl.innerHTML = - 'Drop a leg artifact zip anywhere on this page (or click here to pick the file).'; + 'Drop a leg artifact zip anywhere on this page
(or click here to pick the file).'; statusEl.onclick = () => fileInput.click(); } } From 7807f54b51046d56e79859f856db5f88183d2fe5 Mon Sep 17 00:00:00 2001 From: ethernet Date: Fri, 14 Aug 2026 16:02:50 -0400 Subject: [PATCH 091/227] fix(install-e2e): one time base for every transcript in a leg ts_prefix captured its own start per pipe, so each log's [+MM:SS] was relative to that log's creation - the install log started at +0:00, the app-update log at +0:00 too, and no single offset could align all files with the recording. The drivers now stamp TS_BASE= $SECONDS once at start and ts_prefix stamps every line relative to it (falling back to its own start when unset): all logs in a leg share the driver's clock, and playback.html's one offset slider (recording start vs driver start) aligns every file at once. The ps1 twin was already driver-relative (TsPrefixStart is captured at dot-source time, near the driver's top). Verified: two pipes in one shell - the second starts at [+00:03], continuing the driver clock instead of resetting to [+00:00]. --- tests/install/e2e-assets/ts-prefix.sh | 16 ++++++++++------ tests/install/installer-script-e2e.sh | 5 +++++ tests/install/macos-desktop-e2e.sh | 5 +++++ 3 files changed, 20 insertions(+), 6 deletions(-) diff --git a/tests/install/e2e-assets/ts-prefix.sh b/tests/install/e2e-assets/ts-prefix.sh index b340517dc14f..5fb4a5055e45 100644 --- a/tests/install/e2e-assets/ts-prefix.sh +++ b/tests/install/e2e-assets/ts-prefix.sh @@ -1,10 +1,14 @@ #!/usr/bin/env bash -# ts_prefix: prefix each stdin line with [+MM:SS] relative to the moment -# the pipe started. The playback.html leg player uses these prefixes to -# sync a log file to the screen recording's timeline (the video starts a -# few seconds before the driver, hence the player's offset slider). +# ts_prefix: prefix each stdin line with [+MM:SS] relative to the DRIVER's +# start (TS_BASE), not this pipe's start -- every log in a leg shares one +# time base, so the playback.html sync (one offset slider against the +# recording) aligns all files at once. The playback.html leg player uses +# these prefixes to sync a log file to the screen recording's timeline +# (the video starts a few seconds before the driver, hence the player's +# offset slider). # # Usage (sourced from a driver): +# TS_BASE=$SECONDS # once, at driver start # cmd 2>&1 | ts_prefix > "$LOG_DIR/x.log" # # Must be the LAST consumer in the pipe: with `set -o pipefail` the exit @@ -12,10 +16,10 @@ # SECONDS is bash-specific, so this helper is bash-only. ts_prefix() { - local _start=$SECONDS + local _base="${TS_BASE:-$SECONDS}" local _line _t while IFS= read -r _line; do - _t=$((SECONDS - _start)) + _t=$((SECONDS - _base)) printf '[+%02d:%02d] %s\n' $((_t / 60)) $((_t % 60)) "$_line" done } diff --git a/tests/install/installer-script-e2e.sh b/tests/install/installer-script-e2e.sh index d7f3e77b37f4..54785ba70834 100755 --- a/tests/install/installer-script-e2e.sh +++ b/tests/install/installer-script-e2e.sh @@ -45,6 +45,11 @@ set -euo pipefail +# One time base for every transcript in this leg: ts_prefix stamps lines +# relative to TS_BASE, so all logs share the driver's clock and a single +# playback.html offset slider aligns every file with the recording. +TS_BASE=$SECONDS + INSTALL_METHOD="installer-script" UPDATE_METHOD="" INSTALL_REF="" diff --git a/tests/install/macos-desktop-e2e.sh b/tests/install/macos-desktop-e2e.sh index cc644e9dafb5..9bf8e494c7f3 100755 --- a/tests/install/macos-desktop-e2e.sh +++ b/tests/install/macos-desktop-e2e.sh @@ -32,6 +32,11 @@ set -euo pipefail +# One time base for every transcript in this leg: ts_prefix stamps lines +# relative to TS_BASE, so all logs share the driver's clock and a single +# playback.html offset slider aligns every file with the recording. +TS_BASE=$SECONDS + PHASE="all" UPDATE_METHOD="" INSTALL_REF="" From 2bd1355c4b97a06f9ccfaa35bf1f756e843bdccf Mon Sep 17 00:00:00 2001 From: ethernet Date: Fri, 14 Aug 2026 16:03:20 -0400 Subject: [PATCH 092/227] style(install-e2e): export TS_BASE so shellcheck sees the use --- tests/install/installer-script-e2e.sh | 2 +- tests/install/macos-desktop-e2e.sh | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) diff --git a/tests/install/installer-script-e2e.sh b/tests/install/installer-script-e2e.sh index 54785ba70834..b294fd613ada 100755 --- a/tests/install/installer-script-e2e.sh +++ b/tests/install/installer-script-e2e.sh @@ -48,7 +48,7 @@ set -euo pipefail # One time base for every transcript in this leg: ts_prefix stamps lines # relative to TS_BASE, so all logs share the driver's clock and a single # playback.html offset slider aligns every file with the recording. -TS_BASE=$SECONDS +export TS_BASE=$SECONDS INSTALL_METHOD="installer-script" UPDATE_METHOD="" diff --git a/tests/install/macos-desktop-e2e.sh b/tests/install/macos-desktop-e2e.sh index 9bf8e494c7f3..40945e2bc01f 100755 --- a/tests/install/macos-desktop-e2e.sh +++ b/tests/install/macos-desktop-e2e.sh @@ -35,7 +35,7 @@ set -euo pipefail # One time base for every transcript in this leg: ts_prefix stamps lines # relative to TS_BASE, so all logs share the driver's clock and a single # playback.html offset slider aligns every file with the recording. -TS_BASE=$SECONDS +export TS_BASE=$SECONDS PHASE="all" UPDATE_METHOD="" From cd9110295513b9723e03f60db0053008b1c216db Mon Sep 17 00:00:00 2001 From: ethernet Date: Fri, 14 Aug 2026 16:14:24 -0400 Subject: [PATCH 093/227] 3 default job --- .github/workflows/install-e2e.yml | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/.github/workflows/install-e2e.yml b/.github/workflows/install-e2e.yml index 93d3b3ae5d59..6320a848dcb6 100644 --- a/.github/workflows/install-e2e.yml +++ b/.github/workflows/install-e2e.yml @@ -56,7 +56,7 @@ on: description: 'How many release tags to sample (newest, oldest, and a spread between).' required: false type: string - default: '5' + default: '3' schedule: # Every 12 hours, off the hour to avoid the top-of-hour runner crunch. - cron: '20 7,19 * * *' From 08f6a6de23a3bb5184d1175f829a96b0a1c2272c Mon Sep 17 00:00:00 2001 From: yoniebans Date: Mon, 17 Aug 2026 10:14:39 +0200 Subject: [PATCH 094/227] fix(install-e2e): REPO_DIR -> REPO_ROOT in macos dmg driver self-check The get-url shim self-check referenced REPO_DIR, which is never defined (the script defines REPO_ROOT). Under set -u the stage phase dies immediately, so every macos desktop-installer@latest leg fails before install. Introduced in 368164d7 with the shim itself. Verified: bash -n clean; shim + self-check block replayed standalone with the fix and passes. --- tests/install/macos-desktop-e2e.sh | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/tests/install/macos-desktop-e2e.sh b/tests/install/macos-desktop-e2e.sh index 40945e2bc01f..ccf36a4f9391 100755 --- a/tests/install/macos-desktop-e2e.sh +++ b/tests/install/macos-desktop-e2e.sh @@ -145,7 +145,7 @@ EOF export PATH="$SHIM_DIR:$PATH" # check it worked - observed_git_url="$(git -C "$REPO_DIR" remote get-url origin)" + observed_git_url="$(git -C "$REPO_ROOT" remote get-url origin)" if [[ "$observed_git_url" != "$REPO_URL_HTTPS" ]]; then fail "failed git remote get-url shim: origin resolves to '$observed_git_url', expected '$REPO_URL_HTTPS'" fi From f0e3bcdac92e296cc90d29f91e9ba42879dd97c9 Mon Sep 17 00:00:00 2001 From: yoniebans Date: Tue, 25 Aug 2026 09:47:23 +0300 Subject: [PATCH 095/227] fix(install-e2e): reach Settings through the late-mounting onboarding overlay The overlay mounts in phases (boot-progress card, then the provider picker) and the settings gear reports visible while sitting under it, so a one-shot dismiss probe and a visibility gate both race it. Alternate short-timeout dismiss clicks with settings clicks until a settings click lands; a landed click is proof the overlay is gone. Also pick the window that renders the app UI instead of trusting firstWindow(), which can return a helper webContents. --- tests/install/e2e-assets/drive-update.cjs | 107 ++++++++++++------ tests/install/e2e-assets/launch-from-spec.mjs | 82 ++++++++++---- 2 files changed, 135 insertions(+), 54 deletions(-) diff --git a/tests/install/e2e-assets/drive-update.cjs b/tests/install/e2e-assets/drive-update.cjs index 25d436072d27..a58919b51d32 100644 --- a/tests/install/e2e-assets/drive-update.cjs +++ b/tests/install/e2e-assets/drive-update.cjs @@ -93,7 +93,28 @@ async function main() { timeout: 120_000 }) - const page = await app.firstWindow({ timeout: 120_000 }) + // firstWindow() can grab a helper webContents (wake indicator etc.), not + // the main app window. Pick the window that actually renders UI (has a + // + ) + + const detail = items.find(item => item.id === selected) + + return
+ } + label={t.statusStack.subagents(live.length)} + preview={<>{live.slice(0, 3).map(row)}{live.length > 3 &&

{t.agents.moreAgents(live.length - 3)}

}}> +
{live.map(row)}
+ + {detail &&
+ +
} +
+} diff --git a/apps/desktop/src/components/chat/status-section.tsx b/apps/desktop/src/components/chat/status-section.tsx index 4fbf3152202d..1a9847a83cad 100644 --- a/apps/desktop/src/components/chat/status-section.tsx +++ b/apps/desktop/src/components/chat/status-section.tsx @@ -10,6 +10,8 @@ interface StatusSectionProps { /** Optional inline status next to the label (running spinner, etc). */ collapsedIndicator?: ReactNode defaultCollapsed?: boolean + /** Compact live content stays visible while the full roster is collapsed. */ + preview?: ReactNode /** Optional glyph between the caret and the label (e.g. a `Codicon`). */ icon?: ReactNode label: ReactNode @@ -27,7 +29,8 @@ export function StatusSection({ collapsedIndicator, defaultCollapsed = true, icon, - label + label, + preview }: StatusSectionProps) { const [collapsed, setCollapsed] = useState(defaultCollapsed) @@ -35,6 +38,7 @@ export function StatusSection({
{accessory &&
{accessory}
}
- {!collapsed &&
{children}
} + {(!collapsed || preview) &&
{collapsed ? preview : children}
}
) } diff --git a/apps/desktop/src/i18n/ar.ts b/apps/desktop/src/i18n/ar.ts index 06e0fe375b51..32ec296f75b5 100644 --- a/apps/desktop/src/i18n/ar.ts +++ b/apps/desktop/src/i18n/ar.ts @@ -1080,6 +1080,14 @@ export const ar = defineLocale({ streaming: 'جار البث', files: 'الملفات', moreFiles: count => `+${count} ملفات إضافية`, + moreAgents: count => `${count} وكلاء إضافيون`, + queued: 'في قائمة الانتظار', + waitingActivity: 'بانتظار النشاط', + steer: 'توجيه', + steerPlaceholder: 'تعليمات لهذا الوكيل الفرعي', + steerQueued: 'في انتظار نقطة التحقق التالية', + stopRequested: 'تم طلب الإيقاف', + requestRejected: 'لم يقبل الوكيل الفرعي الطلب', delegation: index => `التفويض ${index}`, workers: count => `${count} عامل`, workersActive: count => `${count} نشط`, diff --git a/apps/desktop/src/i18n/en.ts b/apps/desktop/src/i18n/en.ts index f1f239a4c3d1..11b6d8e6ee3f 100644 --- a/apps/desktop/src/i18n/en.ts +++ b/apps/desktop/src/i18n/en.ts @@ -1593,6 +1593,14 @@ export const en: Translations = { streaming: 'Streaming', files: 'Files', moreFiles: count => `+${count} more files`, + moreAgents: count => `+${count} more agents`, + queued: 'Queued', + waitingActivity: 'Waiting for activity', + steer: 'Steer', + steerPlaceholder: 'Instructions for this subagent', + steerQueued: 'Queued for the next checkpoint', + stopRequested: 'Stop requested', + requestRejected: 'The subagent did not accept the request', delegation: index => `Delegation ${index}`, workers: count => `${count} workers`, workersActive: count => `${count} active`, diff --git a/apps/desktop/src/i18n/ja.ts b/apps/desktop/src/i18n/ja.ts index 0f2b12848046..4da03e279f96 100644 --- a/apps/desktop/src/i18n/ja.ts +++ b/apps/desktop/src/i18n/ja.ts @@ -1406,6 +1406,14 @@ export const ja = defineLocale({ streaming: 'ストリーミング中', files: 'ファイル', moreFiles: count => `+${count} 件のファイル`, + moreAgents: count => `ほか ${count} 件のエージェント`, + queued: '待機中', + waitingActivity: 'アクティビティ待ち', + steer: '指示', + steerPlaceholder: 'このサブエージェントへの指示', + steerQueued: '次のチェックポイントで処理します', + stopRequested: '停止を要求しました', + requestRejected: 'サブエージェントが要求を受け付けませんでした', delegation: index => `委任 ${index}`, workers: count => `${count} ワーカー`, workersActive: count => `${count} アクティブ`, diff --git a/apps/desktop/src/i18n/ru.ts b/apps/desktop/src/i18n/ru.ts index bbeac02270de..ae4716db1237 100644 --- a/apps/desktop/src/i18n/ru.ts +++ b/apps/desktop/src/i18n/ru.ts @@ -1671,6 +1671,14 @@ export const ru = defineLocale({ streaming: 'Стримится', files: 'Файлы', moreFiles: count => `+ещё ${count} ${RU_NOUN(count, 'файл', 'файла', 'файлов')}`, + moreAgents: count => `Ещё ${count} агентов`, + queued: 'В очереди', + waitingActivity: 'Ожидание активности', + steer: 'Направить', + steerPlaceholder: 'Инструкции этому субагенту', + steerQueued: 'В очереди до следующей контрольной точки', + stopRequested: 'Запрошена остановка', + requestRejected: 'Субагент не принял запрос', delegation: index => `Делегирование ${index}`, workers: count => `${count} ${RU_NOUN(count, 'воркер', 'воркера', 'воркеров')}`, workersActive: count => `${count} ${RU_NOUN(count, 'активен', 'активно', 'активных')}`, diff --git a/apps/desktop/src/i18n/types.ts b/apps/desktop/src/i18n/types.ts index 9370ef9668d2..f30043fe7a38 100644 --- a/apps/desktop/src/i18n/types.ts +++ b/apps/desktop/src/i18n/types.ts @@ -1404,6 +1404,14 @@ export interface Translations { streaming: string files: string moreFiles: (count: number) => string + moreAgents: (count: number) => string + queued: string + waitingActivity: string + steer: string + steerPlaceholder: string + steerQueued: string + stopRequested: string + requestRejected: string delegation: (index: number) => string workers: (count: number) => string workersActive: (count: number) => string diff --git a/apps/desktop/src/i18n/zh-hant.ts b/apps/desktop/src/i18n/zh-hant.ts index 9581fbfb1426..10d4eb9f28bc 100644 --- a/apps/desktop/src/i18n/zh-hant.ts +++ b/apps/desktop/src/i18n/zh-hant.ts @@ -1353,6 +1353,14 @@ export const zhHant = defineLocale({ streaming: '串流傳輸中', files: '檔案', moreFiles: count => `還有 ${count} 個檔案`, + moreAgents: count => `還有 ${count} 個子代理`, + queued: '排隊中', + waitingActivity: '等待活動', + steer: '引導', + steerPlaceholder: '此子代理的指令', + steerQueued: '已排隊，等待下一個檢查點', + stopRequested: '已請求停止', + requestRejected: '子代理未接受請求', delegation: index => `派發 ${index}`, workers: count => `${count} 個工作單元`, workersActive: count => `${count} 個活躍`, diff --git a/apps/desktop/src/i18n/zh.ts b/apps/desktop/src/i18n/zh.ts index 6cb00e724b3b..2d04fba130e7 100644 --- a/apps/desktop/src/i18n/zh.ts +++ b/apps/desktop/src/i18n/zh.ts @@ -1765,6 +1765,14 @@ export const zh: Translations = { streaming: '流式传输', files: '文件', moreFiles: count => `还有 ${count} 个文件`, + moreAgents: count => `还有 ${count} 个子代理`, + queued: '排队中', + waitingActivity: '等待活动', + steer: '引导', + steerPlaceholder: '此子代理的指令', + steerQueued: '已排队，等待下一个检查点', + stopRequested: '已请求停止', + requestRejected: '子代理未接受请求', delegation: index => `派发 ${index}`, workers: count => `${count} 个工作单元`, workersActive: count => `${count} 个活跃`, From 98aa27097886b5f07e1c4be2aff5d2d381f5d1ae Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Mon, 7 Sep 2026 21:17:21 -0700 Subject: [PATCH 147/227] feat(desktop): steer and stop subagents through their parent owner --- .../status-stack/subagent-controls.test.tsx | 62 +++++++++++++ .../status-stack/subagent-controls.tsx | 89 +++++++++++++++++++ .../status-stack/subagent-section.test.tsx | 33 +++++-- .../status-stack/subagent-section.tsx | 65 ++++++++++---- 4 files changed, 228 insertions(+), 21 deletions(-) create mode 100644 apps/desktop/src/app/chat/composer/status-stack/subagent-controls.test.tsx create mode 100644 apps/desktop/src/app/chat/composer/status-stack/subagent-controls.tsx diff --git a/apps/desktop/src/app/chat/composer/status-stack/subagent-controls.test.tsx b/apps/desktop/src/app/chat/composer/status-stack/subagent-controls.test.tsx new file mode 100644 index 000000000000..8836b4d3d9e3 --- /dev/null +++ b/apps/desktop/src/app/chat/composer/status-stack/subagent-controls.test.tsx @@ -0,0 +1,62 @@ +import { cleanup, fireEvent, render, screen, waitFor } from '@testing-library/react' +import { afterEach, expect, it, vi } from 'vitest' + +import * as gateway from '@/store/gateway' +import { _resetSessionOwnerHintsForTests, setSessionOwnerHint } from '@/store/session' +import { $subagentsBySession, upsertSubagent } from '@/store/subagents' + +import { SubagentSection } from './subagent-section' + +vi.stubGlobal( + 'ResizeObserver', + class { + disconnect() {} + observe() {} + unobserve() {} + } +) +Element.prototype.animate = vi.fn(() => ({ cancel() {} }) as Animation) +afterEach(() => { + cleanup() + $subagentsBySession.set({}) + _resetSessionOwnerHintsForTests() + vi.restoreAllMocks() +}) + +it('sends steer and stop to the child parent owner, never the active gateway or child transcript', async () => { + const request = vi.spyOn(gateway, 'requestGatewayForAgent').mockResolvedValue({ status: 'queued', found: true }) + setSessionOwnerHint('parent', { connectionId: 'remote-owner', profile: 'research' }) + upsertSubagent('parent', { subagent_id: 'worker', child_session_id: 'child-transcript', goal: 'Owned work' }) + render() + fireEvent.click(screen.getByRole('button', { name: /Owned work/ })) + fireEvent.change(screen.getByRole('textbox'), { target: { value: 'Check the negative control' } }) + fireEvent.click(screen.getByRole('button', { name: 'Steer' })) + await waitFor(() => + expect(request).toHaveBeenCalledWith('remote-owner', 'research', 'subagent.steer', { + session_id: 'parent', + subagent_id: 'worker', + text: 'Check the negative control' + }) + ) + expect(screen.getByText('Queued for the next checkpoint')).toBeTruthy() + fireEvent.click(screen.getByRole('button', { name: 'Stop' })) + await waitFor(() => + expect(request).toHaveBeenCalledWith('remote-owner', 'research', 'subagent.interrupt', { + session_id: 'parent', + subagent_id: 'worker' + }) + ) + expect($subagentsBySession.get().parent?.[0]?.status).toBe('running') +}) + +it('keeps rejected steer text and does not retarget when the owner is unknown', async () => { + const request = vi.spyOn(gateway, 'requestGatewayForAgent').mockResolvedValue({ status: 'rejected' }) + upsertSubagent('unknown', { subagent_id: 'worker', goal: 'Unbound work' }) + render() + fireEvent.click(screen.getByRole('button', { name: /Unbound work/ })) + fireEvent.change(screen.getByRole('textbox'), { target: { value: 'Keep this instruction' } }) + fireEvent.click(screen.getByRole('button', { name: 'Steer' })) + await waitFor(() => expect(screen.getByRole('alert')).toBeTruthy()) + expect(request).not.toHaveBeenCalled() + expect((screen.getByRole('textbox') as HTMLInputElement).value).toBe('Keep this instruction') +}) diff --git a/apps/desktop/src/app/chat/composer/status-stack/subagent-controls.tsx b/apps/desktop/src/app/chat/composer/status-stack/subagent-controls.tsx new file mode 100644 index 000000000000..5f20481c0453 --- /dev/null +++ b/apps/desktop/src/app/chat/composer/status-stack/subagent-controls.tsx @@ -0,0 +1,89 @@ +import { useState } from 'react' + +import { Button } from '@/components/ui/button' +import { Input } from '@/components/ui/input' +import { useI18n } from '@/i18n' +import { requestForOwnedSession } from '@/store/session-states' + +interface SubagentControlsProps { + sessionId: string + subagentId: string +} + +export function SubagentControls({ sessionId, subagentId }: SubagentControlsProps) { + const { t } = useI18n() + const [text, setText] = useState('') + const [pending, setPending] = useState(false) + const [feedback, setFeedback] = useState('') + const [failed, setFailed] = useState(false) + + const send = async (action: 'steer' | 'interrupt') => { + setPending(true) + setFeedback('') + setFailed(false) + + try { + // Unlike global chrome, a child control must NEVER fall back to whichever + // gateway happens to be active, even on a legacy unbound session. + const result = await requestForOwnedSession<{ found?: boolean; status?: string }>( + sessionId, + async () => { + throw new Error(t.agents.requestRejected) + }, + `subagent.${action}`, + { session_id: sessionId, subagent_id: subagentId, ...(action === 'steer' ? { text: text.trim() } : {}) } + ) + + if (action === 'steer' ? result.status !== 'queued' : !result.found) { + throw new Error(t.agents.requestRejected) + } + + setFeedback(action === 'steer' ? t.agents.steerQueued : t.agents.stopRequested) + + if (action === 'steer') { + setText('') + } + } catch { + setFailed(true) + setFeedback(t.agents.requestRejected) + } finally { + setPending(false) + } + } + + return ( +
event.stopPropagation()} + onSubmit={event => { + event.preventDefault() + event.stopPropagation() + + if (text.trim() && !pending) { + void send('steer') + } + }} + > +
+ setText(event.target.value)} + placeholder={t.agents.steerPlaceholder} + value={text} + /> + + +
+ {feedback && ( +

+ {feedback} +

+ )} +
+ ) +} diff --git a/apps/desktop/src/app/chat/composer/status-stack/subagent-section.test.tsx b/apps/desktop/src/app/chat/composer/status-stack/subagent-section.test.tsx index 98f120e121da..acdd0aea10a4 100644 --- a/apps/desktop/src/app/chat/composer/status-stack/subagent-section.test.tsx +++ b/apps/desktop/src/app/chat/composer/status-stack/subagent-section.test.tsx @@ -6,9 +6,18 @@ import { $subagentsBySession, upsertSubagent } from '@/store/subagents' import { ComposerStatusStack } from './index' -vi.stubGlobal('ResizeObserver', class { disconnect() {} observe() {} }) +vi.stubGlobal( + 'ResizeObserver', + class { + disconnect() {} + observe() {} + } +) -afterEach(() => { cleanup(); $subagentsBySession.set({}) }) +afterEach(() => { + cleanup() + $subagentsBySession.set({}) +}) it('automatically previews bounded live work only from the composer session, including queued children', () => { for (let i = 0; i < 5; i++) { @@ -17,7 +26,13 @@ it('automatically previews bounded live work only from the composer session, inc upsertSubagent('other-profile', { subagent_id: 'foreign', goal: 'Private foreign task' }) upsertSubagent('owner', { subagent_id: 'child-0', text: 'Reading actual source' }, false, 'subagent.progress') - const view = render() + + const view = render( + + + + ) + expect(screen.getByText('Task 0')).toBeTruthy() expect(screen.getByText('Reading actual source')).toBeTruthy() expect(screen.queryByText('Task 4')).toBeNull() @@ -25,13 +40,21 @@ it('automatically previews bounded live work only from the composer session, inc expect(screen.getByRole('button', { name: /5 Subagents/ })).toBeTruthy() fireEvent.click(screen.getByRole('button', { name: /5 Subagents/ })) expect(screen.getByText('Task 4')).toBeTruthy() - view.rerender() + view.rerender( + + + + ) expect(screen.queryByText('Task 0')).toBeNull() }) it('retires the live frame only after every child settles, without depending on the parent busy state', () => { upsertSubagent('owner', { subagent_id: 'child', goal: 'Live task' }) - render() + render( + + + + ) expect(screen.getByText('Live task')).toBeTruthy() act(() => upsertSubagent('owner', { subagent_id: 'child', status: 'completed' }, false, 'subagent.complete')) expect(screen.queryByText('Live task')).toBeNull() diff --git a/apps/desktop/src/app/chat/composer/status-stack/subagent-section.tsx b/apps/desktop/src/app/chat/composer/status-stack/subagent-section.tsx index fda6f4c4efa1..a1a2a19768c5 100644 --- a/apps/desktop/src/app/chat/composer/status-stack/subagent-section.tsx +++ b/apps/desktop/src/app/chat/composer/status-stack/subagent-section.tsx @@ -9,6 +9,8 @@ import { useI18n } from '@/i18n' import { useSessionSlice } from '@/lib/use-session-slice' import { $subagentsBySession, type SubagentProgress } from '@/store/subagents' +import { SubagentControls } from './subagent-controls' + interface SubagentSectionProps { sessionId: string } @@ -23,36 +25,67 @@ export function SubagentSection({ sessionId }: SubagentSectionProps) { const hasLive = live.length > 0 useEffect(() => { - if (!hasLive) { return } + if (!hasLive) { + return + } + const timer = setInterval(() => setNowMs(Date.now()), 1000) return () => clearInterval(timer) }, [hasLive]) - if (!hasLive) { return null } + if (!hasLive) { + return null + } const row = (item: SubagentProgress) => ( - ) const detail = items.find(item => item.id === selected) - return
- } - label={t.statusStack.subagents(live.length)} - preview={<>{live.slice(0, 3).map(row)}{live.length > 3 &&

{t.agents.moreAgents(live.length - 3)}

}}> -
{live.map(row)}
- - {detail &&
- -
} -
+ return ( +
+ } + label={t.statusStack.subagents(live.length)} + preview={ + <> + {live.slice(0, 3).map(row)} + {live.length > 3 && ( +

{t.agents.moreAgents(live.length - 3)}

+ )} + + } + > +
{live.map(row)}
+ + {detail && ( +
+ {(detail.status === 'running' || detail.status === 'queued') && ( + + )} + +
+ )} +
+ ) } From fd3932c4164dba6a39c37a624ad8432a3c8bc9f9 Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Mon, 7 Sep 2026 21:27:01 -0700 Subject: [PATCH 148/227] fix(desktop): preserve scoped steering drafts and worker timer origins --- apps/desktop/src/app/agents/index.tsx | 2 +- .../status-stack/subagent-controls.tsx | 5 +- .../status-stack/subagent-lifecycle.test.tsx | 49 +++++++++++++++++++ .../status-stack/subagent-section.tsx | 30 +++++++----- 4 files changed, 70 insertions(+), 16 deletions(-) create mode 100644 apps/desktop/src/app/chat/composer/status-stack/subagent-lifecycle.test.tsx diff --git a/apps/desktop/src/app/agents/index.tsx b/apps/desktop/src/app/agents/index.tsx index 3921d5e98c61..e84458877a75 100644 --- a/apps/desktop/src/app/agents/index.tsx +++ b/apps/desktop/src/app/agents/index.tsx @@ -323,7 +323,7 @@ function StreamLine({ export function SubagentRow({ node, depth = 0, nowMs }: { node: SubagentNode; depth?: number; nowMs: number }) { const { t } = useI18n() const running = node.status === 'running' || node.status === 'queued' - const elapsed = useElapsedSeconds(running, `subagent:${node.id}`) + const elapsed = useElapsedSeconds(running, `subagent:${node.id}`, node.startedAt) const durationSeconds = typeof node.durationSeconds === 'number' ? Math.max(0, Math.round(node.durationSeconds)) : elapsed diff --git a/apps/desktop/src/app/chat/composer/status-stack/subagent-controls.tsx b/apps/desktop/src/app/chat/composer/status-stack/subagent-controls.tsx index 5f20481c0453..678c5442d91b 100644 --- a/apps/desktop/src/app/chat/composer/status-stack/subagent-controls.tsx +++ b/apps/desktop/src/app/chat/composer/status-stack/subagent-controls.tsx @@ -8,11 +8,12 @@ import { requestForOwnedSession } from '@/store/session-states' interface SubagentControlsProps { sessionId: string subagentId: string + text: string + setText: (text: string) => void } -export function SubagentControls({ sessionId, subagentId }: SubagentControlsProps) { +export function SubagentControls({ sessionId, subagentId, text, setText }: SubagentControlsProps) { const { t } = useI18n() - const [text, setText] = useState('') const [pending, setPending] = useState(false) const [feedback, setFeedback] = useState('') const [failed, setFailed] = useState(false) diff --git a/apps/desktop/src/app/chat/composer/status-stack/subagent-lifecycle.test.tsx b/apps/desktop/src/app/chat/composer/status-stack/subagent-lifecycle.test.tsx new file mode 100644 index 000000000000..5b5e79623d18 --- /dev/null +++ b/apps/desktop/src/app/chat/composer/status-stack/subagent-lifecycle.test.tsx @@ -0,0 +1,49 @@ +import { act, cleanup, fireEvent, render, screen } from '@testing-library/react' +import { afterEach, expect, it, vi } from 'vitest' + +import { $subagentsBySession, upsertSubagent } from '@/store/subagents' + +import { SubagentSection } from './subagent-section' + +vi.stubGlobal( + 'ResizeObserver', + class { + disconnect() {} + observe() {} + unobserve() {} + } +) +Element.prototype.animate = vi.fn(() => ({ cancel() {} }) as Animation) +afterEach(() => { + cleanup() + $subagentsBySession.set({}) + vi.restoreAllMocks() +}) + +it('keeps each worker draft while inspecting siblings and removes settled selection', () => { + upsertSubagent('parent', { subagent_id: 'a', goal: 'Worker A' }) + upsertSubagent('parent', { subagent_id: 'b', goal: 'Worker B' }) + render() + fireEvent.click(screen.getByRole('button', { name: /Worker A/ })) + fireEvent.change(screen.getByRole('textbox'), { target: { value: 'Preserve my instruction' } }) + fireEvent.click(screen.getByRole('button', { name: /Worker B/ })) + expect((screen.getByRole('textbox') as HTMLInputElement).value).toBe('') + fireEvent.click(screen.getByRole('button', { name: /Worker A/ })) + expect((screen.getByRole('textbox') as HTMLInputElement).value).toBe('Preserve my instruction') + act(() => upsertSubagent('parent', { subagent_id: 'a', status: 'completed' }, false, 'subagent.complete')) + expect(screen.queryByText('Worker A')).toBeNull() + expect(screen.queryByRole('textbox')).toBeNull() +}) + +it('measures detail elapsed from worker start rather than first inspection', () => { + const now = vi.spyOn(Date, 'now').mockReturnValue(100000) + upsertSubagent('parent', { subagent_id: 'timed', goal: 'Timed worker' }) + now.mockReturnValue(117000) + const { container } = render() + fireEvent.click(screen.getByRole('button', { name: /Timed worker/ })) + expect( + container.querySelector( + '[data-slot="composer-subagent-detail"] [data-slot="tool-block"] > button > span:last-child' + )?.textContent + ).toBe('17s') +}) diff --git a/apps/desktop/src/app/chat/composer/status-stack/subagent-section.tsx b/apps/desktop/src/app/chat/composer/status-stack/subagent-section.tsx index a1a2a19768c5..bac54f320dae 100644 --- a/apps/desktop/src/app/chat/composer/status-stack/subagent-section.tsx +++ b/apps/desktop/src/app/chat/composer/status-stack/subagent-section.tsx @@ -1,10 +1,11 @@ -import { useEffect, useState } from 'react' +import { useState } from 'react' import { SubagentRow } from '@/app/agents' import { ActivityTimerText } from '@/components/chat/activity-timer-text' import { StatusSection } from '@/components/chat/status-section' import { Codicon } from '@/components/ui/codicon' import { GlyphSpinner } from '@/components/ui/glyph-spinner' +import { useViewedInterval } from '@/hooks/use-viewed-interval' import { useI18n } from '@/i18n' import { useSessionSlice } from '@/lib/use-session-slice' import { $subagentsBySession, type SubagentProgress } from '@/store/subagents' @@ -22,17 +23,10 @@ export function SubagentSection({ sessionId }: SubagentSectionProps) { const live = items.filter(item => item.status === 'running' || item.status === 'queued') const [nowMs, setNowMs] = useState(Date.now) const [selected, setSelected] = useState(null) + const [drafts, setDrafts] = useState>({}) const hasLive = live.length > 0 - useEffect(() => { - if (!hasLive) { - return - } - - const timer = setInterval(() => setNowMs(Date.now()), 1000) - - return () => clearInterval(timer) - }, [hasLive]) + useViewedInterval(() => setNowMs(Date.now()), 1000, hasLive) if (!hasLive) { return null @@ -46,7 +40,11 @@ export function SubagentSection({ sessionId }: SubagentSectionProps) { onClick={() => setSelected(selected === item.id ? null : item.id)} type="button" > - + {item.goal} @@ -60,7 +58,7 @@ export function SubagentSection({ sessionId }: SubagentSectionProps) { ) - const detail = items.find(item => item.id === selected) + const detail = live.find(item => item.id === selected) return (
@@ -81,7 +79,13 @@ export function SubagentSection({ sessionId }: SubagentSectionProps) { {detail && (
{(detail.status === 'running' || detail.status === 'queued') && ( - + setDrafts(previous => ({ ...previous, [detail.id]: text }))} + subagentId={detail.id} + text={drafts[detail.id] ?? ''} + /> )}
From 23fa4ae748237ca943324a8dd78bf931384e0710 Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Mon, 7 Sep 2026 21:44:14 -0700 Subject: [PATCH 149/227] docs: explain live subagent monitoring across chat surfaces --- website/docs/user-guide/cli.md | 1 + website/docs/user-guide/desktop.md | 4 ++++ website/docs/user-guide/features/delegation.md | 16 +++++++++++++++- website/docs/user-guide/tui.md | 1 + .../current/user-guide/cli.md | 1 + .../current/user-guide/features/delegation.md | 10 +++++++++- .../current/user-guide/tui.md | 1 + 7 files changed, 32 insertions(+), 2 deletions(-) diff --git a/website/docs/user-guide/cli.md b/website/docs/user-guide/cli.md index cd7355a12e36..df128820e058 100644 --- a/website/docs/user-guide/cli.md +++ b/website/docs/user-guide/cli.md @@ -186,6 +186,7 @@ When resuming a previous session (`hermes -c` or `hermes --resume `), a "Pre | `Ctrl+X Ctrl+E` | Emacs-style alternate binding for the external editor (same behavior as `Ctrl+G`). | | `Ctrl+S` | **Stash the prompt.** Parks the current draft and clears the composer so you can send something else first. Press `Ctrl+S` again on an empty composer to bring the draft back (cursor at the end, attached images restored). Repeated presses build a stack rather than overwriting, so an earlier draft is never silently lost — with two or more stashed, `Ctrl+S` opens a browse panel (`↑`/`↓` to navigate, `Enter` to restore, `D` to discard, `Esc` or `Ctrl+S` to close). A `📌 N` badge in the status bar shows how many drafts are parked. Multi-line drafts round-trip exactly, including blank lines. The stash lives in memory for the session only — nothing is written to disk, since drafts often contain secrets. | | `Ctrl+C` | Interrupt agent (double-press within 2s to force exit) | +| `F6` | Open the full-screen live subagent monitor without losing the composer draft. The live dock appears automatically above the status bar; arrows select a worker, `Enter` shows its recent log, `s` steers, and `x` requests stop with confirmation. See [Monitoring subagents](/user-guide/features/delegation#monitoring-running-subagents-agents). | | `Ctrl+D` | Exit | | `Ctrl+Z` | Suspend Hermes to background (Unix only). Run `fg` in the shell to resume. | | `Tab` | Accept auto-suggestion (ghost text) or autocomplete slash commands | diff --git a/website/docs/user-guide/desktop.md b/website/docs/user-guide/desktop.md index 6200403c434d..364a8396be4a 100644 --- a/website/docs/user-guide/desktop.md +++ b/website/docs/user-guide/desktop.md @@ -121,6 +121,10 @@ A real terminal lives in the right sidebar, next to the file browser: - **Shells persist while hidden.** Closing or hiding the panel doesn't kill your shell — every open terminal stays mounted with its scrollback and running processes intact until you explicitly close it. - **Add to chat** — select terminal output and send it into the composer as context for your next message. +### Live subagents + +While delegated workers are live, a **Subagents** frame appears above the composer with their count, task names, elapsed time, and latest activity. It previews up to three workers; expand the header for the roster, then select a worker for details and **Steer** / **Stop** controls. Each frame belongs to its chat, including in split panes. Steering acknowledges that guidance is queued for a checkpoint, not that the child has already read it. See [Monitoring subagents](/user-guide/features/delegation#monitoring-running-subagents-agents). + ### Git review & worktrees For sessions running inside a Git repository, the app has a built-in source-control surface: diff --git a/website/docs/user-guide/features/delegation.md b/website/docs/user-guide/features/delegation.md index dbe5e21c0897..ef97852ffdcd 100644 --- a/website/docs/user-guide/features/delegation.md +++ b/website/docs/user-guide/features/delegation.md @@ -385,7 +385,21 @@ The TUI ships a `/agents` overlay (alias `/tasks`) that turns recursive `delegat - Kill and pause controls — cancel a specific subagent mid-flight without interrupting its siblings - Post-hoc review: step through each subagent's turn-by-turn history even after they've returned to the parent -The classic CLI just prints `/agents` as a text summary; the TUI is where the overlay shines. See [TUI — Slash commands](/user-guide/tui#slash-commands). +### Live activity above the composer + +The classic CLI, TUI, and Desktop automatically show live subagents above the composer. You can keep writing while watching the live count, task names, elapsed time, and latest activity. The terminal dock limits visible rows according to screen height and shows how many additional workers are hidden; Desktop previews up to three workers. + +| Surface | Expand and inspect | Control a selected worker | +|---|---|---| +| Classic CLI | **F6** opens the full-screen live roster; arrows select, **Enter** opens the transcript tail, **PgUp/PgDn** scroll | **s** opens a separate steering input; **x**, then **y** requests stop | +| TUI | **Ctrl+T** or `/agents` opens the full-height tree; **Enter** opens detail; **t** opens the live transcript tail | **e** opens steering; **x** stops the selected worker; **X** stops its subtree | +| Desktop | Expand **Subagents** above the composer, then select a worker to inspect its activity and details | **Steer** queues guidance; **Stop** requests interruption for that worker | + +Closing the terminal monitor returns to your existing composer draft. Steering uses its own input and acknowledges **queued**, not delivery: the child consumes guidance at a checkpoint. Stop does not interrupt unrelated siblings. + +The live transcript tail is a bounded recent excerpt, not an unlimited conversation browser. A child leaving the live registry leaves the dock; completion messages and the TUI/Desktop history views remain the place to review finished work. Latest activity is an observation, not a percentage-complete estimate. + +The classic CLI's `/agents` and `/tasks` commands still print a text summary; **F6** is the immediate interactive monitor, including while the parent is busy. See [TUI — Slash commands](/user-guide/tui#slash-commands). On the classic CLI and every gateway platform (Telegram, Discord, Slack, ...), `/agents` also lists **background delegations with live per-child activity**, diff --git a/website/docs/user-guide/tui.md b/website/docs/user-guide/tui.md index 04724657f4bb..e601b3a085ab 100644 --- a/website/docs/user-guide/tui.md +++ b/website/docs/user-guide/tui.md @@ -102,6 +102,7 @@ The directory must contain `dist/entry.js`. Keybindings match the [Classic CLI](cli.md#keybindings) exactly. The only behavioral differences: +- **`Ctrl+T`** expands the automatic live-subagent dock into the full-height `/agents` roster. Select a worker to inspect details, press **`t`** for its recent transcript, **`e`** to steer, or **`x`** to stop it. The dock fits its row count to terminal height and preserves your composer draft. See [Monitoring subagents](/user-guide/features/delegation#monitoring-running-subagents-agents). - **Mouse drag** highlights text with a uniform selection background. - **`Cmd+V` / `Ctrl+V`** first tries normal text paste, then falls back to OSC52/native clipboard reads, and finally image attach when the clipboard or pasted payload resolves to an image. - **`/terminal-setup`** installs local VS Code / Cursor / Windsurf terminal bindings for better `Cmd+Enter` and undo/redo parity on macOS. diff --git a/website/i18n/zh-Hans/docusaurus-plugin-content-docs/current/user-guide/cli.md b/website/i18n/zh-Hans/docusaurus-plugin-content-docs/current/user-guide/cli.md index 90393421ccfc..6ff86a5b2421 100644 --- a/website/i18n/zh-Hans/docusaurus-plugin-content-docs/current/user-guide/cli.md +++ b/website/i18n/zh-Hans/docusaurus-plugin-content-docs/current/user-guide/cli.md @@ -102,6 +102,7 @@ hermes -w -z "Fix issue #123" # 在 worktree 中以单次查询模式运行 | `Ctrl+G` | 在 `$EDITOR`（vim/nvim/nano/VS Code 等）中打开当前输入缓冲区。保存并退出后，编辑后的文本将作为下一条 prompt 发送——适合编写长篇多段落 prompt。 | | `Ctrl+X Ctrl+E` | 外部编辑器的 Emacs 风格备用绑定（与 `Ctrl+G` 行为相同）。 | | `Ctrl+C` | 中断 agent（2 秒内双击强制退出） | +| `F6` | 打开全屏实时子智能体监视器，保留输入草稿。方向键选择，`Enter` 查看近期日志，`s` 引导，`x` 请求停止并确认。 | | `Ctrl+D` | 退出 | | `Ctrl+Z` | 将 Hermes 挂起到后台（仅 Unix）。在 shell 中运行 `fg` 恢复。 | | `Tab` | 接受自动建议（ghost text）或自动补全斜杠命令 | diff --git a/website/i18n/zh-Hans/docusaurus-plugin-content-docs/current/user-guide/features/delegation.md b/website/i18n/zh-Hans/docusaurus-plugin-content-docs/current/user-guide/features/delegation.md index b198c0a04adb..0971ef374a23 100644 --- a/website/i18n/zh-Hans/docusaurus-plugin-content-docs/current/user-guide/features/delegation.md +++ b/website/i18n/zh-Hans/docusaurus-plugin-content-docs/current/user-guide/features/delegation.md @@ -254,7 +254,15 @@ TUI 提供 `/agents` 浮层（别名 `/tasks`），将递归 `delegate_task` 扇 - 终止和暂停控制——可在不中断其兄弟智能体的情况下取消特定子智能体 - 事后回顾：即使子智能体已返回父智能体，也可逐轮查看其历史记录 -经典 CLI 仅将 `/agents` 打印为文本摘要；TUI 才是浮层真正发挥作用的地方。参见 [TUI — 斜杠命令](/user-guide/tui#slash-commands)。 +经典 CLI、TUI 和 Desktop 会在输入框上方自动显示正在运行的子智能体，包括总数、任务名称、已运行时间和最近活动。终端根据屏幕高度限制可见行数，并显示隐藏数量；Desktop 最多预览三个工作者。 + +- **经典 CLI：F6** 打开全屏实时列表；方向键选择，**Enter** 查看近期日志，**PgUp/PgDn** 滚动，**s** 输入引导，**x** 后按 **y** 确认停止。关闭后保留原有输入草稿。 +- **TUI：Ctrl+T** 或 `/agents` 打开完整树状列表；**t** 查看实时日志，**e** 输入引导，**x** 停止选中的工作者，**X** 停止其子树。 +- **Desktop：** 展开输入框上方的 **Subagents**，选择工作者查看详情并使用 **Steer** / **Stop**。 + +引导的“已排队”确认不代表子智能体已经读取；它会在检查点接收。日志预览只包含有大小限制的近期内容。工作者结束后离开实时列表，完成消息和已有历史视图仍可用于回顾。 + +经典 CLI 的 `/agents` 和 `/tasks` 仍打印文本摘要；父智能体忙碌时可直接按 **F6** 打开交互式监视器。参见 [TUI — 斜杠命令](/user-guide/tui#slash-commands)。 ## 深度限制与嵌套编排 {#depth-limit-and-nested-orchestration} diff --git a/website/i18n/zh-Hans/docusaurus-plugin-content-docs/current/user-guide/tui.md b/website/i18n/zh-Hans/docusaurus-plugin-content-docs/current/user-guide/tui.md index c7ac811ef18f..c79b3d95d084 100644 --- a/website/i18n/zh-Hans/docusaurus-plugin-content-docs/current/user-guide/tui.md +++ b/website/i18n/zh-Hans/docusaurus-plugin-content-docs/current/user-guide/tui.md @@ -85,6 +85,7 @@ hermes --tui 快捷键与 [Classic CLI](cli.md#keybindings) 完全一致。仅有以下行为差异： +- **`Ctrl+T`** — 将输入框上方的实时子智能体栏展开为完整 `/agents` 列表；**`t`** 查看近期日志，**`e`** 引导，**`x`** 停止选中的工作者。可见行数随终端高度调整，关闭后保留输入草稿。 - **鼠标拖拽** — 以统一选区背景色高亮文本。 - **`Cmd+V` / `Ctrl+V`** — 优先尝试普通文本粘贴，然后回退到 OSC52/原生剪贴板读取，最后在剪贴板或粘贴内容解析为图片时进行图片附件操作。 - **`/terminal-setup`** — 安装本地 VS Code / Cursor / Windsurf 终端绑定，以在 macOS 上获得更好的 `Cmd+Enter` 和撤销/重做一致性。 From 4cd4f395ea75f710306206aaa00055ecb308961e Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Mon, 7 Sep 2026 21:51:54 -0700 Subject: [PATCH 150/227] test: isolate missing first-party import guard fixture --- tests/hermes_cli/test_update_import_guard.py | 2 ++ 1 file changed, 2 insertions(+) diff --git a/tests/hermes_cli/test_update_import_guard.py b/tests/hermes_cli/test_update_import_guard.py index 5318892be6b3..0eeb63671169 100644 --- a/tests/hermes_cli/test_update_import_guard.py +++ b/tests/hermes_cli/test_update_import_guard.py @@ -329,6 +329,8 @@ def test_import_guard_ignores_missing_third_party_dependency(monkeypatch, tmp_pa def test_import_guard_flags_missing_first_party_module(monkeypatch, tmp_path): """A missing *first-party* module IS skew — the update dropped a file.""" + (tmp_path / "tools").mkdir() + (tmp_path / "tools" / "__init__.py").write_text("") (tmp_path / "consumer.py").write_text("import tools.nonexistent_module\n") monkeypatch.setattr(update_cmd, "_UPDATE_CRITICAL_MODULES", ("consumer",)) monkeypatch.setattr(update_cmd_deps, "_UPDATE_CRITICAL_MODULES", ("consumer",)) From d99a63b645eb4dfa3f910dc3d52b31a0e4392dc0 Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Mon, 7 Sep 2026 21:52:35 -0700 Subject: [PATCH 151/227] fix: yield subagent monitor to incoming CLI prompts --- hermes_cli/callbacks.py | 4 +- hermes_cli/cli_modal_mixin.py | 11 ++- hermes_cli/cli_subagent_monitor.py | 18 ++++- hermes_cli/cli_terminal_mixin.py | 3 + hermes_cli/cli_tui_mixin.py | 12 ++- tests/cli/test_subagent_monitor_prompts.py | 89 ++++++++++++++++++++++ 6 files changed, 124 insertions(+), 13 deletions(-) create mode 100644 tests/cli/test_subagent_monitor_prompts.py diff --git a/hermes_cli/callbacks.py b/hermes_cli/callbacks.py index 5c6bf21f7564..c3a9c9742263 100644 --- a/hermes_cli/callbacks.py +++ b/hermes_cli/callbacks.py @@ -10,7 +10,9 @@ def _invalidate(cli) -> None: - if getattr(cli, "_app", None): + if hasattr(cli, "_paint_now"): + cli._paint_now() + elif getattr(cli, "_app", None): cli._app.invalidate() diff --git a/hermes_cli/cli_modal_mixin.py b/hermes_cli/cli_modal_mixin.py index b74c0bca1257..3d6ea5a14800 100644 --- a/hermes_cli/cli_modal_mixin.py +++ b/hermes_cli/cli_modal_mixin.py @@ -853,11 +853,16 @@ def _handle_approval_selection(self) -> None: self._invalidate() def _secret_capture_callback(self, var_name: str, prompt: str, metadata=None) -> dict: - return prompt_for_secret(self, var_name, prompt, metadata) + self._capture_modal_input_snapshot() + try: + return prompt_for_secret(self, var_name, prompt, metadata) + finally: + self._restore_modal_input_snapshot() + self._paint_now() def _capture_modal_input_snapshot(self) -> None: """Temporarily clear the input buffer and save the user's in-progress draft.""" - if self._modal_input_snapshot is not None or not getattr(self, "_app", None): + if getattr(self, "_modal_input_snapshot", None) is not None or not getattr(self, "_app", None): return try: buf = self._app.current_buffer @@ -868,7 +873,7 @@ def _capture_modal_input_snapshot(self) -> None: def _restore_modal_input_snapshot(self) -> None: """Restore any draft text that was present before a modal prompt opened.""" - snapshot = self._modal_input_snapshot + snapshot = getattr(self, "_modal_input_snapshot", None) self._modal_input_snapshot = None if not snapshot or not getattr(self, "_app", None): return diff --git a/hermes_cli/cli_subagent_monitor.py b/hermes_cli/cli_subagent_monitor.py index 6e666e33c745..94ac77ca8f73 100644 --- a/hermes_cli/cli_subagent_monitor.py +++ b/hermes_cli/cli_subagent_monitor.py @@ -105,6 +105,12 @@ def read_tail(path): return 'Live transcript not available yet.' +def modal_prompt_active(cli): + return any(getattr(cli, name, None) for name in ( + '_clarify_state', '_approval_state', '_slash_confirm_state', '_sudo_state', + '_secret_state', '_model_picker_state', '_command_palette_state')) + + def build_monitor_application(monitor, **kwargs): from prompt_toolkit.application import Application from prompt_toolkit.data_structures import Point @@ -235,8 +241,16 @@ def close(event): Window(FormattedTextControl(lambda: _clip(state['notice'], app.output.get_size().columns)), height=1), Window(FormattedTextControl(footer), height=1), ]), focused_element=roster) + def before_render(app): + # Prompts arrive on worker threads; exit on the UI loop, including the + # first frame if a prompt won the race with in_terminal() acquisition. + if modal_prompt_active(monitor.cli) and not app.is_done: + app.exit() + elif state['detail']: + update_tail() + app = Application(layout=layout, key_bindings=kb, full_screen=True, mouse_support=False, - before_render=lambda app: update_tail() if state['detail'] else None, **kwargs) + before_render=before_render, **kwargs) return app @@ -277,5 +291,5 @@ def text(): cli._subagent_dock_widget = ConditionalContainer( Window(FormattedTextControl(text), wrap_lines=False), - filter=Condition(lambda: bool(monitor.entries)), + filter=Condition(lambda: bool(monitor.entries) and not modal_prompt_active(cli)), ) diff --git a/hermes_cli/cli_terminal_mixin.py b/hermes_cli/cli_terminal_mixin.py index 133d5adc0b0b..5002313b7028 100644 --- a/hermes_cli/cli_terminal_mixin.py +++ b/hermes_cli/cli_terminal_mixin.py @@ -103,6 +103,9 @@ def _paint_now(self) -> None: """ if getattr(self, "_terminal_io_broken", False): return + monitor = getattr(self, "_subagent_monitor", None) + if monitor is not None and monitor.app is not None: + monitor.app.invalidate() app = getattr(self, "_app", None) if app is not None: self._app_invalidate(app, "paint_now", swallow=True) diff --git a/hermes_cli/cli_tui_mixin.py b/hermes_cli/cli_tui_mixin.py index 2f5913646bf0..aca7359fda0a 100644 --- a/hermes_cli/cli_tui_mixin.py +++ b/hermes_cli/cli_tui_mixin.py @@ -1456,8 +1456,9 @@ def _tui_enter_overlay(self, event) -> bool: event.app.invalidate() return True if self._secret_state: - self._submit_secret_response(buf.text) + value = buf.text buf.reset() + self._submit_secret_response(value) event.app.invalidate() return True if self._approval_state: @@ -1857,12 +1858,9 @@ def _tui_build_key_bindings(self): kb.add(Keys.BracketedPaste, eager=True)(self._tui_handle_paste) kb.add('c-v')(self._tui_handle_ctrl_v) kb.add('escape', 'v')(self._tui_handle_alt_v) - from hermes_cli.cli_subagent_monitor import open_monitor - kb.add('f6', filter=Condition(lambda: not any( - getattr(self, name, None) for name in ( - '_clarify_state', '_approval_state', '_slash_confirm_state', '_sudo_state', - '_secret_state', '_model_picker_state', '_command_palette_state'))))( - lambda event: open_monitor(self)) + from hermes_cli.cli_subagent_monitor import modal_prompt_active, open_monitor + kb.add('f6', filter=Condition(lambda: not modal_prompt_active(self)))( + lambda event: open_monitor(self)) return kb def _tui_bind_editor_and_stash(self, kb) -> None: diff --git a/tests/cli/test_subagent_monitor_prompts.py b/tests/cli/test_subagent_monitor_prompts.py new file mode 100644 index 000000000000..15f6ba0513ff --- /dev/null +++ b/tests/cli/test_subagent_monitor_prompts.py @@ -0,0 +1,89 @@ +"""Blocking prompts reclaim input from the nested monitor without a keypress.""" +import asyncio +from types import SimpleNamespace + + +def test_prompt_paint_yields_monitor_but_ordinary_paint_does_not(): + from hermes_cli.cli_subagent_monitor import SubagentMonitor, build_monitor_application, install_dock + from hermes_cli.cli_terminal_mixin import CLITerminalMixin + from prompt_toolkit.input import create_pipe_input + from prompt_toolkit.output import DummyOutput + + async def run(): + for name in ('_approval_state', '_clarify_state', '_secret_state', '_sudo_state', + '_slash_confirm_state'): + dock_cli = SimpleNamespace(agent=None) + install_dock(dock_cli) + dock_cli._subagent_monitor.entries = [{}] + assert dock_cli._subagent_dock_widget.filter() + setattr(dock_cli, name, {'pending': True}) + assert not dock_cli._subagent_dock_widget.filter(), 'dock crowds the blocking prompt' + cli = SimpleNamespace(_app=None) + monitor = SubagentMonitor(cli) + cli._subagent_monitor = monitor + with create_pipe_input() as pipe: + app = build_monitor_application(monitor, input=pipe, output=DummyOutput()) + monitor.app = app + rendered = asyncio.Event() + app.after_render += lambda app: rendered.set() + task = asyncio.create_task(app.run_async()) + await asyncio.wait_for(rendered.wait(), 3) + try: + await asyncio.to_thread(CLITerminalMixin._paint_now, cli) + assert not task.done(), 'ordinary paints must not dismiss the monitor' + setattr(cli, name, {'pending': True}) + await asyncio.to_thread(CLITerminalMixin._paint_now, cli) + done, _ = await asyncio.wait({task}, timeout=1) + assert task in done, f'{name} remained hidden behind the monitor' + await task + finally: + if not task.done(): + app.exit() + await task + asyncio.run(run()) + + +def test_secret_callback_yields_monitor_and_restores_composer(): + from hermes_cli.cli_subagent_monitor import SubagentMonitor, build_monitor_application + from hermes_cli.cli_terminal_mixin import CLITerminalMixin + from hermes_cli.cli_modal_mixin import CLIModalMixin + from prompt_toolkit.buffer import Buffer + from prompt_toolkit.document import Document + from prompt_toolkit.input import create_pipe_input + from prompt_toolkit.output import DummyOutput + + class CLI(CLIModalMixin, CLITerminalMixin): + def _ring_bell(self, **kwargs): + pass + + async def run(): + cli = CLI() + draft = Document('keep my draft', 4) + buffer = Buffer(document=draft) + cli._app = SimpleNamespace(current_buffer=buffer, invalidate=lambda: None) + cli._modal_input_snapshot = None + monitor = cli._subagent_monitor = SubagentMonitor(cli) + with create_pipe_input() as pipe: + monitor.app = build_monitor_application(monitor, input=pipe, output=DummyOutput()) + rendered = asyncio.Event() + monitor.app.after_render += lambda app: rendered.set() + task = asyncio.create_task(monitor.app.run_async()) + await asyncio.wait_for(rendered.wait(), 3) + callback = asyncio.create_task(asyncio.to_thread( + cli._secret_capture_callback, 'FIXTURE_SECRET', 'Owned fixture')) + try: + done, _ = await asyncio.wait({task}, timeout=1) + assert task in done, 'secret arrival did not immediately reclaim input' + assert buffer.text == '', 'draft must not be submitted as a secret' + cli._submit_secret_response('') + result = await asyncio.wait_for(callback, 3) + assert result['skipped'] + assert buffer.document == draft + finally: + if not task.done(): + monitor.app.exit() + await task + if not callback.done(): + cli._submit_secret_response('') + await callback + asyncio.run(run()) From 30be948a0cbb807891e34a606cf4ec829c734a40 Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Mon, 7 Sep 2026 21:42:10 -0700 Subject: [PATCH 152/227] fix: keep Ink agent progress monotonic across snapshot hydration --- ui-tui/src/__tests__/agentRoster.test.ts | 21 +++++++++++++++++++++ ui-tui/src/app/agentRoster.ts | 5 +++-- 2 files changed, 24 insertions(+), 2 deletions(-) diff --git a/ui-tui/src/__tests__/agentRoster.test.ts b/ui-tui/src/__tests__/agentRoster.test.ts index 66ffae6866e3..a492a027428c 100644 --- a/ui-tui/src/__tests__/agentRoster.test.ts +++ b/ui-tui/src/__tests__/agentRoster.test.ts @@ -2,6 +2,27 @@ import { expect, it } from 'vitest' import { $agentSnapshot, applyAgentSnapshot, mergeAgentRoster } from '../app/agentRoster.js' import { shouldPassThroughToGlobalHandler } from '../components/textInput.js' +import type { SubagentProgress } from '../types.js' + +it('never rolls event progress back when an older live snapshot arrives', () => { + const event: SubagentProgress = { + id: 'child', goal: 'inspect', depth: 0, index: 0, parentId: null, + notes: [], thinking: [], tools: ['read_file'], taskCount: 1, + status: 'running', toolCount: 5 + } + + const stale = { + subagents: [{ subagent_id: event.id, status: 'queued', tool_count: 1 }], + delegations: [] + } + + for (const status of ['running', 'completed', 'error', 'interrupted'] as const) { + const latest = { ...event, status } + const [row] = mergeAgentRoster([latest], stale) + expect(row?.status).toBe(latest.status) + expect(row?.toolCount).toBe(latest.toolCount) + } +}) it('lets the agents shortcut leave the composer without stealing redo', () => { const key = { ctrl: true, shift: false, meta: false } as Parameters[1] diff --git a/ui-tui/src/app/agentRoster.ts b/ui-tui/src/app/agentRoster.ts index 8885168c43a5..01711d219058 100644 --- a/ui-tui/src/app/agentRoster.ts +++ b/ui-tui/src/app/agentRoster.ts @@ -35,8 +35,9 @@ export function mergeAgentRoster(events: SubagentProgress[], data: SubagentListR delegationId: s.delegation_id ?? previous?.delegationId, model: s.model ?? previous?.model, startedAt: s.started_at != null ? s.started_at * 1000 : previous?.startedAt, - status: s.status === 'queued' ? 'queued' : 'running', - toolCount: s.tool_count ?? previous?.toolCount ?? 0, + // Snapshot replies may predate progress/completion events already rendered. + status: previous && previous.status !== 'queued' ? previous.status : s.status === 'queued' ? 'queued' : 'running', + toolCount: Math.max(s.tool_count ?? 0, previous?.toolCount ?? 0), tools: previous?.tools.length ? previous.tools : s.current_tool ? [s.current_tool] : [] }) } From 8a7b6ffad2d7bcc6f5ca60a44e8194b9ffc74464 Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Mon, 7 Sep 2026 21:42:10 -0700 Subject: [PATCH 153/227] fix: preserve newer Ink controls when status hydration arrives late --- ui-tui/src/__tests__/agentsHydration.test.tsx | 40 +++++++++++++++++++ ui-tui/src/components/agentsOverlay.tsx | 10 ++++- 2 files changed, 48 insertions(+), 2 deletions(-) create mode 100644 ui-tui/src/__tests__/agentsHydration.test.tsx diff --git a/ui-tui/src/__tests__/agentsHydration.test.tsx b/ui-tui/src/__tests__/agentsHydration.test.tsx new file mode 100644 index 000000000000..88ed88b27f53 --- /dev/null +++ b/ui-tui/src/__tests__/agentsHydration.test.tsx @@ -0,0 +1,40 @@ +import { PassThrough } from 'node:stream' + +import { Box, renderSync } from '@hermes/ink' +import React from 'react' +import { expect, it, vi } from 'vitest' + +import { $delegationState, resetDelegationState } from '../app/delegationStore.js' +import { AgentsOverlay } from '../components/agentsOverlay.js' +import type { GatewayClient } from '../gatewayClient.js' +import { DEFAULT_THEME } from '../theme.js' + +it('does not undo an acknowledged pause when opening status resolves late', async () => { + resetDelegationState() + let resolveStatus!: (value: unknown) => void + const status = new Promise(resolve => { resolveStatus = resolve }) + const request = vi.fn(async (method: string) => method === 'delegation.status' ? status : { paused: true }) + const stdout = Object.assign(new PassThrough(), { columns: 80, rows: 16, isTTY: false }) + const stdin = Object.assign(new PassThrough(), { isTTY: true, setRawMode: () => {}, ref: () => {}, unref: () => {} }) + + const view = renderSync( {}} t={DEFAULT_THEME} />, { + stdout: stdout as unknown as NodeJS.WriteStream, + stdin: stdin as unknown as NodeJS.ReadStream, + stderr: new PassThrough() as unknown as NodeJS.WriteStream, + patchConsole: false + }) + + try { + await vi.waitFor(() => expect(request).toHaveBeenCalledWith('delegation.status', {})) + stdin.write('p') + await vi.waitFor(() => expect($delegationState.get().paused).toBe(true)) + resolveStatus({ paused: false, max_spawn_depth: 4 }) + await status + await new Promise(resolve => setTimeout(resolve, 50)) + expect($delegationState.get().paused).toBe(true) + } finally { + view.unmount() + view.cleanup() + resetDelegationState() + } +}) diff --git a/ui-tui/src/components/agentsOverlay.tsx b/ui-tui/src/components/agentsOverlay.tsx index 4ee2eb4103dd..d44bda719d62 100644 --- a/ui-tui/src/components/agentsOverlay.tsx +++ b/ui-tui/src/components/agentsOverlay.tsx @@ -668,10 +668,16 @@ export function AgentsOverlay({ gw, initialHistoryIndex = 0, onClose, t }: Agent }, [cursor, historyIndex, mode]) useEffect(() => { - // Warm caps + paused flag on open. + // A control acknowledgement or newer hydration must win over this request. + const initial = $delegationState.get() + let active = true gw.request('delegation.status', {}) - .then(r => applyDelegationStatus(asRpcResult(r))) + .then(r => { + if (active && $delegationState.get() === initial) {applyDelegationStatus(asRpcResult(r))} + }) .catch(() => {}) + + return () => { active = false } }, [gw]) useEffect(() => { From 924c5ded2eca54070e1e1e239ea8d6a1540aa064 Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Mon, 7 Sep 2026 21:52:42 -0700 Subject: [PATCH 154/227] fix: scope subagent stops and publish authoritative live progress --- tests/tui_gateway/test_subagent_snapshot.py | 43 ++++++++++++++++++++- tui_gateway/AGENTS.md | 13 +++++-- tui_gateway/methods_session.py | 8 ---- tui_gateway/methods_subagents.py | 43 +++++++++++++-------- ui-tui/src/app/agentRoster.ts | 2 +- ui-tui/src/components/agentsOverlay.tsx | 2 +- ui-tui/src/gatewayTypes.ts | 2 +- 7 files changed, 79 insertions(+), 34 deletions(-) diff --git a/tests/tui_gateway/test_subagent_snapshot.py b/tests/tui_gateway/test_subagent_snapshot.py index feaf235ccb70..ca10d98cf108 100644 --- a/tests/tui_gateway/test_subagent_snapshot.py +++ b/tests/tui_gateway/test_subagent_snapshot.py @@ -30,6 +30,7 @@ def test_snapshot_projects_only_this_sessions_runtime_records(runtime): from tools import async_delegation as bg from tools.delegate_tool_child_run import _register_child from tools.delegate_tool_registry import _unregister_subagent + from tools.delegate_tool_progress import _build_child_progress_callback server, owner, transport, call = runtime release = threading.Event() @@ -52,11 +53,16 @@ def run(): foreign = SimpleNamespace(_subagent_id="foreign", _delegate_depth=1, model="test") _register_child(foreign, None, "foreign secret", owner_session_id="other", owner_transport=transport, owner_session_record={}) + progress = _build_child_progress_callback(0, "owned task", + SimpleNamespace(tool_progress_callback=lambda *a, **kw: None), subagent_id="child") + progress("tool.started", "read_file") + progress("tool.completed", "read_file") try: snapshot = call("subagent.list")["result"] assert [s["subagent_id"] for s in snapshot["subagents"]] == ["child"] - assert snapshot["delegations"][0]["delegation_id"] == did - assert snapshot["delegations"][0]["subagent_ids"] == ["child"] + assert snapshot["subagents"][0]["last_tool"] == "read_file" + assert snapshot["delegations"] == [] + assert snapshot["subagents"][0]["tool_count"] == 1 wire = json.dumps(snapshot) assert "private handoff" not in wire and "foreign secret" not in wire assert "owner_transport" not in wire and "session_key" not in wire @@ -109,3 +115,36 @@ def test_live_tail_and_steer_share_exact_owner_and_end_with_child(runtime): "subagent_id": "child", "available": False, "text": "", "truncated": False} finally: _unregister_subagent("child") + + + +def test_interrupt_requires_exact_live_owner_but_direct_helper_stays_legacy(runtime): + from tools.delegate_tool_child_run import _register_child + from tools.delegate_tool_registry import interrupt_subagent, _unregister_subagent + + server, owner, transport, call = runtime + stopped = [] + child = SimpleNamespace(_subagent_id="child", _delegate_depth=1, model="test", + hard_interrupt=lambda message: stopped.append(message)) + _register_child(child, None, "owned", owner_session_id="ui-owner", + owner_transport=transport, owner_session_record=owner) + try: + for params in ({"session_id": ""}, {"session_id": "missing"}, + {"via": SimpleNamespace(write=lambda frame: True)}): + reply = call("subagent.interrupt", subagent_id="child", **params) + assert "error" in reply or not reply["result"]["found"] + assert stopped == [] + server._sessions["foreign"] = {**owner} + assert not call("subagent.interrupt", session_id="foreign", subagent_id="child")["result"]["found"] + server._sessions["ui-owner"] = {**owner} + assert not call("subagent.interrupt", subagent_id="child")["result"]["found"] + assert stopped == [] + server._sessions["ui-owner"] = owner + assert call("subagent.interrupt", subagent_id="child")["result"]["found"] + assert len(stopped) == 1 + assert interrupt_subagent("child") + assert len(stopped) == 2 + _unregister_subagent("child") + assert not call("subagent.interrupt", subagent_id="child")["result"]["found"] + finally: + _unregister_subagent("child") diff --git a/tui_gateway/AGENTS.md b/tui_gateway/AGENTS.md index fda09b48d768..bb571fd34583 100644 --- a/tui_gateway/AGENTS.md +++ b/tui_gateway/AGENTS.md @@ -43,10 +43,11 @@ existing topical sibling, registered in the table — no `if method == ...` chai `subagent.list({session_id})` returns `{subagents, delegations}` for the calling transport's live session. Live child records are pinned to the exact session -record and transport; background records use their captured `origin_ui_session_id` -so compression does not hide ongoing work. Both lists are allowlisted projections: -no dispatch context, results, callbacks, or routing keys are sent. Clients hydrate -from this snapshot on their existing poll and avoid updates when unchanged. +record and transport. `last_tool` is the last started tool, not an in-flight +indicator. Async completion units are not agents and lack exact generation authority; +`delegations` remains an empty array for wire compatibility. No dispatch context, +results, callbacks, or routing keys are sent. Clients hydrate from this snapshot +on their existing poll and avoid updates when unchanged. `subagent.tail({session_id, subagent_id})` returns `{subagent_id, available, text, truncated}`: the last 16 KiB of the live child's @@ -56,6 +57,10 @@ This is live-only, not persisted completion history. Invalid session/transport returns error 4001. `subagent.steer({session_id, subagent_id, text})` remains the shared control: `status: queued` acknowledges acceptance, not delivery; final boundary races are reported by the existing runtime as `missed_steer`. +`subagent.interrupt({session_id, subagent_id})` requires the same exact live +session/transport/generation ownership, including for subtree members. Missing +RPC session authority is rejected; direct in-process `interrupt_subagent(id)` +retains its legacy unscoped contract. ## Slash command flow diff --git a/tui_gateway/methods_session.py b/tui_gateway/methods_session.py index a1eaff2bf4c9..e12cdab1bfbd 100644 --- a/tui_gateway/methods_session.py +++ b/tui_gateway/methods_session.py @@ -2033,14 +2033,6 @@ def _(rid, params: dict) -> dict: return _ok(rid, {"paused": set_spawn_paused(bool(params.get("paused", True)))}) -@method("subagent.interrupt") -def _(rid, params: dict) -> dict: - from tools.delegate_tool import interrupt_subagent - if not (subagent_id := _str_param(params, "subagent_id")): - return _err(rid, 4000, "subagent_id required") - return _ok(rid, {"found": interrupt_subagent(subagent_id), "subagent_id": subagent_id}) - - @method("subagent.steer") def _(rid, params: dict) -> dict: """Queue steering text into a live delegated child (the in-flight tool call is never cut). "queued" diff --git a/tui_gateway/methods_subagents.py b/tui_gateway/methods_subagents.py index 53547df19e44..97b8eb02c4d4 100644 --- a/tui_gateway/methods_subagents.py +++ b/tui_gateway/methods_subagents.py @@ -11,10 +11,7 @@ _SUBAGENT_SNAPSHOT_FIELDS = ( "subagent_id", "parent_id", "depth", "goal", "delegation_id", "model", - "started_at", "status", "tool_count", "current_tool", "accepting_steer", -) -_ASYNC_SNAPSHOT_FIELDS = ( - "delegation_id", "goal", "role", "model", "status", "dispatched_at", "completed_at", "is_batch", + "started_at", "status", "tool_count", "last_tool", "accepting_steer", ) _SUBAGENT_TAIL_BYTES = 16384 @@ -35,25 +32,37 @@ def _(rid, params): transport, owner = _current_session_steer_authority(session_id) if transport is None or owner is None: return _err(rid, 4001, "session not found or not owned by this transport") - from tools.async_delegation import _records, _records_lock - live = _owned_subagent_records(session_id, transport, owner) - # Read only projected fields, without invoking unrelated sessions' progress callbacks. - with _records_lock: - delegations = [] - for record in _records.values(): - if record.get("origin_ui_session_id") != session_id: - continue - item = {key: record.get(key) for key in _ASYNC_SNAPSHOT_FIELDS} - item["subagent_ids"] = [r["subagent_id"] for r in live - if r.get("delegation_id") == record.get("delegation_id")] - delegations.append(item) return _ok(rid, { "subagents": [{key: r.get(key) for key in _SUBAGENT_SNAPSHOT_FIELDS} for r in live], - "delegations": delegations, + "delegations": [], }) +@method("subagent.interrupt") +def _(rid, params): + from agent.interrupt_compat import request_hard_interrupt + + subagent_id = _str_param(params, "subagent_id") + if not subagent_id: + return _err(rid, 4000, "subagent_id required") + session_id = _str_param(params, "session_id") + transport, owner = _current_session_steer_authority(session_id) + if transport is None or owner is None: + return _err(rid, 4001, "session not found or not owned by this transport") + record = next((r for r in _owned_subagent_records(session_id, transport, owner) + if r.get("subagent_id") == subagent_id), None) + agent = record.get("agent") if record else None + # Interrupt the authorized object, never re-resolve a globally recyclable id. + found = False + if agent is not None: + try: + found = bool(request_hard_interrupt(agent, f"Interrupted via TUI ({subagent_id})")) + except Exception: + logger.debug("subagent interrupt failed", exc_info=True) + return _ok(rid, {"found": found, "subagent_id": subagent_id}) + + @method("subagent.tail") def _(rid, params): session_id = _str_param(params, "session_id") diff --git a/ui-tui/src/app/agentRoster.ts b/ui-tui/src/app/agentRoster.ts index 01711d219058..f30ba1c1d57f 100644 --- a/ui-tui/src/app/agentRoster.ts +++ b/ui-tui/src/app/agentRoster.ts @@ -38,7 +38,7 @@ export function mergeAgentRoster(events: SubagentProgress[], data: SubagentListR // Snapshot replies may predate progress/completion events already rendered. status: previous && previous.status !== 'queued' ? previous.status : s.status === 'queued' ? 'queued' : 'running', toolCount: Math.max(s.tool_count ?? 0, previous?.toolCount ?? 0), - tools: previous?.tools.length ? previous.tools : s.current_tool ? [s.current_tool] : [] + tools: previous?.tools.length ? previous.tools : s.last_tool ? [s.last_tool] : [] }) } diff --git a/ui-tui/src/components/agentsOverlay.tsx b/ui-tui/src/components/agentsOverlay.tsx index d44bda719d62..be5505e33eef 100644 --- a/ui-tui/src/components/agentsOverlay.tsx +++ b/ui-tui/src/components/agentsOverlay.tsx @@ -696,7 +696,7 @@ export function AgentsOverlay({ gw, initialHistoryIndex = 0, onClose, t }: Agent } } - const interrupt = (id: string) => gw.request('subagent.interrupt', { subagent_id: id }) + const interrupt = (id: string) => gw.request('subagent.interrupt', { session_id: sid, subagent_id: id }) const killOne = (id: string) => guardLive(() => { diff --git a/ui-tui/src/gatewayTypes.ts b/ui-tui/src/gatewayTypes.ts index 7de5b462a47e..4b0b0be85cd2 100644 --- a/ui-tui/src/gatewayTypes.ts +++ b/ui-tui/src/gatewayTypes.ts @@ -617,7 +617,7 @@ export interface SubagentListResponse { started_at?: number | null status?: string | null tool_count?: number | null - current_tool?: string | null + last_tool?: string | null }[] delegations: AsyncDelegationRecord[] } From 975699714dcd3d7da626d5629d72345488b0e6a6 Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Mon, 7 Sep 2026 21:52:00 -0700 Subject: [PATCH 155/227] feat(desktop): hydrate scoped subagents and read extended live transcripts --- .../app/chat/composer/status-stack/index.tsx | 2 + .../status-stack/subagent-hydration.test.tsx | 94 +++++++++++++++++++ .../status-stack/subagent-section.tsx | 6 +- .../status-stack/subagent-transcript.tsx | 73 ++++++++++++++ .../status-stack/use-subagent-snapshot.ts | 75 +++++++++++++++ apps/desktop/src/i18n/ar.ts | 4 + apps/desktop/src/i18n/en.ts | 4 + apps/desktop/src/i18n/ja.ts | 4 + apps/desktop/src/i18n/ru.ts | 4 + apps/desktop/src/i18n/types.ts | 3 + apps/desktop/src/i18n/zh-hant.ts | 4 + apps/desktop/src/i18n/zh.ts | 4 + .../src/store/subagent-snapshot.test.ts | 32 +++++++ apps/desktop/src/store/subagents.ts | 56 +++++++++++ 14 files changed, 364 insertions(+), 1 deletion(-) create mode 100644 apps/desktop/src/app/chat/composer/status-stack/subagent-hydration.test.tsx create mode 100644 apps/desktop/src/app/chat/composer/status-stack/subagent-transcript.tsx create mode 100644 apps/desktop/src/app/chat/composer/status-stack/use-subagent-snapshot.ts create mode 100644 apps/desktop/src/store/subagent-snapshot.test.ts diff --git a/apps/desktop/src/app/chat/composer/status-stack/index.tsx b/apps/desktop/src/app/chat/composer/status-stack/index.tsx index c7a085450be8..b58e2057026a 100644 --- a/apps/desktop/src/app/chat/composer/status-stack/index.tsx +++ b/apps/desktop/src/app/chat/composer/status-stack/index.tsx @@ -35,6 +35,7 @@ import { SessionControlSections } from './session-control' import { useSessionValue } from './session-control-utils' import { StatusItemRow } from './status-row' import { SubagentSection } from './subagent-section' +import { useSubagentSnapshot } from './use-subagent-snapshot' // Slow safety-net poll for silent exits (processes without notify_on_complete // emit no event when they die). Only armed while a running row is on screen. @@ -93,6 +94,7 @@ interface ComposerStatusStackProps { export function ComposerStatusStack({ onSubmit, queue, sessionId }: ComposerStatusStackProps) { const { t } = useI18n() const navigate = useNavigate() + useSubagentSnapshot(sessionId) // Subscribe to THIS session's slice only. Both maps churn on other // sessions' activity (subagent ticks, background polls, preview updates in // any tile); a whole-map `useStore` re-rendered every mounted stack — one diff --git a/apps/desktop/src/app/chat/composer/status-stack/subagent-hydration.test.tsx b/apps/desktop/src/app/chat/composer/status-stack/subagent-hydration.test.tsx new file mode 100644 index 000000000000..8ce59a6d33c5 --- /dev/null +++ b/apps/desktop/src/app/chat/composer/status-stack/subagent-hydration.test.tsx @@ -0,0 +1,94 @@ +import { act, cleanup, fireEvent, render, screen, waitFor } from '@testing-library/react' +import { MemoryRouter } from 'react-router' +import { afterEach, expect, it, vi } from 'vitest' + +import * as gateway from '@/store/gateway' +import { _resetSessionOwnerHintsForTests, setSessionOwnerHint } from '@/store/session' +import { $subagentsBySession, upsertSubagent } from '@/store/subagents' + +import { ComposerStatusStack } from './index' + +vi.stubGlobal( + 'ResizeObserver', + class { + disconnect() {} + observe() {} + unobserve() {} + } +) +Element.prototype.animate = vi.fn(() => ({ cancel() {} }) as Animation) +afterEach(() => { + cleanup() + $subagentsBySession.set({}) + _resetSessionOwnerHintsForTests() + vi.restoreAllMocks() +}) + +it('hydrates the empty owner composer and exposes an extended owner-routed transcript without leaking on session change', async () => { + const text = 'Full transcript line with preserved whitespace\n'.repeat(100) + + const request = vi.spyOn(gateway, 'requestGatewayForAgent').mockImplementation(async (_c, _p, method) => { + if (method === 'subagent.list') { + return { + subagents: [{ subagent_id: 'worker', goal: 'Recovered work', started_at: 1000, status: 'running' }], + delegations: [] + } as never + } + + if (method === 'subagent.tail') { + return { subagent_id: 'worker', available: true, text, truncated: true } as never + } + + return {} as never + }) + + setSessionOwnerHint('parent', { connectionId: 'remote-owner', profile: 'research' }) + + const view = render( + + + + ) + + await screen.findByText('Recovered work') + expect($subagentsBySession.get().parent[0].startedAt).toBe(1000000) + fireEvent.click(screen.getByRole('button', { name: /Recovered work/ })) + await waitFor(() => expect(document.querySelector('[data-slot="subagent-transcript"]')?.textContent).toContain(text)) + expect(request).toHaveBeenCalledWith('remote-owner', 'research', 'subagent.tail', { + session_id: 'parent', + subagent_id: 'worker' + }) + view.rerender( + + + + ) + expect(document.querySelector('[data-slot="subagent-transcript"]')).toBeNull() +}) + +it('does not resurrect a child completed while the roster snapshot was in flight', async () => { + let resolve!: (value: unknown) => void + vi.spyOn(gateway, 'requestGatewayForAgent').mockImplementation(async (_c, _p, method) => { + if (method === 'subagent.list') { + return (await new Promise(r => { + resolve = r + })) as never + } + + return {} as never + }) + setSessionOwnerHint('parent', { connectionId: 'remote-owner', profile: 'research' }) + upsertSubagent('parent', { subagent_id: 'worker', goal: 'Finishing work' }) + render( + + + + ) + await waitFor(() => expect(resolve).toBeTypeOf('function')) + act(() => upsertSubagent('parent', { subagent_id: 'worker', status: 'completed' }, false, 'subagent.complete')) + await act(async () => + resolve({ subagents: [{ subagent_id: 'worker', goal: 'Finishing work', status: 'running' }], delegations: [] }) + ) + expect(screen.queryByText('Finishing work')).toBeNull() + expect($subagentsBySession.get().parent[0].status).toBe('completed') +}) diff --git a/apps/desktop/src/app/chat/composer/status-stack/subagent-section.tsx b/apps/desktop/src/app/chat/composer/status-stack/subagent-section.tsx index bac54f320dae..230cc7fdfb3a 100644 --- a/apps/desktop/src/app/chat/composer/status-stack/subagent-section.tsx +++ b/apps/desktop/src/app/chat/composer/status-stack/subagent-section.tsx @@ -11,6 +11,7 @@ import { useSessionSlice } from '@/lib/use-session-slice' import { $subagentsBySession, type SubagentProgress } from '@/store/subagents' import { SubagentControls } from './subagent-controls' +import { SubagentTranscript } from './subagent-transcript' interface SubagentSectionProps { sessionId: string @@ -78,7 +79,7 @@ export function SubagentSection({ sessionId }: SubagentSectionProps) { {detail && (
- {(detail.status === 'running' || detail.status === 'queued') && ( + {!detail.id.startsWith('delegation:') && ( )} + {!detail.id.startsWith('delegation:') && ( + + )}
)}
diff --git a/apps/desktop/src/app/chat/composer/status-stack/subagent-transcript.tsx b/apps/desktop/src/app/chat/composer/status-stack/subagent-transcript.tsx new file mode 100644 index 000000000000..542c45b87eb5 --- /dev/null +++ b/apps/desktop/src/app/chat/composer/status-stack/subagent-transcript.tsx @@ -0,0 +1,73 @@ +import { useEffect, useState } from 'react' + +import { useI18n } from '@/i18n' +import { knownOwnerForSession, requestForOwnedSession } from '@/store/session-states' + +import { rejectUnownedSubagentRequest } from './use-subagent-snapshot' + +interface Tail { + available: boolean + text: string + truncated: boolean +} + +export function SubagentTranscript({ sessionId, subagentId }: { sessionId: string; subagentId: string }) { + const { t } = useI18n() + const [tail, setTail] = useState(null) + useEffect(() => { + let cancelled = false + let pending = false + const owner = JSON.stringify(knownOwnerForSession(sessionId)) + + const refresh = async () => { + if (pending || document.visibilityState === 'hidden') { + return + } + + pending = true + + try { + const result = await requestForOwnedSession(sessionId, rejectUnownedSubagentRequest, 'subagent.tail', { + session_id: sessionId, + subagent_id: subagentId + }) + + if (!cancelled && owner === JSON.stringify(knownOwnerForSession(sessionId))) { + setTail({ + available: result.available, + text: typeof result.text === 'string' ? result.text.slice(-16384) : '', + truncated: result.truncated + }) + } + } catch { + if (!cancelled) { + setTail({ available: false, text: '', truncated: false }) + } + } finally { + pending = false + } + } + + void refresh() + const timer = window.setInterval(() => void refresh(), 2000) + + return () => { + cancelled = true + window.clearInterval(timer) + } + }, [sessionId, subagentId]) + + return ( +
+

{t.agents.extendedTranscript}

+ {tail?.truncated &&

{t.agents.transcriptTruncated}

} + {tail?.available ? ( + 
+          {tail.text}
+
+ ) : ( +

{tail ? t.agents.transcriptUnavailable : t.agents.waitingActivity}

+ )} +
+ ) +} diff --git a/apps/desktop/src/app/chat/composer/status-stack/use-subagent-snapshot.ts b/apps/desktop/src/app/chat/composer/status-stack/use-subagent-snapshot.ts new file mode 100644 index 000000000000..8550121423a5 --- /dev/null +++ b/apps/desktop/src/app/chat/composer/status-stack/use-subagent-snapshot.ts @@ -0,0 +1,75 @@ +import { useStore } from '@nanostores/react' +import { useEffect } from 'react' + +import { $gatewayState } from '@/store/session' +import { knownOwnerForSession, requestForOwnedSession } from '@/store/session-states' +import { $subagentsBySession, reconcileSubagentSnapshot, type SubagentPayload } from '@/store/subagents' + +export const rejectUnownedSubagentRequest = async (): Promise => { + throw new Error('Subagent owner unavailable') +} + +/** Hydrate even an empty composer; live events remain authoritative over reads. */ +export function useSubagentSnapshot(sessionId: string | null) { + const gatewayState = useStore($gatewayState) + useEffect(() => { + if (!sessionId) { + return + } + + let cancelled = false + let pending = false + let failures = 0 + + const refresh = async () => { + if (cancelled || pending || failures >= 3) { + return + } + + pending = true + const before = $subagentsBySession.get()[sessionId] + const owner = JSON.stringify(knownOwnerForSession(sessionId)) + + try { + const snapshot = await requestForOwnedSession<{ subagents: SubagentPayload[]; delegations: SubagentPayload[] }>( + sessionId, + rejectUnownedSubagentRequest, + 'subagent.list', + { session_id: sessionId } + ) + + if ( + !cancelled && + owner === JSON.stringify(knownOwnerForSession(sessionId)) && + before === $subagentsBySession.get()[sessionId] && + Array.isArray(snapshot.subagents) + ) { + reconcileSubagentSnapshot(sessionId, snapshot.subagents, snapshot.delegations) + } + + failures = 0 + } catch { + // Older backends retain their event-fed frame; don't hot-loop a missing RPC. + failures++ + } finally { + pending = false + } + } + + void refresh() + const timer = window.setInterval(() => void refresh(), 5000) + + const retry = () => { + failures = 0 + void refresh() + } + + window.addEventListener('focus', retry) + + return () => { + cancelled = true + window.clearInterval(timer) + window.removeEventListener('focus', retry) + } + }, [sessionId, gatewayState]) +} diff --git a/apps/desktop/src/i18n/ar.ts b/apps/desktop/src/i18n/ar.ts index 32ec296f75b5..cbb7452a379d 100644 --- a/apps/desktop/src/i18n/ar.ts +++ b/apps/desktop/src/i18n/ar.ts @@ -1069,6 +1069,10 @@ export const ar = defineLocale({ failedToUpdate: name => `فشل تحديث ${name}` }, agents: { + extendedTranscript: 'سجل موسّع', + transcriptTruncated: 'عرض أحدث 16 KiB', + transcriptUnavailable: 'السجل المباشر غير متاح', + close: 'إغلاق الوكلاء', title: 'شجرة التوليد', subtitle: 'نشاط الوكلاء الفرعيين المباشر للدور الحالي.', diff --git a/apps/desktop/src/i18n/en.ts b/apps/desktop/src/i18n/en.ts index 11b6d8e6ee3f..280d691f19df 100644 --- a/apps/desktop/src/i18n/en.ts +++ b/apps/desktop/src/i18n/en.ts @@ -1582,6 +1582,10 @@ export const en: Translations = { resetToMine: 'Back to my map' }, agents: { + extendedTranscript: 'Extended transcript', + transcriptTruncated: 'Showing the latest 16 KiB', + transcriptUnavailable: 'Live transcript unavailable', + close: 'Close agents', title: 'Spawn tree', subtitle: 'Live subagent activity for the current turn.', diff --git a/apps/desktop/src/i18n/ja.ts b/apps/desktop/src/i18n/ja.ts index 4da03e279f96..08d7d3f28516 100644 --- a/apps/desktop/src/i18n/ja.ts +++ b/apps/desktop/src/i18n/ja.ts @@ -1395,6 +1395,10 @@ export const ja = defineLocale({ emptyDesc: 'Hermes がスキルやメモリを蓄積すると、ここに表示されます。' }, agents: { + extendedTranscript: '詳細な実行ログ', + transcriptTruncated: '最新の 16 KiB を表示', + transcriptUnavailable: 'ライブログは利用できません', + close: 'エージェントを閉じる', title: 'スポーンツリー', subtitle: '現在のターンのライブサブエージェントのアクティビティ。', diff --git a/apps/desktop/src/i18n/ru.ts b/apps/desktop/src/i18n/ru.ts index ae4716db1237..45fc98d7b210 100644 --- a/apps/desktop/src/i18n/ru.ts +++ b/apps/desktop/src/i18n/ru.ts @@ -1660,6 +1660,10 @@ export const ru = defineLocale({ resetToMine: 'Вернуться к моей карте' }, agents: { + extendedTranscript: 'Подробный журнал', + transcriptTruncated: 'Последние 16 КиБ', + transcriptUnavailable: 'Текущий журнал недоступен', + close: 'Закрыть агентов', title: 'Дерево запусков', subtitle: 'Активные субагенты текущего хода в реальном времени.', diff --git a/apps/desktop/src/i18n/types.ts b/apps/desktop/src/i18n/types.ts index f30043fe7a38..6ed488b170f7 100644 --- a/apps/desktop/src/i18n/types.ts +++ b/apps/desktop/src/i18n/types.ts @@ -1393,6 +1393,9 @@ export interface Translations { resetToMine: string } agents: { + extendedTranscript: string + transcriptTruncated: string + transcriptUnavailable: string close: string title: string subtitle: string diff --git a/apps/desktop/src/i18n/zh-hant.ts b/apps/desktop/src/i18n/zh-hant.ts index 10d4eb9f28bc..d5243ce78ed5 100644 --- a/apps/desktop/src/i18n/zh-hant.ts +++ b/apps/desktop/src/i18n/zh-hant.ts @@ -1342,6 +1342,10 @@ export const zhHant = defineLocale({ emptyDesc: '當 Hermes 為你的工作建立技能與記憶時，會顯示在這裡。' }, agents: { + extendedTranscript: '完整記錄尾端', + transcriptTruncated: '顯示最新 16 KiB', + transcriptUnavailable: '即時記錄無法使用', + close: '關閉代理', title: '派生樹', subtitle: '目前回合的子代理即時活動。', diff --git a/apps/desktop/src/i18n/zh.ts b/apps/desktop/src/i18n/zh.ts index 2d04fba130e7..1366057b68ba 100644 --- a/apps/desktop/src/i18n/zh.ts +++ b/apps/desktop/src/i18n/zh.ts @@ -1754,6 +1754,10 @@ export const zh: Translations = { resetToMine: '返回我的图谱' }, agents: { + extendedTranscript: '扩展记录', + transcriptTruncated: '显示最新 16 KiB', + transcriptUnavailable: '实时记录不可用', + close: '关闭代理', title: '派生树', subtitle: '当前回合的子代理实时活动。', diff --git a/apps/desktop/src/store/subagent-snapshot.test.ts b/apps/desktop/src/store/subagent-snapshot.test.ts new file mode 100644 index 000000000000..4605fd74e01a --- /dev/null +++ b/apps/desktop/src/store/subagent-snapshot.test.ts @@ -0,0 +1,32 @@ +import { afterEach, expect, it } from 'vitest' + +import { $subagentsBySession, reconcileSubagentSnapshot } from './subagents' + +afterEach(() => $subagentsBySession.set({})) +it('projects unresolved running delegations once, replaces them with real workers, and retires completed units', () => { + const delegation = { + delegation_id: 'batch', + goal: 'Queued unit', + status: 'running', + dispatched_at: 1000, + subagent_ids: [] + } + reconcileSubagentSnapshot('owner', [], [delegation]) + expect($subagentsBySession.get().owner).toMatchObject([ + { id: 'delegation:batch', goal: 'Queued unit', startedAt: 1000000 } + ]) + const child = { + subagent_id: 'worker', + delegation_id: 'batch', + goal: 'Actual worker', + status: 'running', + started_at: 1001 + } + reconcileSubagentSnapshot('owner', [child], [delegation]) + expect($subagentsBySession.get().owner.map(row => row.id)).toEqual(['worker']) + const before = $subagentsBySession.get().owner + reconcileSubagentSnapshot('owner', [child], [delegation]) + expect($subagentsBySession.get().owner).toBe(before) + reconcileSubagentSnapshot('owner', [], [{ ...delegation, status: 'completed' }]) + expect($subagentsBySession.get().owner).toEqual([]) +}) diff --git a/apps/desktop/src/store/subagents.ts b/apps/desktop/src/store/subagents.ts index ae923aae8d83..fc2103e41f03 100644 --- a/apps/desktop/src/store/subagents.ts +++ b/apps/desktop/src/store/subagents.ts @@ -212,6 +212,62 @@ function toProgress(payload: SubagentPayload, prev: SubagentProgress | undefined } } +/** Reconcile a scoped, race-checked snapshot without replacing stream history. */ +export function reconcileSubagentSnapshot( + sid: string, + children: SubagentPayload[], + delegations: SubagentPayload[] = [] +) { + const represented = new Set(children.map(child => str(child.delegation_id)).filter(Boolean)) + + const payloads: SubagentPayload[] = [ + ...children, + ...delegations + .filter(unit => unit.status === 'running' && !represented.has(str(unit.delegation_id)) && str(unit.delegation_id)) + .map(unit => ({ + ...unit, + subagent_id: `delegation:${str(unit.delegation_id)}`, + started_at: unit.dispatched_at, + status: 'queued' + })) + ] + + const map = $subagentsBySession.get() + const previous = map[sid] ?? [] + const ids = new Set(payloads.map(p => str(p.subagent_id)).filter(Boolean)) + const next = previous.filter(item => TERMINAL.has(item.status) || ids.has(item.id)) + + for (const payload of payloads) { + const id = str(payload.subagent_id) + + if (!id) { + continue + } + + const index = next.findIndex(item => item.id === id) + const prev = next[index] + + if (prev && TERMINAL.has(prev.status)) { + continue + } + + const projected = toProgress(payload, prev) + projected.startedAt = (num(payload.started_at) ?? 0) * 1000 || prev?.startedAt || projected.startedAt + projected.updatedAt = prev?.updatedAt ?? projected.startedAt + projected.currentTool = str(payload.current_tool) || undefined + + if (index < 0) { + next.push(projected) + } else { + next[index] = JSON.stringify(prev) === JSON.stringify(projected) ? prev : projected + } + } + + if (next.length !== previous.length || next.some((item, index) => item !== previous[index])) { + $subagentsBySession.set({ ...map, [sid]: next }) + } +} + export function clearSessionSubagents(sid: string) { const map = $subagentsBySession.get() From 7befa11bf25b94a91edcdd11bb10161acbb03c1b Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Mon, 7 Sep 2026 22:07:59 -0700 Subject: [PATCH 156/227] fix: retain subagent control after live session reattachment --- tests/tui_gateway/test_subagent_snapshot.py | 72 +++++++++++++++++++++ tools/delegate_tool_registry.py | 5 ++ tui_gateway/AGENTS.md | 4 +- tui_gateway/session_lifecycle.py | 14 +++- 4 files changed, 92 insertions(+), 3 deletions(-) diff --git a/tests/tui_gateway/test_subagent_snapshot.py b/tests/tui_gateway/test_subagent_snapshot.py index ca10d98cf108..2d38cd9b0ac8 100644 --- a/tests/tui_gateway/test_subagent_snapshot.py +++ b/tests/tui_gateway/test_subagent_snapshot.py @@ -148,3 +148,75 @@ def test_interrupt_requires_exact_live_owner_but_direct_helper_stays_legacy(runt assert not call("subagent.interrupt", subagent_id="child")["result"]["found"] finally: _unregister_subagent("child") + + + +def test_reattach_preserves_child_controls_including_late_registration(runtime, tmp_path): + from tools.delegate_tool_child_run import _register_child + + server, owner, old, call = runtime + new = type("Transport", (), {"write": lambda self, frame: True})() + transcript = tmp_path / "child.txt" + transcript.write_text("live child output") + steered, stopped = [], [] + + def register(sid): + child = SimpleNamespace(_subagent_id=sid, _delegate_depth=1, model="test", + _live_transcript_path=str(transcript), + steer=lambda text: steered.append(text) or True, + hard_interrupt=lambda text: stopped.append(text)) + _register_child(child, None, "owned", owner_session_id="ui-owner", + owner_transport=old, owner_session_record=owner) + + register("before") + owner["transport"] = server._detached_ws_transport + owner["history_lock"] = threading.Lock() + with server._session_resume_lock, owner["history_lock"]: + assert server._reattach_refusal(1, "ui-owner", owner) is None + server._rebind_live_transport("ui-owner", owner, new) + # A dispatch captured before reload may not construct its child until afterwards. + register("after") + assert {row["subagent_id"] for row in call("subagent.list", via=new)["result"]["subagents"]} == {"before", "after"} + # Closing a second authenticated viewer hands control back to the survivor. + popup = type(new)() + with server._session_resume_lock, owner["history_lock"]: + server._rebind_live_transport("ui-owner", owner, popup) + assert server._close_sessions_for_transport(popup) == (0, 0) + assert owner["transport"] is new + assert {row["subagent_id"] for row in call("subagent.list", via=new)["result"]["subagents"]} == {"before", "after"} + for sid in ("before", "after"): + assert call("subagent.tail", via=new, subagent_id=sid)["result"]["text"] == "live child output" + assert call("subagent.steer", via=new, subagent_id=sid, text=sid)["result"]["status"] == "queued" + assert call("subagent.interrupt", via=new, subagent_id=sid)["result"]["found"] + assert steered == ["before", "after"] and len(stopped) == 2 + for method in ("list", "tail", "steer", "interrupt"): + denied = call("subagent." + method, subagent_id="before", text="old") + assert "error" in denied or denied["result"].get("status") == "rejected" + assert steered == ["before", "after"] and len(stopped) == 2 + + +def test_reattach_does_not_adopt_foreign_or_retired_generations(runtime): + from tools.delegate_tool_child_run import _register_child + + server, owner, old, call = runtime + new = type("Transport", (), {"write": lambda self, frame: True})() + effects = [] + for sid, session_id, record in (("foreign", "other", owner), + ("retired", "ui-owner", {**owner})): + child = SimpleNamespace(_subagent_id=sid, _delegate_depth=1, model="test", + steer=lambda text: effects.append(text) or True, + hard_interrupt=lambda text: effects.append(text)) + _register_child(child, None, "private", owner_session_id=session_id, + owner_transport=old, owner_session_record=record) + with server._session_resume_lock: + assert server._reattach_refusal(1, "ui-owner", {**owner})["error"]["code"] == 4007 + owner["_client_gone_interrupt_requested"] = True + assert server._reattach_refusal(1, "ui-owner", owner)["error"]["code"] == 4009 + del owner["_client_gone_interrupt_requested"] + server._rebind_live_transport("ui-owner", owner, new) + assert call("subagent.list", via=new)["result"]["subagents"] == [] + for sid in ("foreign", "retired"): + assert not call("subagent.tail", via=new, subagent_id=sid)["result"]["available"] + assert call("subagent.steer", via=new, subagent_id=sid, text="deny")["result"]["status"] == "rejected" + assert not call("subagent.interrupt", via=new, subagent_id=sid)["result"]["found"] + assert effects == [] diff --git a/tools/delegate_tool_registry.py b/tools/delegate_tool_registry.py index 43d295d9ec11..dd2a225e12f4 100644 --- a/tools/delegate_tool_registry.py +++ b/tools/delegate_tool_registry.py @@ -53,6 +53,11 @@ def _register_subagent(record: Dict[str, Any]) -> None: return record.setdefault("accepting_steer", True) with _active_subagents_lock: + owner = record.get("owner_session_record") + if owner is not None and record.get("owner_transport") is not None: + # Child construction can finish after its captured dispatch transport + # was replaced. The exact session object retains generation authority. + record["owner_transport"] = owner.get("transport") _active_subagents[sid] = record def _unregister_subagent(subagent_id: str, *, agent: Any = None) -> None: diff --git a/tui_gateway/AGENTS.md b/tui_gateway/AGENTS.md index bb571fd34583..d4181645bd11 100644 --- a/tui_gateway/AGENTS.md +++ b/tui_gateway/AGENTS.md @@ -43,7 +43,9 @@ existing topical sibling, registered in the table — no `if method == ...` chai `subagent.list({session_id})` returns `{subagents, delegations}` for the calling transport's live session. Live child records are pinned to the exact session -record and transport. `last_tool` is the last started tool, not an in-flight +record and transport. Authenticated live reattachment transfers that exact generation's +child authority to the new transport (also for late child registration and surviving +viewers); foreign or retired generations remain inaccessible. `last_tool` is the last started tool, not an in-flight indicator. Async completion units are not agents and lack exact generation authority; `delegations` remains an empty array for wire compatibility. No dispatch context, results, callbacks, or routing keys are sent. Clients hydrate from this snapshot diff --git a/tui_gateway/session_lifecycle.py b/tui_gateway/session_lifecycle.py index 8792565c856e..166abfe247e5 100644 --- a/tui_gateway/session_lifecycle.py +++ b/tui_gateway/session_lifecycle.py @@ -466,7 +466,17 @@ def _reattach_refusal(rid, sid: str, session: dict) -> dict | None: def _rebind_live_transport(sid: str, session: dict, transport: Transport) -> None: """Attach a live peer without displacing existing subscribers (caller holds ``history_lock``).""" - _attach_session_transport(session, transport) + from tools.delegate_tool_registry import _active_subagents, _active_subagents_lock + + # Transfer only this exact live generation's capabilities at the authenticated + # attachment seam, including records spawned through an older dispatch context. + with _active_subagents_lock: + _attach_session_transport(session, transport) + for record in _active_subagents.values(): + if (record.get("owner_session_id") == sid + and record.get("owner_session_record") is session + and record.get("owner_transport") is not None): + record["owner_transport"] = transport # Every transport that showed this session (pop-outs resume the same sid); on disconnect the last # viewer becomes the transport instead of the drop sentinel. session.setdefault("viewers", {})[transport] = time.time() @@ -632,7 +642,7 @@ def _close_sessions_for_transport(transport, *, end_reason: str = "ws_disconnect viewers.pop(transport, None) live = [vt for vt, ts in sorted(viewers.items(), key=lambda kv: kv[1]) if not _transport_is_dead(vt)] if live: - current["transport"] = live[-1] + _rebind_live_transport(sid, current, live[-1]) else: current["transport"] = _detached_ws_transport current.pop("_client_gone_interrupt_requested", None) From c2a0a188c194aabfac0f8940651d7e42407ee417 Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Mon, 7 Sep 2026 21:59:51 -0700 Subject: [PATCH 157/227] fix(desktop): align subagent hydration with owned roster contract --- .../status-stack/subagent-hydration.test.tsx | 8 +++- .../status-stack/subagent-section.tsx | 20 ++++----- .../status-stack/use-subagent-snapshot.ts | 4 +- .../src/store/subagent-snapshot.test.ts | 41 +++++++++---------- apps/desktop/src/store/subagents.ts | 31 ++++---------- 5 files changed, 45 insertions(+), 59 deletions(-) diff --git a/apps/desktop/src/app/chat/composer/status-stack/subagent-hydration.test.tsx b/apps/desktop/src/app/chat/composer/status-stack/subagent-hydration.test.tsx index 8ce59a6d33c5..4f8f47929ee7 100644 --- a/apps/desktop/src/app/chat/composer/status-stack/subagent-hydration.test.tsx +++ b/apps/desktop/src/app/chat/composer/status-stack/subagent-hydration.test.tsx @@ -25,12 +25,14 @@ afterEach(() => { }) it('hydrates the empty owner composer and exposes an extended owner-routed transcript without leaking on session change', async () => { - const text = 'Full transcript line with preserved whitespace\n'.repeat(100) + const text = ' Full transcript\tline with preserved whitespace \n\n'.repeat(100) const request = vi.spyOn(gateway, 'requestGatewayForAgent').mockImplementation(async (_c, _p, method) => { if (method === 'subagent.list') { return { - subagents: [{ subagent_id: 'worker', goal: 'Recovered work', started_at: 1000, status: 'running' }], + subagents: [ + { subagent_id: 'worker', goal: 'Recovered work', started_at: 1000, status: 'running', last_tool: 'read_file' } + ], delegations: [] } as never } @@ -51,6 +53,8 @@ it('hydrates the empty owner composer and exposes an extended owner-routed trans ) await screen.findByText('Recovered work') + expect(screen.getByText('Read File')).toBeTruthy() + expect(request).toHaveBeenCalledWith('remote-owner', 'research', 'subagent.list', { session_id: 'parent' }) expect($subagentsBySession.get().parent[0].startedAt).toBe(1000000) fireEvent.click(screen.getByRole('button', { name: /Recovered work/ })) await waitFor(() => expect(document.querySelector('[data-slot="subagent-transcript"]')?.textContent).toContain(text)) diff --git a/apps/desktop/src/app/chat/composer/status-stack/subagent-section.tsx b/apps/desktop/src/app/chat/composer/status-stack/subagent-section.tsx index 230cc7fdfb3a..e9c3410ab79a 100644 --- a/apps/desktop/src/app/chat/composer/status-stack/subagent-section.tsx +++ b/apps/desktop/src/app/chat/composer/status-stack/subagent-section.tsx @@ -79,19 +79,15 @@ export function SubagentSection({ sessionId }: SubagentSectionProps) { {detail && (
- {!detail.id.startsWith('delegation:') && ( - setDrafts(previous => ({ ...previous, [detail.id]: text }))} - subagentId={detail.id} - text={drafts[detail.id] ?? ''} - /> - )} + setDrafts(previous => ({ ...previous, [detail.id]: text }))} + subagentId={detail.id} + text={drafts[detail.id] ?? ''} + /> - {!detail.id.startsWith('delegation:') && ( - - )} +
)} diff --git a/apps/desktop/src/app/chat/composer/status-stack/use-subagent-snapshot.ts b/apps/desktop/src/app/chat/composer/status-stack/use-subagent-snapshot.ts index 8550121423a5..191f60f130ba 100644 --- a/apps/desktop/src/app/chat/composer/status-stack/use-subagent-snapshot.ts +++ b/apps/desktop/src/app/chat/composer/status-stack/use-subagent-snapshot.ts @@ -31,7 +31,7 @@ export function useSubagentSnapshot(sessionId: string | null) { const owner = JSON.stringify(knownOwnerForSession(sessionId)) try { - const snapshot = await requestForOwnedSession<{ subagents: SubagentPayload[]; delegations: SubagentPayload[] }>( + const snapshot = await requestForOwnedSession<{ subagents: SubagentPayload[] }>( sessionId, rejectUnownedSubagentRequest, 'subagent.list', @@ -44,7 +44,7 @@ export function useSubagentSnapshot(sessionId: string | null) { before === $subagentsBySession.get()[sessionId] && Array.isArray(snapshot.subagents) ) { - reconcileSubagentSnapshot(sessionId, snapshot.subagents, snapshot.delegations) + reconcileSubagentSnapshot(sessionId, snapshot.subagents) } failures = 0 diff --git a/apps/desktop/src/store/subagent-snapshot.test.ts b/apps/desktop/src/store/subagent-snapshot.test.ts index 4605fd74e01a..1533bc0b024c 100644 --- a/apps/desktop/src/store/subagent-snapshot.test.ts +++ b/apps/desktop/src/store/subagent-snapshot.test.ts @@ -1,32 +1,31 @@ import { afterEach, expect, it } from 'vitest' -import { $subagentsBySession, reconcileSubagentSnapshot } from './subagents' +import { $subagentsBySession, reconcileSubagentSnapshot, upsertSubagent } from './subagents' afterEach(() => $subagentsBySession.set({})) -it('projects unresolved running delegations once, replaces them with real workers, and retires completed units', () => { - const delegation = { - delegation_id: 'batch', - goal: 'Queued unit', - status: 'running', - dispatched_at: 1000, - subagent_ids: [] - } - reconcileSubagentSnapshot('owner', [], [delegation]) - expect($subagentsBySession.get().owner).toMatchObject([ - { id: 'delegation:batch', goal: 'Queued unit', startedAt: 1000000 } - ]) +it('hydrates last tool activity without claiming it is active and preserves stream history on refresh', () => { const child = { subagent_id: 'worker', - delegation_id: 'batch', goal: 'Actual worker', status: 'running', - started_at: 1001 + started_at: 1001, + last_tool: 'read_file' } - reconcileSubagentSnapshot('owner', [child], [delegation]) - expect($subagentsBySession.get().owner.map(row => row.id)).toEqual(['worker']) - const before = $subagentsBySession.get().owner - reconcileSubagentSnapshot('owner', [child], [delegation]) - expect($subagentsBySession.get().owner).toBe(before) - reconcileSubagentSnapshot('owner', [], [{ ...delegation, status: 'completed' }]) + + reconcileSubagentSnapshot('owner', [child]) + const first = $subagentsBySession.get().owner + expect(first[0].stream).toMatchObject([{ kind: 'tool', text: 'Read File' }]) + expect(first[0].currentTool).toBeUndefined() + expect(first[0].startedAt).toBe(child.started_at * 1000) + reconcileSubagentSnapshot('owner', [child]) + expect($subagentsBySession.get().owner).toBe(first) + upsertSubagent('other', { subagent_id: 'worker', goal: 'Other owner' }) + const other = $subagentsBySession.get().other + upsertSubagent('owner', { subagent_id: 'worker', text: 'New progress' }, false, 'subagent.progress') + const stream = $subagentsBySession.get().owner[0].stream + reconcileSubagentSnapshot('owner', [child]) + expect($subagentsBySession.get().owner[0].stream).toBe(stream) + reconcileSubagentSnapshot('owner', []) expect($subagentsBySession.get().owner).toEqual([]) + expect($subagentsBySession.get().other).toBe(other) }) diff --git a/apps/desktop/src/store/subagents.ts b/apps/desktop/src/store/subagents.ts index fc2103e41f03..4c43c1890e56 100644 --- a/apps/desktop/src/store/subagents.ts +++ b/apps/desktop/src/store/subagents.ts @@ -213,31 +213,13 @@ function toProgress(payload: SubagentPayload, prev: SubagentProgress | undefined } /** Reconcile a scoped, race-checked snapshot without replacing stream history. */ -export function reconcileSubagentSnapshot( - sid: string, - children: SubagentPayload[], - delegations: SubagentPayload[] = [] -) { - const represented = new Set(children.map(child => str(child.delegation_id)).filter(Boolean)) - - const payloads: SubagentPayload[] = [ - ...children, - ...delegations - .filter(unit => unit.status === 'running' && !represented.has(str(unit.delegation_id)) && str(unit.delegation_id)) - .map(unit => ({ - ...unit, - subagent_id: `delegation:${str(unit.delegation_id)}`, - started_at: unit.dispatched_at, - status: 'queued' - })) - ] - +export function reconcileSubagentSnapshot(sid: string, children: SubagentPayload[]) { const map = $subagentsBySession.get() const previous = map[sid] ?? [] - const ids = new Set(payloads.map(p => str(p.subagent_id)).filter(Boolean)) + const ids = new Set(children.map(p => str(p.subagent_id)).filter(Boolean)) const next = previous.filter(item => TERMINAL.has(item.status) || ids.has(item.id)) - for (const payload of payloads) { + for (const payload of children) { const id = str(payload.subagent_id) if (!id) { @@ -254,7 +236,12 @@ export function reconcileSubagentSnapshot( const projected = toProgress(payload, prev) projected.startedAt = (num(payload.started_at) ?? 0) * 1000 || prev?.startedAt || projected.startedAt projected.updatedAt = prev?.updatedAt ?? projected.startedAt - projected.currentTool = str(payload.current_tool) || undefined + + // A roster records the last tool, not a currently executing call. Seed cold + // activity only; a repeated snapshot must not append over newer live text. + if (!projected.stream.length && str(payload.last_tool)) { + projected.stream = [{ at: projected.updatedAt, kind: 'tool', text: formatTool(str(payload.last_tool)) }] + } if (index < 0) { next.push(projected) From 8a23f14df3e2aeb8e98c98857df43c9ff58ecdf0 Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Mon, 7 Sep 2026 23:38:16 -0700 Subject: [PATCH 158/227] fix: keep classic subagent dock compact and skin-colored --- hermes_cli/cli_subagent_monitor.py | 20 ++++++--- hermes_cli/skin_engine.py | 3 ++ tests/cli/test_subagent_dock_surface.py | 58 +++++++++++++++++++++++++ 3 files changed, 75 insertions(+), 6 deletions(-) create mode 100644 tests/cli/test_subagent_dock_surface.py diff --git a/hermes_cli/cli_subagent_monitor.py b/hermes_cli/cli_subagent_monitor.py index 94ac77ca8f73..bd3c18c66120 100644 --- a/hermes_cli/cli_subagent_monitor.py +++ b/hermes_cli/cli_subagent_monitor.py @@ -78,6 +78,7 @@ def control(self, action, message=None, *, target=None): def dock_text(self, *, columns, rows): if not self.entries: return '' + columns = max(0, columns - 2) count = min(len(self.entries), max(1, min(4, (rows - 10) // 3))) hidden = len(self.entries) - count heading = f' Subagents · {len(self.entries)} live · F6 expand' @@ -89,7 +90,7 @@ def dock_text(self, *, columns, rows): lines.append(_clip(f" ● {_clip(row.get('goal'), goal_width)} · {activity}", columns)) if hidden: lines.append(_clip(f' +{hidden} more · F6 all subagents', columns)) - return '\n'.join(lines) + return '\n'.join(' ' + line for line in lines) def read_tail(path): @@ -132,7 +133,7 @@ def roster_text(): text = f"{'❯' if selected else ' '} {row['elapsed']}s · {row.get('status') or 'starting'} · {row.get('goal') or row['subagent_id']}" if row.get('last_tool'): text += f" · last: {row['last_tool']}" - rows.append(('reverse' if selected else '', _clip(text, size.columns) + '\n')) + rows.append(('class:subagent-dock.selected' if selected else '', _clip(text, size.columns) + '\n')) return rows or [('', 'No live subagents. Results arrive in the conversation.')] def cursor(): @@ -155,7 +156,7 @@ def header(): title = f"Subagents · {len(monitor.entries)} live" if state['detail'] and row: title += f" · {row['subagent_id']} · {row.get('goal') or ''}" - return [('bold', _clip(title, app.output.get_size().columns))] + return [('class:subagent-dock.heading', _clip(title, app.output.get_size().columns))] def footer(): narrow = app.output.get_size().columns < 60 @@ -240,7 +241,7 @@ def close(event): ConditionalContainer(steer, filter=Condition(lambda: state['steering'])), Window(FormattedTextControl(lambda: _clip(state['notice'], app.output.get_size().columns)), height=1), Window(FormattedTextControl(footer), height=1), - ]), focused_element=roster) + ], style='class:subagent-dock'), focused_element=roster) def before_render(app): # Prompts arrive on worker threads; exit on the UI loop, including the # first frame if a prompt won the race with in_terminal() acquisition. @@ -249,6 +250,9 @@ def before_render(app): elif state['detail']: update_tail() + from prompt_toolkit.styles import Style + from hermes_cli.skin_engine import get_prompt_toolkit_style_overrides + kwargs.setdefault('style', Style.from_dict(get_prompt_toolkit_style_overrides())) app = Application(layout=layout, key_bindings=kb, full_screen=True, mouse_support=False, before_render=before_render, **kwargs) return app @@ -287,9 +291,13 @@ def install_dock(cli): def text(): size = get_app().output.get_size() - return [('class:subagent-border', monitor.dock_text(columns=size.columns, rows=size.rows))] + lines = monitor.dock_text(columns=size.columns, rows=size.rows).splitlines() + return [('class:subagent-dock.heading' if i == 0 else '', + line + ('\n' if i < len(lines) - 1 else '')) + for i, line in enumerate(lines)] cli._subagent_dock_widget = ConditionalContainer( - Window(FormattedTextControl(text), wrap_lines=False), + Window(FormattedTextControl(text), wrap_lines=False, dont_extend_height=True, + style='class:subagent-dock'), filter=Condition(lambda: bool(monitor.entries) and not modal_prompt_active(cli)), ) diff --git a/hermes_cli/skin_engine.py b/hermes_cli/skin_engine.py index f4a662855d0a..ca4515616b08 100644 --- a/hermes_cli/skin_engine.py +++ b/hermes_cli/skin_engine.py @@ -485,6 +485,9 @@ def get_active_goodbye(fallback: str = "Goodbye! ⚕") -> str: "status-bar-dim": "bg:{status_bg} {status_dim}", "status-bar-good": "bg:{status_bg} {status_good} bold", "status-bar-warn": "bg:{status_bg} {status_warn} bold", "status-bar-bad": "bg:{status_bg} {status_bad} bold", "status-bar-critical": "bg:{status_bg} {status_critical} bold", + "subagent-dock": "bg:{status_bg} {status_text}", + "subagent-dock.heading": "bg:{status_bg} {status_strong} bold", + "subagent-dock.selected": "bg:{menu_current_bg} {text} bold", "input-rule": "{input_rule}", "image-badge": "{label} bold", "completion-menu": "bg:{menu_bg} {text}", "completion-menu.completion": "bg:{menu_bg} {text}", "completion-menu.completion.current": "bg:{menu_current_bg} {title}", diff --git a/tests/cli/test_subagent_dock_surface.py b/tests/cli/test_subagent_dock_surface.py new file mode 100644 index 000000000000..a00be9f2dd63 --- /dev/null +++ b/tests/cli/test_subagent_dock_surface.py @@ -0,0 +1,58 @@ +"""A passive dock paints a bounded surface without acquiring the editor.""" +import asyncio +from types import SimpleNamespace + +import pytest + + +@pytest.mark.parametrize('columns,rows,skin', [(100, 30, 'default'), (80, 20, 'daylight')]) +def test_passive_dock_fills_rows_but_keeps_input_live(monkeypatch, columns, rows, skin): + from prompt_toolkit.application import Application + from prompt_toolkit.data_structures import Size + from prompt_toolkit.input import create_pipe_input + from prompt_toolkit.layout import HSplit, Layout + from prompt_toolkit.output import DummyOutput + from prompt_toolkit.styles import Style + from prompt_toolkit.widgets import TextArea + from hermes_cli.cli_subagent_monitor import install_dock + from hermes_cli import skin_engine + + monkeypatch.setattr(skin_engine, '_active_skin', skin_engine.load_skin(skin)) + cli = SimpleNamespace(agent=None) + install_dock(cli) + cli._subagent_monitor.entries = [dict(goal=f'Worker {i}', elapsed=12, + last_tool='terminal', status='running') for i in range(6)] + submitted = [] + editor = TextArea(height=1, multiline=False, + accept_handler=lambda buffer: submitted.append(buffer.text)) + output = DummyOutput() + output.get_size = lambda: Size(rows=rows, columns=columns) + + async def run(): + with create_pipe_input() as pipe: + app = Application(layout=Layout(HSplit([cli._subagent_dock_widget, editor]), + focused_element=editor), input=pipe, output=output, + style=Style.from_dict(skin_engine.get_prompt_toolkit_style_overrides())) + painted = asyncio.Event() + app.after_render += lambda _: painted.set() + task = asyncio.create_task(app.run_async()) + await asyncio.wait_for(painted.wait(), 3) + try: + screen = app.renderer._last_screen + dock_height = screen.visible_windows_to_write_positions[editor.window].ypos + assert 2 <= dock_height <= 6 + for y in range(dock_height): + assert screen.data_buffer[y][0].char == ' ' + for x in (0, columns // 2, columns - 1): + attrs = app._merged_style.get_attrs_for_style_str(screen.data_buffer[y][x].style) + assert attrs.bgcolor, 'the entire dock row needs a background' + assert attrs.color != attrs.bgcolor + painted.clear() + pipe.send_text('Follow up while workers run\r') + await asyncio.wait_for(painted.wait(), 3) + assert submitted == ['Follow up while workers run'] + assert app.layout.has_focus(editor) + finally: + app.exit() + await task + asyncio.run(run()) From a0fa236ed464b0d14e9ea91edad2507bb372e231 Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Mon, 7 Sep 2026 23:42:56 -0700 Subject: [PATCH 159/227] fix: retain worker tool activity in narrow classic monitor --- hermes_cli/cli_subagent_monitor.py | 12 ++++++++---- tests/cli/test_subagent_dock_surface.py | 23 +++++++++++++++++++++++ 2 files changed, 31 insertions(+), 4 deletions(-) diff --git a/hermes_cli/cli_subagent_monitor.py b/hermes_cli/cli_subagent_monitor.py index bd3c18c66120..c593a3f1ed86 100644 --- a/hermes_cli/cli_subagent_monitor.py +++ b/hermes_cli/cli_subagent_monitor.py @@ -130,10 +130,14 @@ def roster_text(): rows = [] for row in monitor.entries: selected = row['subagent_id'] == monitor.selected_id - text = f"{'❯' if selected else ' '} {row['elapsed']}s · {row.get('status') or 'starting'} · {row.get('goal') or row['subagent_id']}" - if row.get('last_tool'): - text += f" · last: {row['last_tool']}" - rows.append(('class:subagent-dock.selected' if selected else '', _clip(text, size.columns) + '\n')) + prefix = f"{row['elapsed']}s · {row.get('status') or 'starting'} · " + activity = f" · last: {row['last_tool']}" if row.get('last_tool') else '' + goal_width = max(0, size.columns - 2 - get_cwidth(prefix + activity)) + goal = _clip(row.get('goal') or row['subagent_id'], goal_width) + text = f"{'❯' if selected else ' '} " + _clip(prefix + goal + activity, max(0, size.columns - 2)) + # Pad selection in terminal cells, not codepoints (task names may be wide). + text += ' ' * max(0, size.columns - get_cwidth(text)) + rows.append(('class:subagent-dock.selected' if selected else '', text + '\n')) return rows or [('', 'No live subagents. Results arrive in the conversation.')] def cursor(): diff --git a/tests/cli/test_subagent_dock_surface.py b/tests/cli/test_subagent_dock_surface.py index a00be9f2dd63..c9633e705f86 100644 --- a/tests/cli/test_subagent_dock_surface.py +++ b/tests/cli/test_subagent_dock_surface.py @@ -56,3 +56,26 @@ async def run(): app.exit() await task asyncio.run(run()) + + +def test_expanded_roster_reserves_activity_before_long_task_names(): + from prompt_toolkit.data_structures import Size + from prompt_toolkit.input import create_pipe_input + from prompt_toolkit.output import DummyOutput + from prompt_toolkit.utils import get_cwidth + from hermes_cli.cli_subagent_monitor import SubagentMonitor, build_monitor_application + + monitor = SubagentMonitor(SimpleNamespace()) + monitor.entries = [dict(subagent_id=str(i), goal='Inspect 界 ' * 30, + elapsed=24, status='running', last_tool='terminal') for i in range(6)] + monitor.selected_id = '0' + output = DummyOutput() + with create_pipe_input() as pipe: + for columns, rows in [(100, 30), (80, 20)]: + output.get_size = lambda: Size(rows=rows, columns=columns) + app = build_monitor_application(monitor, input=pipe, output=output) + fragments = app.layout.current_control.text() + assert len(fragments) == len(monitor.entries) + for _, text in fragments: + assert '24s' in text and 'running' in text and 'last: terminal' in text + assert get_cwidth(text.rstrip('\n')) == columns From 4aad9b11b165cffc6b2a60aa067433c4e3af8069 Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Mon, 7 Sep 2026 23:29:41 -0700 Subject: [PATCH 160/227] fix: distinguish the live agent dock with a themed surface --- ui-tui/src/__tests__/agentsDock.test.tsx | 13 ++++++++++++- ui-tui/src/components/agentsPanel.tsx | 3 ++- 2 files changed, 14 insertions(+), 2 deletions(-) diff --git a/ui-tui/src/__tests__/agentsDock.test.tsx b/ui-tui/src/__tests__/agentsDock.test.tsx index 24208783a828..378c28855150 100644 --- a/ui-tui/src/__tests__/agentsDock.test.tsx +++ b/ui-tui/src/__tests__/agentsDock.test.tsx @@ -5,9 +5,11 @@ import React from 'react' import stripAnsi from 'strip-ansi' import { expect, it } from 'vitest' +import { renderToScreen } from '../../packages/hermes-ink/src/ink/render-to-screen.js' +import { cellAtIndex } from '../../packages/hermes-ink/src/ink/screen.js' import { AgentsPanelView } from '../components/agentsPanel.js' import { buildAgentRows, dockRowLimit } from '../lib/agentRows.js' -import { DEFAULT_THEME } from '../theme.js' +import { DARK_THEME, DEFAULT_THEME, LIGHT_THEME } from '../theme.js' import type { SubagentProgress } from '../types.js' const agent = (id: string): SubagentProgress => ({ @@ -56,6 +58,15 @@ it('bounds painted chrome by viewport while retaining true live count and activi } expect(dockRowLimit(14)).toBeLessThan(dockRowLimit(40)) + + for (const t of [DARK_THEME, LIGHT_THEME]) { + const rows = buildAgentRows(agents, [], 45000, dockRowLimit(20)) + const { screen, height } = renderToScreen(, 80) + // Ink marks fills in the low style bit: even blank edge cells must paint. + const filled = Array.from({ length: height * 80 }, (_, i) => cellAtIndex(screen, i).styleId & 1) + + expect(filled.every(Boolean)).toBe(true) + } }) it('deduplicates async batches and hides settled history without dropping live work', () => { diff --git a/ui-tui/src/components/agentsPanel.tsx b/ui-tui/src/components/agentsPanel.tsx index a338925d19c4..76e102a9a4bb 100644 --- a/ui-tui/src/components/agentsPanel.tsx +++ b/ui-tui/src/components/agentsPanel.tsx @@ -6,6 +6,7 @@ import { useAgentRoster } from '../app/agentRoster.js' import { patchOverlayState } from '../app/overlayStore.js' import { $uiState } from '../app/uiStore.js' import { type AgentRows, buildAgentRows, dockRowLimit } from '../lib/agentRows.js' +import { mix } from '../lib/color.js' import { statusGlyph } from '../lib/subagentGlyph.js' import { fmtDuration } from '../lib/subagentTree.js' import { compactPreview } from '../lib/text.js' @@ -15,7 +16,7 @@ export function AgentsPanelView({ cols, hidden, rows, running, t }: AgentRows & if (!running) {return null} return ( - + patchOverlayState({ agents: true })} wrap="truncate-end"> {`▾ ${running} live agents${hidden ? ` · +${hidden} more` : ''} · Ctrl+T expand`} From 3669e17a509445c744e7e066c468477f34dc3738 Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Mon, 7 Sep 2026 23:31:44 -0700 Subject: [PATCH 161/227] fix: keep the compact agent dock passive --- ui-tui/src/components/agentsPanel.tsx | 5 ++--- 1 file changed, 2 insertions(+), 3 deletions(-) diff --git a/ui-tui/src/components/agentsPanel.tsx b/ui-tui/src/components/agentsPanel.tsx index 76e102a9a4bb..0f0ab2f00234 100644 --- a/ui-tui/src/components/agentsPanel.tsx +++ b/ui-tui/src/components/agentsPanel.tsx @@ -3,7 +3,6 @@ import { useStore } from '@nanostores/react' import { useEffect, useState } from 'react' import { useAgentRoster } from '../app/agentRoster.js' -import { patchOverlayState } from '../app/overlayStore.js' import { $uiState } from '../app/uiStore.js' import { type AgentRows, buildAgentRows, dockRowLimit } from '../lib/agentRows.js' import { mix } from '../lib/color.js' @@ -17,11 +16,11 @@ export function AgentsPanelView({ cols, hidden, rows, running, t }: AgentRows & return ( - patchOverlayState({ agents: true })} wrap="truncate-end"> + {`▾ ${running} live agents${hidden ? ` · +${hidden} more` : ''} · Ctrl+T expand`} {rows.map(row => ( - patchOverlayState({ agents: true })}> + {statusGlyph(row.status, t).glyph} {compactPreview(row.goal, Math.max(8, cols - 18))} From b3e5c437fc30aaf412f60c129846504d1eb208a8 Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Mon, 7 Sep 2026 23:33:38 -0700 Subject: [PATCH 162/227] test: exercise dock fills with terminal color enabled --- ui-tui/src/__tests__/agentsDock.test.tsx | 8 +++++++- 1 file changed, 7 insertions(+), 1 deletion(-) diff --git a/ui-tui/src/__tests__/agentsDock.test.tsx b/ui-tui/src/__tests__/agentsDock.test.tsx index 378c28855150..c210bce63ccc 100644 --- a/ui-tui/src/__tests__/agentsDock.test.tsx +++ b/ui-tui/src/__tests__/agentsDock.test.tsx @@ -1,9 +1,10 @@ import { PassThrough } from 'node:stream' import { renderSync } from '@hermes/ink' +import chalk from 'chalk' import React from 'react' import stripAnsi from 'strip-ansi' -import { expect, it } from 'vitest' +import { afterEach, beforeEach, expect, it } from 'vitest' import { renderToScreen } from '../../packages/hermes-ink/src/ink/render-to-screen.js' import { cellAtIndex } from '../../packages/hermes-ink/src/ink/screen.js' @@ -12,6 +13,11 @@ import { buildAgentRows, dockRowLimit } from '../lib/agentRows.js' import { DARK_THEME, DEFAULT_THEME, LIGHT_THEME } from '../theme.js' import type { SubagentProgress } from '../types.js' +const colorLevel = chalk.level + +beforeEach(() => { chalk.level = 3 }) +afterEach(() => { chalk.level = colorLevel }) + const agent = (id: string): SubagentProgress => ({ id, goal: 'Investigate authentication handshake '.repeat(8), From fb5715c19f9ab091a95a210d1b72d4ad7e42550b Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Tue, 8 Sep 2026 00:08:04 -0700 Subject: [PATCH 163/227] fix: preserve subagent controls across attached session peers --- tests/tui_gateway/test_subagent_snapshot.py | 6 +++++- tools/delegate_tool_registry.py | 9 ++++++++- tui_gateway/methods_subagents.py | 4 ++-- tui_gateway/session_lifecycle.py | 2 +- 4 files changed, 16 insertions(+), 5 deletions(-) diff --git a/tests/tui_gateway/test_subagent_snapshot.py b/tests/tui_gateway/test_subagent_snapshot.py index 2d38cd9b0ac8..642db33fb417 100644 --- a/tests/tui_gateway/test_subagent_snapshot.py +++ b/tests/tui_gateway/test_subagent_snapshot.py @@ -181,8 +181,12 @@ def register(sid): popup = type(new)() with server._session_resume_lock, owner["history_lock"]: server._rebind_live_transport("ui-owner", owner, popup) + for peer in (new, popup): + assert {r["subagent_id"] for r in call("subagent.list", via=peer)["result"]["subagents"]} == {"before", "after"} + assert call("subagent.tail", via=peer, subagent_id="before")["result"]["text"] == "live child output" assert server._close_sessions_for_transport(popup) == (0, 0) - assert owner["transport"] is new + assert server._session_transport_contains(owner, new) + assert not server._session_transport_contains(owner, popup) assert {row["subagent_id"] for row in call("subagent.list", via=new)["result"]["subagents"]} == {"before", "after"} for sid in ("before", "after"): assert call("subagent.tail", via=new, subagent_id=sid)["result"]["text"] == "live child output" diff --git a/tools/delegate_tool_registry.py b/tools/delegate_tool_registry.py index dd2a225e12f4..f5beb16d2e99 100644 --- a/tools/delegate_tool_registry.py +++ b/tools/delegate_tool_registry.py @@ -109,6 +109,13 @@ def interrupt_subagent(subagent_id: str) -> bool: logger.debug("interrupt_subagent(%s) failed: %s", subagent_id, exc) return False +def _subagent_transport_matches(record, transport) -> bool: + from tui_gateway.transport import FanoutTransport + + bound = record.get("owner_transport") + return bound is transport or (isinstance(bound, FanoutTransport) and bound.contains(transport)) + + def steer_subagent( subagent_id: str, text: str, *, owner_session_id: Optional[str] = None, owner_transport: Any = None, owner_session_record: Any = None, @@ -130,7 +137,7 @@ def steer_subagent( if owner_session_id is not None and ( record.get("owner_session_id") != owner_session_id or owner_transport is None - or record.get("owner_transport") is not owner_transport + or not _subagent_transport_matches(record, owner_transport) or owner_session_record is None or record.get("owner_session_record") is not owner_session_record ): diff --git a/tui_gateway/methods_subagents.py b/tui_gateway/methods_subagents.py index 97b8eb02c4d4..c316e7f5a16b 100644 --- a/tui_gateway/methods_subagents.py +++ b/tui_gateway/methods_subagents.py @@ -17,12 +17,12 @@ def _owned_subagent_records(session_id, transport, owner): - from tools.delegate_tool_registry import _active_subagents, _active_subagents_lock + from tools.delegate_tool_registry import _active_subagents, _active_subagents_lock, _subagent_transport_matches with _active_subagents_lock: return [dict(r) for r in _active_subagents.values() if r.get("owner_session_id") == session_id - and r.get("owner_transport") is transport + and _subagent_transport_matches(r, transport) and r.get("owner_session_record") is owner] diff --git a/tui_gateway/session_lifecycle.py b/tui_gateway/session_lifecycle.py index 166abfe247e5..30fbc43166e6 100644 --- a/tui_gateway/session_lifecycle.py +++ b/tui_gateway/session_lifecycle.py @@ -476,7 +476,7 @@ def _rebind_live_transport(sid: str, session: dict, transport: Transport) -> Non if (record.get("owner_session_id") == sid and record.get("owner_session_record") is session and record.get("owner_transport") is not None): - record["owner_transport"] = transport + record["owner_transport"] = session["transport"] # Every transport that showed this session (pop-outs resume the same sid); on disconnect the last # viewer becomes the transport instead of the drop sentinel. session.setdefault("viewers", {})[transport] = time.time() From 7ee52894a7bdca054ce02d1a2e12cc7136c5e871 Mon Sep 17 00:00:00 2001 From: "hermes-seaeye[bot]" <307254004+hermes-seaeye[bot]@users.noreply.github.com> Date: Tue, 8 Sep 2026 10:11:57 +0000 Subject: [PATCH 164/227] fmt(js): `npm run fix` on merge (#105729) Co-authored-by: github-actions[bot] --- ui-tui/src/__tests__/agentControls.test.ts | 7 +- ui-tui/src/__tests__/agentRoster.test.ts | 14 +++- ui-tui/src/__tests__/agentsDock.test.tsx | 8 ++- ui-tui/src/__tests__/agentsHydration.test.tsx | 23 ++++--- ui-tui/src/app/agentRoster.ts | 4 +- ui-tui/src/app/useMainApp.ts | 12 ++-- ui-tui/src/components/agentControls.tsx | 27 +++++--- ui-tui/src/components/agentsOverlay.tsx | 65 +++++++++++++++---- ui-tui/src/components/agentsPanel.tsx | 15 ++++- ui-tui/src/lib/agentRows.ts | 4 +- 10 files changed, 132 insertions(+), 47 deletions(-) diff --git a/ui-tui/src/__tests__/agentControls.test.ts b/ui-tui/src/__tests__/agentControls.test.ts index 439b5dd0a74f..812a8638c3e8 100644 --- a/ui-tui/src/__tests__/agentControls.test.ts +++ b/ui-tui/src/__tests__/agentControls.test.ts @@ -24,11 +24,12 @@ it('reports queued acceptance rather than claiming delivery and preserves reject }) it('keeps every roster selection in the visible viewport including short terminals', () => { - for (const height of [10, 14, 24, 40]) - {for (const cursor of [0, 9, 19]) { + for (const height of [10, 14, 24, 40]) { + for (const cursor of [0, 9, 19]) { const view = rosterViewport(height, 20, cursor) expect(view.start).toBeLessThanOrEqual(cursor) expect(view.start + view.rows).toBeGreaterThan(cursor) expect(view.rows + (view.timelineRows ? view.timelineRows + 4 : 0) + 7).toBeLessThanOrEqual(height) - }} + } + } }) diff --git a/ui-tui/src/__tests__/agentRoster.test.ts b/ui-tui/src/__tests__/agentRoster.test.ts index a492a027428c..ae9f4c43d75c 100644 --- a/ui-tui/src/__tests__/agentRoster.test.ts +++ b/ui-tui/src/__tests__/agentRoster.test.ts @@ -6,9 +6,17 @@ import type { SubagentProgress } from '../types.js' it('never rolls event progress back when an older live snapshot arrives', () => { const event: SubagentProgress = { - id: 'child', goal: 'inspect', depth: 0, index: 0, parentId: null, - notes: [], thinking: [], tools: ['read_file'], taskCount: 1, - status: 'running', toolCount: 5 + id: 'child', + goal: 'inspect', + depth: 0, + index: 0, + parentId: null, + notes: [], + thinking: [], + tools: ['read_file'], + taskCount: 1, + status: 'running', + toolCount: 5 } const stale = { diff --git a/ui-tui/src/__tests__/agentsDock.test.tsx b/ui-tui/src/__tests__/agentsDock.test.tsx index c210bce63ccc..66ab13a3c76d 100644 --- a/ui-tui/src/__tests__/agentsDock.test.tsx +++ b/ui-tui/src/__tests__/agentsDock.test.tsx @@ -15,8 +15,12 @@ import type { SubagentProgress } from '../types.js' const colorLevel = chalk.level -beforeEach(() => { chalk.level = 3 }) -afterEach(() => { chalk.level = colorLevel }) +beforeEach(() => { + chalk.level = 3 +}) +afterEach(() => { + chalk.level = colorLevel +}) const agent = (id: string): SubagentProgress => ({ id, diff --git a/ui-tui/src/__tests__/agentsHydration.test.tsx b/ui-tui/src/__tests__/agentsHydration.test.tsx index 88ed88b27f53..adf56a79ab87 100644 --- a/ui-tui/src/__tests__/agentsHydration.test.tsx +++ b/ui-tui/src/__tests__/agentsHydration.test.tsx @@ -12,17 +12,24 @@ import { DEFAULT_THEME } from '../theme.js' it('does not undo an acknowledged pause when opening status resolves late', async () => { resetDelegationState() let resolveStatus!: (value: unknown) => void - const status = new Promise(resolve => { resolveStatus = resolve }) - const request = vi.fn(async (method: string) => method === 'delegation.status' ? status : { paused: true }) + const status = new Promise(resolve => { + resolveStatus = resolve + }) + const request = vi.fn(async (method: string) => (method === 'delegation.status' ? status : { paused: true })) const stdout = Object.assign(new PassThrough(), { columns: 80, rows: 16, isTTY: false }) const stdin = Object.assign(new PassThrough(), { isTTY: true, setRawMode: () => {}, ref: () => {}, unref: () => {} }) - const view = renderSync( {}} t={DEFAULT_THEME} />, { - stdout: stdout as unknown as NodeJS.WriteStream, - stdin: stdin as unknown as NodeJS.ReadStream, - stderr: new PassThrough() as unknown as NodeJS.WriteStream, - patchConsole: false - }) + const view = renderSync( + + {}} t={DEFAULT_THEME} /> + , + { + stdout: stdout as unknown as NodeJS.WriteStream, + stdin: stdin as unknown as NodeJS.ReadStream, + stderr: new PassThrough() as unknown as NodeJS.WriteStream, + patchConsole: false + } + ) try { await vi.waitFor(() => expect(request).toHaveBeenCalledWith('delegation.status', {})) diff --git a/ui-tui/src/app/agentRoster.ts b/ui-tui/src/app/agentRoster.ts index f30ba1c1d57f..43715d0bed05 100644 --- a/ui-tui/src/app/agentRoster.ts +++ b/ui-tui/src/app/agentRoster.ts @@ -14,7 +14,9 @@ export const $agentSnapshot = atom<{ sid: string | null; data: SubagentListRespo export function applyAgentSnapshot(sid: string | null, data: SubagentListResponse = EMPTY) { const previous = $agentSnapshot.get() - if (previous.sid !== sid || JSON.stringify(previous.data) !== JSON.stringify(data)) {$agentSnapshot.set({ sid, data })} + if (previous.sid !== sid || JSON.stringify(previous.data) !== JSON.stringify(data)) { + $agentSnapshot.set({ sid, data }) + } } export function mergeAgentRoster(events: SubagentProgress[], data: SubagentListResponse): SubagentProgress[] { diff --git a/ui-tui/src/app/useMainApp.ts b/ui-tui/src/app/useMainApp.ts index 23aabe976269..59c81427f453 100644 --- a/ui-tui/src/app/useMainApp.ts +++ b/ui-tui/src/app/useMainApp.ts @@ -596,11 +596,15 @@ export function useMainApp(gw: GatewayClient) { const refresh = () => { const sid = ui.sid - gw.request('subagent.list', { session_id: sid }).then(raw => { - const result = asRpcResult(raw) + gw.request('subagent.list', { session_id: sid }) + .then(raw => { + const result = asRpcResult(raw) - if (!stopped && result && getUiState().sid === sid) {applyAgentSnapshot(sid, result)} - }).catch(() => {}) + if (!stopped && result && getUiState().sid === sid) { + applyAgentSnapshot(sid, result) + } + }) + .catch(() => {}) gw.request('session.active_list', { current_session_id: getUiState().sid }) .then(raw => { const result = asRpcResult(raw) diff --git a/ui-tui/src/components/agentControls.tsx b/ui-tui/src/components/agentControls.tsx index e3523d027971..7235c6373edd 100644 --- a/ui-tui/src/components/agentControls.tsx +++ b/ui-tui/src/components/agentControls.tsx @@ -49,18 +49,24 @@ export function AgentSteerForm({ const [feedback, setFeedback] = useState('') const [pending, setPending] = useState(false) useInput((_ch, key) => { - if (key.escape && !pending) {onClose()} + if (key.escape && !pending) { + onClose() + } }) const submit = async () => { - if (!text.trim() || pending) {return} + if (!text.trim() || pending) { + return + } setPending(true) try { const result = await sendAgentSteer(gw, sid, id, text) setFeedback(result.message) - if (result.accepted) {setText('')} + if (result.accepted) { + setText('') + } } catch (error) { setFeedback(`Not queued: ${error instanceof Error ? error.message : String(error)}`) } finally { @@ -98,7 +104,9 @@ export function AgentLiveTail({ gw, sid, id, t }: ControlProps) { let pending = false const refresh = async () => { - if (pending) {return} + if (pending) { + return + } pending = true try { @@ -106,14 +114,17 @@ export function AgentLiveTail({ gw, sid, id, t }: ControlProps) { await gw.request('subagent.tail', { session_id: sid, subagent_id: id }) ) - if (active) - {setTail( + if (active) { + setTail( result?.available ? `${result.truncated ? '[last 16 KiB]\n' : ''}${result.text}` : 'Live transcript unavailable; child may have finished. Progress and output remain below.' - )} + ) + } } catch { - if (active) {setTail('Could not refresh live transcript.')} + if (active) { + setTail('Could not refresh live transcript.') + } } finally { pending = false } diff --git a/ui-tui/src/components/agentsOverlay.tsx b/ui-tui/src/components/agentsOverlay.tsx index be5505e33eef..e2d3beb6d1e5 100644 --- a/ui-tui/src/components/agentsOverlay.tsx +++ b/ui-tui/src/components/agentsOverlay.tsx @@ -628,7 +628,11 @@ export function AgentsOverlay({ gw, initialHistoryIndex = 0, onClose, t }: Agent const selected = rows[cursor] ?? null const cols = stdout?.columns ?? 80 - const { rows: rowsH, start: listWindowStart, timelineRows } = rosterViewport((stdout?.rows ?? 24) - (flash ? 1 : 0), rows.length, cursor) + const { + rows: rowsH, + start: listWindowStart, + timelineRows + } = rosterViewport((stdout?.rows ?? 24) - (flash ? 1 : 0), rows.length, cursor) // ── Effects ──────────────────────────────────────────────────────── @@ -673,11 +677,15 @@ export function AgentsOverlay({ gw, initialHistoryIndex = 0, onClose, t }: Agent let active = true gw.request('delegation.status', {}) .then(r => { - if (active && $delegationState.get() === initial) {applyDelegationStatus(asRpcResult(r))} + if (active && $delegationState.get() === initial) { + applyDelegationStatus(asRpcResult(r)) + } }) .catch(() => {}) - return () => { active = false } + return () => { + active = false + } }, [gw]) useEffect(() => { @@ -696,7 +704,8 @@ export function AgentsOverlay({ gw, initialHistoryIndex = 0, onClose, t }: Agent } } - const interrupt = (id: string) => gw.request('subagent.interrupt', { session_id: sid, subagent_id: id }) + const interrupt = (id: string) => + gw.request('subagent.interrupt', { session_id: sid, subagent_id: id }) const killOne = (id: string) => guardLive(() => { @@ -750,13 +759,21 @@ export function AgentsOverlay({ gw, initialHistoryIndex = 0, onClose, t }: Agent const scrollDetail = (dy: number) => detailScrollRef.current?.scrollBy(dy) useInput((ch, key) => { - if (mode === 'steer') {return} + if (mode === 'steer') { + return + } - if (key.ctrl && ch === 't') {return closeWithCleanup()} + if (key.ctrl && ch === 't') { + return closeWithCleanup() + } - if (ch === 'e' && selected && sid && !replayMode) {return setMode('steer')} + if (ch === 'e' && selected && sid && !replayMode) { + return setMode('steer') + } - if (ch === 't' && selected) {return setMode('tail')} + if (ch === 't' && selected) { + return setMode('tail') + } if (ch === 'q') { return closeWithCleanup() @@ -911,13 +928,17 @@ export function AgentsOverlay({ gw, initialHistoryIndex = 0, onClose, t }: Agent - {mode === 'steer' && selected && sid ? setMode('detail')} sid={sid} t={t} /> : rows.length === 0 ? ( + {mode === 'steer' && selected && sid ? ( + setMode('detail')} sid={sid} t={t} /> + ) : rows.length === 0 ? ( No subagents this turn. Trigger delegate_task to populate the tree. ) : mode === 'list' ? ( - {timelineRows > 0 ? : null} + {timelineRows > 0 ? ( + + ) : null} {rows.slice(listWindowStart, listWindowStart + rowsH).map((node, i) => ( @@ -935,9 +956,19 @@ export function AgentsOverlay({ gw, initialHistoryIndex = 0, onClose, t }: Agent ) : ( - + - {selected && mode === 'tail' && sid && !replayMode ? : selected ? : null} + {selected && mode === 'tail' && sid && !replayMode ? ( + + ) : selected ? ( + + ) : null} @@ -948,8 +979,14 @@ export function AgentsOverlay({ gw, initialHistoryIndex = 0, onClose, t }: Agent )} - Enter detail · e steer · t tail · x stop · Esc back - {flash ? {flash} : null} + + Enter detail · e steer · t tail · x stop · Esc back + + {flash ? ( + + {flash} + + ) : null} {mode === 'list' ? ( diff --git a/ui-tui/src/components/agentsPanel.tsx b/ui-tui/src/components/agentsPanel.tsx index 0f0ab2f00234..b503f3e50e09 100644 --- a/ui-tui/src/components/agentsPanel.tsx +++ b/ui-tui/src/components/agentsPanel.tsx @@ -12,10 +12,17 @@ import { compactPreview } from '../lib/text.js' import type { Theme } from '../theme.js' export function AgentsPanelView({ cols, hidden, rows, running, t }: AgentRows & { cols: number; t: Theme }) { - if (!running) {return null} + if (!running) { + return null + } return ( - + {`▾ ${running} live agents${hidden ? ` · +${hidden} more` : ''} · Ctrl+T expand`} @@ -40,7 +47,9 @@ export function LiveAgentsPanel({ cols }: { cols: number }) { const live = subagents.some(s => s.status === 'running' || s.status === 'queued') const [now, setNow] = useState(Date.now) useEffect(() => { - if (!live) {return} + if (!live) { + return + } const timer = setInterval(() => setNow(Date.now()), 1000) return () => clearInterval(timer) diff --git a/ui-tui/src/lib/agentRows.ts b/ui-tui/src/lib/agentRows.ts index 718769fb8132..7510393f7157 100644 --- a/ui-tui/src/lib/agentRows.ts +++ b/ui-tui/src/lib/agentRows.ts @@ -133,7 +133,9 @@ export const buildAgentRows = ( running += 1 active.push(row) } else { - if (RESULT_READY.has(s.status)) {done += 1} + if (RESULT_READY.has(s.status)) { + done += 1 + } } } From 04b88a4c0a7fe05ebe3bf053a103a39fe6638f11 Mon Sep 17 00:00:00 2001 From: "hermes-seaeye[bot]" <307254004+hermes-seaeye[bot]@users.noreply.github.com> Date: Tue, 8 Sep 2026 10:17:46 +0000 Subject: [PATCH 165/227] fmt(js): `npm run fix` on merge (#105732) Co-authored-by: github-actions[bot] --- ui-tui/src/__tests__/agentsHydration.test.tsx | 2 ++ ui-tui/src/components/agentControls.tsx | 2 ++ ui-tui/src/components/agentsOverlay.tsx | 1 + ui-tui/src/components/agentsPanel.tsx | 1 + 4 files changed, 6 insertions(+) diff --git a/ui-tui/src/__tests__/agentsHydration.test.tsx b/ui-tui/src/__tests__/agentsHydration.test.tsx index adf56a79ab87..d3b48453b8f8 100644 --- a/ui-tui/src/__tests__/agentsHydration.test.tsx +++ b/ui-tui/src/__tests__/agentsHydration.test.tsx @@ -12,9 +12,11 @@ import { DEFAULT_THEME } from '../theme.js' it('does not undo an acknowledged pause when opening status resolves late', async () => { resetDelegationState() let resolveStatus!: (value: unknown) => void + const status = new Promise(resolve => { resolveStatus = resolve }) + const request = vi.fn(async (method: string) => (method === 'delegation.status' ? status : { paused: true })) const stdout = Object.assign(new PassThrough(), { columns: 80, rows: 16, isTTY: false }) const stdin = Object.assign(new PassThrough(), { isTTY: true, setRawMode: () => {}, ref: () => {}, unref: () => {} }) diff --git a/ui-tui/src/components/agentControls.tsx b/ui-tui/src/components/agentControls.tsx index 7235c6373edd..d5f56db27a7a 100644 --- a/ui-tui/src/components/agentControls.tsx +++ b/ui-tui/src/components/agentControls.tsx @@ -58,6 +58,7 @@ export function AgentSteerForm({ if (!text.trim() || pending) { return } + setPending(true) try { @@ -107,6 +108,7 @@ export function AgentLiveTail({ gw, sid, id, t }: ControlProps) { if (pending) { return } + pending = true try { diff --git a/ui-tui/src/components/agentsOverlay.tsx b/ui-tui/src/components/agentsOverlay.tsx index e2d3beb6d1e5..5940979faeee 100644 --- a/ui-tui/src/components/agentsOverlay.tsx +++ b/ui-tui/src/components/agentsOverlay.tsx @@ -628,6 +628,7 @@ export function AgentsOverlay({ gw, initialHistoryIndex = 0, onClose, t }: Agent const selected = rows[cursor] ?? null const cols = stdout?.columns ?? 80 + const { rows: rowsH, start: listWindowStart, diff --git a/ui-tui/src/components/agentsPanel.tsx b/ui-tui/src/components/agentsPanel.tsx index b503f3e50e09..3c16f16e2801 100644 --- a/ui-tui/src/components/agentsPanel.tsx +++ b/ui-tui/src/components/agentsPanel.tsx @@ -50,6 +50,7 @@ export function LiveAgentsPanel({ cols }: { cols: number }) { if (!live) { return } + const timer = setInterval(() => setNow(Date.now()), 1000) return () => clearInterval(timer) From dec10c9177af76c8cddeddda21fbc67a599ac235 Mon Sep 17 00:00:00 2001 From: Lester Liang <153183032+lesterlxt@users.noreply.github.com> Date: Wed, 22 Jul 2026 12:56:04 +1000 Subject: [PATCH 166/227] fix(desktop): keep task status card opaque on scroll --- .../chat/composer/status-stack/index.test.tsx | 46 +++++++++++++++++++ .../app/chat/composer/status-stack/index.tsx | 17 ++++--- 2 files changed, 57 insertions(+), 6 deletions(-) create mode 100644 apps/desktop/src/app/chat/composer/status-stack/index.test.tsx diff --git a/apps/desktop/src/app/chat/composer/status-stack/index.test.tsx b/apps/desktop/src/app/chat/composer/status-stack/index.test.tsx new file mode 100644 index 000000000000..dda159077f1e --- /dev/null +++ b/apps/desktop/src/app/chat/composer/status-stack/index.test.tsx @@ -0,0 +1,46 @@ +import { cleanup, render, screen } from '@testing-library/react' +import { MemoryRouter } from 'react-router-dom' +import { afterEach, beforeEach, describe, expect, it, vi } from 'vitest' + +import { I18nProvider } from '@/i18n' +import { $threadScrolledUp, resetThreadScroll } from '@/store/thread-scroll' + +import { ComposerStatusStack } from './index' + +class TestResizeObserver { + disconnect() {} + observe() {} + unobserve() {} +} + +vi.stubGlobal('ResizeObserver', TestResizeObserver) + +describe('ComposerStatusStack scroll treatment', () => { + beforeEach(() => { + $threadScrolledUp.set(true) + }) + + afterEach(() => { + cleanup() + resetThreadScroll() + }) + + it('dims only the status content while keeping the dock card opaque', () => { + const view = render( + + + Queued task} sessionId={null} /> + + + ) + + const card = view.container.querySelector('[class*="bg-(--composer-fill)"]') + const dimmedContent = screen.getByText('Queued task').closest('.opacity-30') + + expect(card).not.toBeNull() + expect(card?.classList.contains('opacity-30')).toBe(false) + expect(dimmedContent).not.toBeNull() + expect(dimmedContent).not.toBe(card) + expect(card?.contains(dimmedContent)).toBe(true) + }) +}) diff --git a/apps/desktop/src/app/chat/composer/status-stack/index.tsx b/apps/desktop/src/app/chat/composer/status-stack/index.tsx index b58e2057026a..abf972787786 100644 --- a/apps/desktop/src/app/chat/composer/status-stack/index.tsx +++ b/apps/desktop/src/app/chat/composer/status-stack/index.tsx @@ -301,14 +301,19 @@ export function ComposerStatusStack({ onSubmit, queue, sessionId }: ComposerStat composerDockCard('top'), // Inset (mx-2) so the stack reads slightly narrower than the composer // surface below it — the original look. - 'mx-2 overflow-hidden rounded-b-none border-b border-b-transparent pt-0.5', - 'transition-opacity duration-200 ease-out', - scrolledUp ? 'opacity-30 group-hover/composer:opacity-100' : 'opacity-100' + 'mx-2 overflow-hidden rounded-b-none border-b border-b-transparent pt-0.5' )} > - {sections.map(section => ( -
{section.node}
- ))} +
+ {sections.map(section => ( +
{section.node}
+ ))} +
)} From 3dcf0deb29660c23cb5ecf4ed89a50dbe18e949f Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Tue, 8 Sep 2026 03:01:01 -0700 Subject: [PATCH 167/227] test(desktop): adapt status-card regression to current router --- apps/desktop/src/app/chat/composer/status-stack/index.test.tsx | 2 +- .../emails/153183032+lesterlxt@users.noreply.github.com | 2 ++ 2 files changed, 3 insertions(+), 1 deletion(-) create mode 100644 contributors/emails/153183032+lesterlxt@users.noreply.github.com diff --git a/apps/desktop/src/app/chat/composer/status-stack/index.test.tsx b/apps/desktop/src/app/chat/composer/status-stack/index.test.tsx index dda159077f1e..63d18f0af706 100644 --- a/apps/desktop/src/app/chat/composer/status-stack/index.test.tsx +++ b/apps/desktop/src/app/chat/composer/status-stack/index.test.tsx @@ -1,5 +1,5 @@ import { cleanup, render, screen } from '@testing-library/react' -import { MemoryRouter } from 'react-router-dom' +import { MemoryRouter } from 'react-router' import { afterEach, beforeEach, describe, expect, it, vi } from 'vitest' import { I18nProvider } from '@/i18n' diff --git a/contributors/emails/153183032+lesterlxt@users.noreply.github.com b/contributors/emails/153183032+lesterlxt@users.noreply.github.com new file mode 100644 index 000000000000..60a13cf232ba --- /dev/null +++ b/contributors/emails/153183032+lesterlxt@users.noreply.github.com @@ -0,0 +1,2 @@ +lesterlxt +# PR #69069 salvage From f3ccab1f6067d188cdffb35b7a6ad46292a2ca2f Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Tue, 8 Sep 2026 03:14:17 -0700 Subject: [PATCH 168/227] test: define passive one-line subagent dock contract --- tests/cli/test_subagent_dock_collapse.py | 92 ++++++++++++++++++++++++ 1 file changed, 92 insertions(+) create mode 100644 tests/cli/test_subagent_dock_collapse.py diff --git a/tests/cli/test_subagent_dock_collapse.py b/tests/cli/test_subagent_dock_collapse.py new file mode 100644 index 000000000000..4a054b1998fc --- /dev/null +++ b/tests/cli/test_subagent_dock_collapse.py @@ -0,0 +1,92 @@ +"""Collapsing the passive dock never takes space or focus from the editor.""" +import asyncio +from types import SimpleNamespace + + +def test_collapsed_dock_reserves_one_shaded_row_without_changing_editor(): + from prompt_toolkit.application import Application + from prompt_toolkit.data_structures import Size + from prompt_toolkit.document import Document + from prompt_toolkit.input import create_pipe_input + from prompt_toolkit.layout import HSplit, Layout + from prompt_toolkit.output import DummyOutput + from prompt_toolkit.styles import Style + from prompt_toolkit.widgets import TextArea + from hermes_cli import cli_subagent_monitor as dock + from hermes_cli.skin_engine import get_prompt_toolkit_style_overrides + + cli = SimpleNamespace(agent=None) + dock.install_dock(cli) + monitor = cli._subagent_monitor + monitor.entries = [dict(subagent_id=str(i), goal='Worker 界 ' * 20, + elapsed=12, last_tool='terminal', status='running') for i in range(6)] + submitted = [] + editor = TextArea(height=1, multiline=False, + accept_handler=lambda buffer: submitted.append(buffer.text)) + output = DummyOutput() + + async def run(): + with create_pipe_input() as pipe: + app = Application(layout=Layout(HSplit([cli._subagent_dock_widget, editor]), + focused_element=editor), input=pipe, output=output, + style=Style.from_dict(get_prompt_toolkit_style_overrides())) + cli._invalidate = app.invalidate + painted = asyncio.Event() + app.after_render += lambda _: painted.set() + output.get_size = lambda: Size(rows=30, columns=100) + task = asyncio.create_task(app.run_async()) + await asyncio.wait_for(painted.wait(), 3) + try: + def height(): + return app.renderer._last_screen.visible_windows_to_write_positions[editor.window].ypos + + preview_height = height() + assert preview_height > 1 + editor.buffer.document = Document('keep this draft', 5) + dock.toggle_dock(cli) + for columns, rows in [(100, 30), (80, 20), (40, 14)]: + output.get_size = lambda: Size(rows=rows, columns=columns) + painted.clear() + app.invalidate() + await asyncio.wait_for(painted.wait(), 3) + assert height() == 1 + assert app.layout.has_focus(editor) + assert (editor.text, editor.buffer.cursor_position) == ('keep this draft', 5) + screen = app.renderer._last_screen + for x in (0, columns // 2, columns - 1): + attrs = app._merged_style.get_attrs_for_style_str(screen.data_buffer[0][x].style) + assert attrs.bgcolor and attrs.color != attrs.bgcolor + painted.clear() + pipe.send_text('X\r') + await asyncio.wait_for(painted.wait(), 3) + assert submitted == ['keep Xthis draft'] + assert height() == 1 + output.get_size = lambda: Size(rows=30, columns=100) + painted.clear() + dock.toggle_dock(cli) + await asyncio.wait_for(painted.wait(), 3) + assert height() == preview_height + finally: + app.exit() + await task + asyncio.run(run()) + + +def test_collapsed_summary_prioritizes_live_count_and_controls_at_small_widths(): + from prompt_toolkit.utils import get_cwidth + from hermes_cli.cli_subagent_monitor import SubagentMonitor + + monitor = SubagentMonitor(SimpleNamespace()) + monitor.entries = [dict(goal='Inspect 界 ' * 40, elapsed=12, + last_tool='terminal', status='running') for _ in range(6)] + monitor.collapsed = True + for width in (100, 80, 40, 24, 10, 1): + text = monitor.dock_text(columns=width, rows=20) + assert len(text.splitlines()) == 1 + assert get_cwidth(text) <= width + if width >= 24: + assert '6 live' in text and 'F6' in text and 'F7' in text + if width >= 80: + assert 'last: terminal' in text and 'F6 expand' in text and 'F7 restore' in text + monitor.entries.clear() + assert monitor.dock_text(columns=80, rows=20) == '' From 34e539ebeb192a19db8537d3886d7faf6ac5b0bf Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Tue, 8 Sep 2026 03:16:37 -0700 Subject: [PATCH 169/227] feat: collapse Classic CLI subagent dock with F7 --- hermes_cli/cli_subagent_monitor.py | 26 +++++++++++++++++++++++++- hermes_cli/cli_tui_mixin.py | 4 +++- 2 files changed, 28 insertions(+), 2 deletions(-) diff --git a/hermes_cli/cli_subagent_monitor.py b/hermes_cli/cli_subagent_monitor.py index c593a3f1ed86..420c1512d514 100644 --- a/hermes_cli/cli_subagent_monitor.py +++ b/hermes_cli/cli_subagent_monitor.py @@ -29,6 +29,7 @@ def __init__(self, cli): self._last_poll = 0 self.app = None self.opening = False + self.collapsed = False @property def selected(self): @@ -78,10 +79,26 @@ def control(self, action, message=None, *, target=None): def dock_text(self, *, columns, rows): if not self.entries: return '' + if self.collapsed: + count = f'{len(self.entries)} live' + # Keep both controls before spending scarce cells on activity. + headings = ( + f'Subagents · {count} · F6 expand · F7 restore', + f'{count} · F6 expand · F7 restore', + f'{count} · F6 · F7', + count, + ) + width = max(0, columns - 1) + heading = next((text for text in headings if get_cwidth(text) <= width), count) + row = self.entries[0] + activity = f"last: {row['last_tool']}" if row.get('last_tool') else row.get('status') or 'starting' + if get_cwidth(heading + ' · ' + activity) <= width: + heading += ' · ' + activity + return _clip(' ' + heading, max(0, columns)) columns = max(0, columns - 2) count = min(len(self.entries), max(1, min(4, (rows - 10) // 3))) hidden = len(self.entries) - count - heading = f' Subagents · {len(self.entries)} live · F6 expand' + heading = f' Subagents · {len(self.entries)} live · F6 expand · F7 collapse' lines = [_clip(heading, columns)] for row in self.entries[:count]: activity = f"{row['elapsed']}s · " + (f"last: {row['last_tool']}" if row['last_tool'] else row.get('status') or 'starting') @@ -284,6 +301,13 @@ async def run(): asyncio.get_running_loop().create_task(run()) +def toggle_dock(cli): + monitor = getattr(cli, '_subagent_monitor', None) + if monitor is not None: + monitor.collapsed = not monitor.collapsed + cli._invalidate() + + def install_dock(cli): from prompt_toolkit.application import get_app from prompt_toolkit.layout import ConditionalContainer, Window diff --git a/hermes_cli/cli_tui_mixin.py b/hermes_cli/cli_tui_mixin.py index aca7359fda0a..603d0e4db650 100644 --- a/hermes_cli/cli_tui_mixin.py +++ b/hermes_cli/cli_tui_mixin.py @@ -1858,9 +1858,11 @@ def _tui_build_key_bindings(self): kb.add(Keys.BracketedPaste, eager=True)(self._tui_handle_paste) kb.add('c-v')(self._tui_handle_ctrl_v) kb.add('escape', 'v')(self._tui_handle_alt_v) - from hermes_cli.cli_subagent_monitor import modal_prompt_active, open_monitor + from hermes_cli.cli_subagent_monitor import modal_prompt_active, open_monitor, toggle_dock kb.add('f6', filter=Condition(lambda: not modal_prompt_active(self)))( lambda event: open_monitor(self)) + kb.add('f7', filter=Condition(lambda: not modal_prompt_active(self)))( + lambda event: toggle_dock(self)) return kb def _tui_bind_editor_and_stash(self, kb) -> None: From f7fe32bfde392e8a43b916f84836c55d2e27988e Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Tue, 8 Sep 2026 03:21:09 -0700 Subject: [PATCH 170/227] feat: add compact Ink agent dock and Enter live tail --- ui-tui/src/__tests__/agentsCompact.test.tsx | 70 +++++++++++++++++++++ ui-tui/src/app/agentRoster.ts | 3 + ui-tui/src/app/useInputHandlers.ts | 9 ++- ui-tui/src/components/agentsOverlay.tsx | 20 +++--- ui-tui/src/components/agentsPanel.tsx | 22 ++++--- ui-tui/src/components/textInput.tsx | 2 +- ui-tui/src/content/hotkeys.ts | 1 + ui-tui/src/types/hermes-ink.d.ts | 2 +- 8 files changed, 104 insertions(+), 25 deletions(-) create mode 100644 ui-tui/src/__tests__/agentsCompact.test.tsx diff --git a/ui-tui/src/__tests__/agentsCompact.test.tsx b/ui-tui/src/__tests__/agentsCompact.test.tsx new file mode 100644 index 000000000000..f4cb1350fe4d --- /dev/null +++ b/ui-tui/src/__tests__/agentsCompact.test.tsx @@ -0,0 +1,70 @@ +import { PassThrough } from 'node:stream' + +import { Box, renderSync } from '@hermes/ink' +import React from 'react' +import stripAnsi from 'strip-ansi' +import { expect, it, vi } from 'vitest' + +import { renderToScreen } from '../../packages/hermes-ink/src/ink/render-to-screen.js' +import { cellAtIndex } from '../../packages/hermes-ink/src/ink/screen.js' +import { applyAgentSnapshot } from '../app/agentRoster.js' +import { patchUiState, resetUiState } from '../app/uiStore.js' +import { AgentsOverlay } from '../components/agentsOverlay.js' +import { AgentsPanelView } from '../components/agentsPanel.js' +import type { GatewayClient } from '../gatewayClient.js' +import { buildAgentRows } from '../lib/agentRows.js' +import { DEFAULT_THEME } from '../theme.js' + +it('keeps collapsed live chrome to one row without losing count or restore controls', () => { + const rows = buildAgentRows([], [{ delegation_id: 'work', status: 'running', goal: 'Review the boundary' }], 1000) + + for (const cols of [78, 98]) { + const view = renderToScreen(, cols) + expect(view.height).toBe(1) + const text = Array.from({ length: cols }, (_, i) => cellAtIndex(view.screen, i).char).join('') + expect(text).toContain(`${rows.running} live agents`) + expect(text).toContain('Ctrl+T expand') + expect(text).toContain('F7 restore') + expect(renderToScreen(, cols).height).toBeGreaterThan(view.height) + } +}) + +it('opens the selected live transcript on Enter while details remain independently accessible', async () => { + patchUiState({ sid: 'owner' }) + applyAgentSnapshot('owner', { subagents: [{ subagent_id: 'child', goal: 'Inspect ownership', status: 'running' }], delegations: [] }) + + const request = vi.fn(async (method: string) => method === 'subagent.tail' + ? { available: true, text: 'CHILD_TOOL_OUTPUT', truncated: false } + : {}) + + const stdout = Object.assign(new PassThrough(), { columns: 80, rows: 20, isTTY: false }) + const stdin = Object.assign(new PassThrough(), { isTTY: true, setRawMode: () => {}, ref: () => {}, unref: () => {} }) + let output = '' + stdout.on('data', chunk => { output += stripAnsi(chunk.toString()) }) + + const view = renderSync( {}} t={DEFAULT_THEME} />, { + stdout: stdout as unknown as NodeJS.WriteStream, + stdin: stdin as unknown as NodeJS.ReadStream, + stderr: new PassThrough() as unknown as NodeJS.WriteStream, + patchConsole: false + }) + + try { + await vi.waitFor(() => expect(output).toContain('Inspect ownership')) + stdin.write('\r') + await vi.waitFor(() => expect(request).toHaveBeenCalledWith('subagent.tail', { session_id: 'owner', subagent_id: 'child' })) + await vi.waitFor(() => expect(output).toContain('CHILD_TOOL_OUTPUT')) + output = '' + stdin.write('d') + await vi.waitFor(() => expect(output).toContain('Inspect ownership')) + expect(output).not.toContain('CHILD_TOOL_OUTPUT') + output = '' + stdin.write('t') + await vi.waitFor(() => expect(output).toContain('CHILD_TOOL_OUTPUT')) + } finally { + view.unmount() + view.cleanup() + applyAgentSnapshot(null) + resetUiState() + } +}) diff --git a/ui-tui/src/app/agentRoster.ts b/ui-tui/src/app/agentRoster.ts index 43715d0bed05..3668f4eb86a4 100644 --- a/ui-tui/src/app/agentRoster.ts +++ b/ui-tui/src/app/agentRoster.ts @@ -8,6 +8,9 @@ import type { SubagentProgress } from '../types.js' import { useTurnSelector } from './turnStore.js' import { $uiState } from './uiStore.js' +// Session-local presentation only; never persisted to config. +export const $agentDockCollapsed = atom(false) + const EMPTY: SubagentListResponse = { subagents: [], delegations: [] } export const $agentSnapshot = atom<{ sid: string | null; data: SubagentListResponse }>({ sid: null, data: EMPTY }) diff --git a/ui-tui/src/app/useInputHandlers.ts b/ui-tui/src/app/useInputHandlers.ts index 0239befdb6a5..a5f258c18290 100644 --- a/ui-tui/src/app/useInputHandlers.ts +++ b/ui-tui/src/app/useInputHandlers.ts @@ -17,6 +17,7 @@ import { computePrecisionWheelStep, initPrecisionWheel } from '../lib/precisionW import { computeWheelStep, initWheelAccelForHost } from '../lib/wheelAccel.js' import { closeWidget, dispatchWidgetInput } from '../sdk/host.js' +import { $agentDockCollapsed } from './agentRoster.js' import { getInputSelection } from './inputSelectionStore.js' import { type GatewayRpc, @@ -366,7 +367,7 @@ export function useInputHandlers(ctx: InputHandlerContext): InputHandlerResult { // still the dedicated discard (pushes the draft to history so Up recalls it). const lastEscRef = useRef(0) - useInput((ch, key) => { + useInput((ch, key, event) => { const live = getUiState() if (key.escape) { @@ -631,6 +632,12 @@ export function useInputHandlers(ctx: InputHandlerContext): InputHandlerResult { // typed to run the command. Works mid-stream: picking a model writes the // session model (config.set), which the next turn reads while the in-flight // turn keeps streaming. + if (event.keypress.name === 'f7' && !key.ctrl && !key.meta && !key.shift && !key.super) { + $agentDockCollapsed.set(!$agentDockCollapsed.get()) + + return + } + if (isCtrl(key, ch, 't')) { return patchOverlayState({ agents: true, agentsInitialHistoryIndex: 0 }) } diff --git a/ui-tui/src/components/agentsOverlay.tsx b/ui-tui/src/components/agentsOverlay.tsx index 5940979faeee..e61c32f7587a 100644 --- a/ui-tui/src/components/agentsOverlay.tsx +++ b/ui-tui/src/components/agentsOverlay.tsx @@ -772,9 +772,9 @@ export function AgentsOverlay({ gw, initialHistoryIndex = 0, onClose, t }: Agent return setMode('steer') } - if (ch === 't' && selected) { - return setMode('tail') - } + if (ch === 't' && !key.ctrl && selected) {return setMode('tail')} + + if (ch === 'd' && !key.ctrl && selected) {return setMode('detail')} if (ch === 'q') { return closeWithCleanup() @@ -847,7 +847,7 @@ export function AgentsOverlay({ gw, initialHistoryIndex = 0, onClose, t }: Agent // List mode. if ((key.return || key.rightArrow || ch === 'l') && selected) { - return setMode('detail') + return setMode(key.return && !replayMode ? 'tail' : 'detail') } if (key.upArrow || ch === 'k' || key.wheelUp) { @@ -980,18 +980,12 @@ export function AgentsOverlay({ gw, initialHistoryIndex = 0, onClose, t }: Agent )} - - Enter detail · e steer · t tail · x stop · Esc back - - {flash ? ( - - {flash} - - ) : null} + {replayMode ? 'Enter/d detail' : 'Enter/t tail · d detail'} · e steer · x stop · Esc back + {flash ? {flash} : null} {mode === 'list' ? ( - ↑↓/jk move · g/G top/bottom · Enter/→ open detail{controlsHint} · s sort:{SORT_LABEL[sort]} · f filter: + ↑↓/jk move · g/G top/bottom · {replayMode ? 'Enter/→ detail' : 'Enter tail · d/→ detail'}{controlsHint} · s sort:{SORT_LABEL[sort]} · f filter: {FILTER_LABEL[filter]} {history.length > 0 ? ` · [ / ] history ${historyIndex}/${history.length}` : ''} {' · q close'} diff --git a/ui-tui/src/components/agentsPanel.tsx b/ui-tui/src/components/agentsPanel.tsx index 3c16f16e2801..7bf5fa6b3c7e 100644 --- a/ui-tui/src/components/agentsPanel.tsx +++ b/ui-tui/src/components/agentsPanel.tsx @@ -1,8 +1,8 @@ -import { Box, Text, useStdout } from '@hermes/ink' +import { Box, stringWidth, Text, useStdout } from '@hermes/ink' import { useStore } from '@nanostores/react' import { useEffect, useState } from 'react' -import { useAgentRoster } from '../app/agentRoster.js' +import { $agentDockCollapsed, useAgentRoster } from '../app/agentRoster.js' import { $uiState } from '../app/uiStore.js' import { type AgentRows, buildAgentRows, dockRowLimit } from '../lib/agentRows.js' import { mix } from '../lib/color.js' @@ -11,10 +11,13 @@ import { fmtDuration } from '../lib/subagentTree.js' import { compactPreview } from '../lib/text.js' import type { Theme } from '../theme.js' -export function AgentsPanelView({ cols, hidden, rows, running, t }: AgentRows & { cols: number; t: Theme }) { - if (!running) { - return null - } +export function AgentsPanelView({ collapsed = false, cols, hidden, rows, running, t }: AgentRows & { collapsed?: boolean; cols: number; t: Theme }) { + if (!running) {return null} + + const summary = `▸ ${running} live agents` + const hints = ' · Ctrl+T expand · F7 restore' + const activityWidth = cols - stringWidth(summary + hints) - 3 + const activity = rows[0]?.detail && activityWidth >= 12 ? ` · ${compactPreview(rows[0].detail, activityWidth)}` : '' return ( - {`▾ ${running} live agents${hidden ? ` · +${hidden} more` : ''} · Ctrl+T expand`} + {collapsed ? summary + activity + hints : `▾ ${running} live agents${hidden ? ` · +${hidden} more` : ''} · Ctrl+T expand · F7 collapse`} - {rows.map(row => ( + {!collapsed && rows.map(row => ( {statusGlyph(row.status, t).glyph} @@ -42,6 +45,7 @@ export function AgentsPanelView({ cols, hidden, rows, running, t }: AgentRows & export function LiveAgentsPanel({ cols }: { cols: number }) { const { theme } = useStore($uiState) + const collapsed = useStore($agentDockCollapsed) const { stdout } = useStdout() const subagents = useAgentRoster() const live = subagents.some(s => s.status === 'running' || s.status === 'queued') @@ -57,6 +61,6 @@ export function LiveAgentsPanel({ cols }: { cols: number }) { }, [live]) return ( - + ) } diff --git a/ui-tui/src/components/textInput.tsx b/ui-tui/src/components/textInput.tsx index 65b9a45c8d66..176b4fa5064b 100644 --- a/ui-tui/src/components/textInput.tsx +++ b/ui-tui/src/components/textInput.tsx @@ -1350,7 +1350,7 @@ export function TextInput({ // actually get voice toggled instead of a paste (Copilot round-7 // follow-up on #19835). The pass-through predicate is a no-op for // ordinary typing and plain paste when voice is unbound to 'v'. - if (shouldPassThroughToGlobalHandler(inp, k, voiceRecordKey)) { + if (event.keypress.name === 'f7' || shouldPassThroughToGlobalHandler(inp, k, voiceRecordKey)) { flushKeyBurst() return diff --git a/ui-tui/src/content/hotkeys.ts b/ui-tui/src/content/hotkeys.ts index 13cc2628f147..0fda3cd002eb 100644 --- a/ui-tui/src/content/hotkeys.ts +++ b/ui-tui/src/content/hotkeys.ts @@ -26,6 +26,7 @@ export const HOTKEYS: [string, string][] = [ ['↑/↓', 'completions / queue edit / history'], ['Ctrl+X', 'open live session switcher (deletes queued message while editing)'], ['Ctrl+T', 'expand live agents (keeps your draft)'], + ['F7', 'collapse / restore live agent preview'], ['Ctrl+O', 'open model picker (keeps your draft; applies to next turn mid-stream)'], [action + '+A/E', 'home / end of line'], [action + '+Z / ' + action + '+Y', 'undo / redo input edits'], diff --git a/ui-tui/src/types/hermes-ink.d.ts b/ui-tui/src/types/hermes-ink.d.ts index 10a105440514..7e7d9b543798 100644 --- a/ui-tui/src/types/hermes-ink.d.ts +++ b/ui-tui/src/types/hermes-ink.d.ts @@ -28,7 +28,7 @@ declare module '@hermes/ink' { export type InputEvent = { readonly input: string readonly key: Key - readonly keypress: { readonly isPasted?: boolean; readonly raw?: string } + readonly keypress: { readonly isPasted?: boolean; readonly name?: string; readonly raw?: string } } export type InputHandler = (input: string, key: Key, event: InputEvent) => void From 0a3b7fdce29af86b2a3f9bb2f4184099a5accf08 Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Tue, 8 Sep 2026 03:26:53 -0700 Subject: [PATCH 171/227] fix: preserve Ink composer cursor across agent monitors --- ui-tui/src/__tests__/agentsCompact.test.tsx | 13 +++++++++++++ ui-tui/src/components/agentsOverlay.tsx | 2 +- ui-tui/src/components/appLayout.tsx | 13 +++++++++---- ui-tui/src/components/textInput.tsx | 17 +++++++++++++++-- 4 files changed, 38 insertions(+), 7 deletions(-) diff --git a/ui-tui/src/__tests__/agentsCompact.test.tsx b/ui-tui/src/__tests__/agentsCompact.test.tsx index f4cb1350fe4d..0507f378055c 100644 --- a/ui-tui/src/__tests__/agentsCompact.test.tsx +++ b/ui-tui/src/__tests__/agentsCompact.test.tsx @@ -8,9 +8,11 @@ import { expect, it, vi } from 'vitest' import { renderToScreen } from '../../packages/hermes-ink/src/ink/render-to-screen.js' import { cellAtIndex } from '../../packages/hermes-ink/src/ink/screen.js' import { applyAgentSnapshot } from '../app/agentRoster.js' +import { getInputSelection } from '../app/inputSelectionStore.js' import { patchUiState, resetUiState } from '../app/uiStore.js' import { AgentsOverlay } from '../components/agentsOverlay.js' import { AgentsPanelView } from '../components/agentsPanel.js' +import { TextInput } from '../components/textInput.js' import type { GatewayClient } from '../gatewayClient.js' import { buildAgentRows } from '../lib/agentRows.js' import { DEFAULT_THEME } from '../theme.js' @@ -61,6 +63,17 @@ it('opens the selected live transcript on Enter while details remain independent output = '' stdin.write('t') await vi.waitFor(() => expect(output).toContain('CHILD_TOOL_OUTPUT')) + const cursorSnapshotRef = { current: null } + const onChange = vi.fn() + view.rerender() + await vi.waitFor(() => expect(getInputSelection()?.value).toBe('draft')) + stdin.write('\x1b[D') + await vi.waitFor(() => expect(getInputSelection()?.start).toBe(4)) + view.rerender() + view.rerender() + await vi.waitFor(() => expect(getInputSelection()?.start).toBe(4)) + stdin.write('!') + await vi.waitFor(() => expect(onChange).toHaveBeenCalledWith('draf!t')) } finally { view.unmount() view.cleanup() diff --git a/ui-tui/src/components/agentsOverlay.tsx b/ui-tui/src/components/agentsOverlay.tsx index e61c32f7587a..e0467dc69964 100644 --- a/ui-tui/src/components/agentsOverlay.tsx +++ b/ui-tui/src/components/agentsOverlay.tsx @@ -979,7 +979,7 @@ export function AgentsOverlay({ gw, initialHistoryIndex = 0, onClose, t }: Agent )} - + {replayMode ? 'Enter/d detail' : 'Enter/t tail · d detail'} · e steer · x stop · Esc back {flash ? {flash} : null} diff --git a/ui-tui/src/components/appLayout.tsx b/ui-tui/src/components/appLayout.tsx index 0bbe6c92bf91..79e3eb3faf95 100644 --- a/ui-tui/src/components/appLayout.tsx +++ b/ui-tui/src/components/appLayout.tsx @@ -3,7 +3,7 @@ import '../sdk/apps/index.js' import { AlternateScreen, Box, NoSelect, ScrollBox, Text } from '@hermes/ink' import { useStore } from '@nanostores/react' -import { Fragment, memo, useEffect, useMemo, useRef } from 'react' +import { Fragment, memo, type MutableRefObject, useEffect, useMemo, useRef } from 'react' import { useGateway } from '../app/gatewayContext.js' import type { AppLayoutProps } from '../app/interfaces.js' @@ -36,7 +36,7 @@ import { MessageLine } from './messageLine.js' import { PetKitty, PetSprite } from './petSprite.js' import { QueuedMessages } from './queuedMessages.js' import { LiveTodoPanel, StreamingAssistant } from './streamingAssistant.js' -import { TextInput, type TextInputMouseApi } from './textInput.js' +import { type InputCursorSnapshot, TextInput, type TextInputMouseApi } from './textInput.js' // Box geometry, kept here so the transcript's reservation math matches the // rendered overlay exactly. @@ -275,8 +275,9 @@ const TranscriptPane = memo(function TranscriptPane({ const ComposerPane = memo(function ComposerPane({ actions, composer, + cursorSnapshotRef, status -}: Pick) { +}: Pick & { cursorSnapshotRef: MutableRefObject }) { const ui = useStore($uiState) const isBlocked = useStore($isBlocked) const sh = (composer.inputBuf[0] ?? composer.input).startsWith('!') @@ -423,6 +424,7 @@ const ComposerPane = memo(function ComposerPane({ accentColor={ui.theme.color.accent} color={ui.theme.color.text} columns={inputColumns} + cursorSnapshotRef={cursorSnapshotRef} mouseApiRef={inputMouseRef} onChange={composer.updateInput} onPaste={composer.handleTextPaste} @@ -531,6 +533,9 @@ export const AppLayout = memo(function AppLayout({ const overlay = useStore($overlayState) const ui = useStore($uiState) + const cursorSnapshotRef = useRef(null) + useEffect(() => { cursorSnapshotRef.current = null }, [ui.sid]) + // Inline mode skips AlternateScreen so the host terminal's native // scrollback captures rows scrolled off the top; composer + progress // stay anchored via normal flex-column flow. @@ -572,7 +577,7 @@ export const AppLayout = memo(function AppLayout({ - + {SHOW_FPS && ( diff --git a/ui-tui/src/components/textInput.tsx b/ui-tui/src/components/textInput.tsx index 176b4fa5064b..29264205a1eb 100644 --- a/ui-tui/src/components/textInput.tsx +++ b/ui-tui/src/components/textInput.tsx @@ -782,6 +782,7 @@ export function TextInput({ onSubmit, mask, mouseApiRef, + cursorSnapshotRef, voiceRecordKey = DEFAULT_VOICE_RECORD_KEY, placeholder = '', placeholderColor, @@ -789,7 +790,7 @@ export function TextInput({ color, focus = true }: TextInputProps) { - const [cur, setCur] = useState(value.length) + const [cur, setCur] = useState(() => cursorSnapshotRef?.current?.value === value ? cursorSnapshotRef.current.cursor : value.length) const [sel, setSel] = useState(null) const fwdDel = useFwdDelete(focus) const termFocus = useTerminalFocus() @@ -922,7 +923,7 @@ export function TextInput({ const ownEcho = self.current && value === vRef.current self.current = false - if (ownEcho) { + if (ownEcho || value === vRef.current) { return } @@ -936,6 +937,12 @@ export function TextInput({ redo.current = [] }, [value]) + // The composer unmounts while full-screen monitors own input. Keep its + // insertion point with the shell, not with transient steer/secret inputs. + useEffect(() => () => { + if (cursorSnapshotRef) {cursorSnapshotRef.current = { cursor: curRef.current, value: vRef.current }} + }, [cursorSnapshotRef]) + useEffect(() => { if (!focus) { return @@ -1777,12 +1784,18 @@ export interface PasteEvent { value: string } +export interface InputCursorSnapshot { + cursor: number + value: string +} + interface TextInputProps { /** Hex/ansi256 tone for `/skill`, `@ref`, and `[[ token ]]` spans. */ accentColor?: string /** Hex color for typed text (theme text); terminal default when omitted. */ color?: string columns?: number + cursorSnapshotRef?: MutableRefObject focus?: boolean mask?: string mouseApiRef?: MutableRefObject From 3863d13440d61aba6c8abc36d2d763e8364d950e Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Tue, 8 Sep 2026 03:36:32 -0700 Subject: [PATCH 172/227] docs: describe one-line docks and consistent transcript controls --- ui-tui/src/__tests__/agentsCompact.test.tsx | 40 +++++++++++------ ui-tui/src/components/agentsOverlay.tsx | 21 ++++++--- ui-tui/src/components/agentsPanel.tsx | 45 +++++++++++++------ ui-tui/src/components/appLayout.tsx | 15 +++++-- ui-tui/src/components/textInput.tsx | 15 +++++-- website/docs/user-guide/cli.md | 1 + .../docs/user-guide/features/delegation.md | 4 +- website/docs/user-guide/tui.md | 3 +- .../current/user-guide/cli.md | 1 + .../current/user-guide/features/delegation.md | 4 +- .../current/user-guide/tui.md | 3 +- 11 files changed, 109 insertions(+), 43 deletions(-) diff --git a/ui-tui/src/__tests__/agentsCompact.test.tsx b/ui-tui/src/__tests__/agentsCompact.test.tsx index 0507f378055c..59e2cf2c0530 100644 --- a/ui-tui/src/__tests__/agentsCompact.test.tsx +++ b/ui-tui/src/__tests__/agentsCompact.test.tsx @@ -27,34 +27,48 @@ it('keeps collapsed live chrome to one row without losing count or restore contr expect(text).toContain(`${rows.running} live agents`) expect(text).toContain('Ctrl+T expand') expect(text).toContain('F7 restore') - expect(renderToScreen(, cols).height).toBeGreaterThan(view.height) + expect(renderToScreen(, cols).height).toBeGreaterThan( + view.height + ) } }) it('opens the selected live transcript on Enter while details remain independently accessible', async () => { patchUiState({ sid: 'owner' }) - applyAgentSnapshot('owner', { subagents: [{ subagent_id: 'child', goal: 'Inspect ownership', status: 'running' }], delegations: [] }) + applyAgentSnapshot('owner', { + subagents: [{ subagent_id: 'child', goal: 'Inspect ownership', status: 'running' }], + delegations: [] + }) - const request = vi.fn(async (method: string) => method === 'subagent.tail' - ? { available: true, text: 'CHILD_TOOL_OUTPUT', truncated: false } - : {}) + const request = vi.fn(async (method: string) => + method === 'subagent.tail' ? { available: true, text: 'CHILD_TOOL_OUTPUT', truncated: false } : {} + ) const stdout = Object.assign(new PassThrough(), { columns: 80, rows: 20, isTTY: false }) const stdin = Object.assign(new PassThrough(), { isTTY: true, setRawMode: () => {}, ref: () => {}, unref: () => {} }) let output = '' - stdout.on('data', chunk => { output += stripAnsi(chunk.toString()) }) - - const view = renderSync( {}} t={DEFAULT_THEME} />, { - stdout: stdout as unknown as NodeJS.WriteStream, - stdin: stdin as unknown as NodeJS.ReadStream, - stderr: new PassThrough() as unknown as NodeJS.WriteStream, - patchConsole: false + stdout.on('data', chunk => { + output += stripAnsi(chunk.toString()) }) + const view = renderSync( + + {}} t={DEFAULT_THEME} /> + , + { + stdout: stdout as unknown as NodeJS.WriteStream, + stdin: stdin as unknown as NodeJS.ReadStream, + stderr: new PassThrough() as unknown as NodeJS.WriteStream, + patchConsole: false + } + ) + try { await vi.waitFor(() => expect(output).toContain('Inspect ownership')) stdin.write('\r') - await vi.waitFor(() => expect(request).toHaveBeenCalledWith('subagent.tail', { session_id: 'owner', subagent_id: 'child' })) + await vi.waitFor(() => + expect(request).toHaveBeenCalledWith('subagent.tail', { session_id: 'owner', subagent_id: 'child' }) + ) await vi.waitFor(() => expect(output).toContain('CHILD_TOOL_OUTPUT')) output = '' stdin.write('d') diff --git a/ui-tui/src/components/agentsOverlay.tsx b/ui-tui/src/components/agentsOverlay.tsx index e0467dc69964..f7221bf4503a 100644 --- a/ui-tui/src/components/agentsOverlay.tsx +++ b/ui-tui/src/components/agentsOverlay.tsx @@ -772,9 +772,13 @@ export function AgentsOverlay({ gw, initialHistoryIndex = 0, onClose, t }: Agent return setMode('steer') } - if (ch === 't' && !key.ctrl && selected) {return setMode('tail')} + if (ch === 't' && !key.ctrl && selected) { + return setMode('tail') + } - if (ch === 'd' && !key.ctrl && selected) {return setMode('detail')} + if (ch === 'd' && !key.ctrl && selected) { + return setMode('detail') + } if (ch === 'q') { return closeWithCleanup() @@ -980,12 +984,19 @@ export function AgentsOverlay({ gw, initialHistoryIndex = 0, onClose, t }: Agent )} - {replayMode ? 'Enter/d detail' : 'Enter/t tail · d detail'} · e steer · x stop · Esc back - {flash ? {flash} : null} + + {replayMode ? 'Enter/d detail' : 'Enter/t tail · d detail'} · e steer · x stop · Esc back + + {flash ? ( + + {flash} + + ) : null} {mode === 'list' ? ( - ↑↓/jk move · g/G top/bottom · {replayMode ? 'Enter/→ detail' : 'Enter tail · d/→ detail'}{controlsHint} · s sort:{SORT_LABEL[sort]} · f filter: + ↑↓/jk move · g/G top/bottom · {replayMode ? 'Enter/→ detail' : 'Enter tail · d/→ detail'} + {controlsHint} · s sort:{SORT_LABEL[sort]} · f filter: {FILTER_LABEL[filter]} {history.length > 0 ? ` · [ / ] history ${historyIndex}/${history.length}` : ''} {' · q close'} diff --git a/ui-tui/src/components/agentsPanel.tsx b/ui-tui/src/components/agentsPanel.tsx index 7bf5fa6b3c7e..00f4e07aaf5a 100644 --- a/ui-tui/src/components/agentsPanel.tsx +++ b/ui-tui/src/components/agentsPanel.tsx @@ -11,8 +11,17 @@ import { fmtDuration } from '../lib/subagentTree.js' import { compactPreview } from '../lib/text.js' import type { Theme } from '../theme.js' -export function AgentsPanelView({ collapsed = false, cols, hidden, rows, running, t }: AgentRows & { collapsed?: boolean; cols: number; t: Theme }) { - if (!running) {return null} +export function AgentsPanelView({ + collapsed = false, + cols, + hidden, + rows, + running, + t +}: AgentRows & { collapsed?: boolean; cols: number; t: Theme }) { + if (!running) { + return null + } const summary = `▸ ${running} live agents` const hints = ' · Ctrl+T expand · F7 restore' @@ -27,18 +36,21 @@ export function AgentsPanelView({ collapsed = false, cols, hidden, rows, running width={cols} > - {collapsed ? summary + activity + hints : `▾ ${running} live agents${hidden ? ` · +${hidden} more` : ''} · Ctrl+T expand · F7 collapse`} + {collapsed + ? summary + activity + hints + : `▾ ${running} live agents${hidden ? ` · +${hidden} more` : ''} · Ctrl+T expand · F7 collapse`} - {!collapsed && rows.map(row => ( - - - {statusGlyph(row.status, t).glyph} - {compactPreview(row.goal, Math.max(8, cols - 18))} - {row.elapsedSeconds == null ? '' : fmtDuration(row.elapsedSeconds)} - - {` ↳ ${compactPreview(row.detail, cols - 4)}`} - - ))} + {!collapsed && + rows.map(row => ( + + + {statusGlyph(row.status, t).glyph} + {compactPreview(row.goal, Math.max(8, cols - 18))} + {row.elapsedSeconds == null ? '' : fmtDuration(row.elapsedSeconds)} + + {` ↳ ${compactPreview(row.detail, cols - 4)}`} + + ))} ) } @@ -61,6 +73,11 @@ export function LiveAgentsPanel({ cols }: { cols: number }) { }, [live]) return ( - + ) } diff --git a/ui-tui/src/components/appLayout.tsx b/ui-tui/src/components/appLayout.tsx index 79e3eb3faf95..09fbb075a276 100644 --- a/ui-tui/src/components/appLayout.tsx +++ b/ui-tui/src/components/appLayout.tsx @@ -277,7 +277,9 @@ const ComposerPane = memo(function ComposerPane({ composer, cursorSnapshotRef, status -}: Pick & { cursorSnapshotRef: MutableRefObject }) { +}: Pick & { + cursorSnapshotRef: MutableRefObject +}) { const ui = useStore($uiState) const isBlocked = useStore($isBlocked) const sh = (composer.inputBuf[0] ?? composer.input).startsWith('!') @@ -534,7 +536,9 @@ export const AppLayout = memo(function AppLayout({ const ui = useStore($uiState) const cursorSnapshotRef = useRef(null) - useEffect(() => { cursorSnapshotRef.current = null }, [ui.sid]) + useEffect(() => { + cursorSnapshotRef.current = null + }, [ui.sid]) // Inline mode skips AlternateScreen so the host terminal's native // scrollback captures rows scrolled off the top; composer + progress @@ -577,7 +581,12 @@ export const AppLayout = memo(function AppLayout({ - + {SHOW_FPS && ( diff --git a/ui-tui/src/components/textInput.tsx b/ui-tui/src/components/textInput.tsx index 29264205a1eb..a24100169ce4 100644 --- a/ui-tui/src/components/textInput.tsx +++ b/ui-tui/src/components/textInput.tsx @@ -790,7 +790,9 @@ export function TextInput({ color, focus = true }: TextInputProps) { - const [cur, setCur] = useState(() => cursorSnapshotRef?.current?.value === value ? cursorSnapshotRef.current.cursor : value.length) + const [cur, setCur] = useState(() => + cursorSnapshotRef?.current?.value === value ? cursorSnapshotRef.current.cursor : value.length + ) const [sel, setSel] = useState(null) const fwdDel = useFwdDelete(focus) const termFocus = useTerminalFocus() @@ -939,9 +941,14 @@ export function TextInput({ // The composer unmounts while full-screen monitors own input. Keep its // insertion point with the shell, not with transient steer/secret inputs. - useEffect(() => () => { - if (cursorSnapshotRef) {cursorSnapshotRef.current = { cursor: curRef.current, value: vRef.current }} - }, [cursorSnapshotRef]) + useEffect( + () => () => { + if (cursorSnapshotRef) { + cursorSnapshotRef.current = { cursor: curRef.current, value: vRef.current } + } + }, + [cursorSnapshotRef] + ) useEffect(() => { if (!focus) { diff --git a/website/docs/user-guide/cli.md b/website/docs/user-guide/cli.md index df128820e058..61ad61687e6f 100644 --- a/website/docs/user-guide/cli.md +++ b/website/docs/user-guide/cli.md @@ -187,6 +187,7 @@ When resuming a previous session (`hermes -c` or `hermes --resume `), a "Pre | `Ctrl+S` | **Stash the prompt.** Parks the current draft and clears the composer so you can send something else first. Press `Ctrl+S` again on an empty composer to bring the draft back (cursor at the end, attached images restored). Repeated presses build a stack rather than overwriting, so an earlier draft is never silently lost — with two or more stashed, `Ctrl+S` opens a browse panel (`↑`/`↓` to navigate, `Enter` to restore, `D` to discard, `Esc` or `Ctrl+S` to close). A `📌 N` badge in the status bar shows how many drafts are parked. Multi-line drafts round-trip exactly, including blank lines. The stash lives in memory for the session only — nothing is written to disk, since drafts often contain secrets. | | `Ctrl+C` | Interrupt agent (double-press within 2s to force exit) | | `F6` | Open the full-screen live subagent monitor without losing the composer draft. The live dock appears automatically above the status bar; arrows select a worker, `Enter` shows its recent log, `s` steers, and `x` requests stop with confirmation. See [Monitoring subagents](/user-guide/features/delegation#monitoring-running-subagents-agents). | +| `F7` | Toggle the live subagent dock between its multi-row preview and a single summary line without moving composer focus. | | `Ctrl+D` | Exit | | `Ctrl+Z` | Suspend Hermes to background (Unix only). Run `fg` in the shell to resume. | | `Tab` | Accept auto-suggestion (ghost text) or autocomplete slash commands | diff --git a/website/docs/user-guide/features/delegation.md b/website/docs/user-guide/features/delegation.md index ef97852ffdcd..3e0e19eb534c 100644 --- a/website/docs/user-guide/features/delegation.md +++ b/website/docs/user-guide/features/delegation.md @@ -392,11 +392,13 @@ The classic CLI, TUI, and Desktop automatically show live subagents above the co | Surface | Expand and inspect | Control a selected worker | |---|---|---| | Classic CLI | **F6** opens the full-screen live roster; arrows select, **Enter** opens the transcript tail, **PgUp/PgDn** scroll | **s** opens a separate steering input; **x**, then **y** requests stop | -| TUI | **Ctrl+T** or `/agents` opens the full-height tree; **Enter** opens detail; **t** opens the live transcript tail | **e** opens steering; **x** stops the selected worker; **X** stops its subtree | +| TUI | **Ctrl+T** or `/agents` opens the full-height tree; **Enter/t** opens the live transcript tail; **d** opens rich detail (archived/replay Enter still opens detail) | **e** opens steering; **x** stops the selected worker; **X** stops its subtree | | Desktop | Expand **Subagents** above the composer, then select a worker to inspect its activity and details | **Steer** queues guidance; **Stop** requests interruption for that worker | Closing the terminal monitor returns to your existing composer draft. Steering uses its own input and acknowledges **queued**, not delivery: the child consumes guidance at a checkpoint. Stop does not interrupt unrelated siblings. +Press **F7** in the Classic CLI or TUI composer to toggle the dock between its multi-row preview and a single shaded summary line. The summary retains the live count and expand/restore hints, adding activity when space permits. Typing and sending remain available; opening and closing the monitor preserves your draft and insertion point. This is a local presentation choice, not a saved config change. + The live transcript tail is a bounded recent excerpt, not an unlimited conversation browser. A child leaving the live registry leaves the dock; completion messages and the TUI/Desktop history views remain the place to review finished work. Latest activity is an observation, not a percentage-complete estimate. The classic CLI's `/agents` and `/tasks` commands still print a text summary; **F6** is the immediate interactive monitor, including while the parent is busy. See [TUI — Slash commands](/user-guide/tui#slash-commands). diff --git a/website/docs/user-guide/tui.md b/website/docs/user-guide/tui.md index e601b3a085ab..87bed9a239c2 100644 --- a/website/docs/user-guide/tui.md +++ b/website/docs/user-guide/tui.md @@ -102,7 +102,8 @@ The directory must contain `dist/entry.js`. Keybindings match the [Classic CLI](cli.md#keybindings) exactly. The only behavioral differences: -- **`Ctrl+T`** expands the automatic live-subagent dock into the full-height `/agents` roster. Select a worker to inspect details, press **`t`** for its recent transcript, **`e`** to steer, or **`x`** to stop it. The dock fits its row count to terminal height and preserves your composer draft. See [Monitoring subagents](/user-guide/features/delegation#monitoring-running-subagents-agents). +- **`Ctrl+T`** expands the automatic live-subagent dock into the full-height `/agents` roster. Select a worker and press **Enter** (or **`t`**) for its live transcript, **`d`** for rich details, **`e`** to steer, or **`x`** to stop it. The dock fits its row count to terminal height and preserves your composer draft. See [Monitoring subagents](/user-guide/features/delegation#monitoring-running-subagents-agents). +- **`F7`** toggles the live dock between its default preview and one summary line. This does not open the monitor or move composer focus; the choice lasts for this TUI process without changing config. - **Mouse drag** highlights text with a uniform selection background. - **`Cmd+V` / `Ctrl+V`** first tries normal text paste, then falls back to OSC52/native clipboard reads, and finally image attach when the clipboard or pasted payload resolves to an image. - **`/terminal-setup`** installs local VS Code / Cursor / Windsurf terminal bindings for better `Cmd+Enter` and undo/redo parity on macOS. diff --git a/website/i18n/zh-Hans/docusaurus-plugin-content-docs/current/user-guide/cli.md b/website/i18n/zh-Hans/docusaurus-plugin-content-docs/current/user-guide/cli.md index 6ff86a5b2421..6ff9cb00a068 100644 --- a/website/i18n/zh-Hans/docusaurus-plugin-content-docs/current/user-guide/cli.md +++ b/website/i18n/zh-Hans/docusaurus-plugin-content-docs/current/user-guide/cli.md @@ -103,6 +103,7 @@ hermes -w -z "Fix issue #123" # 在 worktree 中以单次查询模式运行 | `Ctrl+X Ctrl+E` | 外部编辑器的 Emacs 风格备用绑定（与 `Ctrl+G` 行为相同）。 | | `Ctrl+C` | 中断 agent（2 秒内双击强制退出） | | `F6` | 打开全屏实时子智能体监视器，保留输入草稿。方向键选择，`Enter` 查看近期日志，`s` 引导，`x` 请求停止并确认。 | +| `F7` | 将实时子智能体栏切换为单行摘要或恢复多行预览，不改变输入焦点。 | | `Ctrl+D` | 退出 | | `Ctrl+Z` | 将 Hermes 挂起到后台（仅 Unix）。在 shell 中运行 `fg` 恢复。 | | `Tab` | 接受自动建议（ghost text）或自动补全斜杠命令 | diff --git a/website/i18n/zh-Hans/docusaurus-plugin-content-docs/current/user-guide/features/delegation.md b/website/i18n/zh-Hans/docusaurus-plugin-content-docs/current/user-guide/features/delegation.md index 0971ef374a23..ee35ae5e817b 100644 --- a/website/i18n/zh-Hans/docusaurus-plugin-content-docs/current/user-guide/features/delegation.md +++ b/website/i18n/zh-Hans/docusaurus-plugin-content-docs/current/user-guide/features/delegation.md @@ -257,11 +257,13 @@ TUI 提供 `/agents` 浮层（别名 `/tasks`），将递归 `delegate_task` 扇 经典 CLI、TUI 和 Desktop 会在输入框上方自动显示正在运行的子智能体，包括总数、任务名称、已运行时间和最近活动。终端根据屏幕高度限制可见行数，并显示隐藏数量；Desktop 最多预览三个工作者。 - **经典 CLI：F6** 打开全屏实时列表；方向键选择，**Enter** 查看近期日志，**PgUp/PgDn** 滚动，**s** 输入引导，**x** 后按 **y** 确认停止。关闭后保留原有输入草稿。 -- **TUI：Ctrl+T** 或 `/agents` 打开完整树状列表；**t** 查看实时日志，**e** 输入引导，**x** 停止选中的工作者，**X** 停止其子树。 +- **TUI：Ctrl+T** 或 `/agents` 打开完整树状列表；**Enter/t** 查看实时日志，**d** 查看详情（历史回放中 Enter 仍打开详情），**e** 输入引导，**x** 停止选中的工作者，**X** 停止其子树。 - **Desktop：** 展开输入框上方的 **Subagents**，选择工作者查看详情并使用 **Steer** / **Stop**。 引导的“已排队”确认不代表子智能体已经读取；它会在检查点接收。日志预览只包含有大小限制的近期内容。工作者结束后离开实时列表，完成消息和已有历史视图仍可用于回顾。 +在经典 CLI 和 TUI 中按 **F7**，可将实时栏折叠为单行摘要，再按一次恢复多行预览。单行保留运行数量和展开/恢复提示，空间允许时显示活动。输入和发送不受影响；关闭监视器后保留草稿及光标位置。此选项不写入配置。 + 经典 CLI 的 `/agents` 和 `/tasks` 仍打印文本摘要；父智能体忙碌时可直接按 **F6** 打开交互式监视器。参见 [TUI — 斜杠命令](/user-guide/tui#slash-commands)。 ## 深度限制与嵌套编排 {#depth-limit-and-nested-orchestration} diff --git a/website/i18n/zh-Hans/docusaurus-plugin-content-docs/current/user-guide/tui.md b/website/i18n/zh-Hans/docusaurus-plugin-content-docs/current/user-guide/tui.md index c79b3d95d084..3c0becd661e3 100644 --- a/website/i18n/zh-Hans/docusaurus-plugin-content-docs/current/user-guide/tui.md +++ b/website/i18n/zh-Hans/docusaurus-plugin-content-docs/current/user-guide/tui.md @@ -85,7 +85,8 @@ hermes --tui 快捷键与 [Classic CLI](cli.md#keybindings) 完全一致。仅有以下行为差异： -- **`Ctrl+T`** — 将输入框上方的实时子智能体栏展开为完整 `/agents` 列表；**`t`** 查看近期日志，**`e`** 引导，**`x`** 停止选中的工作者。可见行数随终端高度调整，关闭后保留输入草稿。 +- **`Ctrl+T`** — 将输入框上方的实时子智能体栏展开为完整 `/agents` 列表；**Enter/t** 查看实时日志，**`d`** 查看详细信息，**`e`** 引导，**`x`** 停止选中的工作者。可见行数随终端高度调整，关闭后保留输入草稿。 +- **`F7`** — 在多行预览和单行摘要之间切换，保留输入焦点，不写入配置。 - **鼠标拖拽** — 以统一选区背景色高亮文本。 - **`Cmd+V` / `Ctrl+V`** — 优先尝试普通文本粘贴，然后回退到 OSC52/原生剪贴板读取，最后在剪贴板或粘贴内容解析为图片时进行图片附件操作。 - **`/terminal-setup`** — 安装本地 VS Code / Cursor / Windsurf 终端绑定，以在 macOS 上获得更好的 `Cmd+Enter` 和撤销/重做一致性。 From d4d4ecfae0c135b7bb52ff4f782ffef17bbc90c7 Mon Sep 17 00:00:00 2001 From: "hermes-seaeye[bot]" <307254004+hermes-seaeye[bot]@users.noreply.github.com> Date: Tue, 8 Sep 2026 10:47:44 +0000 Subject: [PATCH 173/227] fmt(js): `npm run fix` on merge (#105739) Co-authored-by: github-actions[bot] --- ui-tui/src/components/textInput.tsx | 1 + 1 file changed, 1 insertion(+) diff --git a/ui-tui/src/components/textInput.tsx b/ui-tui/src/components/textInput.tsx index a24100169ce4..60e69963cc60 100644 --- a/ui-tui/src/components/textInput.tsx +++ b/ui-tui/src/components/textInput.tsx @@ -793,6 +793,7 @@ export function TextInput({ const [cur, setCur] = useState(() => cursorSnapshotRef?.current?.value === value ? cursorSnapshotRef.current.cursor : value.length ) + const [sel, setSel] = useState(null) const fwdDel = useFwdDelete(focus) const termFocus = useTerminalFocus() From 9e048186a919f04a1f94a50e5a7360ed32c9df23 Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Tue, 8 Sep 2026 03:58:33 -0700 Subject: [PATCH 174/227] fix: make review workers recognizable in live subagent viewers --- agent/review_engine.py | 19 ++++++++------- tests/agent/test_review_engine.py | 24 ++++++++++++------- .../docs/user-guide/features/delegation.md | 2 ++ 3 files changed, 29 insertions(+), 16 deletions(-) diff --git a/agent/review_engine.py b/agent/review_engine.py index 1196bd162df5..18f500b6d5a9 100644 --- a/agent/review_engine.py +++ b/agent/review_engine.py @@ -97,8 +97,15 @@ def collect_parent_loaded_skills(parent_agent, messages: List[Dict[str, Any]], l def build_review_task(snapshot: List[Dict[str, str]], user_prompt: str = "", loaded_skills: Optional[List[str]] = None) -> tuple: - """Compose the reviewer subagent's (goal, context) pair.""" + """Compose a viewer-friendly goal and the complete reviewer briefing.""" + focus = " ".join(user_prompt.split()) + goal = f"Review: {focus}" if focus else "Review recent work" + if len(goal) > 80: + goal = goal[:79].rstrip() + "…" + # The goal is also the live worker label; keep the full instructions in context. lines = [ + _REVIEW_GOAL, + "", "You were spawned by the /review command. The following is an excerpt of the most recent conversation " "between the user and their primary agent. It is your starting evidence — the work to " "review is referenced in it.", @@ -126,7 +133,7 @@ def build_review_task(snapshot: List[Dict[str, str]], user_prompt: str = "", loa "the primary agent and its user. Be direct and specific; do not " "soften findings.", ] - return _REVIEW_GOAL, "\n".join(lines) + return goal, "\n".join(lines) def _load_review_credentials_cfg() -> Optional[Dict[str, Any]]: @@ -176,15 +183,11 @@ def start_review(parent_agent, messages: List[Dict[str, Any]], user_prompt: str def format_dispatch_note(result: Dict[str, Any], user_prompt: str = "") -> str: """Human-facing one-liner for a successful dispatch. Shared by surfaces.""" + if result.get("status") == "dispatched": + return "Review started. Results will return here." model = str(result.get("review_model") or "").strip() model_note = f" on {model}" if model else "" focus_note = f" (focus: {user_prompt.strip()})" if user_prompt.strip() else "" - if result.get("status") == "dispatched": - return ( - f"⚖ Review subagent dispatched{model_note}{focus_note} — it is " - f"investigating the last {DEFAULT_CONTEXT_MESSAGES} messages in " - f"the background and its full review will re-enter this conversation when it finishes." - ) # Synchronous fallback (channels that cannot route async completions). return ( f"⚖ Review completed synchronously{model_note}{focus_note} — " diff --git a/tests/agent/test_review_engine.py b/tests/agent/test_review_engine.py index 547e76f427ef..59748d16b131 100644 --- a/tests/agent/test_review_engine.py +++ b/tests/agent/test_review_engine.py @@ -97,15 +97,21 @@ def test_build_review_task_includes_excerpt_and_prompt(): {"role": "user", "text": "review my PR"}, {"role": "assistant", "text": "PR #99 opened"}, ] - goal, context = build_review_task(snap, "focus on security") - assert "reviewer" in goal.lower() + prompt = "focus on security\n" + "keep these instructions intact " * 20 + goal, context = build_review_task(snap, prompt) + assert goal.startswith("Review: focus on security ") + assert len(goal) <= 80 and "\n" not in goal + assert goal.endswith("…") + assert re_mod._REVIEW_GOAL in context assert "[USER]" in context and "[PRIMARY AGENT]" in context assert "PR #99 opened" in context - assert "focus on security" in context + assert prompt.strip() in context def test_build_review_task_without_prompt_has_no_instruction_block(): goal, context = build_review_task([{"role": "user", "text": "hi"}]) + assert goal == "Review recent work" + assert re_mod._REVIEW_GOAL in context assert "Additional review instructions" not in context @@ -243,7 +249,8 @@ def fake_build(**kw): # The reviewer briefing carries the conversation excerpt + user prompt. assert "PR #77 opened" in built["context"] assert "check the tests" in built["context"] - assert "reviewer" in built["goal"].lower() + assert built["goal"].startswith("Review: ") + assert re_mod._REVIEW_GOAL in built["context"] # The completion re-enters via the shared queue like any subagent. deadline = time.monotonic() + 5.0 @@ -504,12 +511,13 @@ def test_review_registered_in_every_aux_surface(): # --------------------------------------------------------------------------- def test_format_dispatch_note_dispatched(): + prompt = "security\n" * 100 note = format_dispatch_note( - {"status": "dispatched", "review_model": "opus"}, "security" + {"status": "dispatched", "review_model": "opus"}, prompt ) - assert "dispatched on opus" in note - assert "focus: security" in note - assert "re-enter" in note + assert "Review started" in note and "return here" in note + assert "security" not in note and "\n" not in note + assert len(note) < 80 def test_format_dispatch_note_sync_fallback(): diff --git a/website/docs/user-guide/features/delegation.md b/website/docs/user-guide/features/delegation.md index 3e0e19eb534c..7f0832a8dceb 100644 --- a/website/docs/user-guide/features/delegation.md +++ b/website/docs/user-guide/features/delegation.md @@ -262,6 +262,8 @@ What happens: The canonical flow: your main agent opens a PR, you type `/review`, and a second pair of eyes investigates it while you keep working; the review lands back in the chat addressed to the agent that created the PR. +Dispatch prints only “Review started. Results will return here.” The live subagent viewer identifies the worker as **Review: your focus** (or **Review recent work** for bare `/review`), with a shortened single-line label; the reviewer still receives your full instructions. In the classic CLI, the dock above the composer shows elapsed time and latest activity; **F6** opens the roster with its model, transcript, steering, and stop controls. The same review label appears in the TUI and Desktop subagent viewers. + ### Review model By default the reviewer runs on your main model. To pin a dedicated review model, set `auxiliary.review` in `config.yaml`: From ee84ccd8bd13d0025e98bb6be8ceb93303e3bdff Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Tue, 8 Sep 2026 04:06:32 -0700 Subject: [PATCH 175/227] test: align gateway review acknowledgement with shared formatter --- tests/gateway/test_review_command.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/tests/gateway/test_review_command.py b/tests/gateway/test_review_command.py index ddeb5b47b9be..4c59c9221a04 100644 --- a/tests/gateway/test_review_command.py +++ b/tests/gateway/test_review_command.py @@ -103,7 +103,7 @@ def fake_build(**kw): runner = _make_runner(agent) out = await runner._handle_review_command(_Event("check tests")) - assert "dispatched" in out + assert out == re_mod.format_dispatch_note({"status": "dispatched"}) assert "PR #5 opened" in built["context"] assert "check tests" in built["context"] From b4d7cf735d0458fb139228d0352650e0e41fdf4e Mon Sep 17 00:00:00 2001 From: Ben Barclay Date: Tue, 8 Sep 2026 21:25:14 +1000 Subject: [PATCH 176/227] fix(browser): real-profile auth mirror hangs forever on a locked destination MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit A `browser_exec` call could park a thread in `sqlite3_sleep` permanently while mirroring Chrome's auth DBs, holding the agent's turn open. The turn never reaches its `finally`, so no `session.info running=false` settle is emitted and the Desktop composer latches busy — every later message queues and never sends. Captured live: one thread stuck 24+ minutes across two dumps, turn accepted at 15:03 with no `tui turn finished` 46 minutes later. Root cause is the DESTINATION, not the source. `Connection.backup()` retries a busy destination internally and ignores the connection's busy timeout, so `sqlite3.connect(dst, timeout=5)` cannot bound it. A destination left locked by an earlier hung mirror therefore blocks the next mirror forever — and because the tool-level 420s timeout abandons the thread without interrupting a C-level lock wait, the lock is never released and every subsequent launch re-hangs the same way. Self-perpetuating. Two changes: - Back up into a fresh `.new` and `os.replace()` it into place. No other process can hold a file we just created, so there is nothing to contend on, and the swap stays atomic. Measured against a live Chrome with a deliberately locked destination: 0.0006s vs an indefinite hang. - Drop the `mode=ro` (no `immutable=1`) source fallback. 8e746668ba added `immutable=1` to fix exactly this hang but left `mode=ro` as a fallback, keeping the unbounded path one exception away; sqlite's busy timeout does not cover lock negotiation, so nothing bounds it. `immutable=1` is also the semantically correct mode — a committed snapshot of a file another process owns. The bounded plain-copy fallback is unchanged. Tests: three regressions, all mutation-checked (fail on base, pass here). The locked-destination test runs the copy on a worker with a join deadline so the unfixed behaviour fails fast instead of hanging the suite. 199 passing across the browser real-profile and CLI suites. --- hermes_cli/browser_connect.py | 47 +++++++--- tests/tools/test_browser_real_profile.py | 111 +++++++++++++++++++++++ 2 files changed, 144 insertions(+), 14 deletions(-) diff --git a/hermes_cli/browser_connect.py b/hermes_cli/browser_connect.py index c75f9890adae..a825601b1575 100644 --- a/hermes_cli/browser_connect.py +++ b/hermes_cli/browser_connect.py @@ -397,20 +397,39 @@ def _copy_auth_file(src_file: str, dst_file: str) -> bool: under a Windows write lock), falling through to a raw copy; failure only if BOTH fail.""" os.makedirs(os.path.dirname(dst_file), exist_ok=True) if os.path.basename(src_file) in _SQLITE_AUTH_DBS: - # With a live Chrome on macOS, mode=ro WITHOUT immutable=1 can hang connect/backup - # forever (blocked inside lock negotiation, so the busy-timeout never fires). - # immutable=1 reads instantly and is correct: we want a committed snapshot, not - # coordinated writes. A torn read raises → next mode, then the plain-copy fallback. - for uri in (f"file:{src_file}?mode=ro&immutable=1", f"file:{src_file}?mode=ro"): - try: - # Short busy timeout so a truly wedged DB fails fast rather than hanging. - with contextlib.closing(sqlite3.connect(uri, uri=True, timeout=5)) as source: - with contextlib.closing(sqlite3.connect(dst_file)) as out, out: - source.backup(out) - return True - except Exception as e: - logger.debug("real-profile: sqlite-backup of %s failed (%s); trying next mode", - src_file, e) + # With a live Chrome on macOS, mode=ro WITHOUT immutable=1 blocks inside lock + # negotiation and NEVER returns: sqlite's busy timeout does not cover lock + # negotiation, so `timeout=5` cannot rescue it. immutable=1 reads instantly and + # is what we want anyway — a committed snapshot of a file another process owns, + # not coordinated writes. So immutable=1 is the ONLY source mode we attempt. + # + # The DESTINATION is the other half, and the one that actually bit (#hang): + # backup() retries a busy destination internally FOREVER and ignores the + # connection's busy timeout, so a dst left locked by an earlier hung mirror + # parks the thread in sqlite3_sleep permanently, holding the agent's turn open. + # The tool-level timeout abandons that thread but cannot interrupt a C-level + # lock wait, so the lock is never released and every later launch re-hangs — + # the failure is self-perpetuating. + # + # Fix: never back up into the live destination. Write a FRESH temp file (no + # other process can hold it, so there is nothing to contend on) and move it + # into place atomically. Measured against a live Chrome with a deliberately + # locked destination: 0.01s here vs an indefinite hang writing in place. + tmp_dst = f"{dst_file}.new" + try: + with contextlib.suppress(OSError): + os.unlink(tmp_dst) + with contextlib.closing( + sqlite3.connect(f"file:{src_file}?mode=ro&immutable=1", uri=True, timeout=5)) as source: + with contextlib.closing(sqlite3.connect(tmp_dst, timeout=5)) as out, out: + source.backup(out) + os.replace(tmp_dst, dst_file) + return True + except Exception as e: + with contextlib.suppress(OSError): + os.unlink(tmp_dst) + logger.debug("real-profile: sqlite-backup of %s failed (%s); trying plain copy", + src_file, e) try: shutil.copy2(src_file, dst_file) return True diff --git a/tests/tools/test_browser_real_profile.py b/tests/tools/test_browser_real_profile.py index 82661749845c..37829cc4438a 100644 --- a/tests/tools/test_browser_real_profile.py +++ b/tests/tools/test_browser_real_profile.py @@ -1056,6 +1056,117 @@ def test_copy_auth_file_backs_up_db(self, tmp_path): assert bc._copy_auth_file(src, dst) is True assert sqlite3.connect(dst).execute("select count(*) from cookies").fetchone()[0] == 1 + def test_copy_auth_file_never_opens_the_unbounded_ro_mode(self, tmp_path, monkeypatch): + """`mode=ro` without immutable=1 must NEVER be attempted. + + On macOS with a live Chrome that URI blocks inside lock negotiation and never + returns — sqlite's busy timeout does not cover lock negotiation, so the + `timeout=5` argument cannot rescue it. The thread parks in sqlite3_sleep + holding the agent's turn open forever. + + Guard the URI SET rather than the hang itself: reproducing a real indefinite + block needs a live Chrome, but "we never ask for the mode that can hang" is + exactly the invariant and is deterministic. + """ + import hermes_cli.browser_connect as bc + import sqlite3 + + src = str(tmp_path / "Cookies") + con = sqlite3.connect(src) + con.execute("create table cookies(x)") + con.commit() + con.close() + # Make the immutable attempt fail so any surviving fallback is exercised. + dst = str(tmp_path / "out" / "Cookies") + seen: list[str] = [] + real_connect = sqlite3.connect + + def spy(target, *a, **kw): + if isinstance(target, str): + seen.append(target) + if "immutable=1" in target: + raise sqlite3.OperationalError("database is locked") + return real_connect(target, *a, **kw) + + monkeypatch.setattr(bc.sqlite3, "connect", spy) + bc._copy_auth_file(src, dst) + + unbounded = [u for u in seen if "mode=ro" in u and "immutable=1" not in u] + assert not unbounded, f"attempted the mode that can hang forever: {unbounded}" + + def test_copy_auth_file_never_backs_up_into_the_live_destination(self, tmp_path, monkeypatch): + """backup() must target a FRESH temp file, never the destination in place. + + ``backup()`` retries a busy destination internally forever and ignores the + connection's busy timeout, so writing straight into a destination that an + earlier hung mirror still holds parks the thread in sqlite3_sleep permanently. + Verified against a live Chrome with a deliberately locked destination: writing + in place hung indefinitely, temp-file + os.replace completed in 0.01s. + """ + import hermes_cli.browser_connect as bc + import sqlite3 + + src = str(tmp_path / "Cookies") + con = sqlite3.connect(src) + con.execute("create table cookies(x)") + con.execute("insert into cookies values(7)") + con.commit() + con.close() + + dst = str(tmp_path / "out" / "Cookies") + os.makedirs(os.path.dirname(dst), exist_ok=True) + # Hold the destination exclusively, exactly like a previous hung mirror. + holder = sqlite3.connect(dst) + holder.execute("create table t(x)") + holder.execute("begin exclusive") + # Run in a worker with a join deadline: on the unfixed code backup() retries the + # busy destination forever, so an unbounded assertion would hang the SUITE rather + # than fail it. The deadline turns that hang into a clean, fast failure. + import threading + + outcome: dict[str, object] = {} + + def _copy() -> None: + try: + outcome["ok"] = bc._copy_auth_file(src, dst) + except BaseException as exc: # noqa: BLE001 — surfaced via the assert below + outcome["exc"] = exc + + worker = threading.Thread(target=_copy, daemon=True) + worker.start() + worker.join(20) + try: + assert not worker.is_alive(), ( + "backup() blocked on a locked destination — it must write a temp file instead") + assert outcome.get("exc") is None, outcome.get("exc") + assert outcome.get("ok") is True + finally: + holder.rollback() + holder.close() + assert sqlite3.connect(dst).execute("select count(*) from cookies").fetchone()[0] == 1 + assert not os.path.exists(f"{dst}.new"), "temp file left behind" + + def test_copy_auth_file_cleans_up_temp_on_failure(self, tmp_path): + """A failed backup must not strand the .new temp beside the real file.""" + import hermes_cli.browser_connect as bc + + src = str(tmp_path / "Cookies") + open(src, "wb").write(b"not-a-sqlite-db") + dst = str(tmp_path / "out" / "Cookies") + assert bc._copy_auth_file(src, dst) is True # plain-copy fallback + assert not os.path.exists(f"{dst}.new") + + def test_copy_auth_file_falls_back_to_plain_copy_when_backup_fails(self, tmp_path, monkeypatch): + """Dropping the mode=ro fallback must not cost the plain-copy fallback.""" + import hermes_cli.browser_connect as bc + import sqlite3 + + src = str(tmp_path / "Cookies") + open(src, "wb").write(b"not-a-sqlite-db") + dst = str(tmp_path / "out" / "Cookies") + assert bc._copy_auth_file(src, dst) is True + assert open(dst, "rb").read() == b"not-a-sqlite-db" + def test_copy_auth_file_plain_for_non_db(self, tmp_path): import hermes_cli.browser_connect as bc src = str(tmp_path / "Preferences"); open(src, "w").write('{"k":1}') From 58d6d5223af140c7a46794842768666c20f19ca1 Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Tue, 8 Sep 2026 04:50:40 -0700 Subject: [PATCH 177/227] fix(browser): bound auth backups without replacing live SQLite files Preserve Ben Barclay's diagnosis and replace the staging-file approach with SQLite-coordinated writes and a five-second backup callback deadline. A main-file replacement can replay an abandoned destination WAL; immutable source reads can miss committed source WAL. Neither raw copy nor replacement is safe when the destination is locked. Refuse unavailable auth databases without raw-copy fallback, retaining the existing close-browser-and-retry flow. Keep two invariant tests for lock refusal/recovery and source-versus-destination WAL contents. Convert existing text masquerading as database fixtures into real SQLite fixtures. Related: #105754 Related: #96659 --- hermes_cli/browser_connect.py | 59 ++---- tests/tools/test_browser_real_profile.py | 207 +++++++++----------- website/docs/user-guide/features/browser.md | 9 +- 3 files changed, 115 insertions(+), 160 deletions(-) diff --git a/hermes_cli/browser_connect.py b/hermes_cli/browser_connect.py index a825601b1575..1166ab1ff035 100644 --- a/hermes_cli/browser_connect.py +++ b/hermes_cli/browser_connect.py @@ -393,47 +393,28 @@ def _secure_snapshot(path: str, *, contents: bool = False) -> None: def _copy_auth_file(src_file: str, dst_file: str) -> bool: - """Copy one auth file, lock-aware; True on success. SQLite DBs use the online-backup API (works - under a Windows write lock), falling through to a raw copy; failure only if BOTH fail.""" + """Copy auth state; refuse a DB that cannot be snapshotted consistently within five seconds.""" os.makedirs(os.path.dirname(dst_file), exist_ok=True) - if os.path.basename(src_file) in _SQLITE_AUTH_DBS: - # With a live Chrome on macOS, mode=ro WITHOUT immutable=1 blocks inside lock - # negotiation and NEVER returns: sqlite's busy timeout does not cover lock - # negotiation, so `timeout=5` cannot rescue it. immutable=1 reads instantly and - # is what we want anyway — a committed snapshot of a file another process owns, - # not coordinated writes. So immutable=1 is the ONLY source mode we attempt. - # - # The DESTINATION is the other half, and the one that actually bit (#hang): - # backup() retries a busy destination internally FOREVER and ignores the - # connection's busy timeout, so a dst left locked by an earlier hung mirror - # parks the thread in sqlite3_sleep permanently, holding the agent's turn open. - # The tool-level timeout abandons that thread but cannot interrupt a C-level - # lock wait, so the lock is never released and every later launch re-hangs — - # the failure is self-perpetuating. - # - # Fix: never back up into the live destination. Write a FRESH temp file (no - # other process can hold it, so there is nothing to contend on) and move it - # into place atomically. Measured against a live Chrome with a deliberately - # locked destination: 0.01s here vs an indefinite hang writing in place. - tmp_dst = f"{dst_file}.new" - try: - with contextlib.suppress(OSError): - os.unlink(tmp_dst) - with contextlib.closing( - sqlite3.connect(f"file:{src_file}?mode=ro&immutable=1", uri=True, timeout=5)) as source: - with contextlib.closing(sqlite3.connect(tmp_dst, timeout=5)) as out, out: - source.backup(out) - os.replace(tmp_dst, dst_file) - return True - except Exception as e: - with contextlib.suppress(OSError): - os.unlink(tmp_dst) - logger.debug("real-profile: sqlite-backup of %s failed (%s); trying plain copy", - src_file, e) try: - shutil.copy2(src_file, dst_file) + if os.path.basename(src_file) in _SQLITE_AUTH_DBS: + deadline = time.monotonic() + 5.0 + + def check_deadline(_status: int, _remaining: int, _total: int) -> None: + if _status != sqlite3.SQLITE_DONE and time.monotonic() >= deadline: + raise TimeoutError("auth database backup exceeded five seconds") + + # SQLite must coordinate both ends: immutable ignores committed source WAL, + # while replacing only the destination file can replay its abandoned WAL. + # Connection busy timeouts do not bound backup's retry loop; its callback does. + with contextlib.closing(sqlite3.connect( + Path(src_file).resolve().as_uri() + "?mode=ro", uri=True, timeout=0.0)) as source: + with contextlib.closing(sqlite3.connect(dst_file, timeout=0.0)) as out: + source.backup(out, pages=256, progress=check_deadline, sleep=0.1) + else: + shutil.copy2(src_file, dst_file) return True - except OSError as e: + except (OSError, sqlite3.Error) as e: + # A raw DB copy can lose committed WAL or overwrite a locked destination. logger.debug("real-profile: could not copy %s: %s", src_file, e) return False @@ -680,7 +661,7 @@ def snapshot_real_profile(browser: str, src: str | None = None) -> tuple[str | N failed_dbs = _mirror_profile_auth(src, dst, source_profile) if failed_dbs: # even online-backup failed: never launch a silently signed-out session return None, (f"could not read the '{browser}' profile's login data ({failed_dbs} " - f"database(s) locked). Close {browser} and retry, or turn " + f"database(s) unavailable). Close {browser} and retry, or turn " "browser.use_real_profile off.") # Never carry live-instance leftovers into the copy. for leftover in ("SingletonLock", "SingletonSocket", "SingletonCookie"): diff --git a/tests/tools/test_browser_real_profile.py b/tests/tools/test_browser_real_profile.py index 37829cc4438a..2fde6b6803ce 100644 --- a/tests/tools/test_browser_real_profile.py +++ b/tests/tools/test_browser_real_profile.py @@ -20,6 +20,19 @@ from tools import browser_tool_install as bt_install +def _auth_db(path, value=None): + """Store/read a marker in a real auth DB so snapshot fixtures exercise SQLite.""" + import sqlite3 + from contextlib import closing + + with closing(sqlite3.connect(path)) as conn, conn: + if value is not None: + conn.execute("create table if not exists marker(value)") + conn.execute("delete from marker") + conn.execute("insert into marker values(?)", (value,)) + return conn.execute("select value from marker").fetchone()[0] + + class TestRealProfileResolvers: def test_data_dir_windows(self): import hermes_cli.browser_connect as bc @@ -88,9 +101,9 @@ def _make_profile(self, root): (root / "Code Cache" / "js").mkdir(parents=True) (root / "Crashpad").mkdir() (root / "Local State").write_text('{"os_crypt": {}}') - (root / "Default" / "Cookies").write_text("sqlite-cookies") - (root / "Default" / "Network" / "Cookies").write_text("sqlite-net-cookies") - (root / "Default" / "Login Data").write_text("sqlite-logins") + _auth_db((root / "Default" / "Cookies"), "sqlite-cookies") + _auth_db((root / "Default" / "Network" / "Cookies"), "sqlite-net-cookies") + _auth_db((root / "Default" / "Login Data"), "sqlite-logins") (root / "Default" / "Preferences").write_text("{}") (root / "Default" / "Cache" / "Cache_Data" / "big").write_text("x" * 1000) (root / "Code Cache" / "js" / "blob").write_text("y" * 1000) @@ -109,7 +122,7 @@ def test_fresh_snapshot_copies_auth_and_skips_caches(self, tmp_path, monkeypatch assert err is None assert dst == str(home / "browser-profile" / "chrome") # Auth files present - assert (home / "browser-profile" / "chrome" / "Default" / "Cookies").read_text() == "sqlite-cookies" + assert _auth_db((home / "browser-profile" / "chrome" / "Default" / "Cookies")) == "sqlite-cookies" assert (home / "browser-profile" / "chrome" / "Default" / "Network" / "Cookies").exists() assert (home / "browser-profile" / "chrome" / "Default" / "Login Data").exists() assert (home / "browser-profile" / "chrome" / "Local State").exists() @@ -129,13 +142,13 @@ def test_existing_snapshot_refreshes_auth_files_only(self, tmp_path, monkeypatch assert err is None # Simulate: user logs into a new site in their own browser, and the # copy has drifted state that must survive (History not in refresh set). - (src / "Default" / "Cookies").write_text("sqlite-cookies-v2") + _auth_db((src / "Default" / "Cookies"), "sqlite-cookies-v2") copy_history = home / "browser-profile" / "chrome" / "Default" / "History" copy_history.write_text("agent-session-history") dst2, err2 = bc.snapshot_real_profile("chrome", src=str(src)) assert err2 is None and dst2 == dst - assert (home / "browser-profile" / "chrome" / "Default" / "Cookies").read_text() == "sqlite-cookies-v2" + assert _auth_db((home / "browser-profile" / "chrome" / "Default" / "Cookies")) == "sqlite-cookies-v2" assert copy_history.read_text() == "agent-session-history" def test_missing_source_fails_closed(self, tmp_path, monkeypatch): @@ -654,7 +667,7 @@ def test_snapshot_dir_secured(self, tmp_path, monkeypatch): src = tmp_path / "real" / "Default" src.mkdir(parents=True) (tmp_path / "real" / "Local State").write_text("{}") - (src / "Cookies").write_text("db") + _auth_db((src / "Cookies"), "db") monkeypatch.setattr(bc, "get_hermes_home", lambda: tmp_path / "hh") called = {"paths": []} with patch("hermes_cli.config._secure_dir", @@ -678,9 +691,9 @@ def _multi_profile(self, root): '{"profile": {"last_used": "Profile 6"}}' ) # Default is signed OUT (tracking cookies only); Profile 6 has the session. - (root / "Default" / "Cookies").write_text("default-tracking-only") - (root / "Profile 6" / "Cookies").write_text("PROFILE6-SESSION-AUTH") - (root / "Profile 6" / "Login Data").write_text("profile6-logins") + _auth_db((root / "Default" / "Cookies"), "default-tracking-only") + _auth_db((root / "Profile 6" / "Cookies"), "PROFILE6-SESSION-AUTH") + _auth_db((root / "Profile 6" / "Login Data"), "profile6-logins") (root / "Profile 6" / "Preferences").write_text("{}") return root @@ -692,9 +705,9 @@ def test_last_used_profile_lands_in_copy_default(self, tmp_path, monkeypatch): dst, err = bc.snapshot_real_profile("chrome", src=str(src)) assert err is None # The copy's Default must carry PROFILE 6's session, not Default's. - got = (home / "browser-profile" / "chrome" / "Default" / "Cookies").read_text() + got = _auth_db((home / "browser-profile" / "chrome" / "Default" / "Cookies")) assert got == "PROFILE6-SESSION-AUTH" - assert (home / "browser-profile" / "chrome" / "Default" / "Login Data").read_text() == "profile6-logins" + assert _auth_db((home / "browser-profile" / "chrome" / "Default" / "Login Data")) == "profile6-logins" def test_last_used_falls_back_to_default(self, tmp_path): import hermes_cli.browser_connect as bc @@ -716,10 +729,10 @@ def test_refresh_remirrors_last_used(self, tmp_path, monkeypatch): home = tmp_path / "hh" monkeypatch.setattr(bc, "get_hermes_home", lambda: home) bc.snapshot_real_profile("chrome", src=str(src)) # fresh - (src / "Profile 6" / "Cookies").write_text("PROFILE6-REFRESHED") + _auth_db((src / "Profile 6" / "Cookies"), "PROFILE6-REFRESHED") dst, err = bc.snapshot_real_profile("chrome", src=str(src)) # refresh assert err is None - assert (home / "browser-profile" / "chrome" / "Default" / "Cookies").read_text() == "PROFILE6-REFRESHED" + assert _auth_db((home / "browser-profile" / "chrome" / "Default" / "Cookies")) == "PROFILE6-REFRESHED" # ── Bug 3: private-URL sidecar must NOT carry the real profile ── def test_sidecar_never_uses_real_profile(self): @@ -799,8 +812,8 @@ def _multi(self, root): for prof in ("Default", "Profile 6"): (root / prof / "Network").mkdir(parents=True) (root / "Local State").write_text('{"profile": {"last_used": "Profile 6"}}') - (root / "Default" / "Cookies").write_text("default-signed-out") - (root / "Profile 6" / "Cookies").write_text("PROFILE6-SESSION") + _auth_db((root / "Default" / "Cookies"), "default-signed-out") + _auth_db((root / "Profile 6" / "Cookies"), "PROFILE6-SESSION") (root / "Profile 6" / "Preferences").write_text("{}") return root @@ -826,7 +839,7 @@ def test_torn_copy_is_redone_not_overlaid(self, tmp_path, monkeypatch): d, err = bc.snapshot_real_profile("chrome", src=str(src)) assert err is None # Rebuilt from the active profile, not treated as populated. - assert (home / "browser-profile" / "chrome" / "Default" / "Cookies").read_text() == "PROFILE6-SESSION" + assert _auth_db((home / "browser-profile" / "chrome" / "Default" / "Cookies")) == "PROFILE6-SESSION" assert os.path.isfile(os.path.join(dst, bc._SNAPSHOT_DONE_MARKER)) # ── ④ only the active profile is copied, never the others ── @@ -835,14 +848,14 @@ def test_only_active_profile_copied(self, tmp_path, monkeypatch): src = self._multi(tmp_path / "real") # Add a non-active profile with its own cookies — must NOT be copied. (src / "Profile 3").mkdir() - (src / "Profile 3" / "Cookies").write_text("PROFILE3-SHOULD-NOT-COPY") + _auth_db((src / "Profile 3" / "Cookies"), "PROFILE3-SHOULD-NOT-COPY") home = tmp_path / "hh" monkeypatch.setattr(bc, "get_hermes_home", lambda: home) dst, err = bc.snapshot_real_profile("chrome", src=str(src)) assert err is None copy = home / "browser-profile" / "chrome" # Active profile (Profile 6) landed in Default; other profiles absent. - assert (copy / "Default" / "Cookies").read_text() == "PROFILE6-SESSION" + assert _auth_db((copy / "Default" / "Cookies")) == "PROFILE6-SESSION" assert not (copy / "Profile 3").exists() assert not (copy / "Profile 6").exists() @@ -1056,116 +1069,72 @@ def test_copy_auth_file_backs_up_db(self, tmp_path): assert bc._copy_auth_file(src, dst) is True assert sqlite3.connect(dst).execute("select count(*) from cookies").fetchone()[0] == 1 - def test_copy_auth_file_never_opens_the_unbounded_ro_mode(self, tmp_path, monkeypatch): - """`mode=ro` without immutable=1 must NEVER be attempted. - - On macOS with a live Chrome that URI blocks inside lock negotiation and never - returns — sqlite's busy timeout does not cover lock negotiation, so the - `timeout=5` argument cannot rescue it. The thread parks in sqlite3_sleep - holding the agent's turn open forever. - - Guard the URI SET rather than the hang itself: reproducing a real indefinite - block needs a live Chrome, but "we never ask for the mode that can hang" is - exactly the invariant and is deterministic. - """ - import hermes_cli.browser_connect as bc + @pytest.mark.parametrize("locked", ["source", "destination"]) + def test_copy_auth_file_bounds_locks_without_overwriting(self, tmp_path, locked): import sqlite3 - - src = str(tmp_path / "Cookies") - con = sqlite3.connect(src) - con.execute("create table cookies(x)") - con.commit() - con.close() - # Make the immutable attempt fail so any surviving fallback is exercised. - dst = str(tmp_path / "out" / "Cookies") - seen: list[str] = [] - real_connect = sqlite3.connect - - def spy(target, *a, **kw): - if isinstance(target, str): - seen.append(target) - if "immutable=1" in target: - raise sqlite3.OperationalError("database is locked") - return real_connect(target, *a, **kw) - - monkeypatch.setattr(bc.sqlite3, "connect", spy) - bc._copy_auth_file(src, dst) - - unbounded = [u for u in seen if "mode=ro" in u and "immutable=1" not in u] - assert not unbounded, f"attempted the mode that can hang forever: {unbounded}" - - def test_copy_auth_file_never_backs_up_into_the_live_destination(self, tmp_path, monkeypatch): - """backup() must target a FRESH temp file, never the destination in place. - - ``backup()`` retries a busy destination internally forever and ignores the - connection's busy timeout, so writing straight into a destination that an - earlier hung mirror still holds parks the thread in sqlite3_sleep permanently. - Verified against a live Chrome with a deliberately locked destination: writing - in place hung indefinitely, temp-file + os.replace completed in 0.01s. - """ + import subprocess + import sys import hermes_cli.browser_connect as bc - import sqlite3 - - src = str(tmp_path / "Cookies") - con = sqlite3.connect(src) - con.execute("create table cookies(x)") - con.execute("insert into cookies values(7)") - con.commit() - con.close() - dst = str(tmp_path / "out" / "Cookies") - os.makedirs(os.path.dirname(dst), exist_ok=True) - # Hold the destination exclusively, exactly like a previous hung mirror. - holder = sqlite3.connect(dst) - holder.execute("create table t(x)") + src, dst = tmp_path / "Cookies", tmp_path / "out" / "Cookies" + dst.parent.mkdir() + for path, value in ((src, 7), (dst, 99)): + with sqlite3.connect(path) as conn: + conn.execute("create table cookies(x)") + conn.execute("insert into cookies values(?)", (value,)) + conn.close() + holder = sqlite3.connect(src if locked == "source" else dst) holder.execute("begin exclusive") - # Run in a worker with a join deadline: on the unfixed code backup() retries the - # busy destination forever, so an unbounded assertion would hang the SUITE rather - # than fail it. The deadline turns that hang into a clean, fast failure. - import threading - - outcome: dict[str, object] = {} - - def _copy() -> None: - try: - outcome["ok"] = bc._copy_auth_file(src, dst) - except BaseException as exc: # noqa: BLE001 — surfaced via the assert below - outcome["exc"] = exc - - worker = threading.Thread(target=_copy, daemon=True) - worker.start() - worker.join(20) try: - assert not worker.is_alive(), ( - "backup() blocked on a locked destination — it must write a temp file instead") - assert outcome.get("exc") is None, outcome.get("exc") - assert outcome.get("ok") is True + result = subprocess.run( + [sys.executable, "-c", + "from hermes_cli.browser_connect import _copy_auth_file; " + "import sys; print(_copy_auth_file(sys.argv[1], sys.argv[2]))", + str(src), str(dst)], + capture_output=True, text=True, timeout=15, stdin=subprocess.DEVNULL) + assert result.returncode == 0, result.stderr + assert result.stdout.strip() == "False" finally: holder.rollback() holder.close() - assert sqlite3.connect(dst).execute("select count(*) from cookies").fetchone()[0] == 1 - assert not os.path.exists(f"{dst}.new"), "temp file left behind" - - def test_copy_auth_file_cleans_up_temp_on_failure(self, tmp_path): - """A failed backup must not strand the .new temp beside the real file.""" - import hermes_cli.browser_connect as bc - - src = str(tmp_path / "Cookies") - open(src, "wb").write(b"not-a-sqlite-db") - dst = str(tmp_path / "out" / "Cookies") - assert bc._copy_auth_file(src, dst) is True # plain-copy fallback - assert not os.path.exists(f"{dst}.new") - - def test_copy_auth_file_falls_back_to_plain_copy_when_backup_fails(self, tmp_path, monkeypatch): - """Dropping the mode=ro fallback must not cost the plain-copy fallback.""" - import hermes_cli.browser_connect as bc + with sqlite3.connect(dst) as conn: + assert conn.execute("select x from cookies").fetchall() == [(99,)] + conn.close() + assert bc._copy_auth_file(str(src), str(dst)) is True + with sqlite3.connect(dst) as conn: + assert conn.execute("select x from cookies").fetchall() == [(7,)] + conn.close() + + def test_copy_auth_file_preserves_source_wal_not_abandoned_destination_wal(self, tmp_path): import sqlite3 + import subprocess + import sys + import hermes_cli.browser_connect as bc - src = str(tmp_path / "Cookies") - open(src, "wb").write(b"not-a-sqlite-db") - dst = str(tmp_path / "out" / "Cookies") - assert bc._copy_auth_file(src, dst) is True - assert open(dst, "rb").read() == b"not-a-sqlite-db" + src, dst = tmp_path / "Cookies", tmp_path / "out" / "Cookies" + dst.parent.mkdir() + source = sqlite3.connect(src) + source.execute("create table cookies(x)") + source.execute("insert into cookies values(7)") + source.commit() + source.execute("pragma journal_mode=wal") + source.execute("update cookies set x=8") + source.commit() + subprocess.run( + [sys.executable, "-c", + "import sqlite3, os, sys; c=sqlite3.connect(sys.argv[1]); " + "c.execute('create table cookies(x)'); c.commit(); " + "c.execute('pragma journal_mode=wal'); " + "c.execute('insert into cookies values(99)'); c.commit(); os._exit(0)", + str(dst)], check=True, timeout=15, stdin=subprocess.DEVNULL) + assert os.path.exists(str(dst) + "-wal") + try: + assert bc._copy_auth_file(str(src), str(dst)) is True + with sqlite3.connect(dst) as conn: + assert conn.execute("select x from cookies").fetchall() == [(8,)] + conn.close() + finally: + source.close() def test_copy_auth_file_plain_for_non_db(self, tmp_path): import hermes_cli.browser_connect as bc diff --git a/website/docs/user-guide/features/browser.md b/website/docs/user-guide/features/browser.md index 0a5dbf5bb550..7026169fbbb2 100644 --- a/website/docs/user-guide/features/browser.md +++ b/website/docs/user-guide/features/browser.md @@ -224,8 +224,13 @@ open — it fails fast with a "fully quit the browser and retry" message rather than hang or produce a signed-out session. Real-profile browsing on Windows therefore requires the browser **fully quit**, including any background/tray instance (Chrome's "continue running background apps when closed" keeps a -`chrome.exe` alive after you close the window). macOS and Linux can copy the -profile while the browser is running. +`chrome.exe` alive after you close the window). macOS and Linux can usually copy +the profile while the browser is running. On every platform, each authentication +database backup has a five-second retry budget. If the source or snapshot database +stays locked, Hermes stops the launch and asks you to close the browser and retry. +It preserves committed WAL data through SQLite rather than falling back to a raw +file copy, which could silently lose recent logins. Unreadable or corrupt databases +also stop the launch. Set `browser.real_profile_autoclose: true` to let Hermes **offer to close the browser for you** when it's holding the profile. Even with this on, Hermes never From 13fb5e1eceba51fc45a48b5d95a357e144d42689 Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Tue, 8 Sep 2026 05:00:00 -0700 Subject: [PATCH 178/227] test(browser): migrate pinned-profile fixtures to real SQLite --- tests/tools/test_browser_real_profile_pin.py | 15 +++++++-------- 1 file changed, 7 insertions(+), 8 deletions(-) diff --git a/tests/tools/test_browser_real_profile_pin.py b/tests/tools/test_browser_real_profile_pin.py index d0803b32e5c1..65ba699ebd86 100644 --- a/tests/tools/test_browser_real_profile_pin.py +++ b/tests/tools/test_browser_real_profile_pin.py @@ -11,9 +11,8 @@ - pin unset -> native last_used behavior, byte-for-byte """ import json -import os -import pytest +from tests.tools.test_browser_real_profile import _auth_db class TestRealProfilePin: @@ -21,8 +20,8 @@ def _make_profile(self, root, last_used="Profile 2"): """Synthetic Chromium user-data-dir with two profiles + last_used.""" for prof in ("Default", "Profile 2", "Profile 4"): (root / prof / "Network").mkdir(parents=True) - (root / prof / "Cookies").write_text(f"cookies-{prof}") - (root / prof / "Login Data").write_text(f"logins-{prof}") + _auth_db((root / prof / "Cookies"), f"cookies-{prof}") + _auth_db((root / prof / "Login Data"), f"logins-{prof}") (root / prof / "Preferences").write_text("{}") (root / "Crashpad").mkdir() (root / "Local State").write_text( @@ -40,7 +39,7 @@ def test_pin_wins_over_last_used(self, tmp_path, monkeypatch): dst, err = bc.snapshot_real_profile("chrome", src=str(src)) assert err is None and dst - got = (home / "browser-profile" / "chrome" / "Default" / "Cookies").read_text() + got = _auth_db((home / "browser-profile" / "chrome" / "Default" / "Cookies")) assert got == "cookies-Profile 2", "pin must override last_used" def test_bad_pin_fails_closed(self, tmp_path, monkeypatch): @@ -66,7 +65,7 @@ def test_no_pin_keeps_native_last_used(self, tmp_path, monkeypatch): dst, err = bc.snapshot_real_profile("chrome", src=str(src)) assert err is None and dst - got = (home / "browser-profile" / "chrome" / "Default" / "Cookies").read_text() + got = _auth_db((home / "browser-profile" / "chrome" / "Default" / "Cookies")) assert got == "cookies-Profile 4", "no pin = native last_used" def test_re_sync_respects_pin_when_last_used_flips(self, tmp_path, monkeypatch): @@ -86,9 +85,9 @@ def test_re_sync_respects_pin_when_last_used_flips(self, tmp_path, monkeypatch): (src / "Local State").write_text( json.dumps({"os_crypt": {}, "profile": {"last_used": "Profile 4"}}) ) - (src / "Profile 2" / "Cookies").write_text("cookies-Profile 2-v2") + _auth_db((src / "Profile 2" / "Cookies"), "cookies-Profile 2-v2") dst2, err2 = bc.snapshot_real_profile("chrome", src=str(src)) assert err2 is None and dst2 == dst1 - got = (home / "browser-profile" / "chrome" / "Default" / "Cookies").read_text() + got = _auth_db((home / "browser-profile" / "chrome" / "Default" / "Cookies")) assert got == "cookies-Profile 2-v2", "auth re-sync must stay on the pin" From 8aa219ef60ae16bd130e4775a1513e122880654e Mon Sep 17 00:00:00 2001 From: unsupportedpastels Date: Sun, 6 Sep 2026 15:55:30 +0000 Subject: [PATCH 179/227] fix(desktop): expose plugin installation from settings --- .../settings/plugin-install-modal.test.tsx | 112 ++++++++++++++++++ .../src/app/settings/plugin-install-modal.tsx | 50 +++++++- .../src/app/settings/plugins-settings.tsx | 6 + apps/desktop/src/i18n/en.ts | 3 + apps/desktop/src/i18n/ja.ts | 7 ++ apps/desktop/src/i18n/types.ts | 3 + apps/desktop/src/i18n/zh-hant.ts | 7 ++ apps/desktop/src/i18n/zh.ts | 3 + .../src/store/plugin-install-request.ts | 1 + 9 files changed, 186 insertions(+), 6 deletions(-) create mode 100644 apps/desktop/src/app/settings/plugin-install-modal.test.tsx diff --git a/apps/desktop/src/app/settings/plugin-install-modal.test.tsx b/apps/desktop/src/app/settings/plugin-install-modal.test.tsx new file mode 100644 index 000000000000..246b8b8308d0 --- /dev/null +++ b/apps/desktop/src/app/settings/plugin-install-modal.test.tsx @@ -0,0 +1,112 @@ +import { QueryClientProvider } from '@tanstack/react-query' +import { act, cleanup, fireEvent, render, screen, waitFor } from '@testing-library/react' +import { MemoryRouter } from 'react-router' +import { afterEach, beforeEach, describe, expect, it, vi } from 'vitest' + +const { requestGateway } = vi.hoisted(() => ({ requestGateway: vi.fn() })) +vi.mock('@/app/gateway/hooks/use-gateway-request', () => ({ + useGatewayRequest: () => ({ requestGateway }) +})) +vi.mock('@/hermes', async importOriginal => ({ + ...(await importOriginal>()), + getProfiles: async () => ({ profiles: [] }) +})) + +import { queryClient } from '@/lib/query-client' +import { + $pluginInstallRequest, + closePluginInstallRequest, + openPluginInstallRequest +} from '@/store/plugin-install-request' +import { $activeGatewayProfile } from '@/store/profile' +import { $connection, $gatewayState } from '@/store/session' + +import { PluginInstallModal } from './plugin-install-modal' +import { PluginsSettings } from './plugins-settings' + +const probePluginRepo = vi.fn() +const installDesktopPlugin = vi.fn() + +const renderFlow = () => + render( + + + + + + + ) + +beforeEach(() => { + vi.clearAllMocks() + queryClient.clear() + closePluginInstallRequest() + $gatewayState.set('idle') + $activeGatewayProfile.set('default') + probePluginRepo.mockResolvedValue({ ok: true, agent: true, desktop: true, warnings: [] }) + vi.stubGlobal('hermesDesktop', { probePluginRepo, installDesktopPlugin }) +}) +afterEach(() => { + cleanup() + closePluginInstallRequest() + vi.unstubAllGlobals() +}) + +describe('Install from Git entry flow', () => { + it.each(['local', 'remote'] as const)( + 'opens repository entry and reviews without installing in %s mode', + async mode => { + $connection.set({ mode } as NonNullable>) + renderFlow() + fireEvent.click(screen.getByRole('button', { name: 'Install from Git' })) + const input = await screen.findByRole('textbox', { name: 'Repository' }) + const review = screen.getByRole('button', { name: 'Review repository' }) + expect((review as HTMLButtonElement).disabled).toBe(true) + fireEvent.change(input, { target: { value: ' ' } }) + fireEvent.submit(input.closest('form')!) + expect(probePluginRepo).not.toHaveBeenCalled() + fireEvent.change(input, { target: { value: 'https://github.com/example/plugin' } }) + fireEvent.click(review) + await waitFor(() => + expect(probePluginRepo).toHaveBeenCalledWith({ identifier: 'https://github.com/example/plugin' }) + ) + expect(await screen.findByText('This package includes')).toBeTruthy() + expect( + screen.getByText( + mode === 'remote' + ? 'Installs into the connected default backend' + : 'Installs into the default backend (~/.hermes/plugins/)' + ) + ).toBeTruthy() + expect(screen.getByText("Installs into this app's local desktop-plugins folder")).toBeTruthy() + expect(requestGateway).not.toHaveBeenCalled() + expect(installDesktopPlugin).not.toHaveBeenCalled() + fireEvent.click(screen.getByRole('button', { name: 'Cancel' })) + expect($pluginInstallRequest.get()).toBeNull() + } + ) + + it('cancels repository entry and starts fresh when reopened', async () => { + renderFlow() + fireEvent.click(screen.getByRole('button', { name: 'Install from Git' })) + fireEvent.change(await screen.findByRole('textbox', { name: 'Repository' }), { target: { value: 'unfinished' } }) + fireEvent.click(screen.getByRole('button', { name: 'Cancel' })) + expect($pluginInstallRequest.get()).toBeNull() + fireEvent.click(screen.getByRole('button', { name: 'Install from Git' })) + expect(((await screen.findByRole('textbox', { name: 'Repository' })) as HTMLInputElement).value).toBe('') + expect(probePluginRepo).not.toHaveBeenCalled() + expect(installDesktopPlugin).not.toHaveBeenCalled() + }) + + it('preserves prefilled deep-link inspection and legacy selection without auto-install', async () => { + renderFlow() + act(() => openPluginInstallRequest({ repo: 'https://github.com/example/plugin', legacyHint: 'desktop' })) + expect(await screen.findByText('This package includes')).toBeTruthy() + expect(screen.queryByRole('textbox', { name: 'Repository' })).toBeNull() + const boxes = screen.getAllByRole('checkbox') + expect(boxes.map(box => box.getAttribute('aria-checked'))).toEqual(['false', 'true']) + expect(probePluginRepo).toHaveBeenCalledTimes(1) + expect(requestGateway).not.toHaveBeenCalled() + expect(installDesktopPlugin).not.toHaveBeenCalled() + }) +}) diff --git a/apps/desktop/src/app/settings/plugin-install-modal.tsx b/apps/desktop/src/app/settings/plugin-install-modal.tsx index 40012f238c27..1034e203678c 100644 --- a/apps/desktop/src/app/settings/plugin-install-modal.tsx +++ b/apps/desktop/src/app/settings/plugin-install-modal.tsx @@ -15,6 +15,7 @@ import { DialogTitle, preventCloseButtonAutoFocus } from '@/components/ui/dialog' +import { Input } from '@/components/ui/input' import { Switch } from '@/components/ui/switch' import { discoverRuntimePlugins } from '@/contrib/runtime-loader' import { useI18n } from '@/i18n' @@ -26,6 +27,7 @@ import { notify } from '@/store/notifications' import { $pluginInstallRequest, closePluginInstallRequest, + openPluginInstallRequest, type PluginInstallRequest } from '@/store/plugin-install-request' import { $activeGatewayProfile, $profileScope } from '@/store/profile' @@ -47,6 +49,7 @@ export function PluginInstallModal() { const activeProfile = useStore($activeGatewayProfile) const profileScope = useStore($profileScope) + const [repoInput, setRepoInput] = useState('') const [phase, setPhase] = useState('idle') const [probe, setProbe] = useState(null) const [installAgent, setInstallAgent] = useState(true) @@ -58,6 +61,7 @@ export function PluginInstallModal() { const probeToken = useRef(0) const resetState = useCallback(() => { + setRepoInput('') setPhase('idle') setProbe(null) setInstallAgent(true) @@ -142,7 +146,9 @@ export function PluginInstallModal() { return } - void runProbe(request) + if (request.repo) { + void runProbe(request) + } }, [request, resetState, runProbe]) const profileLabel = activeProfile || profileScope || 'default' @@ -258,13 +264,39 @@ export function PluginInstallModal() { }} open={open} > - + {m.title} {m.description} - {request && ( + {request && !request.repo && ( +
{ + event.preventDefault() + const repo = repoInput.trim() + + if (repo) { + openPluginInstallRequest({ ...request, repo }) + } + }} + > + +
+ )} + + {request?.repo && (
@@ -403,9 +435,15 @@ export function PluginInstallModal() { - + {request && !request.repo ? ( + + ) : ( + + )} diff --git a/apps/desktop/src/app/settings/plugins-settings.tsx b/apps/desktop/src/app/settings/plugins-settings.tsx index 69598afe4329..b85c7f917db0 100644 --- a/apps/desktop/src/app/settings/plugins-settings.tsx +++ b/apps/desktop/src/app/settings/plugins-settings.tsx @@ -27,6 +27,7 @@ import { toggleAgentPlugin } from '@/store/agent-plugins' import { notifyError } from '@/store/notifications' +import { openPluginInstallRequest } from '@/store/plugin-install-request' import { $activeGatewayProfile } from '@/store/profile' import { $connection, $gatewayState } from '@/store/session' @@ -374,6 +375,11 @@ export function PluginsSettings() { return ( +
+ +

{p.blurb}

diff --git a/apps/desktop/src/i18n/en.ts b/apps/desktop/src/i18n/en.ts index 280d691f19df..2aa581e093f2 100644 --- a/apps/desktop/src/i18n/en.ts +++ b/apps/desktop/src/i18n/en.ts @@ -469,6 +469,9 @@ export const en: Translations = { sources: { bundled: 'bundled', user: 'user', git: 'git', project: 'project', entrypoint: 'pip' } }, installModal: { + installFromGit: 'Install from Git', + reviewRepository: 'Review repository', + repoPlaceholder: 'https://github.com/owner/repo', title: 'Install plugin', description: 'Review what this repository contains before installing anything.', repoLabel: 'Repository', diff --git a/apps/desktop/src/i18n/ja.ts b/apps/desktop/src/i18n/ja.ts index 08d7d3f28516..2d59ad168895 100644 --- a/apps/desktop/src/i18n/ja.ts +++ b/apps/desktop/src/i18n/ja.ts @@ -299,6 +299,13 @@ export const ja = defineLocale({ }, settings: { + plugins: { + installModal: { + installFromGit: 'Git からインストール', + reviewRepository: 'リポジトリを確認', + repoPlaceholder: 'https://github.com/owner/repo' + } + }, closeSettings: '設定を閉じる', exportConfig: '設定を書き出す', importConfig: '設定を読み込む', diff --git a/apps/desktop/src/i18n/types.ts b/apps/desktop/src/i18n/types.ts index 6ed488b170f7..b6428083441d 100644 --- a/apps/desktop/src/i18n/types.ts +++ b/apps/desktop/src/i18n/types.ts @@ -410,6 +410,9 @@ export interface Translations { sources: Record } installModal: { + installFromGit: string + reviewRepository: string + repoPlaceholder: string title: string description: string repoLabel: string diff --git a/apps/desktop/src/i18n/zh-hant.ts b/apps/desktop/src/i18n/zh-hant.ts index d5243ce78ed5..7818f9ba6b4a 100644 --- a/apps/desktop/src/i18n/zh-hant.ts +++ b/apps/desktop/src/i18n/zh-hant.ts @@ -290,6 +290,13 @@ export const zhHant = defineLocale({ }, settings: { + plugins: { + installModal: { + installFromGit: '從 Git 安裝', + reviewRepository: '檢查儲存庫', + repoPlaceholder: 'https://github.com/owner/repo' + } + }, closeSettings: '關閉設定', exportConfig: '匯出設定', importConfig: '匯入設定', diff --git a/apps/desktop/src/i18n/zh.ts b/apps/desktop/src/i18n/zh.ts index 1366057b68ba..2298c3a0d67c 100644 --- a/apps/desktop/src/i18n/zh.ts +++ b/apps/desktop/src/i18n/zh.ts @@ -456,6 +456,9 @@ export const zh: Translations = { sources: { bundled: '内置', user: '用户', git: 'git', project: '项目', entrypoint: 'pip' } }, installModal: { + installFromGit: '从 Git 安装', + reviewRepository: '检查仓库', + repoPlaceholder: 'https://github.com/owner/repo', title: '安装插件', description: '在安装前查看此仓库包含哪些组件。', repoLabel: '仓库', diff --git a/apps/desktop/src/store/plugin-install-request.ts b/apps/desktop/src/store/plugin-install-request.ts index ff772b5b2b55..8ba245dcbee5 100644 --- a/apps/desktop/src/store/plugin-install-request.ts +++ b/apps/desktop/src/store/plugin-install-request.ts @@ -5,6 +5,7 @@ export type PluginInstallLegacyHint = 'agent' | 'desktop' | null /** Future metrics (opt-in): install count + success/failure, per repo. */ export interface PluginInstallRequest { + /** Empty opens repository entry; a supplied repo goes straight to inspection. */ repo: string enable?: boolean force?: boolean From bc1d1a287966bb4e990c4414461f9f0c0da3e80e Mon Sep 17 00:00:00 2001 From: witcheer <208820874+notwitcheer@users.noreply.github.com> Date: Tue, 8 Sep 2026 12:07:56 +0000 Subject: [PATCH 180/227] docs(delegation): update shipped defaults (250 iterations, 10 concurrent children) and document output_schema The delegation page and the delegation-patterns guide still stated the pre-v2026.8.31 defaults (50 iterations, 3 concurrent subagents). Shipped values: DEFAULT_MAX_ITERATIONS = 250 (tools/delegate_tool.py) and max_concurrent_children: 10 (hermes_cli/config_defaults.py). The per-task output_schema contract (one bounded correction retry, schema_valid / schema_errors on the result) was not documented anywhere on the page. The Max Iterations section also showed max_iterations as a per-call argument; delegate_task ignores caller-supplied values and reads delegation.max_iterations from config. --- website/docs/guides/delegation-patterns.md | 6 +-- .../docs/user-guide/features/delegation.md | 47 ++++++++++++++----- 2 files changed, 38 insertions(+), 15 deletions(-) diff --git a/website/docs/guides/delegation-patterns.md b/website/docs/guides/delegation-patterns.md index d0f5bffb27d0..0b72b96d708b 100644 --- a/website/docs/guides/delegation-patterns.md +++ b/website/docs/guides/delegation-patterns.md @@ -200,14 +200,14 @@ Subagents inherit the parent's enabled toolsets. `delegate_task` does not accept ## Constraints -- **Default 3 parallel tasks**: batches default to 3 concurrent subagents (configurable via `delegation.max_concurrent_children` in config.yaml, no hard ceiling, only a floor of 1) +- **Default 10 parallel tasks**: batches default to 10 concurrent subagents (configurable via `delegation.max_concurrent_children` in config.yaml, no hard ceiling, only a floor of 1) - **Nested delegation is opt-in**: leaf subagents (default) cannot call `delegate_task`, `clarify`, `memory`, or `execute_code`. Orchestrator subagents (`role="orchestrator"`) retain `delegate_task` for further delegation, but only when `delegation.max_spawn_depth` is raised above the default of 1 (floor 1, no ceiling); the other three remain blocked. Disable globally via `delegation.orchestrator_enabled: false`. ### Tuning Concurrency and Depth | Config | Default | Range | Effect | |--------|---------|-------|--------| -| `max_concurrent_children` | 3 | >=1 | Parallel batch size per `delegate_task` call | +| `max_concurrent_children` | 10 | >=1 | Parallel batch size per `delegate_task` call | | `max_spawn_depth` | 1 | >=1 | How many delegation levels can spawn further | Example: running 30 parallel workers with nested subagents: @@ -220,7 +220,7 @@ delegation: - **Separate terminals** — each subagent gets its own terminal session with separate working directory and state - **No conversation history** — subagents see only the `goal` and `context` the parent agent passes when calling `delegate_task` -- **Default 50 iterations** — set `max_iterations` lower for simple tasks to save cost +- **Default 250 iterations** — set `delegation.max_iterations` lower in `config.yaml` for fleets of simple tasks to save cost - **Not durable** — top-level delegation runs in the background and posts its result back later, but it remains tied to the owning session and Hermes process. Session closure, `/stop`, `/new`, or a process restart can cancel or strand in-progress work. Use `cronjob` or `terminal(background=True, notify_on_complete=True)` for work that must survive those boundaries. --- diff --git a/website/docs/user-guide/features/delegation.md b/website/docs/user-guide/features/delegation.md index 7f0832a8dceb..aba9faf860cf 100644 --- a/website/docs/user-guide/features/delegation.md +++ b/website/docs/user-guide/features/delegation.md @@ -42,7 +42,7 @@ delegate_task( ## Parallel Batch -Up to 3 concurrent subagents by default (configurable, no hard ceiling): +Up to 10 concurrent subagents by default (configurable, no hard ceiling): ```python delegate_task(tasks=[ @@ -52,6 +52,29 @@ delegate_task(tasks=[ ]) ``` +## Structured Output (`output_schema`) + +Each task can carry an optional `output_schema`, a JSON Schema object the child's final answer must validate against. The child sees the schema up front as an output contract; when the answer comes back the parent validates it, and on failure sends the child exactly one bounded correction turn carrying the validation errors verbatim (the schema is not re-pasted). The task's result then gains `schema_valid` (true/false) and, on failure, `schema_errors`. + +```python +delegate_task( + tasks=[{ + "goal": "Check which of these three endpoints return 200", + "context": "https://a.example, https://b.example, https://c.example", + "output_schema": { + "type": "object", + "properties": { + "healthy": {"type": "array", "items": {"type": "string"}}, + "failing": {"type": "array", "items": {"type": "string"}} + }, + "required": ["healthy", "failing"] + } + }] +) +``` + +Keep schemas forgiving: require only the fields you will actually read. Tasks without an `output_schema` are unaffected. + ## How Subagent Context Works :::warning Critical: Subagents Know Nothing @@ -161,7 +184,7 @@ This is off by default because every unit is a new turn for the orchestrator: a The dispatch handle lists each unit (`units[].delegation_id`, `group`, `task_indexes`); unit ids are the call's id suffixed `-1`, `-2`, …, and every unit of one call shares a single slot of `delegation.max_concurrent_children`, so grouping never changes capacity accounting (the worker pool grows to the number of live units so no unit waits behind a full pool). An orchestrator subagent waits for its whole batch in the current turn so it can synthesize the results. -- **Maximum concurrency:** 3 tasks by default (configurable via `delegation.max_concurrent_children` or the `DELEGATION_MAX_CONCURRENT_CHILDREN` env var; floor of 1, no hard ceiling). Batches larger than the limit return a tool error rather than being silently truncated. +- **Maximum concurrency:** 10 tasks by default (configurable via `delegation.max_concurrent_children` or the `DELEGATION_MAX_CONCURRENT_CHILDREN` env var; floor of 1, no hard ceiling). Batches larger than the limit return a tool error rather than being silently truncated. - **Thread pool:** Uses `ThreadPoolExecutor` with the configured concurrency limit as max workers - **Progress display:** In CLI mode, a tree-view shows tool calls from each subagent in real-time with per-task completion lines. In gateway mode, progress is batched and relayed to the parent's progress callback. CLI and TUI completion notices use task-first titles such as `Subagent Task Completed: Review changes`; multi-task groups use the group name and task count. Unsuccessful or incomplete work gets a corresponding status label. These compact notices do not replace the full results delivered to the parent agent. - **Result ordering:** Within a unit, results are sorted by task index to match input order regardless of completion order; `TASK i/N` labels index the whole call @@ -294,16 +317,16 @@ Both roles retain `execute_code` (programmatic tool calling) so children can bat ## Max Iterations -Each subagent has an iteration limit (default: 50) that controls how many tool-calling turns it can take: +Each subagent has an iteration limit (default: 250) that controls how many tool-calling turns it can take. The limit is set globally in `config.yaml` and applies to every child; it is not a per-call parameter of `delegate_task`: -```python -delegate_task( - goal="Quick file check", - context="Check if /etc/nginx/nginx.conf exists and print its first 10 lines", - max_iterations=10 # Simple task, don't need many turns -) +```yaml +# In ~/.hermes/config.yaml +delegation: + max_iterations: 60 # lower it for fleets of simple tasks, raise it for long investigations ``` +A child that exhausts its budget returns with `exit_reason: max_iterations` and `truncated: true`, so the parent can tell a budget stop from a completed task. + ## Child Timeout By default there is **no wall-clock timeout** on subagents. Children fail only from what they're actually doing — API errors, tool errors, or hitting their iteration budget — never from a delegation-level stopwatch. Earlier releases shipped a hard cap (300s, later 600s), which kept killing legitimately busy children mid-task: deep code reviews, large research fan-outs, and slow reasoning models routinely need more than 10 minutes while making steady progress the whole time. @@ -565,7 +588,7 @@ error. | **Reasoning** | Full LLM reasoning loop | Just Python code execution | | **Context** | Fresh isolated conversation | No conversation, just script | | **Tool access** | All non-blocked tools with reasoning | 7 tools via RPC, no reasoning | -| **Parallelism** | 3 concurrent subagents by default (configurable) | Single script | +| **Parallelism** | 10 concurrent subagents by default (configurable) | Single script | | **Best for** | Complex tasks needing judgment | Mechanical multi-step pipelines | | **Token cost** | Higher (full LLM loop) | Lower (only stdout returned) | | **User interaction** | None (subagents can't clarify) | None | @@ -577,8 +600,8 @@ error. ```yaml # In ~/.hermes/config.yaml delegation: - max_iterations: 50 # Max turns per child (default: 50) - # max_concurrent_children: 3 # Parallel children per batch (default: 3) + max_iterations: 250 # Max turns per child (default: 250) + # max_concurrent_children: 10 # Parallel children per batch (default: 10) # independent_completions: false # true = each task/group returns as it finishes (default: one message per call) # worktree_isolation: false # Give each child its own git worktree (see Worktree Isolation above) # max_spawn_depth: 1 # Tree depth (floor 1, no ceiling, default 1 = flat). Raise to 2 to allow orchestrator children to spawn leaves; 3+ for deeper trees. From 08ffadc930cc1b85523add3892436d31a177e78c Mon Sep 17 00:00:00 2001 From: witcheer <208820874+notwitcheer@users.noreply.github.com> Date: Tue, 8 Sep 2026 12:08:19 +0000 Subject: [PATCH 181/227] docs(sessions): fix tip box that still says auto-prune ships disabled The Automatic Cleanup section on the same page states that sessions.auto_prune is on by default (since #54189, default flipped to True in hermes_cli/config_defaults.py), while the tip box under the prune commands still said "auto-prune ships disabled. Enable it if...". A reader gets two answers from one page. Rewrites the tip to the shipped default and shows how to turn it off. --- website/docs/user-guide/sessions.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/website/docs/user-guide/sessions.md b/website/docs/user-guide/sessions.md index bd9768d5ec7c..6adc541cd586 100644 --- a/website/docs/user-guide/sessions.md +++ b/website/docs/user-guide/sessions.md @@ -951,5 +951,5 @@ hermes sessions prune --older-than 30 --yes ``` :::tip -The database grows slowly (typical: 10-15 MB for hundreds of sessions) and session history powers `session_search` recall across past conversations, so auto-prune ships disabled. Enable it if you're running a heavy gateway/cron workload where `state.db` is meaningfully affecting performance (observed failure mode: 384 MB state.db with ~1000 sessions slowing down FTS5 inserts and `/resume` listing). Use `hermes sessions prune` for one-off cleanup without turning on the automatic sweep. +Auto-prune is **on by default**: ended sessions that have been inactive for `sessions.retention_days` (default 90) are removed at startup, and active sessions are never touched (see [Automatic Cleanup](#automatic-cleanup) above). Session history powers `session_search` recall across past conversations, so if you want to keep every ended session forever, set `sessions.auto_prune: false` in `config.yaml`, or raise `retention_days`. With auto-prune off, `hermes sessions prune` remains available for one-off cleanup (observed failure mode without any pruning: a 384 MB `state.db` with ~1000 sessions slowing down FTS5 inserts and `/resume` listing). ::: From 0268b706467e6ea1f316b20685911d437fd79256 Mon Sep 17 00:00:00 2001 From: witcheer <208820874+notwitcheer@users.noreply.github.com> Date: Tue, 8 Sep 2026 12:23:03 +0000 Subject: [PATCH 182/227] docs(code-execution): document the session kernel, reset, and stdout spillover The code-execution page describes project vs strict mode, the minimal environment and the RPC channel, but never mentions that execute_code calls share a persistent per-session kernel (tools/code_kernel.py), that a timeout kills the kernel, that reset=true exists, or that stdout beyond 50 KB is spilled to cache/exec with the path returned (tools/code_execution_tool.py). Defaults from hermes_cli/config_defaults.py (kernel_idle_timeout 1800, max_session_kernels 4). Remote-backend kernel and fail-open per-call fallback from tools/code_kernel_remote.py. --- .../user-guide/features/code-execution.md | 25 ++++++++++++++++++- 1 file changed, 24 insertions(+), 1 deletion(-) diff --git a/website/docs/user-guide/features/code-execution.md b/website/docs/user-guide/features/code-execution.md index 1585f415a24f..9cf28e5f4849 100644 --- a/website/docs/user-guide/features/code-execution.md +++ b/website/docs/user-guide/features/code-execution.md @@ -160,7 +160,7 @@ Switching mode changes where scripts run and which interpreter runs them, not wh | Resource | Limit | Notes | |----------|-------|-------| | **Timeout** | 5 minutes (300s) | Script is killed with SIGTERM, then SIGKILL after 5s grace | -| **Stdout** | 50 KB | Output truncated with `[output truncated at 50KB]` notice | +| **Stdout** | 50 KB | Shown head-and-tail inline; the full output is saved to `~/.hermes/cache/exec/` and the path is included in the result | | **Stderr** | 10 KB | Included in output on non-zero exit for debugging | | **Tool calls** | 50 per execution | Error returned when limit reached | @@ -174,6 +174,29 @@ code_execution: max_tool_calls: 50 # Max tool calls per execution (default: 50) ``` +## State Between Calls (the session kernel) + +On the local terminal backend, `execute_code` does not start a fresh interpreter for every call. Each session owns a persistent Python kernel, so variables, imports, and loaded data from one call are available in the next. The agent can load a dataset once and query it across several turns instead of re-reading it every time. Subagents get their own kernel; kernels are never shared across sessions. + +What ends a kernel: + +- **Timeout or interrupt.** A cell that hits the timeout (or is interrupted) kills the kernel process and its state is lost on purpose; the result says so and the next call starts a fresh kernel. +- **`reset=true`.** The agent can pass `reset: true` to discard the kernel's state and start clean. This is also the way to pick up environment changes: a kernel's environment is frozen when it spawns, so a newly allowlisted passthrough variable is invisible until the kernel is reset. +- **Idle timeout and eviction.** Kernels die with the session, after `code_execution.kernel_idle_timeout` idle seconds (default 1800), or when more than `code_execution.max_session_kernels` (default 4) are alive and the oldest is evicted. + +The security envelope is the same as a one-shot script: environment scrubbing, the tool whitelist, and the per-call tool budget all apply to every cell, and tool-call authority (approvals, session, allow-list) is rebound on each cell. + +```yaml +# ~/.hermes/config.yaml +code_execution: + kernel_idle_timeout: 1800 # seconds a kernel may sit idle before it is reaped + max_session_kernels: 4 # kernels kept alive at once; oldest is evicted past this +``` + +**Remote backends** (Docker, SSH, Modal) run a remote session kernel with the same contract. If the kernel cannot be spawned on the backend, Hermes falls back to running each call as a standalone script and says so in the result. + +**Large output.** Stdout over 50 KB is shown head-and-tail inline, and the full text is saved under `~/.hermes/cache/exec/` with the path included in the result, so the agent can page through it with `read_file` instead of re-running the script. + ## How Tool Calls Work Inside Scripts When your script calls a function like `web_search("query")`: From 7ae4acf369fc4dc8ede5100984c0334b2d2e04af Mon Sep 17 00:00:00 2001 From: witcheer <208820874+notwitcheer@users.noreply.github.com> Date: Tue, 8 Sep 2026 12:26:06 +0000 Subject: [PATCH 183/227] docs(mcp): surface device-code login in the remote/headless hosts section hermes mcp login --flow device (RFC 8628, #104891, v2026.9.7) is documented on reference/mcp-config-reference but the user-guide MCP page, where people land when the loopback callback cannot reach their browser, still lists only Desktop relay, paste-back, SSH forward and proxied redirect_uri. Adds one bullet with a link to the reference section. --- website/docs/user-guide/features/mcp.md | 1 + 1 file changed, 1 insertion(+) diff --git a/website/docs/user-guide/features/mcp.md b/website/docs/user-guide/features/mcp.md index a3fe5f0802bb..150753f1fa52 100644 --- a/website/docs/user-guide/features/mcp.md +++ b/website/docs/user-guide/features/mcp.md @@ -275,6 +275,7 @@ On first connect, Hermes prints an authorize URL, opens your browser when possib - **Hermes Desktop (automatic):** when you run the OAuth sign-in from the Desktop app's MCP setup UI against a remote backend, Desktop hosts the callback listener on *your* machine and relays the authorization back to the gateway automatically — no tunnel, paste, or proxy needed. Requires both the Desktop app and the backend to be up to date. - **Paste-back (no setup):** on an interactive terminal Hermes prints "Or paste the redirect URL here…" alongside the authorize URL. Open the URL in your browser, approve, copy the full URL the browser ends up on (the redirect will show a connection error — that's expected), paste it at the prompt. Bare `?code=…&state=…` query strings work too. +- **Device-code login (no callback at all):** if the server's authorization server advertises a device authorization endpoint, run `hermes mcp login --flow device` on the machine running Hermes. It prints a verification URL and a short code; open the URL on any device, enter the code, and Hermes polls for approval. No browser is launched on the host and no callback listener is needed. Set `oauth.flow: device` on the server to make `login` and `reauth` use it by default. Details: [Device-code login](../../reference/mcp-config-reference.md#device-code-login-rfc-8628). - **SSH port forward:** `ssh -N -L :127.0.0.1: user@host` in a separate terminal, then let the redirect flow normally. - **Proxied callback (`redirect_uri`):** when a public HTTPS endpoint forwards to the host (e.g. a Tailscale Funnel or reverse proxy pointed at the callback port), set `oauth.redirect_uri` and the browser redirect reaches Hermes on its own — no tunnel or paste needed: From f77dbc24407dc553a6b9fa41a8a8a53439490ee9 Mon Sep 17 00:00:00 2001 From: witcheer <208820874+notwitcheer@users.noreply.github.com> Date: Tue, 8 Sep 2026 12:23:56 +0000 Subject: [PATCH 184/227] docs(bot-mode): state that same-gateway Group Chats keep running after Desktop closes Since v2026.8.31 (#99007 gateway-owned room authority, #99047 replica takeover, #99099 gateway-side turn driver) a room whose members all live on one gateway continues without any Desktop attached. The Bot Mode page carries the cross-machine and replication beats but never states this durability, so readers still assume the Desktop drives the room. Adds one bullet under Groups and group chats; groups.capabilities.driver flag per tui_gateway/methods_groups.py. --- website/docs/user-guide/bot-mode.md | 1 + 1 file changed, 1 insertion(+) diff --git a/website/docs/user-guide/bot-mode.md b/website/docs/user-guide/bot-mode.md index dfe67f61bded..02a2070b840e 100644 --- a/website/docs/user-guide/bot-mode.md +++ b/website/docs/user-guide/bot-mode.md @@ -100,6 +100,7 @@ Groups are standalone rows in the same activity-ordered roster as Bot DMs. A Bot - Hard caps (10 messages per send, 3 rounds) keep rooms from spinning. - Each member keeps its own persistent `Group: ` session, so room context survives like any other conversation. - **Not every Bot replies to every message.** Speaking is each member's own choice — a Bot replies only when it has something new to add and passes otherwise, and @-mentioning specific members scopes the round to them. Expect the members you addressed (or whoever has something to say) to speak, and the rest to stay quiet. +- **Rooms keep running when you close the Desktop.** When every member of a room lives on the same gateway, that gateway owns turn scheduling through a durable driver: closing Hermes Desktop (or losing its connection) does not stop a room mid-discussion, and the Desktop simply catches up from the room's log when it reconnects. `groups.capabilities` on the gateway reports `driver: true` when this applies. Rooms whose members span several machines are different: each member's turns run on its own gateway, and the cross-connection courier described under *Bot-to-bot messaging* still applies to them. - **Rooms can span machines.** The New Group Chat picker seats Bots from any registered connection; each member's turns run on its own machine, in its own `Group: ` session there. Cross-machine members carry a device badge (`dixie · Mac Mini`) in the room and in other members' transcripts, and the disambiguated `@name-device` handle works in room mentions — so same-named agents on two machines never blur together. ## Bot-to-bot messaging From fb3446a281e4bddc733a04bf92a5ec5f0d6decc9 Mon Sep 17 00:00:00 2001 From: witcheer <208820874+notwitcheer@users.noreply.github.com> Date: Tue, 8 Sep 2026 12:25:32 +0000 Subject: [PATCH 185/227] docs: eight small accuracy fixes from open type/docs issues MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Each fix verified against main @ ee84ccd on 2026-09-08. - windows-native: the troubleshooting entry told users to set HERMES_GATEWAY_FORCE_STARTUP (no code reads it), query a task named HermesGateway (hermes_cli/gateway_windows.py names it Hermes_Gateway), and described the Startup-folder fallback as a cmd.exe shortcut (it writes a .vbs run via wscript.exe). Closes #88077. - faq: 'hermes config set HERMES_MODEL ...' does not change the default model; 'model.default' does. Closes #65855. - messaging/index, slash-commands, irc: gateway settings are read from ~/.hermes/config.yaml (gateway/config.py), not gateway-config.yaml. Closes #65857, closes #78276. - telegram: reaction lifecycle is 👀 then 👍/👎 (plugins/platforms/telegram/adapter.py on_processing_complete), docs said ✅/❌. Closes #78698. - plugins: bundled memory providers win on a name collision (plugins/memory/__init__.py docstring: bundled, user, project, entry point, first seen wins); the page said user plugins override. Model providers keep the documented last-writer-wins. Closes #100281. - architecture, index: terminal backend count is seven (tools/terminal_tool.py: local, docker, singularity, modal, daytona, vercel_sandbox, ssh); two surfaces said six. Closes #78252. - quickstart: the Portal quick path was called 'free'; the login is free, the inference is billed to the subscription. Wording now matches integrations/nous-portal. Closes #78254. --- website/docs/developer-guide/architecture.md | 2 +- website/docs/getting-started/quickstart.md | 2 +- website/docs/index.mdx | 2 +- website/docs/reference/faq.md | 2 +- website/docs/reference/slash-commands.md | 2 +- website/docs/user-guide/features/plugins.md | 2 +- website/docs/user-guide/messaging/index.md | 6 +++--- website/docs/user-guide/messaging/irc.md | 4 ++-- website/docs/user-guide/messaging/telegram.md | 6 +++--- website/docs/user-guide/windows-native.md | 6 +++--- 10 files changed, 17 insertions(+), 17 deletions(-) diff --git a/website/docs/developer-guide/architecture.md b/website/docs/developer-guide/architecture.md index 3640103c3de9..8b826980e7d7 100644 --- a/website/docs/developer-guide/architecture.md +++ b/website/docs/developer-guide/architecture.md @@ -40,7 +40,7 @@ This page is the top-level map of Hermes Agent internals. Use it to orient yours ▼ ▼ ┌───────────────────┐ ┌──────────────────────┐ │ Session Storage │ │ Tool Backends │ -│ (SQLite + FTS5) │ │ Terminal (6 backends) │ +│ (SQLite + FTS5) │ │ Terminal (7 backends) │ │ hermes_state.py │ │ Browser (5 backends) │ │ gateway/session.py│ │ Web (4 backends) │ └───────────────────┘ │ MCP (dynamic) │ diff --git a/website/docs/getting-started/quickstart.md b/website/docs/getting-started/quickstart.md index e0772f02be46..3d424f0ae754 100644 --- a/website/docs/getting-started/quickstart.md +++ b/website/docs/getting-started/quickstart.md @@ -98,7 +98,7 @@ That logs you in, sets Nous as your provider, and turns on the Tool Gateway in o :::info Setup modes On a fresh install, `hermes setup` offers three modes: -- **Quick Setup (Nous Portal)** — free OAuth login, no API keys; sets up a model plus the Tool Gateway tools. The recommended fast path. +- **Quick Setup (Nous Portal)** — OAuth login, no API keys to manage; sets up a model plus the Tool Gateway tools, billed to your [Nous Portal subscription](/integrations/nous-portal). The recommended fast path. - **Full Setup** — walk through every provider, tool, and option yourself (bring your own keys). - **Blank Slate** — everything starts **off** except the bare minimum needed to run an agent: **provider & model, the File Operations toolset, and the Terminal toolset**. No web, browser, code execution, vision, memory, delegation, cron, skills, plugins, or MCP servers — and compression, checkpoints, smart routing, and memory capture are all disabled. After the minimal baseline is applied, you choose one of two paths: **start with everything disabled** (finish now with the minimal agent), or **walk through all configurations** (opt in to tools, skills, plugins, MCP, and messaging). Pick this when you want a minimal, fully-controlled agent and intend to enable only exactly what you need. diff --git a/website/docs/index.mdx b/website/docs/index.mdx index a4f248e8aefa..0634bcdb84ae 100644 --- a/website/docs/index.mdx +++ b/website/docs/index.mdx @@ -134,7 +134,7 @@ It's not a coding copilot tethered to an IDE or a chatbot wrapper around a singl ## Key Features - **A closed learning loop** — Agent-curated memory with periodic nudges, autonomous skill creation, skill self-improvement during use, FTS5 cross-session recall with LLM summarization, and [Honcho](https://github.com/plastic-labs/honcho) dialectic user modeling -- **Runs anywhere, not just your laptop** — 6 terminal backends: local, Docker, SSH, Daytona, Singularity, Modal. Daytona and Modal offer serverless persistence — your environment hibernates when idle, costing nearly nothing +- **Runs anywhere, not just your laptop** — 7 terminal backends: local, Docker, SSH, Daytona, Singularity, Modal, Vercel Sandbox. Daytona and Modal offer serverless persistence — your environment hibernates when idle, costing nearly nothing - **Lives where you do** — CLI, Telegram, Discord, Slack, WhatsApp, Signal, Matrix, Mattermost, Email, SMS, DingTalk, Feishu, WeCom, Weixin, QQ Bot, Yuanbao, BlueBubbles, Home Assistant, Microsoft Teams, Google Chat, and more — 20+ platforms from one gateway - **Built by model trainers** — Created by [Nous Research](https://nousresearch.com), the lab behind Hermes, Nomos, and Psyche. Works with [Nous Portal](https://portal.nousresearch.com), [OpenRouter](https://openrouter.ai), OpenAI, or any endpoint - **Scheduled automations** — Built-in cron with delivery to any platform diff --git a/website/docs/reference/faq.md b/website/docs/reference/faq.md index 7dfc0d3ebdbd..2e7d31b5a645 100644 --- a/website/docs/reference/faq.md +++ b/website/docs/reference/faq.md @@ -284,7 +284,7 @@ Make sure the key matches the provider. An OpenAI key won't work with OpenRouter hermes model # Set a valid model -hermes config set HERMES_MODEL anthropic/claude-opus-4.7 +hermes config set model.default anthropic/claude-opus-4.7 # Or specify per-session hermes chat --model openrouter/meta-llama/llama-3.1-70b-instruct diff --git a/website/docs/reference/slash-commands.md b/website/docs/reference/slash-commands.md index 0bcdcb52cb44..005be52a9a4a 100644 --- a/website/docs/reference/slash-commands.md +++ b/website/docs/reference/slash-commands.md @@ -15,7 +15,7 @@ Installed skills are also exposed as dynamic slash commands on both surfaces. (` ## Permissions and admin/user split -Every messaging platform that supports a per-user allowlist (Telegram, Discord, Slack, Matrix, Mattermost, Signal, …) also supports a two-tier slash command split: **admins** get every registered command, **regular users** only get the names you list in `user_allowed_commands` (plus the always-allowed floor `/help` and `/whoami`). Configure `allow_admin_from` and `user_allowed_commands` (and the per-group equivalents `group_allow_admin_from` / `group_user_allowed_commands`) inside the platform's `extra:` block in `~/.hermes/gateway-config.yaml`. +Every messaging platform that supports a per-user allowlist (Telegram, Discord, Slack, Matrix, Mattermost, Signal, …) also supports a two-tier slash command split: **admins** get every registered command, **regular users** only get the names you list in `user_allowed_commands` (plus the always-allowed floor `/help` and `/whoami`). Configure `allow_admin_from` and `user_allowed_commands` (and the per-group equivalents `group_allow_admin_from` / `group_user_allowed_commands`) inside the platform's `extra:` block in `~/.hermes/config.yaml`. See the per-platform docs for examples — the structure is identical across platforms: diff --git a/website/docs/user-guide/features/plugins.md b/website/docs/user-guide/features/plugins.md index 417850c1c7b1..78b450f262ef 100644 --- a/website/docs/user-guide/features/plugins.md +++ b/website/docs/user-guide/features/plugins.md @@ -144,7 +144,7 @@ Within each source, Hermes also recognizes sub-category directories that route p | `plugins/context_engine//` | Context-compression engines (`ctx.register_context_engine()`) | **Own loader** in `plugins/context_engine/__init__.py` (one active at a time) | | `plugins/model-providers//` | LLM provider profiles (`register_provider(ProviderProfile(...))`) | **Own loader** in `providers/__init__.py` (lazily scanned on first `get_provider_profile()` call) | -User plugins at `~/.hermes/plugins/model-providers//` and `~/.hermes/plugins/memory//` override bundled plugins of the same name — last-writer-wins in `register_provider()` / `register_memory_provider()`. Drop a directory in, and it replaces the built-in without any repo edits. +User plugins at `~/.hermes/plugins/model-providers//` override bundled model providers of the same name (last-writer-wins in `register_provider()`), so you can replace a built-in provider profile without any repo edits. Memory providers resolve the other way round: for `~/.hermes/plugins/memory//` the **bundled** provider wins on a name collision (bundled, then user, then project, then entry points; first seen wins), so a user memory provider needs its own unique name. ## Plugins are opt-in (with a few exceptions) diff --git a/website/docs/user-guide/messaging/index.md b/website/docs/user-guide/messaging/index.md index b792b962aa5d..96446f46e042 100644 --- a/website/docs/user-guide/messaging/index.md +++ b/website/docs/user-guide/messaging/index.md @@ -284,7 +284,7 @@ continuation, not the history loaded when you send a message. ## Per-Channel Model & System Prompt Overrides -Different channels can run different models and personas from a **single gateway** — e.g. a cheap fast model in `#daily` and a frontier model with a specialist prompt in `#dev`. Configure `channel_overrides` under the platform in `~/.hermes/gateway-config.yaml`: +Different channels can run different models and personas from a **single gateway** — e.g. a cheap fast model in `#daily` and a frontier model with a specialist prompt in `#dev`. Configure `channel_overrides` under the platform in `~/.hermes/config.yaml`: ```yaml platforms: @@ -731,7 +731,7 @@ Once upstream is healthy, `/platform resume ` clears the breaker and re-ar ### Restart notifications -When the gateway restarts (or is shut down with in-flight sessions), it can send a one-shot "the agent is back" / "the agent was interrupted" message to each platform's home channel. This is controlled per-platform by the `gateway_restart_notification` flag in `gateway-config.yaml`, which defaults to `true`: +When the gateway restarts (or is shut down with in-flight sessions), it can send a one-shot "the agent is back" / "the agent was interrupted" message to each platform's home channel. This is controlled per-platform by the `gateway_restart_notification` flag in `config.yaml`, which defaults to `true`: ```yaml gateway: @@ -748,7 +748,7 @@ Disable it on noisy or low-priority platforms while leaving it on for your prima ### Typing indicators -While the agent is processing a message, the gateway shows a live typing status on platforms that support it — a "typing…" bubble on Telegram/Discord/Signal, or the "is thinking…" assistant status on Slack. This is controlled per-platform by the `typing_indicator` flag in `gateway-config.yaml`, which defaults to `true`: +While the agent is processing a message, the gateway shows a live typing status on platforms that support it — a "typing…" bubble on Telegram/Discord/Signal, or the "is thinking…" assistant status on Slack. This is controlled per-platform by the `typing_indicator` flag in `config.yaml`, which defaults to `true`: ```yaml gateway: diff --git a/website/docs/user-guide/messaging/irc.md b/website/docs/user-guide/messaging/irc.md index f9fa9d94ceef..4fd21061621f 100644 --- a/website/docs/user-guide/messaging/irc.md +++ b/website/docs/user-guide/messaging/irc.md @@ -15,9 +15,9 @@ IRC is plain text: there is no voice, image, file, thread, reaction, typing, or ## Configure Hermes -You can configure IRC two ways — environment variables (for a quick env-only setup) or the `gateway` block in `~/.hermes/gateway-config.yaml`. +You can configure IRC two ways — environment variables (for a quick env-only setup) or the `gateway` block in `~/.hermes/config.yaml`. -### Option A — gateway-config.yaml +### Option A — config.yaml ```yaml gateway: diff --git a/website/docs/user-guide/messaging/telegram.md b/website/docs/user-guide/messaging/telegram.md index 720217ec7eff..79afe3d1d46c 100644 --- a/website/docs/user-guide/messaging/telegram.md +++ b/website/docs/user-guide/messaging/telegram.md @@ -1224,8 +1224,8 @@ This covers the custom fallback transport layer that Hermes uses for Telegram co The bot can add emoji reactions to messages as visual processing feedback: - 👀 when the bot starts processing your message -- ✅ when the response is delivered successfully -- ❌ if an error occurs during processing +- 👍 when the response is delivered successfully +- 👎 if an error occurs during processing Reactions are **disabled by default**. Enable them in `config.yaml`: @@ -1241,7 +1241,7 @@ TELEGRAM_REACTIONS=true ``` :::note -Unlike Discord (where reactions are additive), Telegram's Bot API replaces all bot reactions in a single call. The transition from 👀 to ✅/❌ happens atomically — you won't see both at once. +Unlike Discord (where reactions are additive), Telegram's Bot API replaces all bot reactions in a single call. The transition from 👀 to 👍/👎 happens atomically — you won't see both at once. ::: :::tip diff --git a/website/docs/user-guide/windows-native.md b/website/docs/user-guide/windows-native.md index f703dfe286a2..69c2726a9de1 100644 --- a/website/docs/user-guide/windows-native.md +++ b/website/docs/user-guide/windows-native.md @@ -176,8 +176,8 @@ hermes gateway install What happens under the hood: -1. `schtasks /Create /SC ONLOGON /RL LIMITED /TN HermesGateway` — registers a task that runs at your login with standard (non-elevated) permissions. No UAC prompt. -2. If schtasks is blocked by group policy, falls back to writing a `start /min cmd.exe /d /c ` shortcut into `%APPDATA%\Microsoft\Windows\Start Menu\Programs\Startup`. Same effect, slightly cruder. +1. `schtasks /Create /SC ONLOGON /RL LIMITED /TN Hermes_Gateway` — registers a task that runs at your login with standard (non-elevated) permissions. No UAC prompt. +2. If schtasks is blocked by group policy, falls back to writing a small `Hermes_Gateway.vbs` launcher (run hidden via `wscript.exe`) into `%APPDATA%\Microsoft\Windows\Start Menu\Programs\Startup`. Same effect, slightly cruder. A VBScript is used rather than a `cmd.exe` shortcut because a console allocated at logon can receive a close event that kills the gateway before it finishes starting. 3. Spawns the gateway **detached via `pythonw.exe`** — not `python.exe`. `pythonw.exe` has no console attached, which immunizes it against `CTRL_C_EVENT` broadcasts from sibling processes (a real issue that used to kill the gateway when you Ctrl+C'd anything in the same process group). Flags used when spawning: `DETACHED_PROCESS | CREATE_NEW_PROCESS_GROUP | CREATE_NO_WINDOW | CREATE_BREAKAWAY_FROM_JOB`. @@ -297,7 +297,7 @@ You hit a shebang-script invocation that bypassed the `.cmd` shim. Hermes resolv Your download of `install.ps1` picked up a UTF-8 BOM. The `irm | iex` form strips BOMs automatically; `[scriptblock]::Create((irm ...))` does not. Re-run with the simple `irm | iex` form, or download the script manually and save it without a BOM via `[IO.File]::WriteAllText($path, $text, (New-Object Text.UTF8Encoding $false))`. **Gateway won't stay running after restart.** -Check `hermes gateway status` — it merges the schtasks entry, the Startup-folder shortcut (if used), and the live PID. If schtasks is registered but not running, group policy may be blocking `ONLOGON` triggers. Run `schtasks /Query /TN HermesGateway /V /FO LIST` to see the task's failure reason, or fall back to the Startup-folder path by uninstalling and reinstalling with `HERMES_GATEWAY_FORCE_STARTUP=1`. +Check `hermes gateway status` — it merges the schtasks entry, the Startup-folder shortcut (if used), and the live PID. If schtasks is registered but not running, group policy may be blocking `ONLOGON` triggers. Run `schtasks /Query /TN Hermes_Gateway /V /FO LIST` (`Hermes_Gateway_` for a named profile) to see the task's failure reason. The Startup-folder fallback engages automatically only when `schtasks` itself fails to register the task; there is no environment variable or flag to force it. **`/edit` still does nothing after setting `$env:EDITOR`.** You set it in the current process only; close and reopen the shell, or set it at User scope in System Properties → Environment Variables. Verify with `echo $env:EDITOR` in a new PowerShell window. From 77a54573433b4c2ad18e16b439f7834e215bffd4 Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Tue, 8 Sep 2026 04:14:55 -0700 Subject: [PATCH 186/227] feat: organize desktop sessions by gateway and profile --- .../src/app/chat/sidebar/filter-menu.tsx | 5 +- .../app/chat/sidebar/gateway-group-model.ts | 47 ++++ .../sidebar/gateway-group-preferences.test.ts | 33 +++ .../chat/sidebar/gateway-group-preferences.ts | 33 +++ .../app/chat/sidebar/gateway-groups.test.tsx | 100 +++++++ .../src/app/chat/sidebar/gateway-groups.tsx | 246 ++++++++++++++++++ apps/desktop/src/app/chat/sidebar/index.tsx | 41 +-- .../chat/sidebar/projects/workspace-groups.ts | 3 + .../src/app/chat/sidebar/sessions-section.tsx | 4 +- apps/desktop/src/i18n/ar.ts | 11 + apps/desktop/src/i18n/en.ts | 11 + apps/desktop/src/i18n/ja.ts | 11 + apps/desktop/src/i18n/ru.ts | 11 + apps/desktop/src/i18n/types.ts | 11 + apps/desktop/src/i18n/zh-hant.ts | 11 + apps/desktop/src/i18n/zh.ts | 11 + 16 files changed, 547 insertions(+), 42 deletions(-) create mode 100644 apps/desktop/src/app/chat/sidebar/gateway-group-model.ts create mode 100644 apps/desktop/src/app/chat/sidebar/gateway-group-preferences.test.ts create mode 100644 apps/desktop/src/app/chat/sidebar/gateway-group-preferences.ts create mode 100644 apps/desktop/src/app/chat/sidebar/gateway-groups.test.tsx create mode 100644 apps/desktop/src/app/chat/sidebar/gateway-groups.tsx diff --git a/apps/desktop/src/app/chat/sidebar/filter-menu.tsx b/apps/desktop/src/app/chat/sidebar/filter-menu.tsx index 9aa683d50839..408ef457d6e0 100644 --- a/apps/desktop/src/app/chat/sidebar/filter-menu.tsx +++ b/apps/desktop/src/app/chat/sidebar/filter-menu.tsx @@ -187,7 +187,8 @@ export function SidebarFilterMenu({ className }: { className?: string }) { const foldCollapsed = foldIds.length > 0 && foldIds.every(id => nodeOpen[id] === false) - const groupingLabel = GROUPINGS.find(option => option.id === grouping)?.label + const groupings = GROUPINGS.map(option => option.id === 'profile' ? {...option, label: t.sidebar.gatewayGroups.grouping} : option) + const groupingLabel = groupings.find(option => option.id === grouping)?.label // Two options are conditional: dragging a row is what picks manual, so it // only appears as a way back out once there's a hand-picked order to leave; @@ -249,7 +250,7 @@ export function SidebarFilterMenu({ className }: { className?: string }) { onValueChange={value => setSidebarGrouping(value as SidebarGrouping)} value={grouping} > - {GROUPINGS.map(option => ( + {groupings.map(option => ( ))} diff --git a/apps/desktop/src/app/chat/sidebar/gateway-group-model.ts b/apps/desktop/src/app/chat/sidebar/gateway-group-model.ts new file mode 100644 index 000000000000..9bf36e1df0f4 --- /dev/null +++ b/apps/desktop/src/app/chat/sidebar/gateway-group-model.ts @@ -0,0 +1,47 @@ +import { useStore } from '@nanostores/react' +import { useMemo } from 'react' + +import type { SessionInfo } from '@/hermes' +import { resolveProfileColor } from '@/lib/profile-color' +import { $connectionsRegistry } from '@/store/connection-registry-state' +import { $profileColors, normalizeProfileKey } from '@/store/profile' + +import type { SidebarSessionGroup } from './projects/workspace-groups' + +/** Group identity never depends on a mutable label, URL, or the active gateway. */ +export function useGatewaySessionGroups(sessions: SessionInfo[], enabled: boolean) { + const registry = useStore($connectionsRegistry) + const colors = useStore($profileColors) + + return useMemo(() => { + if (!enabled) { + return undefined + } + + const groups = new Map() + + for (const session of sessions) { + const profile = normalizeProfileKey(session.profile) + const connectionId = session.connection_id || null + const id = JSON.stringify([connectionId, profile]) + const gateway = registry?.connections.find(connection => connection.id === connectionId) + const label = connectionId ? `${gateway?.label || connectionId} · ${profile}` : profile + + const group: SidebarSessionGroup = groups.get(id) ?? { + id, + label, + connectionId, + profile, + mode: 'profile', + path: null, + color: resolveProfileColor(profile, colors), + sessions: [] + } + + group.sessions.push(session) + groups.set(id, group) + } + + return [...groups.values()].sort((a, b) => a.label.localeCompare(b.label) || a.id.localeCompare(b.id)) + }, [sessions, enabled, registry, colors]) +} diff --git a/apps/desktop/src/app/chat/sidebar/gateway-group-preferences.test.ts b/apps/desktop/src/app/chat/sidebar/gateway-group-preferences.test.ts new file mode 100644 index 000000000000..5d4acb345938 --- /dev/null +++ b/apps/desktop/src/app/chat/sidebar/gateway-group-preferences.test.ts @@ -0,0 +1,33 @@ +// @vitest-environment jsdom +import { expect, it, vi } from 'vitest' + +import { + $gatewayGroupAliases, + $gatewayGroupCollapsed, + $gatewayGroupOrder, + renameGatewayGroup, + reorderGatewayGroups, + toggleGatewayGroup +} from './gateway-group-preferences' + +it('persists identity-scoped edits and reorders newly discovered groups without forgetting hidden groups', async () => { + const local = JSON.stringify(['local', 'default']) + const remote = JSON.stringify(['remote-1', 'default']) + const cloud = JSON.stringify(['cloud-1', 'default']) + $gatewayGroupOrder.set([]) + reorderGatewayGroups([remote, local]) + expect($gatewayGroupOrder.get()).toEqual([remote, local]) + renameGatewayGroup(remote, ' Research lab ') + toggleGatewayGroup(remote) + reorderGatewayGroups([cloud, local]) + expect($gatewayGroupOrder.get()).toEqual([remote, cloud, local]) + expect($gatewayGroupAliases.get()).toEqual({ [remote]: 'Research lab' }) + expect($gatewayGroupCollapsed.get()).toEqual([remote]) + vi.resetModules() + const restored = await import('./gateway-group-preferences') + expect(restored.$gatewayGroupOrder.get()).toEqual([remote, cloud, local]) + expect(restored.$gatewayGroupAliases.get()).toEqual({ [remote]: 'Research lab' }) + expect(restored.$gatewayGroupCollapsed.get()).toEqual([remote]) + restored.renameGatewayGroup(remote, ' ') + expect(restored.$gatewayGroupAliases.get()).toEqual({}) +}) diff --git a/apps/desktop/src/app/chat/sidebar/gateway-group-preferences.ts b/apps/desktop/src/app/chat/sidebar/gateway-group-preferences.ts new file mode 100644 index 000000000000..d1b0ba7a8294 --- /dev/null +++ b/apps/desktop/src/app/chat/sidebar/gateway-group-preferences.ts @@ -0,0 +1,33 @@ +import { Codecs, persistentAtom } from '@/lib/persisted' + +import { mergeVisibleReorder } from './order' + +const PREFIX = 'hermes.desktop.sidebar.gatewayGroups.v1' +export const $gatewayGroupAliases = persistentAtom(`${PREFIX}.aliases`, {}, Codecs.stringRecord) +export const $gatewayGroupOrder = persistentAtom(`${PREFIX}.order`, [], Codecs.stringArray) +export const $gatewayGroupCollapsed = persistentAtom(`${PREFIX}.collapsed`, [], Codecs.stringArray) + +export function renameGatewayGroup(id: string, alias: string) { + const aliases = { ...$gatewayGroupAliases.get() } + const name = alias.trim() + + if (name) { + aliases[id] = name + } else { + delete aliases[id] + } + + $gatewayGroupAliases.set(aliases) +} + +export function reorderGatewayGroups(ids: string[]) { + // A filtered-out or temporarily offline section keeps its place. + const order = $gatewayGroupOrder.get() + const allIds = [...order, ...ids.filter(id => !order.includes(id))] + $gatewayGroupOrder.set(mergeVisibleReorder(allIds, ids)) +} + +export function toggleGatewayGroup(id: string) { + const collapsed = $gatewayGroupCollapsed.get() + $gatewayGroupCollapsed.set(collapsed.includes(id) ? collapsed.filter(key => key !== id) : [...collapsed, id]) +} diff --git a/apps/desktop/src/app/chat/sidebar/gateway-groups.test.tsx b/apps/desktop/src/app/chat/sidebar/gateway-groups.test.tsx new file mode 100644 index 000000000000..c53143c270cd --- /dev/null +++ b/apps/desktop/src/app/chat/sidebar/gateway-groups.test.tsx @@ -0,0 +1,100 @@ +// @vitest-environment jsdom +import { act, cleanup, fireEvent, render, screen, within } from '@testing-library/react' +import { MemoryRouter } from 'react-router' +import { afterEach, expect, it, vi } from 'vitest' + +import { SidebarProvider } from '@/components/ui/sidebar' +import { $connectionsRegistry } from '@/store/connection-registry-state' +import { $sidebarRowMeta, setSidebarGrouping } from '@/store/layout' +import { $newChatRoute, $profiles } from '@/store/profile' +import { $sessionProfilesUsage, $sessions } from '@/store/session' +import { makeSessionInfo } from '@/test/session-info' + +import { ChatSidebar } from './index' + +const noop = () => {} +const resume = vi.fn() + +const mount = () => + render( + + + {}} + /> + + + ) + +afterEach(cleanup) + +it('keeps equal profile names on separate gateways and routes section creation and row navigation to their owner', () => { + mount() + act(() => { + $connectionsRegistry.set({ + version: 2, + primary: 'local', + secureTokenStorage: true, + connections: [ + { id: 'local', label: 'This computer', kind: 'local', tokenSet: false, tokenPreview: null }, + { id: 'remote-1', label: 'Homelab', kind: 'remote', tokenSet: false, tokenPreview: null }, + { id: 'cloud-1', label: 'Cloud workspace', kind: 'cloud', tokenSet: false, tokenPreview: null } + ] + }) + $profiles.set([ + { name: 'default', is_default: true }, + { name: 'work', is_default: false } + ] as typeof $profiles.value) + setSidebarGrouping('profile') + $sidebarRowMeta.set(['cost', 'tokens']) + $sessionProfilesUsage.set({ default: { cost_usd: 3, tokens: 100 } }) + $sessions.set( + ['local', 'remote-1', 'cloud-1'].map(connection_id => + makeSessionInfo({ + id: connection_id, + connection_id, + profile: 'default', + title: `${connection_id} session`, + last_active: Date.now() / 1000 + }) + ) + ) + }) + act(() => + $sessions.set([ + ...$sessions.get(), + makeSessionInfo({ id: 'legacy', profile: 'default', title: 'Legacy session', last_active: Date.now() / 1000 }) + ]) + ) + expect(screen.getAllByText(/\$3\.00/)).toHaveLength(1) + expect( + screen + .getByText(/\$3\.00/) + .closest('[data-gateway-group]') + ?.getAttribute('data-gateway-group') + ).toBe(JSON.stringify([null, 'default'])) + expect(screen.getByText('This computer · default')).toBeTruthy() + expect(screen.getByText('Homelab · default')).toBeTruthy() + expect(screen.getByText('Cloud workspace · default')).toBeTruthy() + fireEvent.click(screen.getByRole('button', { name: 'New session in Homelab · default' })) + expect($newChatRoute.get()).toMatchObject({ connectionId: 'remote-1', profile: 'default' }) + fireEvent.click(screen.getByText('cloud-1 session')) + expect(resume).toHaveBeenLastCalledWith( + 'cloud-1', + expect.objectContaining({ connection_id: 'cloud-1', profile: 'default' }) + ) + const group = screen.getByText('Homelab · default').closest('[data-gateway-group]')! + fireEvent.click(within(group as HTMLElement).getByRole('button', { name: 'Hide Homelab · default sessions' })) + expect(screen.queryByText('remote-1 session')).toBeNull() + expect(screen.getByText('local session')).toBeTruthy() +}) diff --git a/apps/desktop/src/app/chat/sidebar/gateway-groups.tsx b/apps/desktop/src/app/chat/sidebar/gateway-groups.tsx new file mode 100644 index 000000000000..07bc128a6a27 --- /dev/null +++ b/apps/desktop/src/app/chat/sidebar/gateway-groups.tsx @@ -0,0 +1,246 @@ +import type { useSensors } from '@dnd-kit/core' +import { arrayMove } from '@dnd-kit/sortable' +import { useStore } from '@nanostores/react' +import type { ReactNode } from 'react' +import { useState } from 'react' + +import { type NewSessionSplitHandler, startNewSessionDrag } from '@/app/chat/new-session-drag' +import { Button } from '@/components/ui/button' +import { Codicon } from '@/components/ui/codicon' +import { + Dialog, + DialogContent, + DialogDescription, + DialogFooter, + DialogHeader, + DialogTitle +} from '@/components/ui/dialog' +import { DropdownMenu, DropdownMenuContent, DropdownMenuItem, DropdownMenuTrigger } from '@/components/ui/dropdown-menu' +import { Input } from '@/components/ui/input' +import { ProfileGlyph } from '@/components/ui/profile-glyph' +import type { SessionInfo } from '@/hermes' +import { useI18n } from '@/i18n' +import { useStoreSelector } from '@/lib/use-session-slice' +import { newSessionInAgent, newSessionInProfile } from '@/store/profile' +import { $sessionProfilesUsage } from '@/store/session' +import { $sidebarSessionRankIds } from '@/store/sidebar-sort' + +import { SidebarGroupRow, SidebarRowGrab, SidebarRowLink, SidebarRowStack } from './chrome' +import { + $gatewayGroupAliases, + $gatewayGroupCollapsed, + $gatewayGroupOrder, + renameGatewayGroup, + reorderGatewayGroups, + toggleGatewayGroup +} from './gateway-group-preferences' +import { rankSessions } from './order' +import { SIDEBAR_GROUP_PAGE } from './projects/model' +import type { SidebarSessionGroup } from './projects/workspace-groups' +import { WorkspaceAddButton, WorkspaceShowMoreButton } from './projects/workspace-header' +import { ReorderableList, useSortableBindings } from './reorderable-list' + +interface GatewayProfileGroupsProps { + groups: SidebarSessionGroup[] + renderRows: (sessions: SessionInfo[]) => ReactNode + sensors?: ReturnType + onNewSessionSplit?: NewSessionSplitHandler +} + +export function GatewayProfileGroups({ groups, renderRows, sensors, onNewSessionSplit }: GatewayProfileGroupsProps) { + const order = useStore($gatewayGroupOrder) + + const ordered = [...groups].sort((a, b) => { + const left = order.indexOf(a.id) + const right = order.indexOf(b.id) + + return (left < 0 ? Infinity : left) - (right < 0 ? Infinity : right) || 0 + }) + + const ids = ordered.map(group => group.id) + + return ( + + {ordered.map((group, index) => ( + reorderGatewayGroups(arrayMove(ids, index, index + direction))} + onNewSessionSplit={onNewSessionSplit} + renderRows={renderRows} + /> + ))} + + ) +} + +interface GatewayProfileGroupProps { + group: SidebarSessionGroup + renderRows: (sessions: SessionInfo[]) => ReactNode + onMove: (direction: 1 | -1) => void + onNewSessionSplit?: NewSessionSplitHandler + first: boolean + last: boolean +} + +function GatewayProfileGroup({ group, renderRows, onMove, first, last, onNewSessionSplit }: GatewayProfileGroupProps) { + const { t } = useI18n() + const s = t.sidebar + const copy = s.gatewayGroups + const aliases = useStore($gatewayGroupAliases) + const collapsed = useStore($gatewayGroupCollapsed) + const rankIds = useStore($sidebarSessionRankIds) + // Legacy totals are keyed only by profile. Never attribute those figures to + // a registry gateway that happens to expose the same profile name. + const usage = useStoreSelector($sessionProfilesUsage, all => (group.connectionId ? undefined : all[group.profile!])) + const [renaming, setRenaming] = useState(false) + const [draft, setDraft] = useState('') + const [visibleCount, setVisibleCount] = useState(SIDEBAR_GROUP_PAGE) + const sortable = useSortableBindings(group.id) + const label = aliases[group.id] || group.label + const open = !collapsed.includes(group.id) + const sessions = rankSessions(group.sessions, rankIds) + const hiddenCount = Math.max(0, sessions.length - visibleCount) + const route = group.connectionId ? { connectionId: group.connectionId, profile: group.profile! } : undefined + + const startSession = () => { + if (!open) { + toggleGatewayGroup(group.id) + } + + const profile = group.profile! + + if (group.connectionId) { + newSessionInAgent({ connectionId: group.connectionId, profile }) + } else { + newSessionInProfile(profile) + } + } + + return ( + + + + startNewSessionDrag( + placement => { + if (!open) { + toggleGatewayGroup(group.id) + } + + onNewSessionSplit(placement.dir, { + anchor: placement.anchor, + before: placement.before, + profile: group.profile, + route + }) + }, + event, + { label: s.newSessionIn(label), profile: group.profile, route } + ) + : undefined + } + /> + + + + + + { + setDraft(aliases[group.id] || '') + setRenaming(true) + }} + > + {copy.rename} + + renameGatewayGroup(group.id, '')}> + {copy.resetName} + + onMove(-1)}> + {copy.moveUp} + + onMove(1)}> + {copy.moveDown} + + + +
+ } + label={ + toggleGatewayGroup(group.id)}> + {label} + + } + lead={ + + + + } + toggle={{ ariaLabel: s.projects.toggle(label, !open), onToggle: () => toggleGatewayGroup(group.id), open }} + totals={usage ? { costUsd: usage.cost_usd, tokens: usage.tokens } : undefined} + /> + {open && ( + <> + {renderRows(sessions.slice(0, visibleCount))} + {hiddenCount > 0 && ( + setVisibleCount(count => count + SIDEBAR_GROUP_PAGE)} + /> + )} + + )} + + +
{ + event.preventDefault() + renameGatewayGroup(group.id, draft) + setRenaming(false) + }} + > + + {copy.rename} + {copy.aliasHint} + + setDraft(event.target.value)} + placeholder={group.label} + value={draft} + /> + + + + +
+ + + + ) +} diff --git a/apps/desktop/src/app/chat/sidebar/index.tsx b/apps/desktop/src/app/chat/sidebar/index.tsx index 611a5a4d4554..a20b52e70209 100644 --- a/apps/desktop/src/app/chat/sidebar/index.tsx +++ b/apps/desktop/src/app/chat/sidebar/index.tsx @@ -26,7 +26,6 @@ import { useContributions } from '@/contrib/react/use-contributions' import { searchSessions, type SessionInfo, type SessionSearchResult } from '@/hermes' import { useI18n } from '@/i18n' import { comboTokens } from '@/lib/keybinds/combo' -import { resolveProfileColor } from '@/lib/profile-color' import { sessionMatchesSearch } from '@/lib/session-search' import { normalizeSessionSource, sessionSourceLabel } from '@/lib/session-source' import { cn } from '@/lib/utils' @@ -75,7 +74,6 @@ import { import { notifyError } from '@/store/notifications' import { $newChatProfile, - $profileColors, $profiles, $profileScope, ALL_PROFILES, @@ -149,6 +147,7 @@ import { type NewSessionSplitHandler, startNewSessionDrag } from '../new-session import { SidebarSectionAddButton } from './chrome' import { SidebarCronJobsSection } from './cron-jobs-section' import { SidebarFilterMenu } from './filter-menu' +import { useGatewaySessionGroups } from './gateway-group-model' import { SidebarLoadMoreRow } from './load-more-row' import { orderByIds, reconcileOrderIds, resolveManualSessionOrderIds, sameIds } from './order' import { filterSessionsByProfileScope } from './profile-scope' @@ -168,7 +167,6 @@ import { sessionMatchesProjectFilter, sessionRecency as sessionTime, type SidebarProjectTree, - type SidebarSessionGroup, type SidebarWorkspaceTree, sortProjectsForOverview, StartWorkButton, @@ -412,7 +410,6 @@ export function ChatSidebar({ const sessionProfilesTruncated = useStore($sessionProfilesTruncated) const unreadCount = useStore($unreadFinishedSessionIds).length const profiles = useStore($profiles) - const profileColors = useStore($profileColors) const profileScope = useStore($profileScope) const activeConnectionId = useStore($activeConnectionId) @@ -1268,41 +1265,7 @@ export function ChatSidebar({ .sort((a, b) => sessionTime(b.sessions[0]) - sessionTime(a.sessions[0])) }, [visibleMessagingSessions, messagingPlatformTotals, messagingTruncated, isPinnedSession, messagingProfile]) - // Grouping by profile: one collapsible group per profile, color on the header - // (not on every row). Default profile floats to the top, the rest alpha. - // Only reachable while the sidebar is showing every profile — scoped to one, - // it would draw a single group around the whole list. - const profileGrouped = showAllProfiles && grouping === 'profile' - - const profileGroups = useMemo(() => { - if (!profileGrouped) { - return undefined - } - - const groups = new Map() - - for (const session of agentSessions) { - const key = normalizeProfileKey(session.profile) - - const group = groups.get(key) ?? { - color: resolveProfileColor(key, profileColors), - id: key, - label: key, - mode: 'profile', - path: null, - sessions: [] - } - - group.sessions.push(session) - - groups.set(key, group) - } - - // default (root) first, then the rest alphabetically. - return [...groups.values()].sort((a, b) => - a.id === 'default' ? -1 : b.id === 'default' ? 1 : a.label.localeCompare(b.label) - ) - }, [profileGrouped, agentSessions, profileColors]) + const profileGroups = useGatewaySessionGroups(agentSessions, profileScope === ALL_PROFILES && grouping === 'profile') // The flat Sessions list always shows ALL recent sessions; Projects is a // parallel grouped view, not a filter on this one — nothing is hidden here. diff --git a/apps/desktop/src/app/chat/sidebar/projects/workspace-groups.ts b/apps/desktop/src/app/chat/sidebar/projects/workspace-groups.ts index 6a6d8aa8d1d1..3f457c885e39 100644 --- a/apps/desktop/src/app/chat/sidebar/projects/workspace-groups.ts +++ b/apps/desktop/src/app/chat/sidebar/projects/workspace-groups.ts @@ -30,6 +30,9 @@ export interface SidebarSessionGroup { isKanban?: boolean mode?: 'profile' | 'source' | 'workspace' sourceId?: string + // Exact owner for gateway/profile sidebar sections; absent for workspace lanes. + connectionId?: null | string + profile?: string } /** A repo node: holds its branch/worktree lanes (`repo -> lane -> sessions`). */ diff --git a/apps/desktop/src/app/chat/sidebar/sessions-section.tsx b/apps/desktop/src/app/chat/sidebar/sessions-section.tsx index 1104a8d2abd4..5e1aeb303aa8 100644 --- a/apps/desktop/src/app/chat/sidebar/sessions-section.tsx +++ b/apps/desktop/src/app/chat/sidebar/sessions-section.tsx @@ -30,6 +30,7 @@ import { sessionPinId } from '@/store/session' import { $sessionDotStateById, hasLiveTurn } from '@/store/session-dot-state' import { SidebarDateDivider, SidebarSectionMeta } from './chrome' +import { GatewayProfileGroups } from './gateway-groups' import { mergeVisibleReorder, orderRowsWithinGroups, reorderableRowIds } from './order' import { EnteredProjectContent, @@ -524,8 +525,9 @@ export function SidebarSessionsSection({ )} ) + } else if (groups?.length && groups.every(group => group.mode === 'profile' && group.profile)) { + inner = } else if (groups?.length) { - // Profile/source groups never reorder; render them flat with static rows. inner = groups.map(group => ( searchAria: string searchPlaceholder: string diff --git a/apps/desktop/src/i18n/zh-hant.ts b/apps/desktop/src/i18n/zh-hant.ts index 7818f9ba6b4a..0b754c5e50b6 100644 --- a/apps/desktop/src/i18n/zh-hant.ts +++ b/apps/desktop/src/i18n/zh-hant.ts @@ -1970,6 +1970,17 @@ export const zhHant = defineLocale({ }, sidebar: { + gatewayGroups: { + grouping: '閘道與設定檔', + rename: '重新命名群組', + aliasLabel: '顯示名稱', + aliasHint: '僅變更顯示名稱；閘道和設定檔名稱維持不變。', + resetName: '重設名稱', + moveUp: '上移', + moveDown: '下移', + reorder: '調整群組順序', + actions: '群組動作' + }, nav: { 'new-session': '新工作階段', skills: '技能與工具', diff --git a/apps/desktop/src/i18n/zh.ts b/apps/desktop/src/i18n/zh.ts index 2298c3a0d67c..84283ff0436e 100644 --- a/apps/desktop/src/i18n/zh.ts +++ b/apps/desktop/src/i18n/zh.ts @@ -2533,6 +2533,17 @@ export const zh: Translations = { }, sidebar: { + gatewayGroups: { + grouping: '网关与配置', + rename: '重命名分组', + aliasLabel: '显示名称', + aliasHint: '仅更改显示名称；网关和配置档名称保持不变。', + resetName: '重置名称', + moveUp: '上移', + moveDown: '下移', + reorder: '调整分组顺序', + actions: '分组操作' + }, nav: { 'new-session': '新建会话', skills: '技能与工具', From 6909604f971679603b0071e13eef345d4c421def Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Tue, 8 Sep 2026 04:06:08 -0700 Subject: [PATCH 187/227] fix(desktop): reuse saved Cloud gateways from Settings Separate saved Cloud instances from the live window source, use the existing registry activation path instead of repeating sign-in/apply, and persist the chosen instance name while retaining registry identity and custom labels. Naming metadata adapted from IAvecilla's contribution in PR #103224. Co-authored-by: IAvecilla --- .../electron/connection-registry.test.ts | 64 +++++++++++ apps/desktop/electron/connection-registry.ts | 49 ++++++-- apps/desktop/electron/main.ts | 34 +++++- .../app/settings/gateway-settings.test.tsx | 83 +++++++++++++- .../src/app/settings/gateway-settings.tsx | 106 +++++++++++++++--- apps/desktop/src/global.d.ts | 1 + apps/desktop/src/i18n/en.ts | 5 + apps/desktop/src/i18n/ru.ts | 5 + apps/desktop/src/i18n/types.ts | 4 + apps/desktop/src/i18n/zh.ts | 4 + 10 files changed, 325 insertions(+), 30 deletions(-) diff --git a/apps/desktop/electron/connection-registry.test.ts b/apps/desktop/electron/connection-registry.test.ts index d92428af6935..438e91461506 100644 --- a/apps/desktop/electron/connection-registry.test.ts +++ b/apps/desktop/electron/connection-registry.test.ts @@ -48,6 +48,70 @@ function emptyRegistry(): ConnectionRegistry { return normalizeRegistry(null) } +test('Cloud apply upgrades only host labels without changing connection identity', () => { + const url = 'https://agent.example.com' + const name = 'Research cloud' + + const first = reconcileAppliedGlobalConnection(emptyRegistry(), { + mode: 'cloud', + remote: { url, authMode: 'oauth' } + }) + + const named = reconcileAppliedGlobalConnection(first, { + mode: 'cloud', + remote: { url, authMode: 'oauth', name } + }) + + assert.equal(named.primary, first.primary) + assert.equal(named.connections.find(c => c.id === named.primary)?.label, name) + const restored = normalizeRegistry(JSON.parse(JSON.stringify(named))) + assert.equal(restored.connections.find(c => c.id === named.primary)?.name, name) + + const custom = upsertConnection(restored, { + ...restored.connections.find(c => c.id === named.primary)!, + label: 'My device' + }) + + const reapplied = reconcileAppliedGlobalConnection(custom, { + mode: 'cloud', + remote: { url, authMode: 'oauth', name: 'New portal name' } + }) + + assert.equal(reapplied.primary, first.primary) + assert.equal(reapplied.connections.find(c => c.id === first.primary)?.label, 'My device') + + const other = reconcileAppliedGlobalConnection(reapplied, { + mode: 'cloud', + remote: { url: 'https://other.example.com', authMode: 'oauth' } + }) + + assert.notEqual(other.primary, first.primary) + assert.equal(other.connections.find(c => c.id === other.primary)?.name, undefined) +}) + +test('Cloud name survives partial edits but never inherits across gateway URLs', () => { + const registry = reconcileAppliedGlobalConnection(emptyRegistry(), { + mode: 'cloud', + remote: { url: 'https://agent.example.com', authMode: 'oauth', name: 'Research cloud' } + }) + + const existing = registry.connections.find(c => c.id === registry.primary)! + + const renamed = normalizeConnectionInput( + mergeConnectionInput({ id: existing.id, kind: 'cloud', label: 'Mine' }, existing), + registry + ) + + assert.equal(renamed.name, 'Research cloud') + + const retargeted = normalizeConnectionInput( + mergeConnectionInput({ id: existing.id, kind: 'cloud', label: 'Mine', url: 'https://other.example.com' }, existing), + registry + ) + + assert.equal(retargeted.name, undefined) +}) + // --- labels, slugs, handles --- test('labelKey is case-insensitive and trimmed', () => { diff --git a/apps/desktop/electron/connection-registry.ts b/apps/desktop/electron/connection-registry.ts index a6a22c3d90a3..72f52cf5b0e9 100644 --- a/apps/desktop/electron/connection-registry.ts +++ b/apps/desktop/electron/connection-registry.ts @@ -64,6 +64,8 @@ export interface RegistryConnection { headers?: Record /** cloud: portal org slug/id the instance was discovered under. */ org?: string + /** Cloud instance name, separate from the user-editable label. */ + name?: string /** ssh fields (normalizeSshConfig shapes). */ host?: string user?: string @@ -822,6 +824,8 @@ export interface ConnectionInput { token?: unknown headers?: Record org?: string + /** Cloud instance name, separate from the user-editable label. */ + name?: string host?: string user?: string port?: number | string @@ -958,6 +962,12 @@ export function normalizeConnectionInput(input: ConnectionInput, registry: Conne } } + const name = String(input.name || '').trim() + + if (kind === 'cloud' && name) { + entry.name = name + } + const org = String(input.org || '').trim() if (kind === 'cloud' && org) { @@ -995,6 +1005,14 @@ export function mergeConnectionInput(input: ConnectionInput, existing?: null | R inherit('url') inherit('authMode') inherit('org') + + if ( + input.kind === 'cloud' && + (input.url === undefined || normalizeRemoteBaseUrl(input.url) === normalizeRemoteBaseUrl(existing.url)) + ) { + inherit('name') + } + inherit('host') inherit('keyPath') inherit('remoteHermesPath') @@ -1184,6 +1202,12 @@ export function normalizeRegistry(raw: unknown): ConnectionRegistry { clean.headers = storedHeaders } + const name = String(entry.name || '').trim() + + if (kind === 'cloud' && name) { + clean.name = name + } + const org = String(entry.org || '').trim() if (kind === 'cloud' && org) { @@ -1295,6 +1319,12 @@ export function migrateV1ToRegistry(v1: unknown): ConnectionRegistry { entry.headers = v1Headers } + const name = String(block.name || '').trim() + + if (kind === 'cloud' && name) { + entry.name = name + } + const org = String(block.org || '').trim() if (kind === 'cloud' && org) { @@ -1442,7 +1472,7 @@ export function setLastUsedConnection(registry: ConnectionRegistry, id: string): * * Remote-shaped entries are matched by normalized URL across remote/cloud so * changing provenance never duplicates a gateway. Existing identity and - * user-chosen label win; a new entry derives both from the host. Switching to + * user-chosen label win; a Cloud name upgrades only the default host label. Switching to * local keeps registered remotes available while moving primary/last-used * back to This device. */ @@ -1479,12 +1509,16 @@ export function reconcileAppliedGlobalConnection( const kind: ConnectionKind = mode === 'cloud' ? 'cloud' : 'remote' + const hostLabel = hostLabelFromBaseUrl(url) || (kind === 'cloud' ? 'Hermes Cloud' : 'Remote gateway') + const name = kind === 'cloud' ? String(block.name ?? existing?.name ?? '').trim() : '' + const label = - existing?.label || - uniqueLabel( - hostLabelFromBaseUrl(url) || (kind === 'cloud' ? 'Hermes Cloud' : 'Remote gateway'), - registry.connections.map(connection => connection.label) - ) + existing && (!name || existing.label !== hostLabel) + ? existing.label + : uniqueLabel( + name || hostLabel, + registry.connections.filter(connection => connection.id !== existing?.id).map(connection => connection.label) + ) const entry = normalizeConnectionInput( { @@ -1495,7 +1529,8 @@ export function reconcileAppliedGlobalConnection( authMode: block.authMode, token: block.token, headers: block.headers, - org: block.org + org: block.org, + name }, registry ) diff --git a/apps/desktop/electron/main.ts b/apps/desktop/electron/main.ts index dc669cb15ce4..56342dfb5c2d 100644 --- a/apps/desktop/electron/main.ts +++ b/apps/desktop/electron/main.ts @@ -9304,6 +9304,7 @@ function sanitizeConnectionProfiles(raw: Record) { token?: object headers?: object org?: string + name?: string savedSsh?: object } = { mode: modeIsRemoteLike(entry.mode) ? entry.mode : 'local' @@ -9338,6 +9339,12 @@ function sanitizeConnectionProfiles(raw: Record) { // Preserve the Hermes Cloud org tag on cloud-mode entries so Settings can // reopen into the same org for a per-profile cloud connection. if (cleaned.mode === 'cloud') { + const cloudName = String(entry.name || '').trim() + + if (cloudName) { + cleaned.name = cloudName + } + const org = String(entry.org || '').trim() if (org) { @@ -9865,12 +9872,12 @@ async function sanitizeDesktopConnectionConfig(config = readDesktopConnectionCon // `org` (optional) is the Hermes Cloud org slug/id the instance was discovered // under — persisted so Settings can reopen into the same org; omitted from the // block when empty so plain remote connections stay unchanged. -function buildRemoteBlock(remoteUrl, authMode, token, org?: string, headers?: object) { +function buildRemoteBlock(remoteUrl, authMode, token, org?: string, headers?: object, name?: string) { if (authMode !== 'oauth' && !decryptDesktopSecret(token)) { throw new Error('Remote gateway session token is required.') } - const block: { url: string; authMode: string; token: object; headers?: object; org?: string } = { + const block: { url: string; authMode: string; token: object; headers?: object; org?: string; name?: string } = { url: normalizeRemoteBaseUrl(remoteUrl), authMode, token @@ -9882,6 +9889,12 @@ function buildRemoteBlock(remoteUrl, authMode, token, org?: string, headers?: ob block.headers = remoteHeaders } + const nameValue = typeof name === 'string' ? name.trim() : '' + + if (nameValue) { + block.name = nameValue + } + const orgValue = typeof org === 'string' ? org.trim() : '' if (orgValue) { @@ -9919,6 +9932,19 @@ function coerceDesktopConnectionConfig(input: any = {}, existing = readDesktopCo // inherit the saved org. A plain 'remote' connection never carries an org // (switching cloud→remote drops it), so it stays unset unless mode is cloud. const cloudOrg = mode === 'cloud' ? String(input.cloudOrg ?? existingBlock.org ?? '').trim() : '' + + // A saved name belongs to this exact gateway, not another instance in the same org. + const cloudName = + mode === 'cloud' + ? String( + input.cloudName ?? + (existingBlock.url && normalizeRemoteBaseUrl(remoteUrl) === normalizeRemoteBaseUrl(existingBlock.url) + ? existingBlock.name + : '') ?? + '' + ).trim() + : '' + const incomingToken = typeof input.remoteToken === 'string' ? input.remoteToken.trim() : '' const remoteHeaders = @@ -9962,7 +9988,7 @@ function coerceDesktopConnectionConfig(input: any = {}, existing = readDesktopCo if (remoteLike) { profiles[key] = { mode, - ...buildRemoteBlock(remoteUrl, authMode, nextToken, cloudOrg, remoteHeaders) + ...buildRemoteBlock(remoteUrl, authMode, nextToken, cloudOrg, remoteHeaders, cloudName) } } else { const localEntry = localProfileEntry(rawExistingBlock) @@ -9982,7 +10008,7 @@ function coerceDesktopConnectionConfig(input: any = {}, existing = readDesktopCo } const nextRemote = remoteLike - ? buildRemoteBlock(remoteUrl, authMode, nextToken, cloudOrg, remoteHeaders) + ? buildRemoteBlock(remoteUrl, authMode, nextToken, cloudOrg, remoteHeaders, cloudName) : existingMode === 'ssh' ? rawExistingBlock : { url: remoteUrl ? normalizeRemoteBaseUrl(remoteUrl) : remoteUrl, authMode, token: nextToken } diff --git a/apps/desktop/src/app/settings/gateway-settings.test.tsx b/apps/desktop/src/app/settings/gateway-settings.test.tsx index 7a2faf1d350e..a42a6ed77f0c 100644 --- a/apps/desktop/src/app/settings/gateway-settings.test.tsx +++ b/apps/desktop/src/app/settings/gateway-settings.test.tsx @@ -1,9 +1,27 @@ -import { cleanup, render, screen, waitFor } from '@testing-library/react' +import { cleanup, fireEvent, render, screen, waitFor, within } from '@testing-library/react' import { afterEach, beforeEach, describe, expect, it, vi } from 'vitest' // Collect the component graph before the behavioral test deadline starts. import { GatewaySettings } from './gateway-settings' +const { registry, activeId, selectConnection } = vi.hoisted(() => ({ + registry: { value: null as any }, + activeId: { value: 'saved-b' }, + selectConnection: vi.fn().mockResolvedValue(undefined) +})) + +vi.mock('@nanostores/react', () => ({ useStore: (store: any) => store.value })) +vi.mock('@/store/connections', () => ({ + $connectionsRegistry: registry, + $activeConnectionId: activeId, + refreshConnectionsRegistry: vi.fn().mockResolvedValue(null), + selectConnection, + setConnectionsRegistry: vi.fn() +})) +vi.mock('./connections-registry', async importOriginal => ({ + ...(await importOriginal()), + ConnectionsRegistrySection: () => null +})) const getConnectionConfig = vi.fn() const saveConnectionConfig = vi.fn() @@ -39,6 +57,69 @@ afterEach(() => { }) describe('GatewaySettings', () => { + it('keeps saved Cloud instances usable without discovery and marks the live source, not the default', async () => { + getConnectionConfig.mockResolvedValue({ ...localConnection, mode: 'cloud', remoteUrl: 'https://a.example' }) + registry.value = { + connections: [ + { id: 'saved-a', kind: 'cloud', label: 'Research', url: 'https://a.example', authMode: 'oauth' }, + { id: 'saved-b', kind: 'cloud', label: 'Writing', url: 'https://b.example', authMode: 'oauth' } + ] + } + const agentSignIn = vi.fn() + const applyConnectionConfig = vi.fn() + Object.assign(window.hermesDesktop, { + applyConnectionConfig, + cloud: { + status: vi.fn().mockResolvedValue({ signedIn: false }), + agentSignIn + } + }) + render() + const research = await screen.findByText('Research') + const row = research.closest('[data-slot]') ?? research.parentElement!.parentElement! + fireEvent.click(within(row as HTMLElement).getByRole('button', { name: 'Use gateway' })) + await waitFor(() => expect(selectConnection).toHaveBeenCalledWith('saved-a')) + expect(screen.getByText('Active in this window')).toBeTruthy() + expect(agentSignIn).not.toHaveBeenCalled() + expect(applyConnectionConfig).not.toHaveBeenCalled() + registry.value = null + }) + it('authenticates and saves only the chosen discovered instance with its friendly name', async () => { + registry.value = null + getConnectionConfig.mockResolvedValue({ ...localConnection, mode: 'cloud' }) + const agentSignIn = vi.fn().mockResolvedValue({ connected: true }) + const applyConnectionConfig = vi.fn().mockResolvedValue({ ...localConnection, mode: 'cloud' }) + Object.assign(window.hermesDesktop, { + applyConnectionConfig, + cloud: { + status: vi.fn().mockResolvedValue({ signedIn: true }), + agentSignIn, + discover: vi.fn().mockResolvedValue({ + agents: [ + { id: 'new-a', name: 'Research Bot', dashboardUrl: 'https://new-a.example' }, + { id: 'new-b', name: 'Writing Bot', dashboardUrl: 'https://new-b.example' } + ], + org: { id: 'org-a' } + }) + } + }) + render() + const buttons = await screen.findAllByRole('button', { name: 'Connect', exact: true }) + expect(agentSignIn).not.toHaveBeenCalled() + expect(applyConnectionConfig).not.toHaveBeenCalled() + fireEvent.click(buttons[0]) + await waitFor(() => + expect(applyConnectionConfig).toHaveBeenCalledWith({ + mode: 'cloud', + remoteAuthMode: 'oauth', + remoteUrl: 'https://new-a.example', + cloudOrg: 'org-a', + cloudName: 'Research Bot' + }) + ) + expect(agentSignIn).toHaveBeenCalledExactlyOnceWith('https://new-a.example') + expect(applyConnectionConfig).toHaveBeenCalledTimes(1) + }) it('loads the machine-level connection config (no profile scoping)', async () => { render() expect(await screen.findByText('Local gateway')).toBeTruthy() diff --git a/apps/desktop/src/app/settings/gateway-settings.tsx b/apps/desktop/src/app/settings/gateway-settings.tsx index 4de24c278c53..874e5e0b4c83 100644 --- a/apps/desktop/src/app/settings/gateway-settings.tsx +++ b/apps/desktop/src/app/settings/gateway-settings.tsx @@ -1,3 +1,4 @@ +import { useStore } from '@nanostores/react' import { useEffect, useMemo, useRef, useState } from 'react' import { Button } from '@/components/ui/button' @@ -24,6 +25,12 @@ import { import { coerceRemoteUrlScheme } from '@/lib/remote-url' import { selectableCardClass } from '@/lib/selectable-card' import { cn } from '@/lib/utils' +import { + $activeConnectionId, + $connectionsRegistry, + refreshConnectionsRegistry, + selectConnection +} from '@/store/connections' import { notify, notifyError, readableError } from '@/store/notifications' import { ConnectionsRegistrySection } from './connections-registry' @@ -169,7 +176,13 @@ export function GatewaySettings({ embedded = false }: { embedded?: boolean } = { const signingSeq = useRef(0) const cloudConnectSeq = useRef(0) const contextSeq = useRef(0) - const [connectedCloudUrl, setConnectedCloudUrl] = useState('') + const registry = useStore($connectionsRegistry) + const activeConnectionId = useStore($activeConnectionId) + const savedCloudConnections = registry?.connections.filter(connection => connection.kind === 'cloud') ?? [] + + useEffect(() => { + void refreshConnectionsRegistry().catch(err => notifyError(err, g.failedLoad)) + }, [g.failedLoad]) // Opt-in OS-keychain encryption for stored gateway secrets. Read lazily via // IPC (never touches the keychain); flipping it re-encodes stored secrets @@ -215,7 +228,6 @@ export function GatewaySettings({ embedded = false }: { embedded?: boolean } = { const normalized = normalizeGatewaySettingsState(config) setState(normalized) - setConnectedCloudUrl(savedCloudConnectionUrl(normalized)) } // When set, the plain-text opt-in dialog is open; `apply` remembers whether @@ -294,19 +306,29 @@ export function GatewaySettings({ embedded = false }: { embedded?: boolean } = { // prefers a fresh probe result over the saved value. const trimmedUrl = coerceRemoteUrlScheme(state.remoteUrl) - // The dashboardUrl of the currently-connected cloud instance (the saved - // cloud connection's remoteUrl), normalized for comparison against each - // discovered agent's dashboardUrl so we can highlight the active one and hide - // its Connect button. Empty unless the saved connection is a cloud one. - // The saved cloud URL was stored via the main-side normalizeRemoteBaseUrl - // (which lowercases the host through URL.toString()), but a discovered agent's - // dashboardUrl arrives raw from NAS — so normalize both sides the same way - // (trim, drop trailing slash, lowercase) or a host-casing difference would - // silently break the connected-highlight. - const normalizeCloudUrl = (url: string) => url.trim().replace(/\/+$/, '').toLowerCase() + const savedAgent = (agent: DesktopCloudAgent) => + registry?.connections.find( + connection => + (connection.kind === 'cloud' || connection.kind === 'remote') && + connection.url && + agent.dashboardUrl && + savedCloudConnectionUrl({ mode: 'cloud', remoteUrl: connection.url }) === + savedCloudConnectionUrl({ mode: 'cloud', remoteUrl: agent.dashboardUrl }) + ) - const isConnectedAgent = (agent: DesktopCloudAgent) => - Boolean(connectedCloudUrl && agent.dashboardUrl && normalizeCloudUrl(agent.dashboardUrl) === connectedCloudUrl) + const isConnectedAgent = (agent: DesktopCloudAgent) => savedAgent(agent)?.id === activeConnectionId + + const activateSavedCloud = async (id: string) => { + setCloudConnectingId(id) + + try { + await selectConnection(id) + } catch (err) { + notifyError(err, g.cloudConnectFailed) + } finally { + setCloudConnectingId(null) + } + } useEffect(() => { if (state.mode !== 'remote' || !trimmedUrl || !/^https?:\/\//i.test(trimmedUrl)) { @@ -887,6 +909,16 @@ export function GatewaySettings({ embedded = false }: { embedded?: boolean } = { setCloudConnectingId(agent.id) try { + // Saved sources keep their identity, credentials and default gateway. + // The activation path reuses healthy sockets and validates auth on a new dial. + const saved = savedAgent(agent) + + if (saved) { + await selectConnection(saved.id) + + return + } + const result = await desktop.cloud.agentSignIn(agent.dashboardUrl) if (seq !== contextSeq.current) { @@ -911,7 +943,8 @@ export function GatewaySettings({ embedded = false }: { embedded?: boolean } = { mode: 'cloud', remoteAuthMode: 'oauth', remoteUrl: agent.dashboardUrl, - cloudOrg: cloudOrgRef.current ?? undefined + cloudOrg: cloudOrgRef.current ?? undefined, + cloudName: agent.name }) if (seq !== contextSeq.current) { @@ -919,6 +952,7 @@ export function GatewaySettings({ embedded = false }: { embedded?: boolean } = { } acceptSavedConfig(next) + await refreshConnectionsRegistry() notify({ kind: 'success', title: g.cloudConnectedTitle, message: g.cloudConnectedTo(agent.name) }) } catch (err) { if (seq !== contextSeq.current) { @@ -1147,6 +1181,40 @@ export function GatewaySettings({ embedded = false }: { embedded?: boolean } = { connection. Replaces the URL/token form while in cloud mode. */} {state.mode === 'cloud' && !state.envOverride ? (
+ {savedCloudConnections.length > 0 ? ( +
+
+ {g.cloudSavedTitle} +
+

{g.cloudSavedDesc}

+ {savedCloudConnections.map(connection => ( +
+ + + {g.cloudActive} + + ) : ( + + ) + } + description={connection.url} + title={connection.label} + /> +
+ ))} +
+ ) : null} - {g.cloudConnectedPill} + {g.cloudActive} ) : (
) diff --git a/apps/desktop/src/global.d.ts b/apps/desktop/src/global.d.ts index bd02af438c96..641d7bfb6e98 100644 --- a/apps/desktop/src/global.d.ts +++ b/apps/desktop/src/global.d.ts @@ -862,6 +862,7 @@ export interface DesktopConnectionConfigInput { // For a 'cloud' connection: the selected Hermes Cloud org (slug or id) to // persist so Settings can reopen into it. Ignored for remote/local modes. cloudOrg?: string + cloudName?: string sshHost?: string sshUser?: string sshPort?: number | null diff --git a/apps/desktop/src/i18n/en.ts b/apps/desktop/src/i18n/en.ts index 24e2ebf0d884..4fe8bcb34fc2 100644 --- a/apps/desktop/src/i18n/en.ts +++ b/apps/desktop/src/i18n/en.ts @@ -922,6 +922,11 @@ export const en: Translations = { }, cloudRefresh: 'Refresh', cloudConnect: 'Connect', + cloudSavedTitle: 'Saved Cloud gateways', + cloudSavedDesc: + 'Use a saved gateway without changing your default. Sign in below to add instances. Manage names and sign-in in the saved connections list.', + cloudUseSaved: 'Use gateway', + cloudActive: 'Active in this window', cloudConnecting: 'Connecting…', cloudDiscoverFailed: 'Could not load your Hermes Cloud agents', cloudConnectFailed: 'Could not connect to that agent', diff --git a/apps/desktop/src/i18n/ru.ts b/apps/desktop/src/i18n/ru.ts index 554b148f217c..8e7afb5de4d7 100644 --- a/apps/desktop/src/i18n/ru.ts +++ b/apps/desktop/src/i18n/ru.ts @@ -1135,6 +1135,11 @@ export const ru = defineLocale({ }, cloudRefresh: 'Обновить', cloudConnect: 'Подключиться', + cloudSavedTitle: 'Сохранённые облачные шлюзы', + cloudSavedDesc: + 'Используйте сохранённый шлюз без изменения шлюза по умолчанию. Войдите ниже, чтобы добавить экземпляры. Имена и вход — в списке сохранённых подключений.', + cloudUseSaved: 'Использовать шлюз', + cloudActive: 'Активен в этом окне', cloudConnecting: 'Подключение…', cloudDiscoverFailed: 'Не удалось загрузить агентов Hermes Cloud', cloudConnectFailed: 'Не удалось подключиться к этому агенту', diff --git a/apps/desktop/src/i18n/types.ts b/apps/desktop/src/i18n/types.ts index 2f3e9a515903..ae869d34352a 100644 --- a/apps/desktop/src/i18n/types.ts +++ b/apps/desktop/src/i18n/types.ts @@ -789,6 +789,10 @@ export interface Translations { cloudNoAgents: { before: string; linkText: string; after: string } cloudRefresh: string cloudConnect: string + cloudSavedTitle: string + cloudSavedDesc: string + cloudUseSaved: string + cloudActive: string cloudConnecting: string cloudDiscoverFailed: string cloudConnectFailed: string diff --git a/apps/desktop/src/i18n/zh.ts b/apps/desktop/src/i18n/zh.ts index 84283ff0436e..2fc5561b553e 100644 --- a/apps/desktop/src/i18n/zh.ts +++ b/apps/desktop/src/i18n/zh.ts @@ -1117,6 +1117,10 @@ export const zh: Translations = { }, cloudRefresh: '刷新', cloudConnect: '连接', + cloudSavedTitle: '已保存的云网关', + cloudSavedDesc: '使用已保存的网关，不更改默认网关。在下方登录以添加实例。在已保存的连接列表中管理名称和登录。', + cloudUseSaved: '使用网关', + cloudActive: '当前窗口正在使用', cloudConnecting: '正在连接…', cloudDiscoverFailed: '无法加载你的 Hermes Cloud 智能体', cloudConnectFailed: '无法连接到该智能体', From f6449ec37c9d0c8e97e54c753e1d32cf4a223ab2 Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Tue, 8 Sep 2026 04:22:55 -0700 Subject: [PATCH 188/227] docs: explain gateway session groups and saved Cloud switching --- .../user-guide/multi-connection-desktop.md | 21 +++++++++++++++++++ 1 file changed, 21 insertions(+) diff --git a/website/docs/user-guide/multi-connection-desktop.md b/website/docs/user-guide/multi-connection-desktop.md index 9746d02cdc58..d1d6977e0b3f 100644 --- a/website/docs/user-guide/multi-connection-desktop.md +++ b/website/docs/user-guide/multi-connection-desktop.md @@ -81,6 +81,27 @@ cron stay scoped to that gateway; the app-managed window backend is still chosen by the connection-mode controls above. **Primary** is the registry fallback and does not switch the current workspace. +## Organizing session groups + +In the Sessions sidebar's view menu, choose **Gateway & profile** while viewing +all profiles. Each gateway/profile pair gets its own collapsible section, so two +gateways with a `default` profile no longer share a section. Section labels start +as the gateway name followed by the profile name. + +Use a section's menu to **Rename group**, **Reset name**, **Move up**, or +**Move down**. Renaming changes only the sidebar label, not the gateway or profile. +Drag the section's leading icon to reorder it, or focus that handle and use +Space, arrow keys, then Space to place it. Names, order, and collapsed sections +are remembered on this desktop. The section's new-session action targets that +section's gateway and profile. + +The Hermes Cloud panel also lists **Saved Cloud gateways** when portal discovery +is signed out. **Use gateway** selects an existing saved connection without +changing the default gateway; **Active in this window** identifies the current +one. Adding a new instance uses its friendly Cloud name, while existing custom +connection names are preserved. Saved connections still need valid gateway +authentication; manage sign-in from the registered connection controls. + ## Adding a connection, step by step 1. Open **Settings → Gateways** and scroll to the connections registry (or From a529ecfeb59493a05e78b9bfffa2bdfbf39bc150 Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Tue, 8 Sep 2026 04:49:10 -0700 Subject: [PATCH 189/227] feat(desktop): nest profile sessions under gateway sections --- .../app/chat/sidebar/gateway-groups.test.tsx | 31 +++- .../src/app/chat/sidebar/gateway-groups.tsx | 143 +++++++++++++----- .../user-guide/multi-connection-desktop.md | 15 +- 3 files changed, 141 insertions(+), 48 deletions(-) diff --git a/apps/desktop/src/app/chat/sidebar/gateway-groups.test.tsx b/apps/desktop/src/app/chat/sidebar/gateway-groups.test.tsx index c53143c270cd..4176503eabc4 100644 --- a/apps/desktop/src/app/chat/sidebar/gateway-groups.test.tsx +++ b/apps/desktop/src/app/chat/sidebar/gateway-groups.test.tsx @@ -83,18 +83,37 @@ it('keeps equal profile names on separate gateways and routes section creation a .closest('[data-gateway-group]') ?.getAttribute('data-gateway-group') ).toBe(JSON.stringify([null, 'default'])) - expect(screen.getByText('This computer · default')).toBeTruthy() - expect(screen.getByText('Homelab · default')).toBeTruthy() - expect(screen.getByText('Cloud workspace · default')).toBeTruthy() - fireEvent.click(screen.getByRole('button', { name: 'New session in Homelab · default' })) + act(() => + $sessions.set([ + ...$sessions.get(), + makeSessionInfo({ + id: 'remote-work', + connection_id: 'remote-1', + profile: 'work', + title: 'Work session', + last_active: Date.now() / 1000 + }) + ]) + ) + const gateway = screen.getByText('Homelab').closest('[data-gateway-section]') as HTMLElement + expect(within(gateway).getByText('default')).toBeTruthy() + expect(within(gateway).getByText('work')).toBeTruthy() + expect(gateway.querySelectorAll('[data-gateway-group]')).toHaveLength(2) + fireEvent.click(within(gateway).getAllByRole('button', { name: 'New session in default' })[0]) expect($newChatRoute.get()).toMatchObject({ connectionId: 'remote-1', profile: 'default' }) fireEvent.click(screen.getByText('cloud-1 session')) expect(resume).toHaveBeenLastCalledWith( 'cloud-1', expect.objectContaining({ connection_id: 'cloud-1', profile: 'default' }) ) - const group = screen.getByText('Homelab · default').closest('[data-gateway-group]')! - fireEvent.click(within(group as HTMLElement).getByRole('button', { name: 'Hide Homelab · default sessions' })) + const group = within(gateway).getByText('default').closest('[data-gateway-group]')! + fireEvent.click(within(group as HTMLElement).getByRole('button', { name: 'Hide default sessions' })) + expect(screen.queryByText('remote-1 session')).toBeNull() + expect(screen.getByText('Work session')).toBeTruthy() + fireEvent.click(within(gateway).getByRole('button', { name: 'Hide Homelab sessions' })) + expect(screen.queryByText('Work session')).toBeNull() + fireEvent.click(within(gateway).getByRole('button', { name: 'Show Homelab sessions' })) + expect(screen.getByText('Work session')).toBeTruthy() expect(screen.queryByText('remote-1 session')).toBeNull() expect(screen.getByText('local session')).toBeTruthy() }) diff --git a/apps/desktop/src/app/chat/sidebar/gateway-groups.tsx b/apps/desktop/src/app/chat/sidebar/gateway-groups.tsx index 07bc128a6a27..82281dcd4126 100644 --- a/apps/desktop/src/app/chat/sidebar/gateway-groups.tsx +++ b/apps/desktop/src/app/chat/sidebar/gateway-groups.tsx @@ -21,6 +21,7 @@ import { ProfileGlyph } from '@/components/ui/profile-glyph' import type { SessionInfo } from '@/hermes' import { useI18n } from '@/i18n' import { useStoreSelector } from '@/lib/use-session-slice' +import { $connectionsRegistry } from '@/store/connection-registry-state' import { newSessionInAgent, newSessionInProfile } from '@/store/profile' import { $sessionProfilesUsage } from '@/store/session' import { $sidebarSessionRankIds } from '@/store/sidebar-sort' @@ -45,12 +46,49 @@ interface GatewayProfileGroupsProps { renderRows: (sessions: SessionInfo[]) => ReactNode sensors?: ReturnType onNewSessionSplit?: NewSessionSplitHandler + nested?: boolean } -export function GatewayProfileGroups({ groups, renderRows, sensors, onNewSessionSplit }: GatewayProfileGroupsProps) { +export function GatewayProfileGroups({ + groups, + renderRows, + sensors, + onNewSessionSplit, + nested = false +}: GatewayProfileGroupsProps) { + const registry = useStore($connectionsRegistry) const order = useStore($gatewayGroupOrder) + const gatewayProfiles = new Map() + const sections: SidebarSessionGroup[] = [] - const ordered = [...groups].sort((a, b) => { + for (const group of groups) { + // Unknown legacy ownership stays unassigned; never guess a local gateway. + if (nested || !group.connectionId) { + sections.push(nested ? { ...group, label: group.profile! } : group) + + continue + } + + const id = JSON.stringify(['gateway', group.connectionId]) + const profiles = gatewayProfiles.get(id) + + if (profiles) { + profiles.push(group) + } else { + gatewayProfiles.set(id, [group]) + sections.push({ + id, + connectionId: group.connectionId, + label: + registry?.connections.find(connection => connection.id === group.connectionId)?.label || group.connectionId, + mode: 'profile', + path: null, + sessions: [] + }) + } + } + + const ordered = [...sections].sort((a, b) => { const left = order.indexOf(a.id) const right = order.indexOf(b.id) @@ -70,7 +108,19 @@ export function GatewayProfileGroups({ groups, renderRows, sensors, onNewSession onMove={direction => reorderGatewayGroups(arrayMove(ids, index, index + direction))} onNewSessionSplit={onNewSessionSplit} renderRows={renderRows} - /> + > + {gatewayProfiles.has(group.id) && ( +
+ +
+ )} + ))} ) @@ -83,9 +133,18 @@ interface GatewayProfileGroupProps { onNewSessionSplit?: NewSessionSplitHandler first: boolean last: boolean + children?: ReactNode } -function GatewayProfileGroup({ group, renderRows, onMove, first, last, onNewSessionSplit }: GatewayProfileGroupProps) { +function GatewayProfileGroup({ + group, + renderRows, + onMove, + first, + last, + onNewSessionSplit, + children +}: GatewayProfileGroupProps) { const { t } = useI18n() const s = t.sidebar const copy = s.gatewayGroups @@ -120,35 +179,42 @@ function GatewayProfileGroup({ group, renderRows, onMove, first, last, onNewSess } return ( - + - - startNewSessionDrag( - placement => { - if (!open) { - toggleGatewayGroup(group.id) - } - - onNewSessionSplit(placement.dir, { - anchor: placement.anchor, - before: placement.before, - profile: group.profile, - route - }) - }, - event, - { label: s.newSessionIn(label), profile: group.profile, route } - ) - : undefined - } - /> + {group.profile && ( + + startNewSessionDrag( + placement => { + if (!open) { + toggleGatewayGroup(group.id) + } + + onNewSessionSplit(placement.dir, { + anchor: placement.anchor, + before: placement.before, + profile: group.profile, + route + }) + }, + event, + { label: s.newSessionIn(label), profile: group.profile, route } + ) + : undefined + } + /> + )} +
+ ) : roster.length === 0 ? ( + + ) : allBotsHidden && !hiddenExpanded ? ( +
+
+ + {b.roster.allHidden} +
+

{b.roster.allHiddenDesc}

+ +
+ ) : rosterRows.length === 0 && matchingHiddenBots.length === 0 ? ( +
+ +
+ ) : ( +
+
+ {showGatewaySections + ? [ + sortedGroupRows.length ? renderGroupChatSection() : null, + ...gatewaySections.sections.map(renderGatewaySection) + ].filter(Boolean) + : renderUserSections(rosterRows)} + {showHiddenSection ? ( +
+ {hasRosterConstraint ? ( +
+ + Hidden + {matchingHiddenBots.length} +
+ ) : ( + $showHiddenBots.set(!hiddenExpanded)} + > + + Hidden + {hiddenBots.length} + + )} + {showHiddenRows ? ( + matchingHiddenBots.length ? ( + hiddenGatewaySections.sectioned ? ( + hiddenGatewaySections.sections.map(renderHiddenGatewaySection) + ) : ( + matchingHiddenBots.map((bot: RosterRow) => renderBotRow(bot, 'hidden:')) + ) + ) : ( +
{b.roster.noHiddenMatch}
+ ) + ) : null} +
+ ) : null} +
+
+ )} + + ) +} diff --git a/apps/desktop/src/plugins/hermes-bots/roster-pane-derivation.ts b/apps/desktop/src/plugins/hermes-bots/roster-pane-derivation.ts new file mode 100644 index 000000000000..d83f1677de18 --- /dev/null +++ b/apps/desktop/src/plugins/hermes-bots/roster-pane-derivation.ts @@ -0,0 +1,248 @@ +import { botActivitySession } from './data' +import { botRosterKey, filterBots, preferReachableSameNameRows } from './data' +import type { $groupChats } from './group-chat' +import { groupChatMemberBots, groupChatNames, groupLastActivity } from './group-membership' +import { isBotPinned } from './hidden-bots' +import { isBotHidden } from './hidden-bots' +import { filterBotsByGateway, groupMatchesRosterFilters, rosterGatewaySections } from './roster-sections' +import type { rosterGatewayOptions } from './roster-sections' +import { botRosterMeta } from './routing' +import { BOT_ROSTER_SEARCH_THRESHOLD } from './row-helpers' +import { ACTIVE_WINDOW_S, rosterActivityMatches } from './row-helpers' +import type { BotMeta, GroupMember, RosterActivityFilter, RosterKindFilter, RosterRow } from './types' +/** The two row shapes the roster sorts together — `kind` is the discriminant. */ +interface RosterBotRow { + active: boolean + activity: number + bot: RosterRow + kind: 'bot' + pinned: boolean +} +export interface RosterGroupRow { + active: boolean + activity: number + kind: 'group' + members: GroupMember[] + name: string + pinned: boolean +} + +interface RosterRowsInput { + roster: RosterRow[] + allMeta: Record + gatewayFilter: string + query: string + activityFilter: RosterActivityFilter + rowKindFilter: RosterKindFilter + groupRooms: ReturnType + activeRosterKeys: Set + gatewayOptions: ReturnType + activityOf: (bot: RosterRow) => number + isPinned: (bot: RosterRow) => boolean +} + +export function deriveRosterRows({ + roster, + allMeta, + gatewayFilter, + query, + activityFilter, + rowKindFilter, + groupRooms, + activeRosterKeys, + gatewayOptions, + activityOf, + isPinned +}: RosterRowsInput) { + const activeSourceRoster = roster.filter(bot => !bot.remoteSource) + // Hidden rows remain fully alive and recoverable at the bottom. Every + // non-display consumer continues to receive the complete roster. + const hiddenBots = roster.filter(bot => isBotHidden(bot, allMeta)) + const visibleRoster = roster.filter(bot => !isBotHidden(bot, allMeta)) + const gatewayRoster = filterBotsByGateway(visibleRoster, gatewayFilter) + + const filteredRoster = filterBots(gatewayRoster, allMeta, query).filter((bot: RosterRow) => + rosterActivityMatches( + { + activity: activityOf(bot), + active: activeRosterKeys.has(botRosterKey(bot)) + }, + activityFilter + ) + ) + + const filteredHiddenBots = filterBots(filterBotsByGateway(hiddenBots, gatewayFilter), allMeta, query).filter( + (bot: RosterRow) => + rosterActivityMatches( + { + activity: activityOf(bot), + active: activeRosterKeys.has(botRosterKey(bot)) + }, + activityFilter + ) + ) + + const groupNames = groupChatNames(allMeta, groupRooms) + + const groupRows = groupNames + .map(name => ({ + name, + members: groupChatMemberBots(name, roster, allMeta) + })) + .filter(row => groupMatchesRosterFilters(row.name, row.members, allMeta, query, gatewayFilter)) + .map((row): RosterGroupRow => ({ + kind: 'group', + name: row.name, + members: row.members, + pinned: Boolean(groupRooms[row.name]?.pinned), + activity: groupLastActivity(groupRooms[row.name]), + active: + Boolean( + groupLastActivity(groupRooms[row.name]) && + Date.now() - groupLastActivity(groupRooms[row.name]) <= ACTIVE_WINDOW_S * 1000 + ) || row.members.some(member => activeRosterKeys.has(botRosterKey(member))) + })) + .filter(row => rowKindFilter !== 'bots' && rosterActivityMatches(row, activityFilter)) + + const botRows = + rowKindFilter === 'groups' + ? [] + : preferReachableSameNameRows(filteredRoster).map((bot): RosterBotRow => ({ + kind: 'bot', + bot, + pinned: isPinned(bot), + activity: activityOf(bot), + active: activeRosterKeys.has(botRosterKey(bot)) + })) + + const sortRosterRows = (rows: T[]): T[] => + rows.slice().sort((a, b) => { + const pa = a.pinned ? 1 : 0 + const pb = b.pinned ? 1 : 0 + + if (pa !== pb) { + return pb - pa + } + + return b.activity - a.activity + }) + + const rosterRows = sortRosterRows([...botRows, ...groupRows]) + const sortedGroupRows = sortRosterRows(groupRows) + const gatewaySections = rosterGatewaySections(botRows, gatewayOptions, gatewayFilter) + const showGatewaySections = gatewaySections.sectioned && botRows.length > 0 + + return { + activeSourceRoster, + hiddenBots, + visibleRoster, + filteredHiddenBots, + groupNames, + rosterRows, + sortedGroupRows, + gatewaySections, + showGatewaySections + } +} + +interface RosterPresentationInput { + rowKindFilter: RosterKindFilter + activityFilter: RosterActivityFilter + gatewayFilter: string + query: string + filteredHiddenBots: RosterRow[] + hiddenBots: RosterRow[] + hiddenExpanded: boolean + roster: RosterRow[] + groupNames: string[] + visibleRoster: RosterRow[] + gatewayOptions: ReturnType +} + +export function deriveRosterPresentation({ + rowKindFilter, + activityFilter, + gatewayFilter, + query, + filteredHiddenBots, + hiddenBots, + hiddenExpanded, + roster, + groupNames, + visibleRoster, + gatewayOptions +}: RosterPresentationInput) { + const activeFilterCount = + (rowKindFilter === 'all' ? 0 : 1) + (activityFilter === 'all' ? 0 : 1) + (gatewayFilter === 'all' ? 0 : 1) + + const hasRosterConstraint = Boolean(query.trim()) || activeFilterCount > 0 + const matchingHiddenBots = rowKindFilter === 'groups' ? [] : filteredHiddenBots + const showHiddenSection = hiddenBots.length > 0 && (!hasRosterConstraint || matchingHiddenBots.length > 0) + const showHiddenRows = hiddenExpanded || hasRosterConstraint + const rosterItemCount = roster.length + groupNames.length + + const allBotsHidden = + !hasRosterConstraint && visibleRoster.length === 0 && groupNames.length === 0 && hiddenBots.length > 0 + + const showRosterSearch = + gatewayOptions.length > 1 || rosterItemCount >= BOT_ROSTER_SEARCH_THRESHOLD || Boolean(query.trim()) + + const showRosterFilters = + gatewayOptions.length > 1 || + groupNames.length > 0 || + rosterItemCount >= BOT_ROSTER_SEARCH_THRESHOLD || + activeFilterCount > 0 + + const showRosterTools = showRosterSearch || showRosterFilters + + const hiddenGatewaySections = rosterGatewaySections( + matchingHiddenBots.map((bot: RosterRow) => ({ + kind: 'bot', + bot + })), + gatewayOptions, + gatewayFilter + ) + + return { + activeFilterCount, + hasRosterConstraint, + matchingHiddenBots, + showHiddenSection, + showHiddenRows, + allBotsHidden, + showRosterSearch, + showRosterFilters, + showRosterTools, + hiddenGatewaySections + } +} + +export function sortRosterBots(sourceWithSelectedOwner: RosterRow[], allMeta: Record) { + // Messaging-app order: most recent activity first, where "activity" is + // the newest of (bot created, last message in any of its sessions). A + // freshly created bot tops the list until another bot gets a message. + // No special slot for the primary bot — it competes on recency too. + const activityOf = (bot: RosterRow): number => { + const created = botRosterMeta(bot, allMeta)?.created || bot.ui_meta?.['hermes-bots']?.created || 0 + const lastMsg = (botActivitySession(bot)?.last_active || 0) * 1000 + + return Math.max(created, lastMsg) + } + + // Pin is a source-qualified Desktop preference, not gateway profile state. + const isPinned = (bot: RosterRow): boolean => isBotPinned(bot, allMeta) + + const roster = sourceWithSelectedOwner.slice().sort((a, b) => { + const pa = isPinned(a) ? 1 : 0 + const pb = isPinned(b) ? 1 : 0 + + if (pa !== pb) { + return pb - pa + } + + return activityOf(b) - activityOf(a) + }) + + return { roster, activityOf, isPinned } +} diff --git a/apps/desktop/src/plugins/hermes-bots/roster-pane-dialogs.tsx b/apps/desktop/src/plugins/hermes-bots/roster-pane-dialogs.tsx new file mode 100644 index 000000000000..8e1f25b161b7 --- /dev/null +++ b/apps/desktop/src/plugins/hermes-bots/roster-pane-dialogs.tsx @@ -0,0 +1,161 @@ +import { ConfirmDialog, host } from '@hermes/plugin-sdk' +import type { useI18n } from '@hermes/plugin-sdk' + +import { CreateAgentDialog, CreateGroupChatDialog, GroupDialog } from './create-dialog' +import type { useRoster } from './data' +import { EditProfileDialog } from './edit-profile-dialog' +import { disbandGroupChat, openGroupChat } from './group-chat-view' +import type { useBots } from './i18n' +import { deleteBot } from './profile-ops' +import type { GroupMember, RosterRow } from './types' +import { createBotSection, renameBotSection } from './user-sections' +import { SectionNameDialog } from './user-sections-ui' + +interface renderRosterDialogsProps { + b: ReturnType + t: ReturnType['t'] + createOpen: boolean + setCreateOpen: (value: boolean) => void + groupCreateOpen: boolean + setGroupCreateOpen: (value: boolean) => void + editing: RosterRow | null + setEditing: (value: RosterRow | null) => void + deleting: (RosterRow & { path?: string }) | null + setDeleting: (value: (RosterRow & { path?: string }) | null) => void + deletingGroup: { members: GroupMember[]; name: string } | null + setDeletingGroup: (value: { members: GroupMember[]; name: string } | null) => void + grouping: RosterRow | null + setGrouping: (value: RosterRow | null) => void + sectionDialog: null | { bot?: RosterRow; mode: 'create' } | { id: string; mode: 'rename'; name: string } + setSectionDialog: ( + value: null | { bot?: RosterRow; mode: 'create' } | { id: string; mode: 'rename'; name: string } + ) => void + roster: RosterRow[] + activeSourceRoster: RosterRow[] + refetch: ReturnType['refetch'] +} + +export function renderRosterDialogs({ + b, + t, + createOpen, + setCreateOpen, + groupCreateOpen, + setGroupCreateOpen, + editing, + setEditing, + deleting, + setDeleting, + deletingGroup, + setDeletingGroup, + grouping, + setGrouping, + sectionDialog, + setSectionDialog, + roster, + activeSourceRoster, + refetch +}: renderRosterDialogsProps) { + return ( + <> + { + setCreateOpen(false) + void refetch() + }} + open={createOpen} + roster={activeSourceRoster} + /> + setGroupCreateOpen(false)} + onCreated={groupName => openGroupChat(groupName)} + open={groupCreateOpen} // Full multi-source roster: group chats can seat bots from other + // registered connections — their turns route to their own machines. + roster={roster} + /> + { + if (!open) { + setSectionDialog(null) + } + }} + onSubmit={name => { + if (sectionDialog?.mode === 'rename') { + renameBotSection(sectionDialog.id, name) + } else { + createBotSection(name, sectionDialog?.bot ? [sectionDialog.bot] : []) + } + }} + open={Boolean(sectionDialog)} + /> + { + setEditing(null) + void refetch() + }} + open={Boolean(editing)} + /> + {grouping ? setGrouping(null)} /> : null} + + {'This will permanently delete the bot '} + {deleting.name} + {' and its associated Hermes profile at '} + {deleting.path}. This cannot be undone. + + ) : null + } + destructive + doneLabel="Deleted" + onClose={() => setDeleting(null)} + onConfirm={async () => { + if (!deleting) { + return + } + + const name = deleting.name + await deleteBot(deleting) + await refetch() + host.notify({ + kind: 'success', + message: `Deleted profile ${name}` + }) + }} + open={Boolean(deleting)} + title={b.bot.deleteTitle} + /> + setDeletingGroup(null)} + onConfirm={async () => { + if (!deletingGroup) { + return + } + + await disbandGroupChat(deletingGroup.name, deletingGroup.members) + host.notify({ + kind: 'success', + message: `Deleted group “${deletingGroup.name}”` + }) + }} + open={Boolean(deletingGroup)} + title={b.group.deleteTitle} + /> + + ) +} diff --git a/apps/desktop/src/plugins/hermes-bots/roster-pane-lifecycle.ts b/apps/desktop/src/plugins/hermes-bots/roster-pane-lifecycle.ts new file mode 100644 index 000000000000..680cd76b68b9 --- /dev/null +++ b/apps/desktop/src/plugins/hermes-bots/roster-pane-lifecycle.ts @@ -0,0 +1,54 @@ +import { atom, host } from '@hermes/plugin-sdk' +import { useEffect } from 'react' + +import { $lastRoster } from './data' +import type { useRoster } from './data' +import { displayName } from './labels' +import { mergeServerMeta, pullServerAvatars } from './profile-ops' +import { trackInboundActivity } from './roster-actions' +import { botRosterMeta, botWorkspaceOwnerKey } from './routing' +import { backfillMessagingProtocol } from './soul' +import type { GatewaySource } from './types' +import type { BotMeta, RosterRow } from './types' + +/** Last source inventory returned by the desktop-wide agent roster. */ +export const $lastSources = atom([]) + +interface RosterSnapshotInput { + data: ReturnType['data'] + live: RosterRow[] | null + roster: RosterRow[] + allMeta: Record + activeSourceRoster: RosterRow[] +} + +export function usePublishRosterSnapshot({ data, live, roster, allMeta, activeSourceRoster }: RosterSnapshotInput) { + useEffect(() => { + if (!live) { + return + } + + // Offline-owner ghosts belong only to this render. Shared roster state + // feeds merge caching, group membership, creation, and durable sync. These + // writes must settle after render: other subscribers of the same atoms + // would otherwise be updated while BotsPane was still rendering. + $lastRoster.set(roster.filter(row => !row?.ghost)) + // Tabs caption a bot chat by its bot (#99152); republished with the + // roster so a rename follows and tiles restored at boot resolve. + roster.forEach(bot => { + host.setWorkspaceOwnerLabel?.(botWorkspaceOwnerKey(bot), displayName(bot, botRosterMeta(bot, allMeta))) + }) + + if (Array.isArray(data?.sources)) { + $lastSources.set(data.sources) + } + + mergeServerMeta(activeSourceRoster, data?.fetchedAt || 0) + pullServerAvatars(activeSourceRoster) + trackInboundActivity(roster) + backfillMessagingProtocol(activeSourceRoster) + // React Query owns the stable server snapshot; derived arrays intentionally + // follow that snapshot rather than retriggering on their own atom writes. + // eslint-disable-next-line react-hooks/exhaustive-deps + }, [data]) +} diff --git a/apps/desktop/src/plugins/hermes-bots/roster-pane-sections.tsx b/apps/desktop/src/plugins/hermes-bots/roster-pane-sections.tsx new file mode 100644 index 000000000000..3c63207871e0 --- /dev/null +++ b/apps/desktop/src/plugins/hermes-bots/roster-pane-sections.tsx @@ -0,0 +1,195 @@ +import { host } from '@hermes/plugin-sdk' +import type { ReactNode } from 'react' + +import { botRosterKey } from './data' +import type { useBots } from './i18n' +import type { RosterGroupRow } from './roster-pane-derivation' +import { GatewayKindGlyph, GatewaySectionHeading, RosterSectionHeader } from './roster-sections' +import type { ResolvedRosterGatewaySection } from './roster-sections' +import type { BotMeta, GroupMember, RosterRow } from './types' +import type { $botSections } from './user-sections' +import { + deleteBotSection, + groupRowsBySection, + moveBotSection, + moveBotsToSection, + UNASSIGNED_SECTION_KEY +} from './user-sections' +import { SectionDropZone, UserSectionHeader } from './user-sections-ui' + +interface RosterSectionRenderersProps { + b: ReturnType + userSections: ReturnType + roster: RosterRow[] + allMeta: Record + dragging: string | null + rosterSectionCollapsed: (id: string) => boolean + toggleRosterSection: (id: string) => void + setSectionDialog: ( + value: null | { bot?: RosterRow; mode: 'create' } | { id: string; mode: 'rename'; name: string } + ) => void + renderBotRow: (bot: RosterRow, keyPrefix?: string) => ReactNode + renderGroupRow: (row: { members: GroupMember[]; name: string }) => ReactNode + sortedGroupRows: RosterGroupRow[] +} + +export function rosterSectionRenderers({ + b, + userSections, + roster, + allMeta, + dragging, + rosterSectionCollapsed, + toggleRosterSection, + setSectionDialog, + renderBotRow, + renderGroupRow, + sortedGroupRows +}: RosterSectionRenderersProps) { + const removeSection = (id: string) => { + const name = userSections.find(section => section.id === id)?.name || '' + const { members, undo } = deleteBotSection(id, roster) + + // No confirmation: nothing is lost (the bots fall back to Unassigned) and + // the toast's Undo puts the section and its members back. + host.notify({ + action: { label: b.sections.undo, onClick: undo }, + durationMs: 8_000, + kind: 'info', + message: b.sections.deleted(name, members.length) + }) + } + + // USER SECTIONS — composed with the gateway sections, not instead of them. + // The gateway headings own the top level whenever the roster shows more + // than one connection (that axis answers "where does this run", which no + // folder name can, and a bot's membership lives in its profile on THAT + // gateway); user sections group the rows INSIDE each connection bucket, + // indented under it, and group the flat list when there is only one. + // `keyPrefix` keeps row keys unique across the gateway buckets. + type UserSectionRow = { bot: RosterRow; kind?: 'bot' } | RosterGroupRow + + const renderUserSections = (rows: UserSectionRow[], keyPrefix = '') => { + // No sections made: the plain list, exactly as before this feature. + if (!userSections.length) { + return rows.map(row => (row.kind === 'group' ? renderGroupRow(row) : renderBotRow(row.bot, keyPrefix))) + } + + const nested = Boolean(keyPrefix) + const blocks = groupRowsBySection(rows, userSections, allMeta) + + return ( + blocks + // An empty Unassigned is not worth a heading; an empty NAMED section + // is, because it is somewhere the user made and is about to drop into. + // Inside a gateway bucket the same empty section would repeat under + // every connection, so there it only appears while a drag is in flight + // (as the drop target it exists for); the row menu files into it + // regardless. + .filter(block => block.rows.length || (block.id && (!nested || dragging))) + .map(block => { + const key = `${keyPrefix}${block.id ? `user-section:${block.id}` : UNASSIGNED_SECTION_KEY}` + const collapsed = rosterSectionCollapsed(key) + const order = userSections.findIndex(section => section.id === block.id) + + return ( + row.kind !== 'group' && botRosterKey(row.bot) === dragging) + } + key={key} + nested={nested} + onDropBot={rosterKey => { + const bot = roster.find(row => botRosterKey(row) === rosterKey) + + // `block.id` is null for Unassigned, which is exactly the value + // moveBotsToSection wants for "clear the assignment". + if (bot) { + void moveBotsToSection([bot], block.id) + } + }} + > + = 0 && order < userSections.length - 1} + canMoveUp={order > 0} + collapsed={collapsed} + count={block.rows.length} + id={block.id} + name={block.name} + onDelete={() => block.id && removeSection(block.id)} + onMove={delta => block.id && moveBotSection(block.id, delta)} + onRename={() => block.id && setSectionDialog({ id: block.id, mode: 'rename', name: block.name })} + onToggle={() => toggleRosterSection(key)} + /> + {collapsed ? null : block.rows.length ? ( +
+ {block.rows.map(row => + row.kind === 'group' ? renderGroupRow(row) : renderBotRow(row.bot, `${key}:`) + )} +
+ ) : ( + // Empty section: a quiet dashed slot that says what it is for, + // and doubles as a roomy drop target. +
+ {b.sections.emptyHint} +
+ )} + + ) + }) + ) + } + + const renderGatewaySection = (section: ResolvedRosterGatewaySection) => { + const sectionId = `gateway:${section.id}` + const collapsed = rosterSectionCollapsed(sectionId) + + return ( +
+ toggleRosterSection(sectionId)} + option={section.option} + /> + {collapsed ? null : ( +
{renderUserSections(section.rows, `${section.id}:`)}
+ )} +
+ ) + } + + const renderGroupChatSection = () => { + const sectionId = 'group-chats' + const collapsed = rosterSectionCollapsed(sectionId) + + return ( +
+ toggleRosterSection(sectionId)} + tip={`${sortedGroupRows.length} global group chat${sortedGroupRows.length === 1 ? '' : 's'}`} + /> + {collapsed ? null :
{sortedGroupRows.map(renderGroupRow)}
} +
+ ) + } + + const renderHiddenGatewaySection = (section: ResolvedRosterGatewaySection) => ( +
+
+ + + {section.option?.label || section.option?.connectionId || 'Current gateway'} + + {section.rows.length} +
+ {section.rows.map(row => renderBotRow(row.bot, `hidden:${section.id}:`))} +
+ ) + + return { renderUserSections, renderGatewaySection, renderGroupChatSection, renderHiddenGatewaySection } +} diff --git a/apps/desktop/src/plugins/hermes-bots/roster-pane-toolbar.tsx b/apps/desktop/src/plugins/hermes-bots/roster-pane-toolbar.tsx new file mode 100644 index 000000000000..ca366664868a --- /dev/null +++ b/apps/desktop/src/plugins/hermes-bots/roster-pane-toolbar.tsx @@ -0,0 +1,227 @@ +import { + Button, + cn, + Codicon, + DropdownMenu, + DropdownMenuContent, + DropdownMenuItem, + DropdownMenuSeparator, + DropdownMenuTrigger, + SearchField, + Tip +} from '@hermes/plugin-sdk' + +import { botSourceStatus } from './data' +import type { useBots } from './i18n' +import { setActivityToasts } from './roster-actions' +import { GatewayKindGlyph } from './roster-sections' +import type { rosterGatewayOptions } from './roster-sections' +import type { RosterActivityFilter, RosterKindFilter, RosterRow } from './types' + +interface renderRosterToolbarProps { + b: ReturnType + activityToasts: boolean + activeSourceRoster: RosterRow[] + setCreateOpen: (value: boolean) => void + setGroupCreateOpen: (value: boolean) => void + setSectionDialog: ( + value: null | { bot?: RosterRow; mode: 'create' } | { id: string; mode: 'rename'; name: string } + ) => void + showRosterTools: boolean + showRosterSearch: boolean + showRosterFilters: boolean + query: string + setQuery: (value: string) => void + activeFilterCount: number + gatewayOptions: ReturnType + rowKindFilter: RosterKindFilter + setRowKindFilter: (value: RosterKindFilter) => void + activityFilter: RosterActivityFilter + setActivityFilter: (value: RosterActivityFilter) => void + gatewayFilter: string + setGatewayFilter: (value: string) => void +} + +export function renderRosterToolbar({ + b, + activityToasts, + activeSourceRoster, + setCreateOpen, + setGroupCreateOpen, + setSectionDialog, + showRosterTools, + showRosterSearch, + showRosterFilters, + query, + setQuery, + activeFilterCount, + gatewayOptions, + rowKindFilter, + setRowKindFilter, + activityFilter, + setActivityFilter, + gatewayFilter, + setGatewayFilter +}: renderRosterToolbarProps) { + return ( + <> +
+ + Bots + +
+ + + + + + + + + + + setCreateOpen(true)}> + + {b.bot.newTitle} + + setGroupCreateOpen(true)}> + + {b.group.newTitle} + + + setSectionDialog({ mode: 'create' })}> + + {b.sections.newSection} + + + +
+
+ {showRosterTools ? ( +
+ {showRosterSearch ? ( + + ) : ( + + )} + {showRosterFilters ? ( + + + + + + + + {( + [ + ['all', b.roster.botsAndGroups], + ['bots', b.roster.botsOnly], + ['groups', b.roster.groupsOnly] + ] as [RosterKindFilter, string][] + ).map(([value, label]) => ( + setRowKindFilter(value)}> + {label} + {rowKindFilter === value ? : null} + + ))} + + {( + [ + ['all', b.roster.anyActivity], + ['active', b.roster.activeNow], + ['recent', b.roster.recentlyActive], + ['older', b.roster.older] + ] as [RosterActivityFilter, string][] + ).map(([value, label]) => ( + setActivityFilter(value)}> + {label} + {activityFilter === value ? : null} + + ))} + {gatewayOptions.length > 1 ? : null} + {gatewayOptions.length > 1 ? ( + setGatewayFilter('all')}> + + All gateways + {gatewayFilter === 'all' ? : null} + + ) : null} + {gatewayOptions.length > 1 + ? gatewayOptions.map(option => { + const status = botSourceStatus({ + sourceError: option.error, + sourceReachable: option.reachable + }) + + return ( + setGatewayFilter(option.connectionId)} + > + + {option.label || option.connectionId} + + {option.count} + + {gatewayFilter === option.connectionId ? : null} + + ) + }) + : []} + {activeFilterCount ? : null} + {activeFilterCount ? ( + { + setRowKindFilter('all') + setActivityFilter('all') + setGatewayFilter('all') + }} + > + {b.roster.clearFilters} + + ) : null} + + + ) : null} +
+ ) : null} + + ) +} diff --git a/apps/desktop/src/plugins/hermes-bots/roster-pane.tsx b/apps/desktop/src/plugins/hermes-bots/roster-pane.tsx index cd94ee134abc..1401c3504252 100644 --- a/apps/desktop/src/plugins/hermes-bots/roster-pane.tsx +++ b/apps/desktop/src/plugins/hermes-bots/roster-pane.tsx @@ -1,33 +1,4 @@ -/** - * The Bots pane itself: the roster's selection reconciliation, the - * workspace-ownership reads its lifecycle keys off, and the pane that lists - * every bot and group chat. - * - * The top of the roster stack. It composes the rows, the section headings and - * the dialogs; nothing in Bot Mode imports it except the plugin entry point. - */ - -import { - atom, - Button, - cn, - Codicon, - ConfirmDialog, - DisclosureCaret, - DropdownMenu, - DropdownMenuContent, - DropdownMenuItem, - DropdownMenuSeparator, - DropdownMenuTrigger, - GlyphSpinner, - host, - PanelEmpty, - RowButton, - SearchField, - Tip, - useI18n, - useValue -} from '@hermes/plugin-sdk' +import { host, useI18n, useValue } from '@hermes/plugin-sdk' import { useEffect, useRef, useState } from 'react' import { BotRow, GroupRow } from './bot-row' @@ -42,60 +13,43 @@ import { parseRosterKey, saveSelectedRosterBot } from './bot-state' -import { CreateAgentDialog, CreateGroupChatDialog, GroupDialog } from './create-dialog' +/** + * The Bots pane itself: the roster's selection reconciliation, the + * workspace-ownership reads its lifecycle keys off, and the pane that lists + * every bot and group chat. + * + * The top of the roster stack. It composes the rows, the section headings and + * the dialogs; nothing in Bot Mode imports it except the plugin entry point. + */ import { $botMeta, $lastRoster, annotateBotSource, - botActivitySession, botRosterKey, botSourceStatus, - filterBots, - preferReachableSameNameRows, sourceByConnection, useRoster } from './data' -import { EditProfileDialog } from './edit-profile-dialog' import { $groupChats, $groupChatWorkspace, $groupClarify, $groupNeedsYou } from './group-chat' -import { disbandGroupChat, GroupChatWorkspace, openGroupChat } from './group-chat-view' -import { groupChatMemberBots, groupChatNames, groupLastActivity } from './group-membership' +import { GroupChatWorkspace, openGroupChat } from './group-chat-view' +import { groupChatMemberBots } from './group-membership' import { $groupMainTabsRev, shouldRenderGroupChatInPane } from './group-panes' import { groupHasPendingClarify } from './group-turns' -import { $showHiddenBots, isBotHidden, isBotPinned } from './hidden-bots' +import { $showHiddenBots, isBotHidden } from './hidden-bots' import { useBots } from './i18n' -import { displayName } from './labels' -import { deleteBot, mergeServerMeta, pullServerAvatars } from './profile-ops' -import { $activityToasts, setActivityToasts, trackInboundActivity } from './roster-actions' -import { - botNeedsHandleLabel, - filterBotsByGateway, - GatewayKindGlyph, - GatewaySectionHeading, - groupMatchesRosterFilters, - rosterGatewayOptions, - rosterGatewaySections, - RosterSectionHeader -} from './roster-sections' -import type { ResolvedRosterGatewaySection } from './roster-sections' -import { botRosterMeta, botWorkspaceOwnerKey, setBotsWorkspaceOwner } from './routing' -import { ACTIVE_WINDOW_S, activeBots, BOT_ROSTER_SEARCH_THRESHOLD, rosterActivityMatches } from './row-helpers' -import { backfillMessagingProtocol } from './soul' +import { $activityToasts } from './roster-actions' +import { renderRosterContent } from './roster-pane-content' +import { deriveRosterPresentation, deriveRosterRows, sortRosterBots } from './roster-pane-derivation' +import { renderRosterDialogs } from './roster-pane-dialogs' +import { $lastSources, usePublishRosterSnapshot } from './roster-pane-lifecycle' +import { rosterSectionRenderers } from './roster-pane-sections' +import { renderRosterToolbar } from './roster-pane-toolbar' +import { botNeedsHandleLabel, rosterGatewayOptions } from './roster-sections' +import { botWorkspaceOwnerKey, setBotsWorkspaceOwner } from './routing' +import { activeBots } from './row-helpers' import type { BotMeta, GatewaySource, GroupMember, RosterActivityFilter, RosterKindFilter, RosterRow } from './types' -import { - $botSections, - $draggingBot, - createBotSection, - deleteBotSection, - groupRowsBySection, - moveBotSection, - moveBotsToSection, - renameBotSection, - UNASSIGNED_SECTION_KEY -} from './user-sections' -import { SectionDropZone, SectionNameDialog, useEscapeCancelsBotDrag, UserSectionHeader } from './user-sections-ui' - -/** Last source inventory returned by the desktop-wide agent roster. */ -const $lastSources = atom([]) +import { $botSections, $draggingBot } from './user-sections' +import { useEscapeCancelsBotDrag } from './user-sections-ui' // ── roster pane ────────────────────────────────────────────────────────────── @@ -234,21 +188,35 @@ export function releaseStaleOpenBotChat(focusedStoredId: null | string | undefin } } -/** The two row shapes the roster sorts together — `kind` is the discriminant. */ -interface RosterBotRow { - active: boolean - activity: number - bot: RosterRow - kind: 'bot' - pinned: boolean -} -interface RosterGroupRow { - active: boolean - activity: number - kind: 'group' - members: GroupMember[] - name: string - pinned: boolean +function useReconcileRosterOwner( + data: ReturnType['data'], + error: ReturnType['error'], + selectionHydrated: boolean, + roster: RosterRow[], + sourceSnapshot: GatewaySource[], + allMeta: Record +) { + // The roster has ANSWERED once data or a terminal error exists — that, not + // row count, is what lets this pane stop showing its loading state (an empty + // answer is a real answer; a pending one must not flash "No bots"). Keep the + // persisted-selection writes out of render: React may replay a render, but + // an abandoned render must never become a storage mutation. + useEffect(() => { + if (!data && !error) { + return + } + + $rosterHydrated.set(true) + + if (selectionHydrated) { + reconcileRosterSelection(roster, sourceSnapshot, allMeta) + const selected = selectedRosterBot(roster, $selectedRosterKey.get()) + + if ($botsPaneVisible.get() && !$groupChatWorkspace.get() && selected) { + setBotsWorkspaceOwner(botWorkspaceOwnerKey(selected), selected) + } + } + }, [data, error, selectionHydrated, roster, sourceSnapshot, allMeta]) } export function BotsPane() { @@ -306,19 +274,6 @@ export function BotsPane() { }, [gatewayUp, refetch]) const allMeta = useValue($botMeta) - // Messaging-app order: most recent activity first, where "activity" is - // the newest of (bot created, last message in any of its sessions). A - // freshly created bot tops the list until another bot gets a message. - // No special slot for the primary bot — it competes on recency too. - const activityOf = (bot: RosterRow): number => { - const created = botRosterMeta(bot, allMeta)?.created || bot.ui_meta?.['hermes-bots']?.created || 0 - const lastMsg = (botActivitySession(bot)?.last_active || 0) * 1000 - - return Math.max(created, lastMsg) - } - - // Pin is a source-qualified Desktop preference, not gateway profile state. - const isPinned = (bot: RosterRow): boolean => isBotPinned(bot, allMeta) // Resilience (@wesleysimplicio, #13): a failed refresh must not erase a // roster the user already had — mixed local+cloud gateways and remotes // waking from sleep fail transiently. Render the last good snapshot with @@ -330,16 +285,7 @@ export function BotsPane() { const sourceWithSelectedOwner = selectionHydrated && rosterHydrated ? rosterWithSelectedOwner(source, sourceSnapshot, selectedRosterKey) : source - const roster = sourceWithSelectedOwner.slice().sort((a, b) => { - const pa = isPinned(a) ? 1 : 0 - const pb = isPinned(b) ? 1 : 0 - - if (pa !== pb) { - return pb - pa - } - - return activityOf(b) - activityOf(a) - }) + const { roster, activityOf, isPinned } = sortRosterBots(sourceWithSelectedOwner, allMeta) // React Query can briefly report neither loading nor data while the plugin // and the persisted connection registry hydrate. Keep that transition in a @@ -354,118 +300,59 @@ export function BotsPane() { setGatewayFilter('all') } }, [gatewayFilterExists]) - const activeSourceRoster = roster.filter(bot => !bot.remoteSource) - // Hidden rows remain fully alive and recoverable at the bottom. Every - // non-display consumer continues to receive the complete roster. const hiddenExpanded = useValue($showHiddenBots) - const hiddenBots = roster.filter(bot => isBotHidden(bot, allMeta)) - const visibleRoster = roster.filter(bot => !isBotHidden(bot, allMeta)) - const gatewayRoster = filterBotsByGateway(visibleRoster, gatewayFilter) - - const filteredRoster = filterBots(gatewayRoster, allMeta, query).filter((bot: RosterRow) => - rosterActivityMatches( - { - activity: activityOf(bot), - active: activeRosterKeys.has(botRosterKey(bot)) - }, - activityFilter - ) - ) - - const filteredHiddenBots = filterBots(filterBotsByGateway(hiddenBots, gatewayFilter), allMeta, query).filter( - (bot: RosterRow) => - rosterActivityMatches( - { - activity: activityOf(bot), - active: activeRosterKeys.has(botRosterKey(bot)) - }, - activityFilter - ) - ) - - const groupNames = groupChatNames(allMeta, groupRooms) - - const groupRows = groupNames - .map(name => ({ - name, - members: groupChatMemberBots(name, roster, allMeta) - })) - .filter(row => groupMatchesRosterFilters(row.name, row.members, allMeta, query, gatewayFilter)) - .map((row): RosterGroupRow => ({ - kind: 'group', - name: row.name, - members: row.members, - pinned: Boolean(groupRooms[row.name]?.pinned), - activity: groupLastActivity(groupRooms[row.name]), - active: - Boolean( - groupLastActivity(groupRooms[row.name]) && - Date.now() - groupLastActivity(groupRooms[row.name]) <= ACTIVE_WINDOW_S * 1000 - ) || row.members.some(member => activeRosterKeys.has(botRosterKey(member))) - })) - .filter(row => rowKindFilter !== 'bots' && rosterActivityMatches(row, activityFilter)) - - const botRows = - rowKindFilter === 'groups' - ? [] - : preferReachableSameNameRows(filteredRoster).map((bot): RosterBotRow => ({ - kind: 'bot', - bot, - pinned: isPinned(bot), - activity: activityOf(bot), - active: activeRosterKeys.has(botRosterKey(bot)) - })) - - const sortRosterRows = (rows: T[]): T[] => - rows.slice().sort((a, b) => { - const pa = a.pinned ? 1 : 0 - const pb = b.pinned ? 1 : 0 - - if (pa !== pb) { - return pb - pa - } - - return b.activity - a.activity - }) - - const rosterRows = sortRosterRows([...botRows, ...groupRows]) - const sortedGroupRows = sortRosterRows(groupRows) - const gatewaySections = rosterGatewaySections(botRows, gatewayOptions, gatewayFilter) - const showGatewaySections = gatewaySections.sectioned && botRows.length > 0 - const activeFilterCount = - (rowKindFilter === 'all' ? 0 : 1) + (activityFilter === 'all' ? 0 : 1) + (gatewayFilter === 'all' ? 0 : 1) - - const hasRosterConstraint = Boolean(query.trim()) || activeFilterCount > 0 - const matchingHiddenBots = rowKindFilter === 'groups' ? [] : filteredHiddenBots - const showHiddenSection = hiddenBots.length > 0 && (!hasRosterConstraint || matchingHiddenBots.length > 0) - const showHiddenRows = hiddenExpanded || hasRosterConstraint - const rosterItemCount = roster.length + groupNames.length - - const allBotsHidden = - !hasRosterConstraint && visibleRoster.length === 0 && groupNames.length === 0 && hiddenBots.length > 0 - - const showRosterSearch = - gatewayOptions.length > 1 || rosterItemCount >= BOT_ROSTER_SEARCH_THRESHOLD || Boolean(query.trim()) + const { + activeSourceRoster, + hiddenBots, + visibleRoster, + filteredHiddenBots, + groupNames, + rosterRows, + sortedGroupRows, + gatewaySections, + showGatewaySections + } = deriveRosterRows({ + roster, + allMeta, + gatewayFilter, + query, + activityFilter, + rowKindFilter, + groupRooms, + activeRosterKeys, + gatewayOptions, + activityOf, + isPinned + }) - const showRosterFilters = - gatewayOptions.length > 1 || - groupNames.length > 0 || - rosterItemCount >= BOT_ROSTER_SEARCH_THRESHOLD || - activeFilterCount > 0 + const { + activeFilterCount, + hasRosterConstraint, + matchingHiddenBots, + showHiddenSection, + showHiddenRows, + allBotsHidden, + showRosterSearch, + showRosterFilters, + showRosterTools, + hiddenGatewaySections + } = deriveRosterPresentation({ + rowKindFilter, + activityFilter, + gatewayFilter, + query, + filteredHiddenBots, + hiddenBots, + hiddenExpanded, + roster, + groupNames, + visibleRoster, + gatewayOptions + }) - const showRosterTools = showRosterSearch || showRosterFilters const rosterSectionCollapsed = (id: string): boolean => !hasRosterConstraint && collapsedRosterSections.has(id) - const hiddenGatewaySections = rosterGatewaySections( - matchingHiddenBots.map((bot: RosterRow) => ({ - kind: 'bot', - bot - })), - gatewayOptions, - gatewayFilter - ) - const toggleRosterSection = (id: string): void => { setCollapsedRosterSections(previous => { const next = new Set(previous) @@ -493,56 +380,9 @@ export function BotsPane() { return () => cancelAnimationFrame(frame) }, [hiddenExpanded, hasRosterConstraint]) - useEffect(() => { - if (!live) { - return - } - - // Offline-owner ghosts belong only to this render. Shared roster state - // feeds merge caching, group membership, creation, and durable sync. These - // writes must settle after render: other subscribers of the same atoms - // would otherwise be updated while BotsPane was still rendering. - $lastRoster.set(roster.filter(row => !row?.ghost)) - // Tabs caption a bot chat by its bot (#99152); republished with the - // roster so a rename follows and tiles restored at boot resolve. - roster.forEach(bot => { - host.setWorkspaceOwnerLabel?.(botWorkspaceOwnerKey(bot), displayName(bot, botRosterMeta(bot, allMeta))) - }) - - if (Array.isArray(data?.sources)) { - $lastSources.set(data.sources) - } + usePublishRosterSnapshot({ data, live, roster, allMeta, activeSourceRoster }) - mergeServerMeta(activeSourceRoster, data?.fetchedAt || 0) - pullServerAvatars(activeSourceRoster) - trackInboundActivity(roster) - backfillMessagingProtocol(activeSourceRoster) - // React Query owns the stable server snapshot; derived arrays intentionally - // follow that snapshot rather than retriggering on their own atom writes. - // eslint-disable-next-line react-hooks/exhaustive-deps - }, [data]) - - // The roster has ANSWERED once data or a terminal error exists — that, not - // row count, is what lets this pane stop showing its loading state (an empty - // answer is a real answer; a pending one must not flash "No bots"). Keep the - // persisted-selection writes out of render: React may replay a render, but - // an abandoned render must never become a storage mutation. - useEffect(() => { - if (!data && !error) { - return - } - - $rosterHydrated.set(true) - - if (selectionHydrated) { - reconcileRosterSelection(roster, sourceSnapshot, allMeta) - const selected = selectedRosterBot(roster, $selectedRosterKey.get()) - - if ($botsPaneVisible.get() && !$groupChatWorkspace.get() && selected) { - setBotsWorkspaceOwner(botWorkspaceOwnerKey(selected), selected) - } - } - }, [data, error, selectionHydrated, roster, sourceSnapshot, allMeta]) + useReconcileRosterOwner(data, error, selectionHydrated, roster, sourceSnapshot, allMeta) const staleNotice = error && !live && roster.length @@ -580,509 +420,95 @@ export function BotsPane() { /> ) - const removeSection = (id: string) => { - const name = userSections.find(section => section.id === id)?.name || '' - const { members, undo } = deleteBotSection(id, roster) - - // No confirmation: nothing is lost (the bots fall back to Unassigned) and - // the toast's Undo puts the section and its members back. - host.notify({ - action: { label: b.sections.undo, onClick: undo }, - durationMs: 8_000, - kind: 'info', - message: b.sections.deleted(name, members.length) + const { renderUserSections, renderGatewaySection, renderGroupChatSection, renderHiddenGatewaySection } = + rosterSectionRenderers({ + b, + userSections, + roster, + allMeta, + dragging, + rosterSectionCollapsed, + toggleRosterSection, + setSectionDialog, + renderBotRow, + renderGroupRow, + sortedGroupRows }) - } - - // USER SECTIONS — composed with the gateway sections, not instead of them. - // The gateway headings own the top level whenever the roster shows more - // than one connection (that axis answers "where does this run", which no - // folder name can, and a bot's membership lives in its profile on THAT - // gateway); user sections group the rows INSIDE each connection bucket, - // indented under it, and group the flat list when there is only one. - // `keyPrefix` keeps row keys unique across the gateway buckets. - type UserSectionRow = { bot: RosterRow; kind?: 'bot' } | RosterGroupRow - - const renderUserSections = (rows: UserSectionRow[], keyPrefix = '') => { - // No sections made: the plain list, exactly as before this feature. - if (!userSections.length) { - return rows.map(row => (row.kind === 'group' ? renderGroupRow(row) : renderBotRow(row.bot, keyPrefix))) - } - - const nested = Boolean(keyPrefix) - const blocks = groupRowsBySection(rows, userSections, allMeta) - - return ( - blocks - // An empty Unassigned is not worth a heading; an empty NAMED section - // is, because it is somewhere the user made and is about to drop into. - // Inside a gateway bucket the same empty section would repeat under - // every connection, so there it only appears while a drag is in flight - // (as the drop target it exists for); the row menu files into it - // regardless. - .filter(block => block.rows.length || (block.id && (!nested || dragging))) - .map(block => { - const key = `${keyPrefix}${block.id ? `user-section:${block.id}` : UNASSIGNED_SECTION_KEY}` - const collapsed = rosterSectionCollapsed(key) - const order = userSections.findIndex(section => section.id === block.id) - - return ( - row.kind !== 'group' && botRosterKey(row.bot) === dragging) - } - key={key} - nested={nested} - onDropBot={rosterKey => { - const bot = roster.find(row => botRosterKey(row) === rosterKey) - - // `block.id` is null for Unassigned, which is exactly the value - // moveBotsToSection wants for "clear the assignment". - if (bot) { - void moveBotsToSection([bot], block.id) - } - }} - > - = 0 && order < userSections.length - 1} - canMoveUp={order > 0} - collapsed={collapsed} - count={block.rows.length} - id={block.id} - name={block.name} - onDelete={() => block.id && removeSection(block.id)} - onMove={delta => block.id && moveBotSection(block.id, delta)} - onRename={() => block.id && setSectionDialog({ id: block.id, mode: 'rename', name: block.name })} - onToggle={() => toggleRosterSection(key)} - /> - {collapsed ? null : block.rows.length ? ( -
- {block.rows.map(row => - row.kind === 'group' ? renderGroupRow(row) : renderBotRow(row.bot, `${key}:`) - )} -
- ) : ( - // Empty section: a quiet dashed slot that says what it is for, - // and doubles as a roomy drop target. -
- {b.sections.emptyHint} -
- )} - - ) - }) - ) - } - - const renderGatewaySection = (section: ResolvedRosterGatewaySection) => { - const sectionId = `gateway:${section.id}` - const collapsed = rosterSectionCollapsed(sectionId) - - return ( -
- toggleRosterSection(sectionId)} - option={section.option} - /> - {collapsed ? null : ( -
{renderUserSections(section.rows, `${section.id}:`)}
- )} -
- ) - } - - const renderGroupChatSection = () => { - const sectionId = 'group-chats' - const collapsed = rosterSectionCollapsed(sectionId) - - return ( -
- toggleRosterSection(sectionId)} - tip={`${sortedGroupRows.length} global group chat${sortedGroupRows.length === 1 ? '' : 's'}`} - /> - {collapsed ? null :
{sortedGroupRows.map(renderGroupRow)}
} -
- ) - } - - const renderHiddenGatewaySection = (section: ResolvedRosterGatewaySection) => ( -
-
- - - {section.option?.label || section.option?.connectionId || 'Current gateway'} - - {section.rows.length} -
- {section.rows.map(row => renderBotRow(row.bot, `hidden:${section.id}:`))} -
- ) return (
-
- - Bots - -
- - - - - - - - - - - setCreateOpen(true)}> - - {b.bot.newTitle} - - setGroupCreateOpen(true)}> - - {b.group.newTitle} - - - setSectionDialog({ mode: 'create' })}> - - {b.sections.newSection} - - - -
-
- {showRosterTools ? ( -
- {showRosterSearch ? ( - - ) : ( - - )} - {showRosterFilters ? ( - - - - - - - - {( - [ - ['all', b.roster.botsAndGroups], - ['bots', b.roster.botsOnly], - ['groups', b.roster.groupsOnly] - ] as [RosterKindFilter, string][] - ).map(([value, label]) => ( - setRowKindFilter(value)}> - {label} - {rowKindFilter === value ? : null} - - ))} - - {( - [ - ['all', b.roster.anyActivity], - ['active', b.roster.activeNow], - ['recent', b.roster.recentlyActive], - ['older', b.roster.older] - ] as [RosterActivityFilter, string][] - ).map(([value, label]) => ( - setActivityFilter(value)}> - {label} - {activityFilter === value ? : null} - - ))} - {gatewayOptions.length > 1 ? : null} - {gatewayOptions.length > 1 ? ( - setGatewayFilter('all')}> - - All gateways - {gatewayFilter === 'all' ? : null} - - ) : null} - {gatewayOptions.length > 1 - ? gatewayOptions.map(option => { - const status = botSourceStatus({ - sourceError: option.error, - sourceReachable: option.reachable - }) - - return ( - setGatewayFilter(option.connectionId)} - > - - {option.label || option.connectionId} - - {option.count} - - {gatewayFilter === option.connectionId ? : null} - - ) - }) - : []} - {activeFilterCount ? : null} - {activeFilterCount ? ( - { - setRowKindFilter('all') - setActivityFilter('all') - setGatewayFilter('all') - }} - > - {b.roster.clearFilters} - - ) : null} - - - ) : null} -
- ) : null} - {staleNotice ? ( -
- {staleNotice} -
- ) : null} - {(isLoading || initialRosterLoading) && !roster.length ? ( -
- -
- ) : error && !roster.length ? ( -
-
- {gatewayUp - ? b.roster.rosterUnavailable(error instanceof Error ? error.message : 'gateway error') - : b.roster.waitingForGateway} -
- -
- ) : roster.length === 0 ? ( - - ) : allBotsHidden && !hiddenExpanded ? ( -
-
- - {b.roster.allHidden} -
-

{b.roster.allHiddenDesc}

- -
- ) : rosterRows.length === 0 && matchingHiddenBots.length === 0 ? ( -
- -
- ) : ( -
-
- {showGatewaySections - ? [ - sortedGroupRows.length ? renderGroupChatSection() : null, - ...gatewaySections.sections.map(renderGatewaySection) - ].filter(Boolean) - : renderUserSections(rosterRows)} - {showHiddenSection ? ( -
- {hasRosterConstraint ? ( -
- - Hidden - {matchingHiddenBots.length} -
- ) : ( - $showHiddenBots.set(!hiddenExpanded)} - > - - Hidden - {hiddenBots.length} - - )} - {showHiddenRows ? ( - matchingHiddenBots.length ? ( - hiddenGatewaySections.sectioned ? ( - hiddenGatewaySections.sections.map(renderHiddenGatewaySection) - ) : ( - matchingHiddenBots.map((bot: RosterRow) => renderBotRow(bot, 'hidden:')) - ) - ) : ( -
{b.roster.noHiddenMatch}
- ) - ) : null} -
- ) : null} -
-
- )} - { - setCreateOpen(false) - void refetch() - }} - open={createOpen} - roster={activeSourceRoster} - /> - setGroupCreateOpen(false)} - onCreated={groupName => openGroupChat(groupName)} - open={groupCreateOpen} // Full multi-source roster: group chats can seat bots from other - // registered connections — their turns route to their own machines. - roster={roster} - /> - { - if (!open) { - setSectionDialog(null) - } - }} - onSubmit={name => { - if (sectionDialog?.mode === 'rename') { - renameBotSection(sectionDialog.id, name) - } else { - createBotSection(name, sectionDialog?.bot ? [sectionDialog.bot] : []) - } - }} - open={Boolean(sectionDialog)} - /> - { - setEditing(null) - void refetch() - }} - open={Boolean(editing)} - /> - {grouping ? setGrouping(null)} /> : null} - - {'This will permanently delete the bot '} - {deleting.name} - {' and its associated Hermes profile at '} - {deleting.path}. This cannot be undone. - - ) : null - } - destructive - doneLabel="Deleted" - onClose={() => setDeleting(null)} - onConfirm={async () => { - if (!deleting) { - return - } - - const name = deleting.name - await deleteBot(deleting) - await refetch() - host.notify({ - kind: 'success', - message: `Deleted profile ${name}` - }) - }} - open={Boolean(deleting)} - title={b.bot.deleteTitle} - /> - setDeletingGroup(null)} - onConfirm={async () => { - if (!deletingGroup) { - return - } - - await disbandGroupChat(deletingGroup.name, deletingGroup.members) - host.notify({ - kind: 'success', - message: `Deleted group “${deletingGroup.name}”` - }) - }} - open={Boolean(deletingGroup)} - title={b.group.deleteTitle} - /> + {renderRosterToolbar({ + b, + activityToasts, + activeSourceRoster, + setCreateOpen, + setGroupCreateOpen, + setSectionDialog, + showRosterTools, + showRosterSearch, + showRosterFilters, + query, + setQuery, + activeFilterCount, + gatewayOptions, + rowKindFilter, + setRowKindFilter, + activityFilter, + setActivityFilter, + gatewayFilter, + setGatewayFilter + })} + {renderRosterContent({ + b, + staleNotice, + isLoading, + initialRosterLoading, + roster, + error, + gatewayUp, + refetch, + allBotsHidden, + hiddenExpanded, + rosterRows, + matchingHiddenBots, + query, + selectedGateway, + showGatewaySections, + sortedGroupRows, + gatewaySections, + showHiddenSection, + hiddenSectionRef, + hasRosterConstraint, + hiddenBots, + showHiddenRows, + hiddenGatewaySections, + renderBotRow, + renderGroupChatSection, + renderGatewaySection, + renderUserSections, + renderHiddenGatewaySection + })} + {renderRosterDialogs({ + b, + t, + createOpen, + setCreateOpen, + groupCreateOpen, + setGroupCreateOpen, + editing, + setEditing, + deleting, + setDeleting, + deletingGroup, + setDeletingGroup, + grouping, + setGrouping, + sectionDialog, + setSectionDialog, + roster, + activeSourceRoster, + refetch + })}
) } From 960dee7fe5d73f09897df382f40fee0d4636186d Mon Sep 17 00:00:00 2001 From: brooklyn! Date: Tue, 8 Sep 2026 15:20:07 -0500 Subject: [PATCH 206/227] fix(desktop): Hide tabs works on the sessions sidebar MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit hideOnly chrome pinned the Sessions/Bots strip on at any tab count, so never was a silent no-op. The panes stay; ⌘⌥T brings the strip back. Co-authored-by: Cursor --- .../tree/hide-only-strip-tabs.test.ts | 4 ++-- .../collapse-restore-affordance.test.tsx | 8 +++----- .../tree/renderer/strip-visibility.test.ts | 15 +++++---------- .../tree/renderer/strip-visibility.ts | 18 +++++------------- 4 files changed, 15 insertions(+), 30 deletions(-) diff --git a/apps/desktop/src/components/pane-shell/tree/hide-only-strip-tabs.test.ts b/apps/desktop/src/components/pane-shell/tree/hide-only-strip-tabs.test.ts index de8fb9bf14b2..e18913aaea7d 100644 --- a/apps/desktop/src/components/pane-shell/tree/hide-only-strip-tabs.test.ts +++ b/apps/desktop/src/components/pane-shell/tree/hide-only-strip-tabs.test.ts @@ -104,14 +104,14 @@ describe('hide-only strip tabs', () => { expect(hideOnlyZoneTabs('g-main')).toEqual([]) }) - it('refuses to hide the strip that is the only handle for hide-only tabs', () => { + it('honors never on the sessions/Bots strip', () => { sessionsBotsTree() setTreeGroupTabStrip('g-side', 'never') const side = $layoutTree.get() const group = side && side.type === 'split' ? side.children[0] : side - expect(group && group.type === 'group' ? tabStripVisibleForGroup(group) : false).toBe(true) + expect(group && group.type === 'group' ? tabStripVisibleForGroup(group) : true).toBe(false) }) it('excludes hide-only tabs from every close verb', () => { diff --git a/apps/desktop/src/components/pane-shell/tree/renderer/collapse-restore-affordance.test.tsx b/apps/desktop/src/components/pane-shell/tree/renderer/collapse-restore-affordance.test.tsx index 0209ae872810..02faeb78ec1f 100644 --- a/apps/desktop/src/components/pane-shell/tree/renderer/collapse-restore-affordance.test.tsx +++ b/apps/desktop/src/components/pane-shell/tree/renderer/collapse-restore-affordance.test.tsx @@ -135,15 +135,13 @@ describe('Sessions/Bots strip — #91223', () => { expect(tabEl('hermes-bots:pane')).toBeTruthy() }) - it('an explicit never still paints the strip — hide-only chrome has no other handle', () => { + it('an explicit never hides the sessions/Bots strip', () => { setTreeGroupTabStrip('g-side', 'never') - expect(tabStripVisibleForGroup(zoneAt(0))).toBe(true) + expect(tabStripVisibleForGroup(zoneAt(0))).toBe(false) render() - expect(tablist()).toBeTruthy() - expect(tabEl('sessions')).toBeTruthy() - expect(tabEl('hermes-bots:pane')).toBeTruthy() + expect(tablist()).toBeNull() }) }) diff --git a/apps/desktop/src/components/pane-shell/tree/renderer/strip-visibility.test.ts b/apps/desktop/src/components/pane-shell/tree/renderer/strip-visibility.test.ts index 668c85bf6226..c12973042fed 100644 --- a/apps/desktop/src/components/pane-shell/tree/renderer/strip-visibility.test.ts +++ b/apps/desktop/src/components/pane-shell/tree/renderer/strip-visibility.test.ts @@ -9,7 +9,6 @@ const tile = (): StripPane => ({ collapsePane: false, placement: 'main' }) const workspace = (): StripPane => ({ collapsePane: false, placement: 'main', uncloseable: true }) const toolPanel = (): StripPane => ({ collapsePane: true, placement: 'bottom' }) const sideChrome = (): StripPane => ({ collapsePane: false, placement: 'right' }) -const hideOnlyChrome = (): StripPane => ({ collapsePane: false, hideOnly: true, placement: 'left' }) describe('auto (no stored choice)', () => { it('gives a lone workspace no strip and a stack of two a strip', () => { @@ -45,18 +44,14 @@ describe('no dead zone', () => { expect(resolveTabStripVisible({ mode: 'never', shown: [toolPanel()] })).toBe(true) }) - it('keeps the strip for hide-only chrome even when the zone says never', () => { - // Sessions / Bots: the Show/Hide rows and the chips themselves live on - // the strip. Hiding it is the #91223 trap — nothing left to click. - expect(resolveTabStripVisible({ mode: 'never', shown: [hideOnlyChrome()] })).toBe(true) - expect(resolveTabStripVisible({ mode: 'never', shown: [hideOnlyChrome(), hideOnlyChrome()] })).toBe(true) - }) - it('still hides a zone that cannot strand anything', () => { - // The workspace is uncloseable, and a stack is reachable by tab cycling — - // the invariant protects handles, it does not veto hiding as such. + // The workspace is uncloseable, a stack is reachable by tab cycling, and + // hide-only chrome (sessions / Bots) keeps its panes + ⌘⌥T — the invariant + // protects handles, it does not veto hiding as such. expect(resolveTabStripVisible({ mode: 'never', shown: [workspace()] })).toBe(false) expect(resolveTabStripVisible({ mode: 'never', shown: [toolPanel(), toolPanel()] })).toBe(false) + expect(resolveTabStripVisible({ mode: 'never', shown: [sideChrome()] })).toBe(false) + expect(resolveTabStripVisible({ mode: 'never', shown: [sideChrome(), sideChrome()] })).toBe(false) }) }) diff --git a/apps/desktop/src/components/pane-shell/tree/renderer/strip-visibility.ts b/apps/desktop/src/components/pane-shell/tree/renderer/strip-visibility.ts index 201d6275893a..573e8a9879ab 100644 --- a/apps/desktop/src/components/pane-shell/tree/renderer/strip-visibility.ts +++ b/apps/desktop/src/components/pane-shell/tree/renderer/strip-visibility.ts @@ -19,9 +19,6 @@ import { paneChrome } from './track-model' export interface StripPane { /** A tool panel (terminal / logs) that collapses rather than closes. */ collapsePane: boolean - /** Standing chrome (sessions / Bots) whose only handle is the strip: - * show/hide replaces Close, and the Show/Hide rows live on the strip. */ - hideOnly?: boolean /** Contribution placement — `'main'` marks a docked tile (session, page, * preview) as opposed to standing side chrome. */ placement?: string @@ -42,9 +39,11 @@ export interface StripZone { /** * A pane is STRANDED without a strip when the strip is the only thing carrying * its handle: a lone closeable tile needs its ✕, a lone tool panel needs a chip - * to grab, hide-only chrome (sessions / Bots) needs the chip that show/hide - * lives on. The uncloseable workspace is not strandable — it cannot be closed - * or lost, so a lone chat is free to be chromeless. + * to grab. The uncloseable workspace is not strandable — it cannot be closed + * or lost, so a lone chat is free to be chromeless. Hide-only chrome (sessions + * / Bots) is the same: the panes stay, Show/Hide is a separate verb, and a + * hidden strip comes back via ⌘⌥T. Treating it as stranded at any count made + * Hide tabs a silent no-op on the sessions sidebar. * * This outranks an explicit `never` on purpose. "Hide the strip" is a request * about chrome, never a request to make a surface unreachable, and a zone that @@ -59,12 +58,6 @@ export interface StripZone { * both the menu row and ⌘⌥T became silent no-ops. */ function stranded(shown: readonly StripPane[]): boolean { - // Hide-only chrome is stranded at ANY count: it has no close verb at all, and - // both the chips and the Show/Hide rows that replace one live on the strip. - if (shown.some(pane => pane.hideOnly)) { - return true - } - if (shown.length !== 1) { return false } @@ -123,7 +116,6 @@ export function tabStripVisibleForZone(zone: { return { collapsePane: zone.isCollapsePane(id), - hideOnly: chrome.hideOnly, placement: chrome.placement, uncloseable: chrome.uncloseable } From be4bd24a5048ebf70017f9ef042400f01a5ec9a7 Mon Sep 17 00:00:00 2001 From: brooklyn! Date: Tue, 8 Sep 2026 15:20:07 -0500 Subject: [PATCH 207/227] fix(desktop): take Import session off the sidebar nav The importer stays in the command palette. The labeled nav row was clutter next to New session / Capabilities / Messaging. Co-authored-by: Cursor --- apps/desktop/src/app/chat/sidebar/index.tsx | 8 -------- apps/desktop/src/app/types.ts | 2 +- apps/desktop/src/i18n/ar.ts | 3 +-- apps/desktop/src/i18n/en.ts | 3 +-- apps/desktop/src/i18n/ja.ts | 3 +-- apps/desktop/src/i18n/ru.ts | 3 +-- apps/desktop/src/i18n/zh-hant.ts | 3 +-- apps/desktop/src/i18n/zh.ts | 3 +-- website/docs/user-guide/sessions.md | 4 ++-- 9 files changed, 9 insertions(+), 23 deletions(-) diff --git a/apps/desktop/src/app/chat/sidebar/index.tsx b/apps/desktop/src/app/chat/sidebar/index.tsx index a20b52e70209..d01f31e90fc9 100644 --- a/apps/desktop/src/app/chat/sidebar/index.tsx +++ b/apps/desktop/src/app/chat/sidebar/index.tsx @@ -136,7 +136,6 @@ import { ARTIFACTS_ROUTE, CRON_ROUTE, MESSAGING_ROUTE, - SESSION_IMPORT_ROUTE, SIDEBAR_NAV_AREA, type SidebarNavContribution, SKILLS_ROUTE @@ -230,12 +229,6 @@ const SIDEBAR_NAV: SidebarNavItem[] = [ icon: props => , route: CRON_ROUTE, keybindActionId: 'nav.cron' - }, - { - id: 'session-import', - label: '', - icon: props => , - route: SESSION_IMPORT_ROUTE } ] @@ -1469,7 +1462,6 @@ export function ChatSidebar({ (item.id === 'messaging' && currentView === 'messaging') || (item.id === 'artifacts' && currentView === 'artifacts') || (item.id === 'cron' && currentView === 'cron') || - (item.id === 'session-import' && currentView === 'session-import') || // Contributed rows light up at their own route. (currentView === 'extension' && Boolean(item.route) && pathname === item.route) diff --git a/apps/desktop/src/app/types.ts b/apps/desktop/src/app/types.ts index 4d3f7545f11b..3bf09f66b150 100644 --- a/apps/desktop/src/app/types.ts +++ b/apps/desktop/src/app/types.ts @@ -163,7 +163,7 @@ export type CommandDispatchResponse = | PrefillCommandDispatchResponse export type SidebarNavId = - 'artifacts' | 'command-center' | 'cron' | 'messaging' | 'new-session' | 'session-import' | 'settings' | 'skills' + 'artifacts' | 'command-center' | 'cron' | 'messaging' | 'new-session' | 'settings' | 'skills' export interface SidebarNavItem { /** Built-in view id, or a contributed row's namespaced contribution id. */ diff --git a/apps/desktop/src/i18n/ar.ts b/apps/desktop/src/i18n/ar.ts index 1ce7297289a5..42e2bced5652 100644 --- a/apps/desktop/src/i18n/ar.ts +++ b/apps/desktop/src/i18n/ar.ts @@ -1754,8 +1754,7 @@ export const ar = defineLocale({ chat: 'المحادثة', settings: 'الإعدادات', cron: 'المهام المجدولة', - agents: 'الوكلاء', - 'session-import': 'استيراد جلسة' + agents: 'الوكلاء' }, searchAria: 'البحث في الجلسات', searchPlaceholder: 'البحث في الجلسات...', diff --git a/apps/desktop/src/i18n/en.ts b/apps/desktop/src/i18n/en.ts index 4fe8bcb34fc2..0053466ece8c 100644 --- a/apps/desktop/src/i18n/en.ts +++ b/apps/desktop/src/i18n/en.ts @@ -2388,8 +2388,7 @@ export const en: Translations = { skills: 'Capabilities', messaging: 'Messaging', artifacts: 'Artifacts', - cron: 'Scheduled jobs', - 'session-import': 'Import session' + cron: 'Scheduled jobs' }, searchAria: 'Search sessions', searchPlaceholder: 'Search sessions…', diff --git a/apps/desktop/src/i18n/ja.ts b/apps/desktop/src/i18n/ja.ts index 38f2611b35bb..8acb78e25e32 100644 --- a/apps/desktop/src/i18n/ja.ts +++ b/apps/desktop/src/i18n/ja.ts @@ -2061,8 +2061,7 @@ export const ja = defineLocale({ skills: 'スキルとツール', messaging: 'メッセージング', artifacts: 'アーティファクト', - cron: 'スケジュール済みジョブ', - 'session-import': 'セッションを取り込む' + cron: 'スケジュール済みジョブ' }, searchAria: 'セッションを検索', searchPlaceholder: 'セッションを検索…', diff --git a/apps/desktop/src/i18n/ru.ts b/apps/desktop/src/i18n/ru.ts index 8e7afb5de4d7..7a6ee0c3cd88 100644 --- a/apps/desktop/src/i18n/ru.ts +++ b/apps/desktop/src/i18n/ru.ts @@ -2421,8 +2421,7 @@ export const ru = defineLocale({ skills: 'Возможности', messaging: 'Сообщения', artifacts: 'Артефакты', - cron: 'Запланированные задачи', - 'session-import': 'Импортировать сессию' + cron: 'Запланированные задачи' }, searchAria: 'Поиск сеансов', searchPlaceholder: 'Поиск сеансов…', diff --git a/apps/desktop/src/i18n/zh-hant.ts b/apps/desktop/src/i18n/zh-hant.ts index 0b754c5e50b6..04017abe19fa 100644 --- a/apps/desktop/src/i18n/zh-hant.ts +++ b/apps/desktop/src/i18n/zh-hant.ts @@ -1986,8 +1986,7 @@ export const zhHant = defineLocale({ skills: '技能與工具', messaging: '訊息平台', artifacts: '成品', - cron: '排程工作', - 'session-import': '匯入工作階段' + cron: '排程工作' }, searchAria: '搜尋工作階段', searchPlaceholder: '搜尋工作階段…', diff --git a/apps/desktop/src/i18n/zh.ts b/apps/desktop/src/i18n/zh.ts index 2fc5561b553e..a20d7a7ba6cb 100644 --- a/apps/desktop/src/i18n/zh.ts +++ b/apps/desktop/src/i18n/zh.ts @@ -2553,8 +2553,7 @@ export const zh: Translations = { skills: '技能与工具', messaging: '消息平台', artifacts: '产物', - cron: '定时任务', - 'session-import': '导入会话' + cron: '定时任务' }, searchAria: '搜索会话', searchPlaceholder: '搜索会话…', diff --git a/website/docs/user-guide/sessions.md b/website/docs/user-guide/sessions.md index 6adc541cd586..3943dcf93a25 100644 --- a/website/docs/user-guide/sessions.md +++ b/website/docs/user-guide/sessions.md @@ -663,8 +663,8 @@ the id plus a ready-to-paste `hermes --resume ` command. `--resume @claude` / `--resume @codex` show the same picker and drop you straight into the imported conversation. -**Hermes Desktop** has the same importer under **Import session** in the -sidebar (also in the command palette). It lists the logs on the machine the +**Hermes Desktop** has the same importer in the command palette (**Import +session**). It lists the logs on the machine the connected backend runs on — not the computer running the app — shows a read-only preview, and **Continue in Hermes** copies the conversation into the selected profile. Browsing never writes to your session store, importing never From 3c0a84f94b2b682997a228eac84690d3d94dfef1 Mon Sep 17 00:00:00 2001 From: brooklyn! Date: Tue, 8 Sep 2026 15:28:55 -0500 Subject: [PATCH 208/227] chore: map MacBook-Air contributor email Co-authored-by: Cursor --- contributors/emails/brooklyn@Brooklyns-MacBook-Air.local | 1 + 1 file changed, 1 insertion(+) create mode 100644 contributors/emails/brooklyn@Brooklyns-MacBook-Air.local diff --git a/contributors/emails/brooklyn@Brooklyns-MacBook-Air.local b/contributors/emails/brooklyn@Brooklyns-MacBook-Air.local new file mode 100644 index 000000000000..cb0a5260d752 --- /dev/null +++ b/contributors/emails/brooklyn@Brooklyns-MacBook-Air.local @@ -0,0 +1 @@ +OutThisLife From acbe72ed1feaf979622f732450bf1adf29357eeb Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Tue, 8 Sep 2026 10:19:40 -0700 Subject: [PATCH 209/227] fix: keep Bot Mode pet selection rings inside the gallery --- .../src/plugins/hermes-bots/pet.test.tsx | 51 ++++++++++++++++++- apps/desktop/src/plugins/hermes-bots/pet.tsx | 4 +- 2 files changed, 53 insertions(+), 2 deletions(-) diff --git a/apps/desktop/src/plugins/hermes-bots/pet.test.tsx b/apps/desktop/src/plugins/hermes-bots/pet.test.tsx index 6e913cb28b09..a3d08e8728e9 100644 --- a/apps/desktop/src/plugins/hermes-bots/pet.test.tsx +++ b/apps/desktop/src/plugins/hermes-bots/pet.test.tsx @@ -12,7 +12,7 @@ * failure must be evicted; a success must not be refetched. */ -import { render, waitFor } from '@testing-library/react' +import { fireEvent, render, waitFor } from '@testing-library/react' import { afterEach, beforeEach, describe, expect, it, vi } from 'vitest' const { hostMock, UnboundedCache, useQueryMock } = vi.hoisted(() => ({ @@ -81,6 +81,55 @@ afterEach(() => { vi.restoreAllMocks() }) +describe('the pet gallery', () => { + it('reserves paint space around boundary tiles inside the bounded scroller', async () => { + stubFetch(async () => ({ blob: async () => new Blob() })) + const PetTab = await loadPetTab() + const view = render() + await waitFor(() => expect(view.container.querySelector('img')).toBeTruthy()) + const tile = view.getByText('Axolotl').closest('button')! + const scroller = tile.parentElement!.parentElement! + + // The selection ring paints one pixel outside each tile. The scrollport + // must leave room for it even at the first/last row and outer columns. + const style = getComputedStyle(scroller) + + for (const side of ['paddingTop', 'paddingRight', 'paddingBottom', 'paddingLeft'] as const) { + expect(parseFloat(style[side]) || 0).toBeGreaterThanOrEqual(1) + } + + expect(parseFloat(style.maxHeight)).toBeGreaterThan(0) + view.unmount() + }) + + it('keeps selection while scrolling for more and resets the search window', async () => { + useQueryMock.mockReturnValue({ data: { pets: Array.from({ length: 60 }, (_, i) => ({ + displayName: `Pet ${i}`, slug: `pet-${i}`, spritesheetUrl: SHEET + })) } }) + stubFetch(async () => ({ blob: async () => new Blob() })) + const PetTab = await loadPetTab() + const onImage = vi.fn() + const view = render() + const first = view.getByText('Pet 0').closest('button')! + fireEvent.click(first) + await waitFor(() => expect(onImage).toHaveBeenCalledWith('data:image/png;base64,ok')) + const scroller = first.parentElement!.parentElement! + Object.defineProperties(scroller, { + clientHeight: { value: 220 }, scrollHeight: { value: 600 }, scrollTop: { value: 400 } + }) + fireEvent.scroll(scroller) + expect(view.getByText('Pet 47')).toBeTruthy() + expect(view.queryByText('Pet 48')).toBeNull() + expect(first.className).toContain('ring-1') + fireEvent.change(view.getByRole('textbox'), { target: { value: 'Pet 59' } }) + expect(view.getByText('Pet 59')).toBeTruthy() + fireEvent.change(view.getByRole('textbox'), { target: { value: '' } }) + expect(view.queryByText('Pet 24')).toBeNull() + expect(onImage).toHaveBeenCalledTimes(1) + view.unmount() + }) +}) + describe('the sprite-frame cache', () => { it('never leaves a failed fetch parked in the cache', async () => { stubFetch(async () => { diff --git a/apps/desktop/src/plugins/hermes-bots/pet.tsx b/apps/desktop/src/plugins/hermes-bots/pet.tsx index 700e1b98f46b..b14fded2783a 100644 --- a/apps/desktop/src/plugins/hermes-bots/pet.tsx +++ b/apps/desktop/src/plugins/hermes-bots/pet.tsx @@ -230,7 +230,9 @@ export function PetTab({ image, onImage }: PetTabProps) { className="overflow-y-auto" onScroll={onScroll} style={{ - maxHeight: 220 + // Leave room for the selection ring outside boundary tiles. + maxHeight: 220, + padding: 2 }} >
From 9d661c2c924dcfa419f70166063cf45a5a4957db Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Tue, 8 Sep 2026 10:50:25 -0700 Subject: [PATCH 210/227] fix(prompt): keep memory guidance within available tools --- agent/prompt_builder.py | 22 +++++++++-------- agent/system_prompt.py | 13 +++++----- tests/agent/test_system_prompt.py | 24 +++++++++++++++++++ tests/run_agent/test_run_agent.py | 11 +++++---- .../docs/developer-guide/prompt-assembly.md | 8 +++---- 5 files changed, 53 insertions(+), 25 deletions(-) diff --git a/agent/prompt_builder.py b/agent/prompt_builder.py index f7f6956f344d..aedf7f43484d 100644 --- a/agent/prompt_builder.py +++ b/agent/prompt_builder.py @@ -158,15 +158,11 @@ def _strip_yaml_frontmatter(content: str) -> str: ) -# Memory guidance (#95681, consolidated): ONE block from ONE builder. The opening frame adapts to which -# stores config enables; everything else is written exactly once. Leads with the positive posture (save -# proactively, replace when full) — the routing rules come after, as refinements, not as the headline. WHAT -# belongs in memory is the memory tool schema's job and is never re-taught here. -def build_memory_guidance(memory_enabled: bool = True, profile_enabled: bool = True) -> str: - """ONE memory-guidance block whose opening frame adapts to the enabled store(s); "" when both are off. - - Positive posture first, routing rules as refinements. WHAT belongs in memory is the tool schema's job. - """ +# Keep the every-session memory scope even when task knowledge cannot be saved as a skill. +def build_memory_guidance( + memory_enabled: bool = True, profile_enabled: bool = True, *, skill_manage_available: bool = True, +) -> str: + """Adapt store and skill-write guidance without widening what belongs in memory.""" if not memory_enabled and not profile_enabled: return "" if memory_enabled: @@ -180,11 +176,17 @@ def build_memory_guidance(memory_enabled: bool = True, profile_enabled: bool = T "loaded into each new session's context; save durable facts about the user with the " "memory tool (target='user') — the built-in notes store is disabled, so never target='memory'. " ) - return frame + ( + skill_routing = ( "Skills come first: when you learn something while doing a task — a " "procedure, a pitfall, and the user's preferences and corrections " "for that kind of work — record it in the skill you used or built " "for the task (skill_manage), where it loads only when relevant. " + if skill_manage_available else + "Task-specific knowledge — procedures, pitfalls, and the user's preferences " + "and corrections for that kind of work — belongs in skills, not in memory, " + "even when skill writing is unavailable. " + ) + return frame + skill_routing + ( "Memory is the narrow exception for facts that apply to EVERY " "session regardless of task (who the user is, environment facts, " "standing conventions with no task home); it has a hard character " diff --git a/agent/system_prompt.py b/agent/system_prompt.py index e7f04271d056..34338b16d540 100644 --- a/agent/system_prompt.py +++ b/agent/system_prompt.py @@ -19,8 +19,8 @@ from agent.prompt_builder import ( DEFAULT_AGENT_IDENTITY, EXECUTION_GUIDANCE_MODELS, GOOGLE_MODEL_OPERATIONAL_GUIDANCE, - HERMES_AGENT_HELP_GUIDANCE, HERMES_AGENT_HELP_GUIDANCE_NO_SKILLS, KANBAN_GUIDANCE, MEMORY_GUIDANCE, - USER_PROFILE_GUIDANCE, PARALLEL_TOOL_CALL_GUIDANCE, PLATFORM_HINTS, SESSION_SEARCH_GUIDANCE, + HERMES_AGENT_HELP_GUIDANCE, HERMES_AGENT_HELP_GUIDANCE_NO_SKILLS, KANBAN_GUIDANCE, + PARALLEL_TOOL_CALL_GUIDANCE, PLATFORM_HINTS, SESSION_SEARCH_GUIDANCE, SKILLS_GUIDANCE, STEER_CHANNEL_NOTE, TASK_COMPLETION_GUIDANCE, TELEGRAM_RICH_MESSAGES_HINT, TOOL_USE_ENFORCEMENT_GUIDANCE, TOOL_USE_ENFORCEMENT_MODELS, drain_truncation_warnings, ) @@ -277,10 +277,11 @@ def _tool_guidance_block(agent: Any) -> Optional[str]: # available"; with only USER.md enabled the narrower block is used. memory_guidance = None if "memory" in names: - if getattr(agent, "_memory_enabled", True): - memory_guidance = MEMORY_GUIDANCE - elif getattr(agent, "_user_profile_enabled", True): - memory_guidance = USER_PROFILE_GUIDANCE + memory_guidance = _pb.build_memory_guidance( + getattr(agent, "_memory_enabled", True), + getattr(agent, "_user_profile_enabled", True), + skill_manage_available="skill_manage" in names, + ) # Kanban lifecycle: resolved once at __init__ (_kanban_worker_guidance); # the kanban_show fallback covers code paths that bypass agent_init. _kanban_guidance = getattr(agent, "_kanban_worker_guidance", None) diff --git a/tests/agent/test_system_prompt.py b/tests/agent/test_system_prompt.py index 7f0862cd49e3..0618ddf1cad2 100644 --- a/tests/agent/test_system_prompt.py +++ b/tests/agent/test_system_prompt.py @@ -6,6 +6,8 @@ from unittest.mock import patch from zoneinfo import ZoneInfo +import pytest + from agent.system_prompt import build_system_prompt, build_system_prompt_parts @@ -55,6 +57,28 @@ def fake_context_files( return captured["cwd"] +@pytest.mark.parametrize("stores", [(True, True), (False, True), (True, False), (False, False)]) +@pytest.mark.parametrize("names", [ + set(), {"memory"}, {"memory", "skill_view", "skills_list"}, + {"memory", "skill_view", "skills_list", "skill_manage"}, +]) +def test_memory_guidance_respects_available_writes(stores, names, monkeypatch, tmp_path): + monkeypatch.chdir(tmp_path) + agent = _make_agent(valid_tool_names=names, skip_context_files=True, + _memory_enabled=stores[0], _user_profile_enabled=stores[1]) + prompt = build_system_prompt(agent) + enabled = "memory" in names and any(stores) + assert ("Memory is the narrow exception" in prompt) == enabled + assert ("(skill_manage)" in prompt) == (enabled and "skill_manage" in names) + if enabled: + assert "EVERY session regardless of task" in prompt + assert "procedures and workflows belong in skills" in prompt + if "skill_manage" not in names: + assert "not in memory" in prompt + if enabled and not stores[0]: + assert "never target='memory'" in prompt + + class TestContextFileCwd: def test_none_when_terminal_cwd_unset(self, monkeypatch): # Unset → None, so discovery falls back to the launch dir inside diff --git a/tests/run_agent/test_run_agent.py b/tests/run_agent/test_run_agent.py index 40bc52a06f7b..a807fb7a6a56 100644 --- a/tests/run_agent/test_run_agent.py +++ b/tests/run_agent/test_run_agent.py @@ -862,11 +862,10 @@ def test_can_use_soul_identity_even_when_context_files_are_skipped(self): def test_memory_guidance_when_memory_tool_loaded(self, agent_with_memory_tool): - from agent.prompt_builder import MEMORY_GUIDANCE - agent_with_memory_tool._memory_enabled = True prompt = agent_with_memory_tool._build_system_prompt() - assert MEMORY_GUIDANCE in prompt + assert "Memory is the narrow exception" in prompt + assert "(skill_manage)" not in prompt def test_no_memory_guidance_when_both_builtin_stores_disabled( self, agent_with_memory_tool @@ -895,13 +894,15 @@ def test_profile_guidance_when_only_user_profile_enabled( MEMORY.md store that does not exist in this configuration, so the profile-specific block is injected instead. """ - from agent.prompt_builder import MEMORY_GUIDANCE, USER_PROFILE_GUIDANCE + from agent.prompt_builder import MEMORY_GUIDANCE agent_with_memory_tool._memory_enabled = False agent_with_memory_tool._user_profile_enabled = True prompt = agent_with_memory_tool._build_system_prompt() assert MEMORY_GUIDANCE not in prompt - assert USER_PROFILE_GUIDANCE in prompt + assert "memory tool (target='user')" in prompt + assert "never target='memory'" in prompt + assert "(skill_manage)" not in prompt diff --git a/website/docs/developer-guide/prompt-assembly.md b/website/docs/developer-guide/prompt-assembly.md index aeae73d6a028..41ababfdb1c4 100644 --- a/website/docs/developer-guide/prompt-assembly.md +++ b/website/docs/developer-guide/prompt-assembly.md @@ -67,10 +67,10 @@ You value correctness, clarity, and efficiency. ... # Layer 2: Tool-aware behavior guidance -You have persistent memory across sessions. Save durable facts using -the memory tool: user preferences, environment details, tool quirks, -and stable conventions. Memory is injected into every turn, so keep -it compact and focused on facts that will still matter later. +Task-learned procedures, pitfalls, and task-specific preferences belong +in skills. Memory is the narrow exception for facts that apply to EVERY +session regardless of task. Skill-writing instructions appear here only +when skill_manage is available; its absence does not widen memory's scope. ... When the user references something from a past conversation or you suspect relevant cross-session context exists, use session_search From 19cd839d54c88a89995dd0cdff27dc7495aa1dc1 Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Tue, 8 Sep 2026 10:54:51 -0700 Subject: [PATCH 211/227] fix(compression): keep lean tails lean after auxiliary feasibility Lowering the session trigger must not replace the window-relative lean selection budget with threshold times target_ratio. Invalidate the lean cache through the existing property while preserving explicit legacy and external-engine fallback behavior. Narrow adaptation of the aux-sync diagnosis and invariants in #93576, without adding a required recalibration method to context engines. Related: #95681, #93576 Co-authored-by: Turgut Kural <58116817+TurgutKural@users.noreply.github.com> --- agent/conversation_compression.py | 9 ++-- .../run_agent/test_compression_feasibility.py | 41 +++++++++++++++++++ .../context-compression-and-caching.md | 11 +++++ 3 files changed, 58 insertions(+), 3 deletions(-) diff --git a/agent/conversation_compression.py b/agent/conversation_compression.py index 7a82999937ed..12359913edb0 100644 --- a/agent/conversation_compression.py +++ b/agent/conversation_compression.py @@ -1697,13 +1697,16 @@ def _lower_threshold_to_aux_context( ) -> None: """Lower the live threshold to the aux model's window and tell the user how to fix config. The summariser sends one user prompt (no system/tools), so threshold == aux_context is safe. - tail_token_budget and threshold_percent are kept in lockstep (as update_model does) or the 1.5x tail - ceiling exceeds the trigger and re-fires.""" + Retention is recalibrated through its selected policy: lean is window-relative; + only legacy follows the lowered threshold.""" compressor = agent.context_compressor old_threshold = compressor.threshold_tokens new_threshold = compressor.threshold_tokens = aux_context summary_target_ratio = getattr(compressor, "summary_target_ratio", None) - if isinstance(summary_target_ratio, (int, float)): + if getattr(compressor, "tail_mode", None) == "lean": + # Keep the window-relative policy owned by the compressor property. + compressor._tail_token_budget = None + elif isinstance(summary_target_ratio, (int, float)): compressor.tail_token_budget = int(new_threshold * summary_target_ratio) main_ctx = compressor.context_length if main_ctx: diff --git a/tests/run_agent/test_compression_feasibility.py b/tests/run_agent/test_compression_feasibility.py index 9017ea67d9c3..a6ba637e87c8 100644 --- a/tests/run_agent/test_compression_feasibility.py +++ b/tests/run_agent/test_compression_feasibility.py @@ -66,6 +66,47 @@ def _make_agent( return agent +@pytest.mark.parametrize("main_context,aux_context", [(1_000_000, 512_000), (400_000, 80_000)]) +def test_aux_sync_keeps_lean_tail_policy(main_context, aux_context): + """Lowering only the trigger must not change window-relative retention.""" + agent = _make_agent(main_context=main_context) + compressor = agent.context_compressor = ContextCompressor( + "test-main-model", config_context_length=main_context, + threshold_percent=0.85, quiet_mode=True, + ) + before = compressor.tail_token_budget + agent._emit_status = lambda message: None + client = MagicMock(base_url="http://localhost/v1", api_key="test-key") + with patch("agent.auxiliary_client.get_text_auxiliary_client", return_value=(client, "aux")), \ + patch("agent.model_metadata.get_model_context_length", return_value=aux_context): + agent._check_compression_model_feasibility() + assert compressor.threshold_tokens == aux_context + assert compressor.tail_token_budget == before + # Repeated feasibility and subsequent model recalibration retain policy. + agent._check_compression_model_feasibility() + assert compressor.tail_token_budget == before + compressor.update_model("test-main-model", context_length=main_context) + assert compressor.tail_token_budget == before + + +def test_aux_sync_legacy_tail_follows_lowered_threshold(): + """Explicit legacy retention follows the current trigger, not its old cache.""" + agent = _make_agent(main_context=1_000_000) + compressor = agent.context_compressor = ContextCompressor( + "test-main-model", config_context_length=1_000_000, + threshold_percent=0.85, tail_mode="legacy", quiet_mode=True, + ) + before = compressor.tail_token_budget + agent._emit_status = lambda message: None + client = MagicMock(base_url="http://localhost/v1", api_key="test-key") + with patch("agent.auxiliary_client.get_text_auxiliary_client", return_value=(client, "aux")), \ + patch("agent.model_metadata.get_model_context_length", return_value=512_000): + agent._check_compression_model_feasibility() + assert compressor.threshold_tokens == 512_000 + assert compressor.tail_token_budget < before + assert compressor.tail_token_budget == int(compressor.threshold_tokens * compressor.summary_target_ratio) + + # ── Core warning logic ────────────────────────────────────────────── diff --git a/website/docs/developer-guide/context-compression-and-caching.md b/website/docs/developer-guide/context-compression-and-caching.md index 0d08bdcc57a1..9221b2ccffff 100644 --- a/website/docs/developer-guide/context-compression-and-caching.md +++ b/website/docs/developer-guide/context-compression-and-caching.md @@ -216,6 +216,17 @@ Consumers observe the mode rather than diffing session ids: Set `in_place: false` to restore the legacy rotating path, where each compaction commits a new session id linked to the previous one via `parent_session_id`. +### Auxiliary feasibility and tail retention + +A smaller auxiliary compression model can lower the live compression trigger without +changing the selected tail policy. In `lean` mode the selection budget remains based +on the **main model's context window**: 2.5%, clamped to 10K–25K tokens. For example, +a 1M main model with a 512K auxiliary model retains a 25K selection budget even when +feasibility lowers its trigger from 850K to 512K. Explicit `legacy` mode instead +recomputes `threshold_tokens × target_ratio` (102,400 tokens at 512K × 0.20). +These are tail-selection budgets, not strict limits on the entire compacted context: +protected messages, boundary alignment, summaries, and anchors can add tokens. + ### Per-model threshold overrides `compression.model_thresholds` lets you trigger compaction at different points From 9d66e76c5d1611d5a5216249b7f4d362aefcf1c4 Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Tue, 8 Sep 2026 10:20:38 -0700 Subject: [PATCH 212/227] feat(desktop): let users order Group Chat rooms Add Move up/down controls for actual rooms without changing bot or folder ordering. Preserve default pin/activity ordering until an explicit move, retain hidden room slots, and persist Desktop-local order through room updates, mirror merges, and hydration. No membership or routing writes. Adapted narrowly from the group ordering idea in archived NousResearch/Hermes-Bot-Mode#105 by @onuraycicek; rename already exists. Co-authored-by: Onur Aycicek --- apps/desktop/package.json | 2 +- .../src/plugins/hermes-bots/group-chat.ts | 4 ++ .../plugins/hermes-bots/group-order.test.ts | 29 ++++++++ .../src/plugins/hermes-bots/group-order.ts | 34 +++++++++ .../src/plugins/hermes-bots/plugin.tsx | 2 + .../src/plugins/hermes-bots/roster-pane.tsx | 69 +++++++++++++------ apps/desktop/src/plugins/hermes-bots/types.ts | 2 + contributors/emails/onur.m.aycicek@gmail.com | 2 + package-lock.json | 2 +- website/docs/user-guide/bot-mode.md | 2 + 10 files changed, 126 insertions(+), 22 deletions(-) create mode 100644 apps/desktop/src/plugins/hermes-bots/group-order.test.ts create mode 100644 apps/desktop/src/plugins/hermes-bots/group-order.ts create mode 100644 contributors/emails/onur.m.aycicek@gmail.com diff --git a/apps/desktop/package.json b/apps/desktop/package.json index 03d9a9f78570..ac9293c32365 100644 --- a/apps/desktop/package.json +++ b/apps/desktop/package.json @@ -2,7 +2,7 @@ "name": "hermes", "productName": "Hermes", "private": true, - "version": "0.17.1", + "version": "0.17.2", "description": "Native desktop shell for Hermes Agent.", "author": "Nous Research", "repository": { diff --git a/apps/desktop/src/plugins/hermes-bots/group-chat.ts b/apps/desktop/src/plugins/hermes-bots/group-chat.ts index 9949c57824b1..92684e5cfabd 100644 --- a/apps/desktop/src/plugins/hermes-bots/group-chat.ts +++ b/apps/desktop/src/plugins/hermes-bots/group-chat.ts @@ -749,6 +749,8 @@ export function durableGroupChatRooms(all: Record = $groupCha // already carries. roomId: typeof room.roomId === 'string' && room.roomId ? room.roomId : null, image: room.image || null, + rosterOrder: room.rosterOrder, + pinned: room.pinned, syncRevision: Math.max(0, Number(room.syncRevision || 0)) } } @@ -1348,6 +1350,8 @@ export function updateGroupChat( roomId: typeof room.roomId === 'string' && room.roomId ? room.roomId : null, // Room picture (small data URL, same normalization as bot avatars). image: room.image || null, + rosterOrder: room.rosterOrder, + pinned: room.pinned, syncRevision: Math.max(0, Number(room.syncRevision || 0)) } } diff --git a/apps/desktop/src/plugins/hermes-bots/group-order.test.ts b/apps/desktop/src/plugins/hermes-bots/group-order.test.ts new file mode 100644 index 000000000000..1dbc43a3388a --- /dev/null +++ b/apps/desktop/src/plugins/hermes-bots/group-order.test.ts @@ -0,0 +1,29 @@ +import { describe, expect, it } from 'vitest' + +import { reorderGroupRows, sortGroupRosterRows } from './group-order' + +const rows = [ + { kind: 'group' as const, name: 'Older', activity: 1, pinned: false }, + { kind: 'bot' as const, name: 'Bot', activity: 2, pinned: false }, + { kind: 'group' as const, name: 'Newer', activity: 3, pinned: false }, + { kind: 'group' as const, name: 'Pinned', activity: 0, pinned: true } +] + +describe('room display order', () => { + it('keeps legacy recency until explicitly ordered, then only replaces room slots within pin bands', () => { + expect(sortGroupRosterRows(rows, {}).map(row => row.name)).toEqual(['Pinned', 'Newer', 'Bot', 'Older']) + const rooms = { Older: { rosterOrder: 0 }, Newer: { rosterOrder: 1 } } + expect(sortGroupRosterRows(rows, rooms).map(row => row.name)).toEqual(['Pinned', 'Older', 'Bot', 'Newer']) + expect(sortGroupRosterRows(rows.map(row => ({ ...row, activity: row.name === 'Newer' ? 999 : row.activity })), rooms).filter(row => row.kind === 'group').map(row => row.name)).toEqual(['Pinned', 'Older', 'Newer']) + expect(rows[0].name).toBe('Older') + }) + + it('moves only visible same-band rooms while retaining hidden slots and ignoring stale targets', () => { + const ordered = sortGroupRosterRows(rows.filter(row => row.kind === 'group'), {}) + expect(reorderGroupRows(ordered, 'Older', -1)).toEqual(['Pinned', 'Older', 'Newer']) + expect(reorderGroupRows(ordered, 'Newer', -1)).toBeNull() + expect(reorderGroupRows(ordered, 'deleted', 1)).toBeNull() + const hidden = { kind: 'group' as const, name: 'Hidden', activity: 2, pinned: false } + expect(reorderGroupRows([ordered[0], ordered[1], hidden, ordered[2]], 'Older', -1, ['Newer', 'Older'])).toEqual(['Pinned', 'Older', 'Hidden', 'Newer']) + }) +}) diff --git a/apps/desktop/src/plugins/hermes-bots/group-order.ts b/apps/desktop/src/plugins/hermes-bots/group-order.ts new file mode 100644 index 000000000000..e3a47fc7cc7b --- /dev/null +++ b/apps/desktop/src/plugins/hermes-bots/group-order.ts @@ -0,0 +1,34 @@ +interface OrderRow { + activity: number + kind: 'bot' | 'group' + name?: string + pinned: boolean +} + +/** Reorder room slots, not bots or folders. Pinning remains the outer band. */ +export function sortGroupRosterRows(rows: T[], rooms: Record): T[] { + const legacy = rows.slice().sort((a, b) => Number(b.pinned) - Number(a.pinned) || b.activity - a.activity) + + const groups = legacy.filter(row => row.kind === 'group').sort((a, b) => + Number(b.pinned) - Number(a.pinned) || + (rooms[a.name!]?.rosterOrder ?? Infinity) - (rooms[b.name!]?.rosterOrder ?? Infinity) + ) + + let index = 0 + + return legacy.map(row => row.kind === 'group' ? groups[index++] : row) +} + +/** Swap visible neighbours without dropping filtered-out rooms from the order. */ +export function reorderGroupRows(rows: OrderRow[], name: string, delta: -1 | 1, visible?: string[]): string[] | null { + const row = rows.find(row => row.name === name) + const band = rows.filter(candidate => candidate.kind === 'group' && candidate.pinned === row?.pinned && (!visible || visible.includes(candidate.name!))) + const index = band.findIndex(candidate => candidate.name === name) + const neighbour = index >= 0 ? band[index + delta] : undefined + + if (!neighbour) { + return null + } + + return rows.filter(row => row.kind === 'group').map(row => row.name === name ? neighbour.name! : row.name === neighbour.name ? name : row.name!) +} diff --git a/apps/desktop/src/plugins/hermes-bots/plugin.tsx b/apps/desktop/src/plugins/hermes-bots/plugin.tsx index 761776fa2325..8032e16134ff 100644 --- a/apps/desktop/src/plugins/hermes-bots/plugin.tsx +++ b/apps/desktop/src/plugins/hermes-bots/plugin.tsx @@ -246,6 +246,8 @@ export default { members: Array.isArray(room.members) ? room.members : [], roomId: typeof room.roomId === 'string' && room.roomId ? room.roomId : null, image: typeof room.image === 'string' && room.image ? room.image : null, + rosterOrder: Number.isFinite(room.rosterOrder) ? room.rosterOrder : undefined, + pinned: Boolean(room.pinned), syncRevision: Math.max(0, Number(room.syncRevision || 0)), epoch: 0, running: false diff --git a/apps/desktop/src/plugins/hermes-bots/roster-pane.tsx b/apps/desktop/src/plugins/hermes-bots/roster-pane.tsx index 8d82d7dd366a..eb169ec2c983 100644 --- a/apps/desktop/src/plugins/hermes-bots/roster-pane.tsx +++ b/apps/desktop/src/plugins/hermes-bots/roster-pane.tsx @@ -56,9 +56,10 @@ import { useRoster } from './data' import { EditProfileDialog } from './edit-profile-dialog' -import { $groupChats, $groupChatWorkspace, $groupNeedsYou } from './group-chat' +import { $groupChats, $groupChatWorkspace, $groupNeedsYou, updateGroupChat } from './group-chat' import { disbandGroupChat, GroupChatWorkspace, openGroupChat } from './group-chat-view' import { groupChatMemberBots, groupChatNames, groupLastActivity } from './group-membership' +import { reorderGroupRows, sortGroupRosterRows } from './group-order' import { $groupMainTabsRev, shouldRenderGroupChatInPane } from './group-panes' import { $showHiddenBots, isBotHidden, isBotPinned } from './hidden-bots' import { useBots } from './i18n' @@ -414,20 +415,32 @@ export function BotsPane() { active: activeRosterKeys.has(botRosterKey(bot)) })) - const sortRosterRows = (rows: T[]): T[] => - rows.slice().sort((a, b) => { - const pa = a.pinned ? 1 : 0 - const pb = b.pinned ? 1 : 0 + const rosterRows = sortGroupRosterRows([...botRows, ...groupRows], groupRooms) + const sortedGroupRows = sortGroupRosterRows(groupRows, groupRooms) - if (pa !== pb) { - return pb - pa - } + const moveRoom = (name: string, delta: -1 | 1) => { + // Read at the gesture, not the last render: a sync or disband may have + // replaced this room in the meantime. Ordering never writes bot metadata. + const current = $groupChats.get() + + if (current[name]?.roomId !== groupRooms[name]?.roomId || current[name]?.tombstone) { + return + } + + const rows = groupChatNames($botMeta.get(), current).map(name => ({ + kind: 'group' as const, + name, + pinned: Boolean(current[name]?.pinned), + activity: groupLastActivity(current[name]) + })) + + const order = reorderGroupRows(sortGroupRosterRows(rows, current), name, delta, sortedGroupRows.map(row => row.name)) - return b.activity - a.activity + order?.forEach((name, rosterOrder) => { + updateGroupChat(name, room => ({ ...room, rosterOrder }), { sync: false }) }) + } - const rosterRows = sortRosterRows([...botRows, ...groupRows]) - const sortedGroupRows = sortRosterRows(groupRows) const gatewaySections = rosterGatewaySections(botRows, gatewayOptions, gatewayFilter) const showGatewaySections = gatewaySections.sectioned && botRows.length > 0 @@ -567,15 +580,31 @@ export function BotsPane() { ) const renderGroupRow = (row: { members: GroupMember[]; name: string }) => ( - +
+ +
+ {([-1, 1] as const).map(delta => ( + + + + ))} +
+
) const removeSection = (id: string) => { diff --git a/apps/desktop/src/plugins/hermes-bots/types.ts b/apps/desktop/src/plugins/hermes-bots/types.ts index 7770c1b5d9e5..7ee6aad189c3 100644 --- a/apps/desktop/src/plugins/hermes-bots/types.ts +++ b/apps/desktop/src/plugins/hermes-bots/types.ts @@ -176,6 +176,8 @@ export interface GroupChat { syncRevision?: number /** Left behind when a room is disbanded, so sync can't resurrect it. */ tombstone?: boolean + /** Local display order, deliberately excluded from the gateway mirror. */ + rosterOrder?: number /** Read when ordering rooms; no write site in the plugin today. */ pinned?: boolean /** How far each `::` has read into `log`. Required: unlike diff --git a/contributors/emails/onur.m.aycicek@gmail.com b/contributors/emails/onur.m.aycicek@gmail.com new file mode 100644 index 000000000000..f336e1632fb9 --- /dev/null +++ b/contributors/emails/onur.m.aycicek@gmail.com @@ -0,0 +1,2 @@ +onuraycicek +# Group room ordering, Hermes-Bot-Mode#105 diff --git a/package-lock.json b/package-lock.json index 2d9efa2e38b0..84d808002d0a 100644 --- a/package-lock.json +++ b/package-lock.json @@ -65,7 +65,7 @@ }, "apps/desktop": { "name": "hermes", - "version": "0.17.1", + "version": "0.17.2", "dependencies": { "@assistant-ui/core": "0.2.23", "@assistant-ui/react": "0.14.24", diff --git a/website/docs/user-guide/bot-mode.md b/website/docs/user-guide/bot-mode.md index 02a2070b840e..c36d4174a529 100644 --- a/website/docs/user-guide/bot-mode.md +++ b/website/docs/user-guide/bot-mode.md @@ -92,6 +92,8 @@ Right-click a local Bot → **Manage groups** to add or remove it from any numbe Groups are standalone rows in the same activity-ordered roster as Bot DMs. A Bot keeps one DM row even when it belongs to several groups, while every group gets its own room row with member count, latest-message preview, timestamp, and needs-you state. +Use the **Move up** and **Move down** arrows beside a room to choose its position among rooms. Until the first move, the existing pinned-first, recent-activity order is unchanged. After a move, room order is saved on this Desktop and survives reloads; new rooms follow the explicitly ordered rooms within their pinned or unpinned band. Moves cannot cross the pinned boundary, and filtering does not discard hidden rooms from the saved order. These controls reorder actual Group Chat rooms, not user-created Bot folders, and do not change membership or gateway ownership. + **Open chat** on any group row (2–6 Bots) opens a shared room where the whole group coordinates: - **One visible conversation.** Public messages and each member's reply stay readable in arrival order, with the speaker's name and timestamp. Starting another topic does not collapse earlier replies. **Reply in thread** continues that topic without reordering the room; **Activity** is a secondary status view, not a replacement for messages. Private Bot Chats remain separate. From aa83c6d61474ddebd745842530c4d7c429182ed6 Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Tue, 8 Sep 2026 10:57:41 -0700 Subject: [PATCH 213/227] fix: hide inactive grouping options from delegation schema --- evals/delegation_group_schema/README.md | 48 +++++++++++++++++ evals/delegation_group_schema/probe.py | 51 +++++++++++++++++++ tests/tools/test_delegate_group_schema.py | 47 +++++++++++++++++ tools/delegate_tool.py | 29 +++++++++-- .../docs/user-guide/features/delegation.md | 4 +- 5 files changed, 173 insertions(+), 6 deletions(-) create mode 100644 evals/delegation_group_schema/README.md create mode 100644 evals/delegation_group_schema/probe.py create mode 100644 tests/tools/test_delegate_group_schema.py diff --git a/evals/delegation_group_schema/README.md b/evals/delegation_group_schema/README.md new file mode 100644 index 000000000000..29d0901c79b7 --- /dev/null +++ b/evals/delegation_group_schema/README.md @@ -0,0 +1,48 @@ +# Delegation grouping schema receipts + +`probe.py TREE OUTPUT [--independent]` assembles real eager CLI tool definitions +in a fresh, credential-free temporary home with networking forbidden. Run it +in separate interpreters against base and fix with the same arguments. +Requires `tiktoken` (`o200k_base`); counts use compact OpenAI function JSON. +This measures schema footprint, not model quality or billed savings. + +Base: `b2aa855b626ff8688eb34b95c60ee8b6a4af3679`. + +| Policy | Base delegate tokens | Fixed delegate tokens | Change | +| --- | ---: | ---: | ---: | +| Default off | 927 | 826 | -101 | +| Explicit on | 927 | 913 | -14 | + +Only `delegate_task` changed in either same-config comparison. Total eager +CLI schemas were 7586 → 7485 (off), 7586 → 7572 (on). The field remains +available when enabled; its parameter schema is unchanged. Repeated assembly +is byte-stable. The two invariants also check static/previous schema immutability, +legacy task normalization, and grouped/ungrouped delivery partitioning. +The default-off exposure assertion failed on base before implementation. + +## Real-model N=1 child smoke + +Pinned to configured Nous `google/gemini-3.7-flash`; no provider fallback. +A separate temporary home held only copied Nous authentication and minimal +explicit configuration (no personal skills/memory). One forced parent tool call +using the real default-off registry schema produced a one-task array without +`group`. The actual delegation handler then built and ran one real AIAgent child, +which completed in one API call with `GROUP_GATE_OK`. Parent tool bytes were +unchanged across child execution. This is a schema-acceptance/execution smoke, +not a comparative quality evaluation or proof of asynchronous delivery timing. + +Before inference, the live catalog was read and an approximately $0.016 budget +estimated for two calls (20k input + 1024 output allowance). + +| Inference | Input | Output | Reasoning (included in output) | API-reported `usage.cost` | +| --- | ---: | ---: | ---: | ---: | +| Parent schema call | 796 | 104 | 75 | $0.000987 | +| Child | 708 | 42 | 37 | $0.0006885 | +| Total | 1504 | 146 | 112 | $0.0016755 | + +Parent request ID: `gen-1788889571-fBLF8eStu3UAevfGtvgD`. +The raw provider usage exposes the same values as `upstream_inference_cost`; +these are API-reported costs, not an independently reconciled Portal invoice. +The child result's local estimator instead reported `$0.000551` with +`cost_status: estimated`; do not substitute that estimate for the raw receipt. +Local full receipt: `/tmp/delegation-group-live-receipt.json`. diff --git a/evals/delegation_group_schema/probe.py b/evals/delegation_group_schema/probe.py new file mode 100644 index 000000000000..ba79fe54d5f2 --- /dev/null +++ b/evals/delegation_group_schema/probe.py @@ -0,0 +1,51 @@ +"""Offline same-config schema probe. Run in a fresh interpreter for each tree/policy. + +python probe.py /path/to/tree /tmp/receipt.json [--independent] +Requires tiktoken; no model calls or model-quality claims. +""" + +import json +import os +from pathlib import Path +import socket +import sys +import tempfile + +root, destination = sys.argv[1:3] +sys.path.insert(0, root) +os.chdir(root) +with tempfile.TemporaryDirectory(prefix="delegate-schema-") as home: + os.environ.clear() + os.environ.update(HOME=home, HERMES_HOME=home, PATH="/usr/bin:/bin", HERMES_PLATFORM="cli") + config = {"tools": {"tool_search": {"defer": []}}} + if "--independent" in sys.argv: + config["delegation"] = {"independent_completions": True} + Path(home, "config.yaml").write_text(json.dumps(config), encoding="utf-8") + + def deny_network(*args, **kwargs): + raise RuntimeError("Offline schema probe forbids network access") + + socket.socket.connect = deny_network + import tiktoken + from model_tools import get_tool_definitions + + encoder = tiktoken.get_encoding("o200k_base") + definitions = get_tool_definitions(enabled_toolsets=["hermes-cli"], quiet_mode=True) + serialized = json.dumps(definitions, separators=(",", ":")) + repeated = get_tool_definitions(enabled_toolsets=["hermes-cli"], quiet_mode=True) + assert json.dumps(repeated, separators=(",", ":")) == serialized + counts = { + definition["function"]["name"]: len(encoder.encode(json.dumps(definition, separators=(",", ":")))) + for definition in definitions + } + delegate = next(d for d in definitions if d["function"]["name"] == "delegate_task") + receipt = { + "config": config, + "tokens": counts, + "total_tokens": sum(counts.values()), + "delegate": delegate, + "repeated_assembly_byte_stable": True, + "model_calls": 0, + } + Path(destination).write_text(json.dumps(receipt, indent=2), encoding="utf-8") + print(json.dumps({"destination": destination, "total_tokens": receipt["total_tokens"], "delegate_tokens": counts["delegate_task"]})) diff --git a/tests/tools/test_delegate_group_schema.py b/tests/tools/test_delegate_group_schema.py new file mode 100644 index 000000000000..c625446f7941 --- /dev/null +++ b/tests/tools/test_delegate_group_schema.py @@ -0,0 +1,47 @@ +"""Grouping is advertised only when the delivery policy consumes it.""" + +import json +from dataclasses import fields + +from tools.delegate_tool import DELEGATE_TASK_SCHEMA, _strip_model_hidden_task_fields +from tools.delegate_tool_dispatch import _Batch, _units_of +from tools.delegate_tool_tasks import _normalize_task_list +from tools.registry import registry + + +def test_group_schema_tracks_delivery_policy_without_mutating_previous_definitions(monkeypatch): + from tools import delegate_tool_config + + original = json.dumps(DELEGATE_TASK_SCHEMA) + snapshots = [] + for config in ({}, {"independent_completions": True}, {"independent_completions": False}): + monkeypatch.setattr(delegate_tool_config, "_cfg", lambda: config) + definition = registry.get_definitions({"delegate_task"})[0] + enabled = config.get("independent_completions", False) + task = definition["function"]["parameters"]["properties"]["tasks"]["items"] + assert ("group" in task["properties"]) == enabled + assert ("group" in definition["function"]["description"]) == enabled + assert json.dumps(registry.get_definitions({"delegate_task"})[0]) == json.dumps(definition) + for previous, serialized in snapshots: + assert json.dumps(previous) == serialized + snapshots.append((definition, json.dumps(definition))) + assert json.dumps(DELEGATE_TASK_SCHEMA) == original + + +def test_legacy_group_replay_remains_accepted_and_delivery_policy_controls_units(monkeypatch): + from tools import delegate_tool_config + + tasks = [{"goal": "Review first module", "group": "join"}, {"goal": "Review second module", "group": "join"}, {"goal": "Review third module"}] + assert _strip_model_hidden_task_fields(tasks) is tasks + normalized, error = _normalize_task_list(None, None, tasks, None, "leaf", 3) + assert error is None and normalized == tasks + batch = _Batch(**{field.name: None for field in fields(_Batch)}) + batch.children = [(i, task, None) for i, task in enumerate(tasks)] + for enabled in (False, True): + monkeypatch.setattr(delegate_tool_config, "_cfg", lambda: {"independent_completions": enabled}) + units = _units_of(batch) + if enabled: + assert [[i for i, _, _ in unit.children] for unit in units] == [[0, 1], [2]] + assert [unit.group for unit in units] == ["join", None] + else: + assert units == [batch] diff --git a/tools/delegate_tool.py b/tools/delegate_tool.py index 4fe93157783e..6e23f2c799f5 100644 --- a/tools/delegate_tool.py +++ b/tools/delegate_tool.py @@ -498,7 +498,7 @@ def delegate_task( # ── OpenAI function-calling schema ────────────────────────────────────────── -def _build_top_level_description() -> str: +def _build_top_level_description(*, independent_completions=None) -> str: """delegate_task description: ONLY guidance stated nowhere else in the schema (limits live in the 'tasks' parameter description, rebuilt per get_definitions()).""" try: @@ -515,15 +515,22 @@ def _build_top_level_description() -> str: ) else: restrictions_rule = "- Children cannot call delegate_task, clarify, memory, or cronjob.\n" - return _DESCRIPTION_HEAD + restrictions_rule + _DESCRIPTION_TAIL + from tools.delegate_tool_config import _get_independent_completions + + if independent_completions is None: + independent_completions = _get_independent_completions() + delivery = ( + "each ungrouped task / `group` returns on its own" + if independent_completions else "one message per call" + ) + return _DESCRIPTION_HEAD.format(delivery=delivery) + restrictions_rule + _DESCRIPTION_TAIL _DESCRIPTION_HEAD = ( "Spawn subagents in isolated contexts; each gets its own conversation, terminal session, and toolset, and only its " "final summary returns to you. Pass every task in `tasks` — one entry spawns one subagent, several run in parallel " "(limit in the tasks description).\n\n" "Runs in the background: dispatch returns immediately with live transcript paths, and the call's results re-enter " - "the conversation as a new message when its subagents finish (one message per call by default; with " - "delegation.independent_completions each ungrouped task / `group` returns on its own). Results are delivered only " + "the conversation as a new message when its subagents finish ({delivery}). Results are delivered only " "BETWEEN your turns: finish whatever does not depend on them, then give a one-line status and END YOUR TURN. Never " "wait or poll on transcripts, artifact files, or CI for a child. " "While children run, `action` (list/steer/stop) controls them live — steer when a transcript shows a " @@ -564,12 +571,24 @@ def _build_tasks_param_description() -> str: def _build_dynamic_schema_overrides() -> dict: """Per-call schema overrides (ToolEntry.dynamic_schema_overrides): every get_definitions() pass rewrites the descriptions to the user's actual limits.""" + from tools.delegate_tool_config import _get_independent_completions + + independent_completions = _get_independent_completions() overrides_params = {**DELEGATE_TASK_SCHEMA["parameters"]} # Copy properties so the static schema dict is never mutated. overrides_params["properties"] = {k: dict(v) for k, v in DELEGATE_TASK_SCHEMA["parameters"]["properties"].items()} overrides_params["properties"]["tasks"]["description"] = _build_tasks_param_description() - return {"description": _build_top_level_description(), "parameters": overrides_params} + if not independent_completions: + tasks = overrides_params["properties"]["tasks"] + tasks["items"] = {**tasks["items"], "properties": { + k: v for k, v in tasks["items"]["properties"].items() if k != "group" + }} + + return { + "description": _build_top_level_description(independent_completions=independent_completions), + "parameters": overrides_params, + } def _p(type_: str, description: str, **extra) -> dict: return {"type": type_, **extra, "description": description} diff --git a/website/docs/user-guide/features/delegation.md b/website/docs/user-guide/features/delegation.md index 607d7bd9bfde..3d2c49bd66ae 100644 --- a/website/docs/user-guide/features/delegation.md +++ b/website/docs/user-guide/features/delegation.md @@ -165,7 +165,9 @@ When a top-level agent provides a `tasks` array, Hermes returns one background h ### Independent completions (opt-in) -Set `delegation.independent_completions: true` to have results land **per completion unit** as each finishes instead: +Set `delegation.independent_completions: true` to have results land **per completion unit** as each finishes instead. The model-facing `group` field and grouping guidance are only advertised when this option is enabled. Start a new session after changing it so the tool schema can reflect the setting without changing an existing conversation's cached prefix. Old calls containing `group` remain accepted; with the option off, the whole call still returns together. + +When independent completions are enabled: - Omit `group` when each result is useful to act on separately. Each task reports as soon as it finishes. - Use the same `group` string when you want to review outputs together: comparison, synthesis, or one coordinated decision. The group returns **one** consolidated message after all its tasks finish. Even independently executable tasks can belong in one group when their results inform the same decision. From ab6f4c54088dfcc7837c4eeefa0979272a97e818 Mon Sep 17 00:00:00 2001 From: Adolanium <94890352+Adolanium@users.noreply.github.com> Date: Tue, 8 Sep 2026 10:32:25 -0700 Subject: [PATCH 214/227] fix(desktop): show the focused bot's working think pose Port Adolanium's focused-turn pose from Hermes-Bot-Mode#101 and hermes-agent#88134 to the current typed Bot Mode implementation. Match the busy signal's connection-qualified focused owner rather than the gateway socket, retain worker activity, and ease transitions in elapsed time on the existing shared face clock. Includes owner-isolation and animated-pose invariants, both proven red on origin/main, and native Electron before/after verification against a real temporary Hermes backend with held loopback inference. Co-authored-by: Teknium <127238744+teknium1@users.noreply.github.com> --- .../hermes-bots/avatar.face-clock.test.ts | 21 ++++++ .../src/plugins/hermes-bots/avatar.tsx | 67 +++++++++++++++---- .../src/plugins/hermes-bots/bot-row.tsx | 17 +---- .../src/plugins/hermes-bots/roster-pane.tsx | 11 ++- .../plugins/hermes-bots/row-helpers.test.ts | 26 ++++--- .../src/plugins/hermes-bots/row-helpers.ts | 54 +++++++++++---- apps/desktop/src/plugins/hermes-bots/types.ts | 2 +- website/docs/user-guide/bot-mode.md | 4 +- 8 files changed, 147 insertions(+), 55 deletions(-) diff --git a/apps/desktop/src/plugins/hermes-bots/avatar.face-clock.test.ts b/apps/desktop/src/plugins/hermes-bots/avatar.face-clock.test.ts index f17ae64d59a9..d3b65d63f553 100644 --- a/apps/desktop/src/plugins/hermes-bots/avatar.face-clock.test.ts +++ b/apps/desktop/src/plugins/hermes-bots/avatar.face-clock.test.ts @@ -220,6 +220,27 @@ describe('the hand-rolled rAF path (older shells)', () => { }) describe('the SDK budgeted-loop path', () => { + it('paints a thinking gaze and eases back without a discontinuity', async () => { + const { captured } = captureLoop() + const { startFaceClock } = await loadClock() + const face = mountFace() + face.innerHTML = '' + face.setAttribute('data-hb-mood', 'think') + startFaceClock() + captured.draw!(1000) + observer!.emit([{ isIntersecting: true, target: face }]) + captured.draw!(2000) + const eye = face.querySelector('ellipse')! + const gaze = eye.getAttribute('cy') + expect(Number(face.querySelector('circle')!.getAttribute('opacity'))).toBeGreaterThan(0) + face.setAttribute('data-hb-mood', 'idle') + captured.draw!(2067) + expect(eye.getAttribute('cy')).toBe(gaze) + captured.draw!(2667) + expect(Number(eye.getAttribute('cy'))).toBeCloseTo(17.2) + expect(Number(face.querySelector('circle')!.getAttribute('opacity'))).toBe(0) + }) + interface CapturedLoop { draw: (now: number) => void idleWhen: () => boolean diff --git a/apps/desktop/src/plugins/hermes-bots/avatar.tsx b/apps/desktop/src/plugins/hermes-bots/avatar.tsx index b8aaae9c8e3e..addaaa6ae805 100644 --- a/apps/desktop/src/plugins/hermes-bots/avatar.tsx +++ b/apps/desktop/src/plugins/hermes-bots/avatar.tsx @@ -601,8 +601,22 @@ interface FacePose { turn: number } -/** Grok-style pose. thinking/working lean and sway. idle is a small sine. */ +/** Working poses lean and sway; idle stays small. */ function facePose(mood: FaceMood | string, t: number): FacePose { + if (mood === 'think') { + return { + turn: -18 + Math.sin(t * 0.55) * 14, + tilt: Math.sin(t * 0.48) * 12 + Math.sin(t * 1.35) * 3, + roll: Math.sin(t * 0.95) * 10, + gazeX: Math.sin(t * 0.7) * 3.6, + gazeY: -2.2 + Math.sin(t * 0.4) * 2.2, + blink: t % 1.45 > 1.26, + d0: 0.2 + 0.8 * Math.max(0, Math.sin(t * 2.4)), + d1: 0.2 + 0.8 * Math.max(0, Math.sin(t * 2.4 - 0.7)), + d2: 0.2 + 0.8 * Math.max(0, Math.sin(t * 2.4 - 1.4)) + } + } + if (mood === 'work') { return { turn: -11 + Math.sin(t * 0.48) * 8, @@ -637,10 +651,41 @@ interface NumericAttrNode { setAttribute(name: string, value: number | string): void } +const faceTransitions = new WeakMap() + +/** Blend from the last painted pose in elapsed time, even after a paused clock. */ +function settlePose(svg: SVGSVGElement, mood: string, target: FacePose, t: number): FacePose { + let state = faceTransitions.get(svg) + + if (!state) { + state = { mood, pose: target, from: target, since: t - 0.4 } + faceTransitions.set(svg, state) + } + + if (state.mood !== mood) { + state.mood = mood + state.from = state.pose + state.since = t + } + + const progress = Math.min(1, Math.max(0, (t - state.since) / 0.4)) + const blend = 1 - (1 - progress) ** 2 + const pose = { ...target } + const keys = ['turn', 'tilt', 'roll', 'gazeX', 'gazeY', 'd0', 'd1', 'd2'] as const + + for (const key of keys) { + pose[key] = state.from[key] + (target[key] - state.from[key]) * blend + } + + state.pose = pose + + return pose +} + function paintMathFace(svg: SVGSVGElement, t: number) { const mood = svg.getAttribute('data-hb-mood') || 'idle' const shape = svg.getAttribute('data-hb-shape') || 'circle' - const pose = facePose(mood, t) + const pose = settlePose(svg, mood, facePose(mood, t), t) const body = svg.querySelector('[data-hb-body]') const open = svg.querySelector('[data-hb-open]') const shut = svg.querySelector('[data-hb-shut]') @@ -1016,7 +1061,7 @@ export function BotFace({ shape, color, image, size = 36, name = 'agent', mood = ) } - const working = mood === 'work' + const working = mood === 'work' || mood === 'think' const eyeFill = isDarkColor(color) ? 'rgba(232,220,195,0.95)' : 'rgba(0,0,0,0.85)' // Catchlight contrast follows the pupil, not the body: dark pupils get the // white sparkle, light (cream) pupils on dark bodies get a dark one — a @@ -1024,7 +1069,7 @@ export function BotFace({ shape, color, image, size = 36, name = 'agent', mood = // maroon/ink/oxblood avatars. const hlFill = isDarkColor(color) ? 'rgba(0,0,0,0.6)' : 'rgba(255,255,255,0.85)' const ring = sampleFaceRing(shape) - const rest = facePose(working ? 'work' : 'idle', 0) + const rest = facePose(mood, 0) // Shape-aware initial eye line — the cloud body sits lower, so its eyes // (and their catchlights) start at the cloud position instead of jumping // there on the first clock paint. @@ -1036,7 +1081,7 @@ export function BotFace({ shape, color, image, size = 36, name = 'agent', mood = className="block overflow-visible" data-bot-face={name} data-hb-math="1" - data-hb-mood={working ? 'work' : 'idle'} + data-hb-mood={mood} data-hb-shape={shape || 'circle'} height={size} viewBox="0 0 40 44" @@ -1066,13 +1111,11 @@ export function BotFace({ shape, color, image, size = 36, name = 'agent', mood = strokeLinecap="round" strokeWidth={2} /> - {working ? ( - - - - - - ) : null} + + + + + ) } diff --git a/apps/desktop/src/plugins/hermes-bots/bot-row.tsx b/apps/desktop/src/plugins/hermes-bots/bot-row.tsx index 48f60acc4d5a..36afdb3bdf29 100644 --- a/apps/desktop/src/plugins/hermes-bots/bot-row.tsx +++ b/apps/desktop/src/plugins/hermes-bots/bot-row.tsx @@ -49,7 +49,6 @@ import { botRosterKey, botSelectionKey, botSourceStatus, - isActiveRosterBot, isDefaultBot, newBotChat, ROSTER_KEY, @@ -63,7 +62,7 @@ import { displayName, stripPreviewMarkdown } from './labels' import { duplicateBot } from './profile-ops' import { openRosterBot } from './roster-actions' import { botRosterMeta, botWorkspaceOwnerKey, setBotsWorkspaceOwner } from './routing' -import { A2A_PREFIX_RE, botCanonicalSessionId, botRowOwnsWorkspace, previewKind, workerActiveAt } from './row-helpers' +import { A2A_PREFIX_RE, botCanonicalSessionId, botRowOwnsWorkspace, botWorkingMood, previewKind, useTurnBusy, workerActiveAt } from './row-helpers' import type { GroupMember, RosterRow, SidebarRowLabels } from './types' import { $botSections, $draggingBot, BOT_DRAG_MIME, botSectionId, moveBotsToSection } from './user-sections' @@ -93,7 +92,6 @@ interface BotRowProps { export function BotRow({ bot, onDelete, onEdit, onGroup, onNewSection, showHandle }: BotRowProps) { const { t } = useI18n() const b = useBots() - const activeProfile = useValue(host.state.profile) const focusedOwner = focusedRosterOwner(useValue($focusedBotOwner)) const selectedRosterKey = useValue($selectedRosterKey) const botChatFocused = useValue($botChatFocused) @@ -118,19 +116,10 @@ export function BotRow({ bot, onDelete, onEdit, onGroup, onNewSection, showHandl // can highlight a remote row, which has no focusable local chat. const isActive = botRowOwnsWorkspace(bot, activeGroup, botChatFocused, focusedOwner, selectedRosterKey) - // Turn-busy is a SOCKET fact: only the gateway-home profile can be mid-turn. - const isGatewayHome = - !bot.remoteSource && - bot.name === activeProfile && - isActiveRosterBot(bot, { - name: activeProfile, - connectionId: activeConnectionId - }) - const { shape, color, image } = botAppearance(bot.name, meta) // Keep user photos/pets. Drop the 160px SVG backfill so the math face can move. const photo = Boolean(image && !isBackfilledFacePng(image)) - const gatewayState = useValue(host.state.gateway) + const turnBusy = useTurnBusy() // Preview identity must match click identity (#88200): when the backend // resolved the pinned canonical chat, preview THAT session — not the // profile's most recent (but unrelated) activity. Activity signals @@ -147,7 +136,7 @@ export function BotRow({ bot, onDelete, onEdit, onGroup, onNewSection, showHandl ? Math.max(activitySession?.last_active || 0, bot.worker_session?.last_active || 0) : activitySession?.last_active || 0 - const botMood = workerActive || (isGatewayHome && gatewayState === 'busy') ? 'work' : 'idle' + const botMood = botWorkingMood(bot, focusedOwner, turnBusy, activeConnectionId) // Status keys off the canonical Bot Chat — the very session this row opens, // so the dot and the click can never describe different conversations. const canonicalSessionId = botCanonicalSessionId(bot) diff --git a/apps/desktop/src/plugins/hermes-bots/roster-pane.tsx b/apps/desktop/src/plugins/hermes-bots/roster-pane.tsx index eb169ec2c983..e8f7738973ab 100644 --- a/apps/desktop/src/plugins/hermes-bots/roster-pane.tsx +++ b/apps/desktop/src/plugins/hermes-bots/roster-pane.tsx @@ -34,11 +34,13 @@ import { BotRow, GroupRow } from './bot-row' import { $botChatFocused, $botsPaneVisible, + $focusedBotOwner, $openBotChat, $rosterHydrated, $selectedRosterHydrated, $selectedRosterKey, clearSelectedRosterKey, + focusedRosterOwner, parseRosterKey, saveSelectedRosterBot } from './bot-state' @@ -78,7 +80,7 @@ import { } from './roster-sections' import type { ResolvedRosterGatewaySection } from './roster-sections' import { botRosterMeta, botWorkspaceOwnerKey, setBotsWorkspaceOwner } from './routing' -import { ACTIVE_WINDOW_S, activeBots, BOT_ROSTER_SEARCH_THRESHOLD, rosterActivityMatches } from './row-helpers' +import { ACTIVE_WINDOW_S, activeBots, BOT_ROSTER_SEARCH_THRESHOLD, rosterActivityMatches, useTurnBusy } from './row-helpers' import { backfillMessagingProtocol } from './soul' import type { BotMeta, GatewaySource, GroupMember, RosterActivityFilter, RosterKindFilter, RosterRow } from './types' import { @@ -257,7 +259,10 @@ export function BotsPane() { const { data, error, isLoading, refetch } = useRoster() const gatewayState = useValue(host.state.gateway) const gatewayUp = gatewayState === 'open' - const activeProfile = (useValue(host.state.profile) || 'default').trim() || 'default' + + const turnBusy = useTurnBusy() + const workingOwner = focusedRosterOwner(useValue($focusedBotOwner)) + const activeConnectionId = host.state.connectionId?.get?.() || 'local' const [createOpen, setCreateOpen] = useState(false) const [groupCreateOpen, setGroupCreateOpen] = useState(false) const [editing, setEditing] = useState(null) @@ -344,7 +349,7 @@ export function BotsPane() { // and the persisted connection registry hydrate. Keep that transition in a // neutral loading state instead of flashing the first-run "No bots" copy. const initialRosterLoading = !data && !error && roster.length === 0 - const activeRosterKeys = new Set(activeBots(roster, activeProfile, gatewayState).map(botRosterKey)) + const activeRosterKeys = new Set(activeBots(roster, workingOwner, turnBusy, Date.now(), activeConnectionId).map(botRosterKey)) const gatewayOptions = rosterGatewayOptions(sourceSnapshot, roster) const selectedGateway = gatewayOptions.find(option => option.connectionId === gatewayFilter) const gatewayFilterExists = gatewayFilter === 'all' || Boolean(selectedGateway) diff --git a/apps/desktop/src/plugins/hermes-bots/row-helpers.test.ts b/apps/desktop/src/plugins/hermes-bots/row-helpers.test.ts index 9182249e00b8..241fa64dd654 100644 --- a/apps/desktop/src/plugins/hermes-bots/row-helpers.test.ts +++ b/apps/desktop/src/plugins/hermes-bots/row-helpers.test.ts @@ -97,13 +97,19 @@ describe('which bots are working right now', () => { row({ name: 'analyst' }) ] - it('includes the gateway-busy profile before its first response lands', () => { - // analyst has no session at all — a busy turn must still show it. - expect(activeBots(roster, 'analyst', 'busy', NOW).map(bot => bot.name)).toContain('analyst') + it('counts the focused live turn only for its connection-qualified owner', () => { + const local = row({ name: 'analyst', connectionId: 'local' }) + const remote = row({ name: 'analyst', connectionId: 'remote', remoteSource: true }) + const owner = { name: 'analyst', connectionId: 'remote', authoritative: true } + expect(activeBots([local, remote], owner, true, NOW)).toEqual([remote]) + expect(activeBots([local, remote], owner, false, NOW)).toEqual([]) + expect(activeBots([local, remote], null, true, NOW)).toEqual([]) + expect(activeBots([local, remote], { ...owner, authoritative: false }, true, NOW)).toEqual([]) + expect(activeBots([row({ name: 'analyst', remoteSource: true })], { ...owner, connectionId: '' }, true, NOW)).toEqual([]) }) it('includes activity inside the liveness window and excludes activity outside it', () => { - const names = activeBots(roster, 'default', 'open', NOW).map(bot => bot.name) + const names = activeBots(roster, null, false, NOW).map(bot => bot.name) expect(names).toContain('researcher') expect(names).not.toContain('scribe') @@ -113,9 +119,9 @@ describe('which bots are working right now', () => { }) it('returns an empty list when nothing is active, and tolerates no roster', () => { - expect(activeBots(roster.slice(1), 'default', 'open', NOW)).toEqual([]) - expect(activeBots(null, 'default', 'open', NOW)).toEqual([]) - expect(activeBots([], 'default', 'open', NOW)).toEqual([]) + expect(activeBots(roster.slice(1), null, false, NOW)).toEqual([]) + expect(activeBots(null, null, false, NOW)).toEqual([]) + expect(activeBots([], null, false, NOW)).toEqual([]) }) it('counts Bot Chat activity that last_session cannot see', () => { @@ -127,7 +133,7 @@ describe('which bots are working right now', () => { }) ] - expect(activeBots(bots, 'other', 'open', NOW).map(bot => bot.name)).toContain('default') + expect(activeBots(bots, null, false, NOW).map(bot => bot.name)).toContain('default') }) it('counts a live kanban/tool worker heartbeat (#90268)', () => { @@ -138,7 +144,7 @@ describe('which bots are working right now', () => { worker_session: { id: 'w1', last_active: secondsAgo(30), source: 'kanban' } }) - expect(activeBots([working], 'other', 'open', NOW).map(bot => bot.name)).toContain('coding') + expect(activeBots([working], null, false, NOW).map(bot => bot.name)).toContain('coding') expect(workerActiveAt(working, NOW)).toBe(true) }) @@ -149,7 +155,7 @@ describe('which bots are working right now', () => { worker_session: { id: 'w1', last_active: secondsAgo(3600), source: 'kanban' } }) - expect(activeBots([finished], 'other', 'open', NOW)).toEqual([]) + expect(activeBots([finished], null, false, NOW)).toEqual([]) // Workers get a wider window than chat activity to bridge one missed // heartbeat — but not an hour's worth. expect(workerActiveAt(finished, NOW)).toBe(false) diff --git a/apps/desktop/src/plugins/hermes-bots/row-helpers.ts b/apps/desktop/src/plugins/hermes-bots/row-helpers.ts index 2c92daeafda5..899b7e5dafb5 100644 --- a/apps/desktop/src/plugins/hermes-bots/row-helpers.ts +++ b/apps/desktop/src/plugins/hermes-bots/row-helpers.ts @@ -3,11 +3,12 @@ * message, whether a bot counts as live, and whether its row owns the * highlight. * - * Pure below the surfaces. Every one of these takes a roster row and returns - * a value, so the bot row, the roster pane and the open path can share one - * answer instead of each deriving its own. + * Shared by the bot row, roster pane and open path rather than deriving + * competing answers in each surface. */ +import { atom, host, useValue } from '@hermes/plugin-sdk' + import { botActivitySession, botHandle, botRosterKey, isActiveRosterBot } from './data' import type { RosterActivityFilter, RosterRow } from './types' @@ -84,23 +85,50 @@ export function workerActiveAt(bot: null | RosterRow | undefined, now = Date.now return Boolean(ts && now / 1000 - ts < WORKER_ACTIVE_WINDOW_S) } -/** Bots that are working right now: the profile the gateway is running a - * turn for (busy), any bot whose last message landed inside the liveness - * window, plus any bot with a live kanban/tool worker. Pure — output - * follows the input roster's order, so presence never reorders or hides - * the normal list. */ +/** Older shells without the live-turn atom remain idle rather than reading socket state. */ +const $idleTurn = atom(false) + +export function useTurnBusy(): boolean { + return useValue(host.state.busy || $idleTurn) +} + +interface WorkingOwner { + authoritative: boolean + connectionId: string + name: string +} + +/** The busy atom follows tile focus, not the gateway socket or row selection. */ +export function botWorkingMood( + bot: RosterRow, + owner: WorkingOwner | null, + turnBusy: boolean, + activeConnectionId = 'local', + now = Date.now() +): 'idle' | 'think' | 'work' { + const botConnectionId = bot.connectionId || (bot.remoteSource ? '' : activeConnectionId) + + if (turnBusy && owner?.authoritative && owner.connectionId && owner.name === bot.name && owner.connectionId === botConnectionId) { + return 'think' + } + + return workerActiveAt(bot, now) ? 'work' : 'idle' +} + +/** Focused turns, recent messages and live workers, preserving roster order. */ export function activeBots( roster: null | RosterRow[] | undefined, - activeProfile: string, - gatewayState: string, - now = Date.now() + owner: WorkingOwner | null, + turnBusy: boolean, + now = Date.now(), + activeConnectionId = 'local' ): RosterRow[] { return (roster || []).filter(bot => { - const busyTurn = !bot.remoteSource && bot.name === activeProfile && gatewayState === 'busy' + const busyTurn = botWorkingMood(bot, owner, turnBusy, activeConnectionId, now) !== 'idle' const last = botActivitySession(bot)?.last_active || 0 const inWindow = Boolean(last && now / 1000 - last < ACTIVE_WINDOW_S) - return busyTurn || inWindow || workerActiveAt(bot, now) + return busyTurn || inWindow }) } diff --git a/apps/desktop/src/plugins/hermes-bots/types.ts b/apps/desktop/src/plugins/hermes-bots/types.ts index 7ee6aad189c3..5b9ad0cc00d9 100644 --- a/apps/desktop/src/plugins/hermes-bots/types.ts +++ b/apps/desktop/src/plugins/hermes-bots/types.ts @@ -288,7 +288,7 @@ export type AvatarShape = 'circle' | 'cloud' | 'drop' | 'hexagon' | 'pill' | 'sq export type BlobKind = 'boxy' | 'capsule' | 'cloud' | 'droplet' | 'hexagon' | 'nub' | 'organic' | 'round' | 'sun' | 'triangle' -export type FaceMood = 'idle' | 'work' +export type FaceMood = 'idle' | 'think' | 'work' export interface AvatarAppearance { /** `null` when nothing is picked — the name's deterministic hue stands in. diff --git a/website/docs/user-guide/bot-mode.md b/website/docs/user-guide/bot-mode.md index c36d4174a529..4ed32b7672ea 100644 --- a/website/docs/user-guide/bot-mode.md +++ b/website/docs/user-guide/bot-mode.md @@ -18,7 +18,7 @@ There is no new primitive to learn: a Bot **is** a Hermes profile — isolated c The roster shows one row per agent profile: avatar, latest-message preview, and timestamp. - **Click a Bot** to land in its chat — every Bot has a canonical, persistent **Bot Chat** conversation that is created (and pinned) the moment the Bot is born. A row click always opens that Bot Chat (the same conversation the row previews), even when you have other tabs open for the Bot; those tabs stay open beside it. In the tab strip the Bot Chat is captioned with the Bot's name, so two open Bots are told apart at a glance. -- **Active now** — a presence strip above the roster shows every Bot currently working: the gateway-busy profile plus any Bot that wrote within the last 90 seconds. Each chip opens that Bot's chat. The strip never reorders the roster and disappears when the fleet is idle. +- **Active now** — the roster's activity filter includes the owner of the focused live turn, Bots that wrote within the last 90 seconds, and Bots with a recent worker heartbeat. A connected gateway alone does not mean a Bot is working. - **Search** filters the roster as you type. - **Hide a Bot** — right-click a row → **Hide Bot** to take a Bot you don't use out of the roster and the Active-now strip. Hiding is display-only: @mentions still resolve, group-chat memberships are untouched, and routines keep running. Once at least one Bot is hidden, an **eye toggle** appears in the pane header — click it to reveal hidden Bots dimmed in place, then right-click → **Unhide Bot** to bring one back. Hidden Bots never toast, but they accumulate unread activity silently and the eye badges a dot so you know something happened. Hidden state is saved in the Bot's profile metadata, so it follows the Bot to every desktop connected to that backend. @@ -71,7 +71,7 @@ Remote-creation notes: Every Bot gets a face: - **Blob faces** (default) — a deterministic soft-body face drawn from the Bot's name: same name, same face, forever. While you type a name in New Agent the face follows it live; hit **Randomize** to re-roll, **Lock face** to keep the one you like even if the name changes, or pin one of the six silhouettes (round, organic, boxy, nub, cloud, sun) while everything else still comes from the name. -- **Geometric faces** — the classic 7 shapes × 10 colors, with blinking eyes that scan while the Bot works. +- **Geometric faces** — the classic 7 shapes × 10 colors. During a focused live turn, the owning Bot leans and looks upward with three animated dots, then eases back to idle when the turn ends. Ownership includes the connection, so same-named Bots on different gateways do not borrow the pose. Background workers keep their existing working animation; photos, blob faces and sigils keep their own rendering. - **An uploaded image** — any picture you like. - **An AI-generated portrait** — when an image backend is configured, generated in place (this rides the standard `image.generate` RPC and works over both local and remote gateways). - **A pixel pet** — a companion from the [petdex gallery](./features/pets.md) that bounces beside the avatar while the Bot is busy. Run `hermes pets` in a terminal to explore the gallery. From 78afbc3c37a8a0c4646eb6063375a43fd01b2b7e Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Tue, 8 Sep 2026 10:53:52 -0700 Subject: [PATCH 215/227] fix(desktop): load property card guidance only on demand --- agent/prompt_builder.py | 7 +- evals/prompt_footprint/property_guidance.py | 76 +++++++++++ .../productivity/property-listings/SKILL.md | 105 +++++++++++++++ tests/skills/test_property_listings_skill.py | 38 ++++++ .../docs/reference/optional-skills-catalog.md | 1 + .../productivity-property-listings.md | 121 ++++++++++++++++++ website/sidebars.ts | 1 + 7 files changed, 343 insertions(+), 6 deletions(-) create mode 100644 evals/prompt_footprint/property_guidance.py create mode 100644 optional-skills/productivity/property-listings/SKILL.md create mode 100644 tests/skills/test_property_listings_skill.py create mode 100644 website/docs/user-guide/skills/optional/productivity/productivity-property-listings.md diff --git a/agent/prompt_builder.py b/agent/prompt_builder.py index aedf7f43484d..a3fe15b316af 100644 --- a/agent/prompt_builder.py +++ b/agent/prompt_builder.py @@ -667,12 +667,7 @@ def hud_surface_note(valid_tool_names: "set[str] | None" = None) -> str: "height live, width from the content's first measured span — lay content flush left with no centering wrappers " "or it measures full-bleed. Widgets talk back: data-hermes-send=\"prompt\" on any clickable element (or " "window.hermes.send(\"prompt\")) sends that prompt as a hidden user turn — answer it by updating the widget's " - "file, not with prose. Property/rental listings render as browsable cards: emit a ```listing fence " - "holding JSON — one object, or an array to compare several — with address (required), price, beds, " - "baths, size, note (why it is worth a look), facts[] (short specs), catches[] (risks to verify), " - "images[] (direct https photo URLs, in listing order — the first is the hero), and links[] " - "({label, url} detail pages, never a search-results URL). Use it for every property you present, " - "including follow-ups and re-rankings, so listings stay comparable." + "file, not with prose." ), "sms": ( "You are communicating via SMS. Keep responses concise and use plain text only — no markdown, no " diff --git a/evals/prompt_footprint/property_guidance.py b/evals/prompt_footprint/property_guidance.py new file mode 100644 index 000000000000..0f7b38af0598 --- /dev/null +++ b/evals/prompt_footprint/property_guidance.py @@ -0,0 +1,76 @@ +"""Offline prompt A/B and real optional catalog -> skill_view probe. +Run with the repository venv; tiktoken must be available. No model API calls. +""" +import json +import os +from pathlib import Path +import subprocess +import sys +import tempfile +from types import SimpleNamespace + +ROOT = Path(__file__).resolve().parents[2] +sys.path.insert(0, str(ROOT)) + + +def main(): + with tempfile.TemporaryDirectory(prefix="property-guidance-") as temporary: + os.environ["HERMES_HOME"] = temporary + os.environ["TERMINAL_CWD"] = temporary + os.chdir(temporary) + import tiktoken + from agent import prompt_builder + from agent.system_prompt import build_system_prompt + from tools.skills_hub_official import OptionalSkillSource + from tools.skills_tool import skill_view + + agent = SimpleNamespace( + load_soul_identity=False, skip_context_files=True, valid_tool_names=[], + _task_completion_guidance=False, _tool_use_enforcement=False, + _environment_probe=False, _kanban_worker_guidance="", _memory_store=None, + _memory_manager=None, model="", provider="", platform="desktop", + pass_session_id=False, session_id="", _emit_status=lambda *args: None, + ) + fixed_hint = prompt_builder.PLATFORM_HINTS["desktop"] + # The original clause is taken from the pinned base, never synthesized. + source = subprocess.check_output( + ["git", "show", f"{sys.argv[1]}:agent/prompt_builder.py"], cwd=ROOT, + text=True, stdin=subprocess.DEVNULL, + ) + import ast + tree = ast.parse(source) + mapping = next(n.value for n in tree.body if isinstance(n, ast.Assign) + and any(isinstance(t, ast.Name) and t.id == "PLATFORM_HINTS" for t in n.targets)) + base_hint = next(ast.literal_eval(value) for key, value in zip(mapping.keys, mapping.values) + if isinstance(key, ast.Constant) and key.value == "desktop") + prompt_builder.PLATFORM_HINTS["desktop"] = base_hint + before = build_system_prompt(agent) + prompt_builder.PLATFORM_HINTS["desktop"] = fixed_hint + after = build_system_prompt(agent) + assert before.replace(base_hint, fixed_hint) == after + result = {"base": sys.argv[1], "tokenizers": {}} + for name in ("cl100k_base", "o200k_base"): + enc = tiktoken.get_encoding(name) + counts = [len(enc.encode(s)) for s in (base_hint, fixed_hint, before, after)] + result["tokenizers"][name] = dict(zip( + ("hint_before", "hint_after", "prompt_before", "prompt_after"), counts)) + optional = OptionalSkillSource() + matches = [m for m in optional.list_local() if "property" in m.tags and "rental" in m.tags] + assert matches + bundle = optional.fetch(matches[0].identifier) + assert bundle + dest = Path(temporary) / "skills" / bundle.name + dest.mkdir(parents=True) + for name, data in bundle.files.items(): + (dest / name).write_bytes(data if isinstance(data, bytes) else data.encode("utf-8")) + loaded = json.loads(skill_view(bundle.name)) + assert loaded["success"], loaded + example = loaded["content"].split("```listing\n", 1)[1].split("```", 1)[0] + assert json.loads(example)["address"] + result.update(identifier=matches[0].identifier, skill_view_success=True, + non_property_prompt_byte_parity=True, example=json.loads(example)) + print(json.dumps(result, indent=2)) + + +if __name__ == "__main__": + main() diff --git a/optional-skills/productivity/property-listings/SKILL.md b/optional-skills/productivity/property-listings/SKILL.md new file mode 100644 index 000000000000..42c45ec5b516 --- /dev/null +++ b/optional-skills/productivity/property-listings/SKILL.md @@ -0,0 +1,105 @@ +--- +name: property-listings +description: Present property and rental listings as desktop cards. +version: 0.1.0 +author: Teknium (teknium1), Hermes Agent +license: MIT +platforms: [linux, macos, windows] +metadata: + hermes: + tags: [property, rental, real-estate, listings, desktop, cards] + category: productivity + related_skills: [] +--- + +# Property Listings Skill + +Present researched properties as browsable cards in the Hermes desktop transcript. +This is a presentation recipe, not a listing search service or an investment valuation. + +## When to Use + +- Presenting property or rental search results, comparing a shortlist, or re-ranking properties. +- Following up on a property already shown: keep using cards so the shortlist stays comparable. +- Outside the desktop app, use ordinary Markdown with source links instead; other clients need not render listing fences. + +## Prerequisites + +- A Hermes desktop conversation for native cards; the backend may be local or remote. +- Property details supplied by the user or verified through `web_search`, `web_extract`, or the browser tools available in this session. +- No additional API keys or dependencies are required for card formatting. + +## How to Run + +Install this optional skill through the Skills catalog, or use `terminal`: + +```text +hermes skills install official/productivity/property-listings +``` + +Load it with `skill_view(name="property-listings")` when presenting listings. +Installing does not retrofit the running conversation's skill index; start a new +conversation for automatic discovery, or explicitly load the installed skill now. + +## Quick Reference + +Emit a fenced code block whose language is `listing` and whose body is valid JSON. +Use one object, an array of objects, or `{ "listings": [...] }` for a comparison. + +| Field | Shape and meaning | +|---|---| +| `address` | Required nonempty street address or property headline. | +| `price` | Formatted string including currency and rental period, if applicable. | +| `beds`, `baths` | Positive numeric counts; omit unknown values. | +| `size` | Formatted area including units. | +| `note` | Why this property is worth a look. | +| `facts` | Array of short verified specs or amenities. | +| `catches` | Array of risks or questions to verify before a tour. | +| `images` | Direct HTTPS photo URLs in listing order; the first is the hero. | +| `links` | Array of `{ "label": "Source", "url": "https://..." }` detail-page links, not search-result URLs. | + +## Procedure + +1. Gather the address, price, specs, photos and canonical detail URL. Distinguish + verified facts from unknowns; do not invent prices, amenities, or photo URLs. +2. Deduplicate portal mirrors of the same property into one card, retaining useful + source links. Keep source dates and availability caveats in the surrounding prose. +3. Emit the `listing` fence for every property presented, including follow-ups and + re-rankings. Keep facts short and put unresolved concerns in `catches`. +4. Check the JSON before sending. This fictional format example illustrates all fields; + replace its values and example URLs with verified listing data: + +```listing +{ + "address": "12 Example Lane", + "price": "$2,400/mo", + "beds": 3, + "baths": 2.5, + "size": "1,600 sqft", + "note": "Fits the requested space and budget.", + "facts": ["12-month lease", "Covered parking"], + "catches": ["Verify pet policy and total move-in fees"], + "images": ["https://example.com/property/front.jpg", "https://example.com/property/kitchen.jpg"], + "links": [{"label": "Listing details", "url": "https://example.com/property/12"}] +} +``` + +## Pitfalls + +- Cards are authored from gathered data, not fetched from a listing URL or embedded portal page. +- A sparse card needs only an address. Omit unknown fields rather than filling them with guesses. +- Use direct remote image URLs, not local paths, data URLs, or search-result pages. + Expired or blocked images disappear from the gallery; the text and links still matter. +- Keep a fence to at most 24 properties, 40 images per property, and 12 entries in + facts, catches and links. Text fields are truncated to 400 characters by the renderer. +- Malformed JSON or a card without identity falls back to a plain code block. + A valid card is not proof that the underlying listing is current or accurate. + +## Verification + +- Every presented property has an address and a verified source link; unknowns are explicit. +- Desktop displays the address, price, specs, facts, catches and links as a native card. +- Photos form a gallery; selecting a photo opens the lightbox. Three or more photos + use a hero-and-supporting-frames mosaic; additional photos remain browsable there. +- If the card fails to render, validate the fence language and JSON, then preserve a + readable Markdown fallback with the same facts and links. diff --git a/tests/skills/test_property_listings_skill.py b/tests/skills/test_property_listings_skill.py new file mode 100644 index 000000000000..76833c474e5b --- /dev/null +++ b/tests/skills/test_property_listings_skill.py @@ -0,0 +1,38 @@ +"""Property card recipes are optional, discoverable, and loadable on demand.""" +import json +from pathlib import Path + +from agent.prompt_builder import PLATFORM_HINTS +from tools.skills_hub_official import OptionalSkillSource + + +ROOT = Path(__file__).resolve().parents[2] + + +def test_property_recipe_is_not_paid_for_by_unrelated_desktop_sessions(): + hint = PLATFORM_HINTS["desktop"] + assert "```listing" not in hint + assert "MEDIA:" in hint and "::preview" in hint + + +def test_optional_catalog_fetch_preserves_a_usable_property_recipe(tmp_path, monkeypatch): + source = OptionalSkillSource() + source._optional_dir = ROOT / "optional-skills" + matches = [m for m in source.list_local() if "property" in m.tags and "rental" in m.tags] + assert matches, "Property tasks must be discoverable in the optional catalog" + bundle = source.fetch(matches[0].identifier) + assert bundle is not None + from tools.skills_tool import skill_view + + monkeypatch.setenv("HERMES_HOME", str(tmp_path)) + destination = tmp_path / "skills" / bundle.name + destination.mkdir(parents=True) + for name, data in bundle.files.items(): + (destination / name).write_bytes(data if isinstance(data, bytes) else data.encode("utf-8")) + loaded = json.loads(skill_view(bundle.name)) + assert loaded["success"], loaded + content = loaded["content"] + example = content.split("```listing\n", 1)[1].split("```", 1)[0] + listing = json.loads(example) + assert listing["address"] and listing["links"] + assert {"price", "beds", "baths", "size", "note", "facts", "catches", "images"} <= listing.keys() diff --git a/website/docs/reference/optional-skills-catalog.md b/website/docs/reference/optional-skills-catalog.md index 2c6cce8fcd44..ea74ac0976d5 100644 --- a/website/docs/reference/optional-skills-catalog.md +++ b/website/docs/reference/optional-skills-catalog.md @@ -207,6 +207,7 @@ hermes skills uninstall | [**decision-questionnaire**](/docs/user-guide/skills/optional/productivity/productivity-decision-questionnaire) | Turn an unanswerable decision into a questionnaire doc. | | [**here-now**](/docs/user-guide/skills/optional/productivity/productivity-here-now) | Publish sites to {slug}.here.now and store files in Drives. | | [**memento-flashcards**](/docs/user-guide/skills/optional/productivity/productivity-memento-flashcards) | Spaced-repetition flashcards: create, review, quiz, export. | +| [**property-listings**](/docs/user-guide/skills/optional/productivity/productivity-property-listings) | Present property and rental listings as desktop cards. | | [**shop**](/docs/user-guide/skills/optional/productivity/productivity-shop) | Shop catalog search, checkout, order tracking, returns. | | [**shopify**](/docs/user-guide/skills/optional/productivity/productivity-shopify) | Query Shopify Admin/Storefront GraphQL APIs via curl. | | [**siyuan**](/docs/user-guide/skills/optional/productivity/productivity-siyuan) | Query and edit a SiYuan knowledge base via its API. | diff --git a/website/docs/user-guide/skills/optional/productivity/productivity-property-listings.md b/website/docs/user-guide/skills/optional/productivity/productivity-property-listings.md new file mode 100644 index 000000000000..664697b200bb --- /dev/null +++ b/website/docs/user-guide/skills/optional/productivity/productivity-property-listings.md @@ -0,0 +1,121 @@ +--- +title: "Property Listings — Present property and rental listings as desktop cards" +sidebar_label: "Property Listings" +description: "Present property and rental listings as desktop cards" +--- + +{/* This page is auto-generated from the skill's SKILL.md by website/scripts/generate-skill-docs.py. Edit the source SKILL.md, not this page. */} + +# Property Listings + +Present property and rental listings as desktop cards. + +## Skill metadata + +| | | +|---|---| +| Source | Optional — install with `hermes skills install official/productivity/property-listings` | +| Path | `optional-skills/productivity/property-listings` | +| Version | `0.1.0` | +| Author | Teknium (teknium1), Hermes Agent | +| License | MIT | +| Platforms | linux, macos, windows | +| Tags | `property`, `rental`, `real-estate`, `listings`, `desktop`, `cards` | + +## Reference: full SKILL.md + +:::info +The following is the complete skill definition that Hermes loads when this skill is triggered. This is what the agent sees as instructions when the skill is active. +::: + +# Property Listings Skill + +Present researched properties as browsable cards in the Hermes desktop transcript. +This is a presentation recipe, not a listing search service or an investment valuation. + +## When to Use + +- Presenting property or rental search results, comparing a shortlist, or re-ranking properties. +- Following up on a property already shown: keep using cards so the shortlist stays comparable. +- Outside the desktop app, use ordinary Markdown with source links instead; other clients need not render listing fences. + +## Prerequisites + +- A Hermes desktop conversation for native cards; the backend may be local or remote. +- Property details supplied by the user or verified through `web_search`, `web_extract`, or the browser tools available in this session. +- No additional API keys or dependencies are required for card formatting. + +## How to Run + +Install this optional skill through the Skills catalog, or use `terminal`: + +```text +hermes skills install official/productivity/property-listings +``` + +Load it with `skill_view(name="property-listings")` when presenting listings. +Installing does not retrofit the running conversation's skill index; start a new +conversation for automatic discovery, or explicitly load the installed skill now. + +## Quick Reference + +Emit a fenced code block whose language is `listing` and whose body is valid JSON. +Use one object, an array of objects, or `{ "listings": [...] }` for a comparison. + +| Field | Shape and meaning | +|---|---| +| `address` | Required nonempty street address or property headline. | +| `price` | Formatted string including currency and rental period, if applicable. | +| `beds`, `baths` | Positive numeric counts; omit unknown values. | +| `size` | Formatted area including units. | +| `note` | Why this property is worth a look. | +| `facts` | Array of short verified specs or amenities. | +| `catches` | Array of risks or questions to verify before a tour. | +| `images` | Direct HTTPS photo URLs in listing order; the first is the hero. | +| `links` | Array of `{ "label": "Source", "url": "https://..." }` detail-page links, not search-result URLs. | + +## Procedure + +1. Gather the address, price, specs, photos and canonical detail URL. Distinguish + verified facts from unknowns; do not invent prices, amenities, or photo URLs. +2. Deduplicate portal mirrors of the same property into one card, retaining useful + source links. Keep source dates and availability caveats in the surrounding prose. +3. Emit the `listing` fence for every property presented, including follow-ups and + re-rankings. Keep facts short and put unresolved concerns in `catches`. +4. Check the JSON before sending. This fictional format example illustrates all fields; + replace its values and example URLs with verified listing data: + +```listing +{ + "address": "12 Example Lane", + "price": "$2,400/mo", + "beds": 3, + "baths": 2.5, + "size": "1,600 sqft", + "note": "Fits the requested space and budget.", + "facts": ["12-month lease", "Covered parking"], + "catches": ["Verify pet policy and total move-in fees"], + "images": ["https://example.com/property/front.jpg", "https://example.com/property/kitchen.jpg"], + "links": [{"label": "Listing details", "url": "https://example.com/property/12"}] +} +``` + +## Pitfalls + +- Cards are authored from gathered data, not fetched from a listing URL or embedded portal page. +- A sparse card needs only an address. Omit unknown fields rather than filling them with guesses. +- Use direct remote image URLs, not local paths, data URLs, or search-result pages. + Expired or blocked images disappear from the gallery; the text and links still matter. +- Keep a fence to at most 24 properties, 40 images per property, and 12 entries in + facts, catches and links. Text fields are truncated to 400 characters by the renderer. +- Malformed JSON or a card without identity falls back to a plain code block. + A valid card is not proof that the underlying listing is current or accurate. + +## Verification + +- Every presented property has an address and a verified source link; unknowns are explicit. +- Desktop displays the address, price, specs, facts, catches and links as a native card. +- Photos form a gallery; selecting a photo opens the lightbox. Three or more photos + use a hero-and-supporting-frames mosaic; additional photos remain browsable there. +- If the card fails to render, validate the fence language and JSON, then preserve a + readable Markdown fallback with the same facts and links. diff --git a/website/sidebars.ts b/website/sidebars.ts index ba0598f507f4..e4e2d43f232a 100644 --- a/website/sidebars.ts +++ b/website/sidebars.ts @@ -538,6 +538,7 @@ const sidebars: SidebarsConfig = { 'user-guide/skills/optional/productivity/productivity-decision-questionnaire', 'user-guide/skills/optional/productivity/productivity-here-now', 'user-guide/skills/optional/productivity/productivity-memento-flashcards', + 'user-guide/skills/optional/productivity/productivity-property-listings', 'user-guide/skills/optional/productivity/productivity-shop', 'user-guide/skills/optional/productivity/productivity-shopify', 'user-guide/skills/optional/productivity/productivity-siyuan', From 22488b8c62d3c92f25149053ae8df68fb0afcb35 Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Tue, 8 Sep 2026 12:04:30 -0700 Subject: [PATCH 216/227] fix(cli): keep monitor repaints safe during prompt handoff --- hermes_cli/cli_subagent_monitor.py | 11 +++++++- hermes_cli/cli_terminal_mixin.py | 4 +-- tests/cli/test_subagent_monitor_prompts.py | 32 ++++++++++++++++++++-- 3 files changed, 42 insertions(+), 5 deletions(-) diff --git a/hermes_cli/cli_subagent_monitor.py b/hermes_cli/cli_subagent_monitor.py index 48aad8776e1d..064c415c3b23 100644 --- a/hermes_cli/cli_subagent_monitor.py +++ b/hermes_cli/cli_subagent_monitor.py @@ -56,6 +56,15 @@ def refresh(self, now=None): self.selected_id = entries[0]['subagent_id'] if entries else None return changed + def invalidate(self): + from hermes_cli.cli_terminal_mixin import _run_on_app_loop + + app = self.app + if app is not None: + # Teardown clears app.loop; don't let it interleave with a worker's + # invalidate call, which reads the loop more than once. + _run_on_app_loop(app, app.invalidate) + def tick(self): now = time.monotonic() if now - self._last_poll < 1: @@ -63,7 +72,7 @@ def tick(self): self._last_poll = now if self.refresh(): if self.app is not None: - self.app.invalidate() + self.invalidate() else: self.cli._invalidate() diff --git a/hermes_cli/cli_terminal_mixin.py b/hermes_cli/cli_terminal_mixin.py index 5002313b7028..dcfaa60182b5 100644 --- a/hermes_cli/cli_terminal_mixin.py +++ b/hermes_cli/cli_terminal_mixin.py @@ -104,8 +104,8 @@ def _paint_now(self) -> None: if getattr(self, "_terminal_io_broken", False): return monitor = getattr(self, "_subagent_monitor", None) - if monitor is not None and monitor.app is not None: - monitor.app.invalidate() + if monitor is not None: + monitor.invalidate() app = getattr(self, "_app", None) if app is not None: self._app_invalidate(app, "paint_now", swallow=True) diff --git a/tests/cli/test_subagent_monitor_prompts.py b/tests/cli/test_subagent_monitor_prompts.py index 15f6ba0513ff..5462a2cf1f19 100644 --- a/tests/cli/test_subagent_monitor_prompts.py +++ b/tests/cli/test_subagent_monitor_prompts.py @@ -1,9 +1,13 @@ """Blocking prompts reclaim input from the nested monitor without a keypress.""" import asyncio +import threading from types import SimpleNamespace +import pytest -def test_prompt_paint_yields_monitor_but_ordinary_paint_does_not(): + +@pytest.mark.parametrize('paint_source', ['modal', 'tick']) +def test_prompt_paint_yields_monitor_but_ordinary_paint_does_not(monkeypatch, paint_source): from hermes_cli.cli_subagent_monitor import SubagentMonitor, build_monitor_application, install_dock from hermes_cli.cli_terminal_mixin import CLITerminalMixin from prompt_toolkit.input import create_pipe_input @@ -29,10 +33,34 @@ async def run(): task = asyncio.create_task(app.run_async()) await asyncio.wait_for(rendered.wait(), 3) try: + rendered.clear() await asyncio.to_thread(CLITerminalMixin._paint_now, cli) + await asyncio.wait_for(rendered.wait(), 3) assert not task.done(), 'ordinary paints must not dismiss the monitor' + loop = asyncio.get_running_loop() + ui_thread = threading.get_ident() + stopped = threading.Event() + task.add_done_callback(lambda _: stopped.set()) + schedule = loop.call_soon_threadsafe + + def schedule_with_teardown(callback, *args, **kwargs): + handle = schedule(callback, *args, **kwargs) + if (callback == app.on_invalidate.fire + and threading.get_ident() != ui_thread): + # Let an already-queued frame see the modal and finish + # teardown while the worker is still in invalidate(). + schedule(app._redraw) + assert stopped.wait(3), 'monitor did not finish teardown' + return handle + setattr(cli, name, {'pending': True}) - await asyncio.to_thread(CLITerminalMixin._paint_now, cli) + with monkeypatch.context() as patch: + patch.setattr(loop, 'call_soon_threadsafe', schedule_with_teardown) + if paint_source == 'tick': + patch.setattr(monitor, 'refresh', lambda: True) + await asyncio.to_thread(monitor.tick) + else: + await asyncio.to_thread(CLITerminalMixin._paint_now, cli) done, _ = await asyncio.wait({task}, timeout=1) assert task in done, f'{name} remained hidden behind the monitor' await task From e9bb6e86fb9ccac56519be4456f3cd00f467a1b3 Mon Sep 17 00:00:00 2001 From: "hermes-seaeye[bot]" <307254004+hermes-seaeye[bot]@users.noreply.github.com> Date: Tue, 8 Sep 2026 21:23:02 +0000 Subject: [PATCH 217/227] fmt(js): `npm run fix` on merge (#106039) Co-authored-by: github-actions[bot] --- apps/desktop/src/app/types.ts | 3 +- .../src/plugins/hermes-bots/bot-row.tsx | 10 +- .../plugins/hermes-bots/group-membership.ts | 4 +- .../plugins/hermes-bots/group-order.test.ts | 21 ++- .../src/plugins/hermes-bots/group-order.ts | 27 +++- .../plugins/hermes-bots/group-turns.test.ts | 151 ++++++++++++++---- .../src/plugins/hermes-bots/pet.test.tsx | 16 +- .../hermes-bots/roster-pane-groups.tsx | 7 +- .../src/plugins/hermes-bots/roster-pane.tsx | 4 +- .../plugins/hermes-bots/row-helpers.test.ts | 4 +- .../src/plugins/hermes-bots/row-helpers.ts | 8 +- tests-js/install-known-failures.test.ts | 4 + tests-js/install-process-close.test.ts | 1 + tests-js/scripts/mock-server.ts | 5 + 14 files changed, 208 insertions(+), 57 deletions(-) diff --git a/apps/desktop/src/app/types.ts b/apps/desktop/src/app/types.ts index 3bf09f66b150..57e85a0a9cae 100644 --- a/apps/desktop/src/app/types.ts +++ b/apps/desktop/src/app/types.ts @@ -162,8 +162,7 @@ export type CommandDispatchResponse = | SendCommandDispatchResponse | PrefillCommandDispatchResponse -export type SidebarNavId = - 'artifacts' | 'command-center' | 'cron' | 'messaging' | 'new-session' | 'settings' | 'skills' +export type SidebarNavId = 'artifacts' | 'command-center' | 'cron' | 'messaging' | 'new-session' | 'settings' | 'skills' export interface SidebarNavItem { /** Built-in view id, or a contributed row's namespaced contribution id. */ diff --git a/apps/desktop/src/plugins/hermes-bots/bot-row.tsx b/apps/desktop/src/plugins/hermes-bots/bot-row.tsx index 36afdb3bdf29..07152ac510e9 100644 --- a/apps/desktop/src/plugins/hermes-bots/bot-row.tsx +++ b/apps/desktop/src/plugins/hermes-bots/bot-row.tsx @@ -62,7 +62,15 @@ import { displayName, stripPreviewMarkdown } from './labels' import { duplicateBot } from './profile-ops' import { openRosterBot } from './roster-actions' import { botRosterMeta, botWorkspaceOwnerKey, setBotsWorkspaceOwner } from './routing' -import { A2A_PREFIX_RE, botCanonicalSessionId, botRowOwnsWorkspace, botWorkingMood, previewKind, useTurnBusy, workerActiveAt } from './row-helpers' +import { + A2A_PREFIX_RE, + botCanonicalSessionId, + botRowOwnsWorkspace, + botWorkingMood, + previewKind, + useTurnBusy, + workerActiveAt +} from './row-helpers' import type { GroupMember, RosterRow, SidebarRowLabels } from './types' import { $botSections, $draggingBot, BOT_DRAG_MIME, botSectionId, moveBotsToSection } from './user-sections' diff --git a/apps/desktop/src/plugins/hermes-bots/group-membership.ts b/apps/desktop/src/plugins/hermes-bots/group-membership.ts index bfbe5814f244..e8d9a86fd86f 100644 --- a/apps/desktop/src/plugins/hermes-bots/group-membership.ts +++ b/apps/desktop/src/plugins/hermes-bots/group-membership.ts @@ -28,8 +28,8 @@ export function followGroupChat(group: string, onRename: (name: string) => void) return } - const moved = Object.entries(rooms).find(([, room]) => - !room.tombstone && (prior.roomId ? room.roomId === prior.roomId : room === prior) + const moved = Object.entries(rooms).find( + ([, room]) => !room.tombstone && (prior.roomId ? room.roomId === prior.roomId : room === prior) ) if (!moved) { diff --git a/apps/desktop/src/plugins/hermes-bots/group-order.test.ts b/apps/desktop/src/plugins/hermes-bots/group-order.test.ts index 1dbc43a3388a..8c4e44cd4114 100644 --- a/apps/desktop/src/plugins/hermes-bots/group-order.test.ts +++ b/apps/desktop/src/plugins/hermes-bots/group-order.test.ts @@ -14,16 +14,31 @@ describe('room display order', () => { expect(sortGroupRosterRows(rows, {}).map(row => row.name)).toEqual(['Pinned', 'Newer', 'Bot', 'Older']) const rooms = { Older: { rosterOrder: 0 }, Newer: { rosterOrder: 1 } } expect(sortGroupRosterRows(rows, rooms).map(row => row.name)).toEqual(['Pinned', 'Older', 'Bot', 'Newer']) - expect(sortGroupRosterRows(rows.map(row => ({ ...row, activity: row.name === 'Newer' ? 999 : row.activity })), rooms).filter(row => row.kind === 'group').map(row => row.name)).toEqual(['Pinned', 'Older', 'Newer']) + expect( + sortGroupRosterRows( + rows.map(row => ({ ...row, activity: row.name === 'Newer' ? 999 : row.activity })), + rooms + ) + .filter(row => row.kind === 'group') + .map(row => row.name) + ).toEqual(['Pinned', 'Older', 'Newer']) expect(rows[0].name).toBe('Older') }) it('moves only visible same-band rooms while retaining hidden slots and ignoring stale targets', () => { - const ordered = sortGroupRosterRows(rows.filter(row => row.kind === 'group'), {}) + const ordered = sortGroupRosterRows( + rows.filter(row => row.kind === 'group'), + {} + ) expect(reorderGroupRows(ordered, 'Older', -1)).toEqual(['Pinned', 'Older', 'Newer']) expect(reorderGroupRows(ordered, 'Newer', -1)).toBeNull() expect(reorderGroupRows(ordered, 'deleted', 1)).toBeNull() const hidden = { kind: 'group' as const, name: 'Hidden', activity: 2, pinned: false } - expect(reorderGroupRows([ordered[0], ordered[1], hidden, ordered[2]], 'Older', -1, ['Newer', 'Older'])).toEqual(['Pinned', 'Older', 'Hidden', 'Newer']) + expect(reorderGroupRows([ordered[0], ordered[1], hidden, ordered[2]], 'Older', -1, ['Newer', 'Older'])).toEqual([ + 'Pinned', + 'Older', + 'Hidden', + 'Newer' + ]) }) }) diff --git a/apps/desktop/src/plugins/hermes-bots/group-order.ts b/apps/desktop/src/plugins/hermes-bots/group-order.ts index e3a47fc7cc7b..0fce0456355a 100644 --- a/apps/desktop/src/plugins/hermes-bots/group-order.ts +++ b/apps/desktop/src/plugins/hermes-bots/group-order.ts @@ -6,23 +6,32 @@ interface OrderRow { } /** Reorder room slots, not bots or folders. Pinning remains the outer band. */ -export function sortGroupRosterRows(rows: T[], rooms: Record): T[] { +export function sortGroupRosterRows( + rows: T[], + rooms: Record +): T[] { const legacy = rows.slice().sort((a, b) => Number(b.pinned) - Number(a.pinned) || b.activity - a.activity) - const groups = legacy.filter(row => row.kind === 'group').sort((a, b) => - Number(b.pinned) - Number(a.pinned) || - (rooms[a.name!]?.rosterOrder ?? Infinity) - (rooms[b.name!]?.rosterOrder ?? Infinity) - ) + const groups = legacy + .filter(row => row.kind === 'group') + .sort( + (a, b) => + Number(b.pinned) - Number(a.pinned) || + (rooms[a.name!]?.rosterOrder ?? Infinity) - (rooms[b.name!]?.rosterOrder ?? Infinity) + ) let index = 0 - return legacy.map(row => row.kind === 'group' ? groups[index++] : row) + return legacy.map(row => (row.kind === 'group' ? groups[index++] : row)) } /** Swap visible neighbours without dropping filtered-out rooms from the order. */ export function reorderGroupRows(rows: OrderRow[], name: string, delta: -1 | 1, visible?: string[]): string[] | null { const row = rows.find(row => row.name === name) - const band = rows.filter(candidate => candidate.kind === 'group' && candidate.pinned === row?.pinned && (!visible || visible.includes(candidate.name!))) + const band = rows.filter( + candidate => + candidate.kind === 'group' && candidate.pinned === row?.pinned && (!visible || visible.includes(candidate.name!)) + ) const index = band.findIndex(candidate => candidate.name === name) const neighbour = index >= 0 ? band[index + delta] : undefined @@ -30,5 +39,7 @@ export function reorderGroupRows(rows: OrderRow[], name: string, delta: -1 | 1, return null } - return rows.filter(row => row.kind === 'group').map(row => row.name === name ? neighbour.name! : row.name === neighbour.name ? name : row.name!) + return rows + .filter(row => row.kind === 'group') + .map(row => (row.name === name ? neighbour.name! : row.name === neighbour.name ? name : row.name!)) } diff --git a/apps/desktop/src/plugins/hermes-bots/group-turns.test.ts b/apps/desktop/src/plugins/hermes-bots/group-turns.test.ts index bf85c51a62c5..0c278f9ecd48 100644 --- a/apps/desktop/src/plugins/hermes-bots/group-turns.test.ts +++ b/apps/desktop/src/plugins/hermes-bots/group-turns.test.ts @@ -369,7 +369,8 @@ describe('clarify and approvals (#90694)', () => { // on $groupNeedsYou/$groupClarify AFTER the turn lands proves nothing: // the clarify has already resolved and its mirror is gone by then. onResumePoll: () => { - sawPendingAttention = sawPendingAttention || live!.turns.groupHasPendingClarify(live!.chat.$groupClarify.get(), 'Core') + sawPendingAttention = + sawPendingAttention || live!.turns.groupHasPendingClarify(live!.chat.$groupClarify.get(), 'Core') }, turn: () => 'targeting staging' }) @@ -506,17 +507,24 @@ describe('clarify and approvals (#90694)', () => { it('keeps late prompt snapshots on the live room and never revives a disbanded room', async ({ onTestFinished }) => { for (const roomId of ['stable-room', undefined]) { for (const disband of [false, true]) { - const room = await loadRoom({ turn: ({ n }) => n === 1 ? 'Completed reply' : '(pass)' }) + const room = await loadRoom({ turn: ({ n }) => (n === 1 ? 'Completed reply' : '(pass)') }) const view = await import('./group-chat-view') const member = { name: 'research', title: '' } room.chat.updateGroupChat('Core', current => ({ - ...current, roomId, running: true, epoch: 1, + ...current, + roomId, + running: true, + epoch: 1, log: [{ id: 'input', at: 1, from: { kind: 'user', name: 'You' }, text: '@research check', thread: 'thread' }] })) let entered!: () => void let release!: () => void - const polled = new Promise(resolve => { entered = resolve }) - const held = new Promise(resolve => { release = resolve }) + const polled = new Promise(resolve => { + entered = resolve + }) + const held = new Promise(resolve => { + release = resolve + }) const original = host.request as (method: string, params: Record) => Promise let submitted = false let answered = false @@ -525,12 +533,19 @@ describe('clarify and approvals (#90694)', () => { host.request = async (method: string, params: Record) => { const result = await original(method, params) - if (method === 'prompt.submit') {submitted = true} + if (method === 'prompt.submit') { + submitted = true + } - if (method === 'clarify.respond') {answered = true} + if (method === 'clarify.respond') { + answered = true + } if (method === 'session.resume' && submitted && !answered) { - if (++polls === 1) { entered(); await held } + if (++polls === 1) { + entered() + await held + } return { ...result, pending_clarify: CLARIFY } } @@ -546,14 +561,19 @@ describe('clarify and approvals (#90694)', () => { release() await drive expect(Object.values(room.chat.$groupClarify.get())).toHaveLength(0) - expect(room.chat.$groupChats.get().Core === undefined || room.chat.$groupChats.get().Core.tombstone).toBe(true) + expect(room.chat.$groupChats.get().Core === undefined || room.chat.$groupChats.get().Core.tombstone).toBe( + true + ) expect(room.chat.$groupChats.get().Core?.log || []).toHaveLength(0) } else { await view.renameGroupChat('Core', 'Renamed', []) const mirrored = new Promise(resolve => { const stop = room.chat.$groupClarify.listen(entries => { - if (Object.keys(entries).length) { stop(); resolve() } + if (Object.keys(entries).length) { + stop() + resolve() + } }) }) @@ -566,7 +586,12 @@ describe('clarify and approvals (#90694)', () => { expect(correctRoom).toBe('Renamed') expect(Object.keys(room.chat.$groupChats.get())).toEqual(['Renamed']) expect(room.chat.$groupChats.get().Renamed.running).toBe(false) - expect(room.chat.$groupChats.get().Renamed.log.filter(entry => entry.from.kind === 'member').map(entry => entry.text)).toEqual(['Completed reply']) + expect( + room.chat.$groupChats + .get() + .Renamed.log.filter(entry => entry.from.kind === 'member') + .map(entry => entry.text) + ).toEqual(['Completed reply']) expect(Object.values(room.chat.$groupClarify.get())).toHaveLength(0) } } @@ -581,10 +606,20 @@ describe('clarify and approvals (#90694)', () => { const data = await import('./data') const members = [{ name: 'research' }, { name: 'ops' }] room.chat.updateGroupChat('Core', current => ({ - ...current, roomId: 'old-rejection-room', running: true, + ...current, + roomId: 'old-rejection-room', + running: true, log: continuation - ? [{ id: 'handoff', at: 1, from: { kind: 'member', name: 'research' }, text: '@ops check', thread: 'thread' }, - { id: 'input', at: 2, from: { kind: 'user', name: 'You' }, text: '@research check', thread: 'thread' }] + ? [ + { + id: 'handoff', + at: 1, + from: { kind: 'member', name: 'research' }, + text: '@ops check', + thread: 'thread' + }, + { id: 'input', at: 2, from: { kind: 'user', name: 'You' }, text: '@research check', thread: 'thread' } + ] : [{ id: 'input', at: 1, from: { kind: 'user', name: 'You' }, text: 'check', thread: 'thread' }], watermarks: { 'thread::research': continuation ? 2 : 0 } })) @@ -592,23 +627,47 @@ describe('clarify and approvals (#90694)', () => { // handoff is driven by the continuation phase. let phaseEntered!: () => void let release!: () => void - const entered = new Promise(resolve => { phaseEntered = resolve }) - const held = new Promise(resolve => { release = resolve }) + const entered = new Promise(resolve => { + phaseEntered = resolve + }) + const held = new Promise(resolve => { + release = resolve + }) const original = host.request as (method: string, params: Record) => Promise + host.request = async (method: string, params: Record) => { - if (method === rejectedMethod) { phaseEntered(); await held; throw new Error('401 unauthorized late rejection') } + if (method === rejectedMethod) { + phaseEntered() + await held + throw new Error('401 unauthorized late rejection') + } return original(method, params) } + const drive = room.rounds.runGroupChatRounds('Core', members, 'thread') await entered await view.disbandGroupChat('Core', []) - room.chat.updateGroupChat('Core', current => ({ ...current, roomId: 'replacement-rejection-room', tombstone: false })) + room.chat.updateGroupChat('Core', current => ({ + ...current, + roomId: 'replacement-rejection-room', + tombstone: false + })) room.chat.appendGroupChatEntry('Core', { kind: 'member', name: 'research' }, '@user replacement needs you') - const before = structuredClone({ rooms: room.chat.$groupChats.get(), activity: activity.$groupActivity.get(), attention: data.$botAttention.get(), needsYou: room.chat.$groupNeedsYou.get() }) + const before = structuredClone({ + rooms: room.chat.$groupChats.get(), + activity: activity.$groupActivity.get(), + attention: data.$botAttention.get(), + needsYou: room.chat.$groupNeedsYou.get() + }) release() await drive - expect({ rooms: room.chat.$groupChats.get(), activity: activity.$groupActivity.get(), attention: data.$botAttention.get(), needsYou: room.chat.$groupNeedsYou.get() }).toEqual(before) + expect({ + rooms: room.chat.$groupChats.get(), + activity: activity.$groupActivity.get(), + attention: data.$botAttention.get(), + needsYou: room.chat.$groupNeedsYou.get() + }).toEqual(before) expect(room.gateway.rpcFor('prompt.submit')).toHaveLength(0) } } @@ -624,8 +683,12 @@ describe('clarify and approvals (#90694)', () => { room.chat.updateGroupChat('Core', current => ({ ...current, roomId: 'retired-room' })) let entered!: () => void let release!: () => void - const polled = new Promise(resolve => { entered = resolve }) - const held = new Promise(resolve => { release = resolve }) + const polled = new Promise(resolve => { + entered = resolve + }) + const held = new Promise(resolve => { + release = resolve + }) const original = host.request as (method: string, params: Record) => Promise let submitted = false @@ -638,7 +701,9 @@ describe('clarify and approvals (#90694)', () => { const result = await original(method, params) - if (method === 'prompt.submit') { submitted = true } + if (method === 'prompt.submit') { + submitted = true + } return result } @@ -667,30 +732,47 @@ describe('clarify and approvals (#90694)', () => { const view = await import('./group-chat-view') const members = [{ name: 'research' }, { name: 'ops' }] room.chat.updateGroupChat('Core', current => ({ - ...current, roomId: 'old-harvest-room', running: true, + ...current, + roomId: 'old-harvest-room', + running: true, stranded: { research: 0, ops: 0 } })) let tick!: () => void const previousWindow = globalThis.window - vi.stubGlobal('window', { setTimeout: (callback: () => void) => { tick = callback; + vi.stubGlobal('window', { + setTimeout: (callback: () => void) => { + tick = callback - return 0 } }) - onTestFinished(() => { vi.stubGlobal('window', previousWindow) }) + return 0 + } + }) + onTestFinished(() => { + vi.stubGlobal('window', previousWindow) + }) let entered!: () => void let release!: () => void - const polled = new Promise(resolve => { entered = resolve }) - const held = new Promise(resolve => { release = resolve }) + const polled = new Promise(resolve => { + entered = resolve + }) + const held = new Promise(resolve => { + release = resolve + }) let background = false const backgroundProfiles: unknown[] = [] const original = host.request as (method: string, params: Record) => Promise host.request = async (method: string, params: Record) => { - if (method !== 'session.resume') { return original(method, params) } + if (method !== 'session.resume') { + return original(method, params) + } if (background) { backgroundProfiles.push(params.profile) - if (params.profile === 'research') { entered(); await held } + if (params.profile === 'research') { + entered() + await held + } } return { running: true, pending_clarify: CLARIFY } @@ -704,7 +786,9 @@ describe('clarify and approvals (#90694)', () => { const { setImmediate } = await import('node:timers/promises') await setImmediate() room.chat.updateGroupChat('Core', current => ({ - ...current, roomId: 'new-harvest-room', stranded: { research: 0, ops: 0 } + ...current, + roomId: 'new-harvest-room', + stranded: { research: 0, ops: 0 } })) const roomsBefore = structuredClone(room.chat.$groupChats.get()) const promptsBefore = structuredClone(room.chat.$groupClarify.get()) @@ -723,7 +807,8 @@ describe('clarify and approvals (#90694)', () => { const room = await loadRoom({ approvalUntil: { research: { payload: APPROVAL, until: 3 } }, onResumePoll: () => { - sawPendingAttention = sawPendingAttention || live!.turns.groupHasPendingClarify(live!.chat.$groupClarify.get(), 'Core') + sawPendingAttention = + sawPendingAttention || live!.turns.groupHasPendingClarify(live!.chat.$groupClarify.get(), 'Core') }, turn: () => 'build cleaned' }) diff --git a/apps/desktop/src/plugins/hermes-bots/pet.test.tsx b/apps/desktop/src/plugins/hermes-bots/pet.test.tsx index a3d08e8728e9..2206c519529a 100644 --- a/apps/desktop/src/plugins/hermes-bots/pet.test.tsx +++ b/apps/desktop/src/plugins/hermes-bots/pet.test.tsx @@ -103,9 +103,15 @@ describe('the pet gallery', () => { }) it('keeps selection while scrolling for more and resets the search window', async () => { - useQueryMock.mockReturnValue({ data: { pets: Array.from({ length: 60 }, (_, i) => ({ - displayName: `Pet ${i}`, slug: `pet-${i}`, spritesheetUrl: SHEET - })) } }) + useQueryMock.mockReturnValue({ + data: { + pets: Array.from({ length: 60 }, (_, i) => ({ + displayName: `Pet ${i}`, + slug: `pet-${i}`, + spritesheetUrl: SHEET + })) + } + }) stubFetch(async () => ({ blob: async () => new Blob() })) const PetTab = await loadPetTab() const onImage = vi.fn() @@ -115,7 +121,9 @@ describe('the pet gallery', () => { await waitFor(() => expect(onImage).toHaveBeenCalledWith('data:image/png;base64,ok')) const scroller = first.parentElement!.parentElement! Object.defineProperties(scroller, { - clientHeight: { value: 220 }, scrollHeight: { value: 600 }, scrollTop: { value: 400 } + clientHeight: { value: 220 }, + scrollHeight: { value: 600 }, + scrollTop: { value: 400 } }) fireEvent.scroll(scroller) expect(view.getByText('Pet 47')).toBeTruthy() diff --git a/apps/desktop/src/plugins/hermes-bots/roster-pane-groups.tsx b/apps/desktop/src/plugins/hermes-bots/roster-pane-groups.tsx index 417aa6566636..cc96cb6ec312 100644 --- a/apps/desktop/src/plugins/hermes-bots/roster-pane-groups.tsx +++ b/apps/desktop/src/plugins/hermes-bots/roster-pane-groups.tsx @@ -43,7 +43,12 @@ export function RosterGroupRowView({ activity: groupLastActivity(current[name]) })) - const order = reorderGroupRows(sortGroupRosterRows(rows, current), name, delta, sortedGroupRows.map(row => row.name)) + const order = reorderGroupRows( + sortGroupRosterRows(rows, current), + name, + delta, + sortedGroupRows.map(row => row.name) + ) order?.forEach((name, rosterOrder) => { updateGroupChat(name, room => ({ ...room, rosterOrder }), { sync: false }) diff --git a/apps/desktop/src/plugins/hermes-bots/roster-pane.tsx b/apps/desktop/src/plugins/hermes-bots/roster-pane.tsx index ceaeae3145e9..4ea913641169 100644 --- a/apps/desktop/src/plugins/hermes-bots/roster-pane.tsx +++ b/apps/desktop/src/plugins/hermes-bots/roster-pane.tsx @@ -296,7 +296,9 @@ export function BotsPane() { // and the persisted connection registry hydrate. Keep that transition in a // neutral loading state instead of flashing the first-run "No bots" copy. const initialRosterLoading = !data && !error && roster.length === 0 - const activeRosterKeys = new Set(activeBots(roster, workingOwner, turnBusy, Date.now(), activeConnectionId).map(botRosterKey)) + const activeRosterKeys = new Set( + activeBots(roster, workingOwner, turnBusy, Date.now(), activeConnectionId).map(botRosterKey) + ) const gatewayOptions = rosterGatewayOptions(sourceSnapshot, roster) const selectedGateway = gatewayOptions.find(option => option.connectionId === gatewayFilter) const gatewayFilterExists = gatewayFilter === 'all' || Boolean(selectedGateway) diff --git a/apps/desktop/src/plugins/hermes-bots/row-helpers.test.ts b/apps/desktop/src/plugins/hermes-bots/row-helpers.test.ts index 241fa64dd654..bf1a3a746a53 100644 --- a/apps/desktop/src/plugins/hermes-bots/row-helpers.test.ts +++ b/apps/desktop/src/plugins/hermes-bots/row-helpers.test.ts @@ -105,7 +105,9 @@ describe('which bots are working right now', () => { expect(activeBots([local, remote], owner, false, NOW)).toEqual([]) expect(activeBots([local, remote], null, true, NOW)).toEqual([]) expect(activeBots([local, remote], { ...owner, authoritative: false }, true, NOW)).toEqual([]) - expect(activeBots([row({ name: 'analyst', remoteSource: true })], { ...owner, connectionId: '' }, true, NOW)).toEqual([]) + expect( + activeBots([row({ name: 'analyst', remoteSource: true })], { ...owner, connectionId: '' }, true, NOW) + ).toEqual([]) }) it('includes activity inside the liveness window and excludes activity outside it', () => { diff --git a/apps/desktop/src/plugins/hermes-bots/row-helpers.ts b/apps/desktop/src/plugins/hermes-bots/row-helpers.ts index 899b7e5dafb5..09d7f5bc562d 100644 --- a/apps/desktop/src/plugins/hermes-bots/row-helpers.ts +++ b/apps/desktop/src/plugins/hermes-bots/row-helpers.ts @@ -108,7 +108,13 @@ export function botWorkingMood( ): 'idle' | 'think' | 'work' { const botConnectionId = bot.connectionId || (bot.remoteSource ? '' : activeConnectionId) - if (turnBusy && owner?.authoritative && owner.connectionId && owner.name === bot.name && owner.connectionId === botConnectionId) { + if ( + turnBusy && + owner?.authoritative && + owner.connectionId && + owner.name === bot.name && + owner.connectionId === botConnectionId + ) { return 'think' } diff --git a/tests-js/install-known-failures.test.ts b/tests-js/install-known-failures.test.ts index 7799f27d7a5b..acc081e54cd7 100644 --- a/tests-js/install-known-failures.test.ts +++ b/tests-js/install-known-failures.test.ts @@ -8,11 +8,13 @@ import { describe, expect, it } from 'vitest' const { matchKnownFailure, rules } = createRequire(import.meta.url)('../tests/install/e2e-assets/known-failures.cjs') const classifier = path.resolve(import.meta.dirname, '../tests/install/e2e-assets/known-failures.cjs') + const lockedLog = [ 'error: failed to remove file `C:/install/venv/Lib/site-packages/../../Scripts/hermes.exe`: Access is denied. (os error 5)', 'File "C:/install/venv/Scripts/hermes.exe/__main__.py", line 10, in ', "subprocess.CalledProcessError: Command '['uv', 'pip', 'install', '-e', '.', '--quiet']' returned non-zero exit status 2.", ].join('\n') + const base = { platform: 'windows', phase: 'update', commit: 'a370ab8391ca5f8de7ebbc449f05cb0df36ade7c', installMethod: 'installer-script', updateMethod: 'hermes-update', @@ -42,6 +44,7 @@ describe('known install failures', () => { error: 'E2E ASSERTION FAILED: app driven via captured hermes desktop spec; update completed', logs: { desktop: '[hermes] [updates] no staged updater; surfacing manual `hermes update` for CLI install at C:/install\n[hermes] [updates] manual: hermes update\n' }, } + expect(matchKnownFailure(sample)?.id).toBe('windows-july-manual-app-update') expect(matchKnownFailure({ ...sample, installMethod: 'desktop-installer@latest' })).toBeNull() expect(matchKnownFailure({ ...sample, error: 'onboarding timed out' })).toBeNull() @@ -50,6 +53,7 @@ describe('known install failures', () => { it('CLI writes a receipt and exits zero only on a confirmed match', () => { const root = mkdtempSync(path.join(os.tmpdir(), 'known-install-')) + try { mkdirSync(path.join(root, 'logs')) writeFileSync(path.join(root, 'shas.json'), '\uFEFF' + JSON.stringify({ old: base.commit, current: 'f'.repeat(40), old_ref: 'v2026.3.12' })) diff --git a/tests-js/install-process-close.test.ts b/tests-js/install-process-close.test.ts index ce8b83c9b96b..38f69d2d1e51 100644 --- a/tests-js/install-process-close.test.ts +++ b/tests-js/install-process-close.test.ts @@ -24,6 +24,7 @@ it('waits for native close, not exit, and retains a close observed before hand-o it('fails if the launched process never closes', async () => { vi.useFakeTimers() + try { const waitForClose = observeProcessClose(Object.assign(new EventEmitter(), { stdio: [], exitCode: null, signalCode: null })) const completion = expect(waitForClose(2_000)).rejects.toThrow('Electron process did not close') diff --git a/tests-js/scripts/mock-server.ts b/tests-js/scripts/mock-server.ts index 9f49873803cf..d9e0806c0c53 100644 --- a/tests-js/scripts/mock-server.ts +++ b/tests-js/scripts/mock-server.ts @@ -505,6 +505,7 @@ export function startMockServer(options: MockServerOptions = {}): Promise typeof message?.content === 'string' && message.content.includes(VERIFICATION_STOP_TRIGGER), ) @@ -517,7 +518,9 @@ export function startMockServer(options: MockServerOptions = {}): Promise { if (stream) { streamScriptedTurn(res, model, turn) @@ -533,6 +536,7 @@ export function startMockServer(options: MockServerOptions = {}): Promise Date: Tue, 8 Sep 2026 21:28:40 +0000 Subject: [PATCH 218/227] fmt(js): `npm run fix` on merge (#106065) Co-authored-by: github-actions[bot] --- .../src/plugins/hermes-bots/group-order.test.ts | 1 + .../desktop/src/plugins/hermes-bots/group-order.ts | 2 ++ .../src/plugins/hermes-bots/group-turns.test.ts | 14 ++++++++++++++ .../src/plugins/hermes-bots/roster-pane.tsx | 2 ++ 4 files changed, 19 insertions(+) diff --git a/apps/desktop/src/plugins/hermes-bots/group-order.test.ts b/apps/desktop/src/plugins/hermes-bots/group-order.test.ts index 8c4e44cd4114..4d52ee8300cb 100644 --- a/apps/desktop/src/plugins/hermes-bots/group-order.test.ts +++ b/apps/desktop/src/plugins/hermes-bots/group-order.test.ts @@ -30,6 +30,7 @@ describe('room display order', () => { rows.filter(row => row.kind === 'group'), {} ) + expect(reorderGroupRows(ordered, 'Older', -1)).toEqual(['Pinned', 'Older', 'Newer']) expect(reorderGroupRows(ordered, 'Newer', -1)).toBeNull() expect(reorderGroupRows(ordered, 'deleted', 1)).toBeNull() diff --git a/apps/desktop/src/plugins/hermes-bots/group-order.ts b/apps/desktop/src/plugins/hermes-bots/group-order.ts index 0fce0456355a..1d596231467a 100644 --- a/apps/desktop/src/plugins/hermes-bots/group-order.ts +++ b/apps/desktop/src/plugins/hermes-bots/group-order.ts @@ -28,10 +28,12 @@ export function sortGroupRosterRows( /** Swap visible neighbours without dropping filtered-out rooms from the order. */ export function reorderGroupRows(rows: OrderRow[], name: string, delta: -1 | 1, visible?: string[]): string[] | null { const row = rows.find(row => row.name === name) + const band = rows.filter( candidate => candidate.kind === 'group' && candidate.pinned === row?.pinned && (!visible || visible.includes(candidate.name!)) ) + const index = band.findIndex(candidate => candidate.name === name) const neighbour = index >= 0 ? band[index + delta] : undefined diff --git a/apps/desktop/src/plugins/hermes-bots/group-turns.test.ts b/apps/desktop/src/plugins/hermes-bots/group-turns.test.ts index 0c278f9ecd48..7b5cde48c9a3 100644 --- a/apps/desktop/src/plugins/hermes-bots/group-turns.test.ts +++ b/apps/desktop/src/plugins/hermes-bots/group-turns.test.ts @@ -519,12 +519,15 @@ describe('clarify and approvals (#90694)', () => { })) let entered!: () => void let release!: () => void + const polled = new Promise(resolve => { entered = resolve }) + const held = new Promise(resolve => { release = resolve }) + const original = host.request as (method: string, params: Record) => Promise let submitted = false let answered = false @@ -627,12 +630,15 @@ describe('clarify and approvals (#90694)', () => { // handoff is driven by the continuation phase. let phaseEntered!: () => void let release!: () => void + const entered = new Promise(resolve => { phaseEntered = resolve }) + const held = new Promise(resolve => { release = resolve }) + const original = host.request as (method: string, params: Record) => Promise host.request = async (method: string, params: Record) => { @@ -654,12 +660,14 @@ describe('clarify and approvals (#90694)', () => { tombstone: false })) room.chat.appendGroupChatEntry('Core', { kind: 'member', name: 'research' }, '@user replacement needs you') + const before = structuredClone({ rooms: room.chat.$groupChats.get(), activity: activity.$groupActivity.get(), attention: data.$botAttention.get(), needsYou: room.chat.$groupNeedsYou.get() }) + release() await drive expect({ @@ -683,12 +691,15 @@ describe('clarify and approvals (#90694)', () => { room.chat.updateGroupChat('Core', current => ({ ...current, roomId: 'retired-room' })) let entered!: () => void let release!: () => void + const polled = new Promise(resolve => { entered = resolve }) + const held = new Promise(resolve => { release = resolve }) + const original = host.request as (method: string, params: Record) => Promise let submitted = false @@ -751,12 +762,15 @@ describe('clarify and approvals (#90694)', () => { }) let entered!: () => void let release!: () => void + const polled = new Promise(resolve => { entered = resolve }) + const held = new Promise(resolve => { release = resolve }) + let background = false const backgroundProfiles: unknown[] = [] const original = host.request as (method: string, params: Record) => Promise diff --git a/apps/desktop/src/plugins/hermes-bots/roster-pane.tsx b/apps/desktop/src/plugins/hermes-bots/roster-pane.tsx index 4ea913641169..e52f51ee4c95 100644 --- a/apps/desktop/src/plugins/hermes-bots/roster-pane.tsx +++ b/apps/desktop/src/plugins/hermes-bots/roster-pane.tsx @@ -296,9 +296,11 @@ export function BotsPane() { // and the persisted connection registry hydrate. Keep that transition in a // neutral loading state instead of flashing the first-run "No bots" copy. const initialRosterLoading = !data && !error && roster.length === 0 + const activeRosterKeys = new Set( activeBots(roster, workingOwner, turnBusy, Date.now(), activeConnectionId).map(botRosterKey) ) + const gatewayOptions = rosterGatewayOptions(sourceSnapshot, roster) const selectedGateway = gatewayOptions.find(option => option.connectionId === gatewayFilter) const gatewayFilterExists = gatewayFilter === 'all' || Boolean(selectedGateway) From ed6671db83c55998ed48bd995a4494ef5c84c23b Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Tue, 8 Sep 2026 14:05:48 -0700 Subject: [PATCH 219/227] test(desktop): retire model catalog fixture jobs at teardown --- apps/desktop/src/app/shell/model-catalog-menu.test.tsx | 3 +++ 1 file changed, 3 insertions(+) diff --git a/apps/desktop/src/app/shell/model-catalog-menu.test.tsx b/apps/desktop/src/app/shell/model-catalog-menu.test.tsx index 6e8e1494bf4b..7c67285f6b44 100644 --- a/apps/desktop/src/app/shell/model-catalog-menu.test.tsx +++ b/apps/desktop/src/app/shell/model-catalog-menu.test.tsx @@ -52,6 +52,9 @@ beforeEach(() => { afterEach(() => { cleanup() + // The backend mock echoes this snapshot; retire fixture jobs before jsdom + // disappears so an in-flight app-level poll cannot schedule another tick. + $localRuntimeJobs.set([]) vi.clearAllMocks() }) From 7568fb67278b5420171fcb515ecb8a183e408fa7 Mon Sep 17 00:00:00 2001 From: unsupportedpastels Date: Tue, 8 Sep 2026 18:03:51 +0000 Subject: [PATCH 220/227] fix(desktop): let subagent header collapse roster and details --- .../status-stack/subagent-section.test.tsx | 41 +++++++++++++++-- .../status-stack/subagent-section.tsx | 45 ++++++++++--------- 2 files changed, 61 insertions(+), 25 deletions(-) diff --git a/apps/desktop/src/app/chat/composer/status-stack/subagent-section.test.tsx b/apps/desktop/src/app/chat/composer/status-stack/subagent-section.test.tsx index acdd0aea10a4..f67d564985b2 100644 --- a/apps/desktop/src/app/chat/composer/status-stack/subagent-section.test.tsx +++ b/apps/desktop/src/app/chat/composer/status-stack/subagent-section.test.tsx @@ -6,11 +6,14 @@ import { $subagentsBySession, upsertSubagent } from '@/store/subagents' import { ComposerStatusStack } from './index' +vi.mock('@/lib/use-enter-animation', () => ({ useEnterAnimation: () => undefined })) + vi.stubGlobal( 'ResizeObserver', class { disconnect() {} observe() {} + unobserve() {} } ) @@ -19,7 +22,7 @@ afterEach(() => { $subagentsBySession.set({}) }) -it('automatically previews bounded live work only from the composer session, including queued children', () => { +it('shows live work only from the composer session and keeps it hidden after collapse and progress', () => { for (let i = 0; i < 5; i++) { upsertSubagent('owner', { subagent_id: `child-${i}`, goal: `Task ${i}`, status: i ? 'queued' : 'running' }) } @@ -35,11 +38,18 @@ it('automatically previews bounded live work only from the composer session, inc expect(screen.getByText('Task 0')).toBeTruthy() expect(screen.getByText('Reading actual source')).toBeTruthy() - expect(screen.queryByText('Task 4')).toBeNull() expect(screen.queryByText('Private foreign task')).toBeNull() - expect(screen.getByRole('button', { name: /5 Subagents/ })).toBeTruthy() - fireEvent.click(screen.getByRole('button', { name: /5 Subagents/ })) + const header = screen.getByRole('button', { name: /5 Subagents/ }) + fireEvent.click(header) + expect(screen.queryByText('Task 0')).toBeNull() + expect(screen.queryByText('Task 4')).toBeNull() + expect(header.getAttribute('aria-expanded')).toBe('false') + act(() => upsertSubagent('owner', { subagent_id: 'child-0', text: 'More progress' }, false, 'subagent.progress')) + expect(screen.queryByText('Task 0')).toBeNull() + fireEvent.click(header) + expect(screen.getByText('Task 0')).toBeTruthy() expect(screen.getByText('Task 4')).toBeTruthy() + expect(screen.getByText('More progress')).toBeTruthy() view.rerender( @@ -48,6 +58,29 @@ it('automatically previews bounded live work only from the composer session, inc expect(screen.queryByText('Task 0')).toBeNull() }) +it('collapses a single worker and its selected detail using the caret, preserving the steering draft', () => { + upsertSubagent('owner', { subagent_id: 'child', goal: 'Single task' }) + + const view = render( + + + + ) + + fireEvent.click(screen.getByRole('button', { name: /Single task/ })) + expect(view.container.querySelector('[data-slot="composer-subagent-detail"]')).toBeTruthy() + const draft = screen.getByRole('textbox') + fireEvent.change(draft, { target: { value: 'Keep this draft' } }) + const header = screen.getByRole('button', { name: /1 Subagent/ }) + fireEvent.click(header.firstElementChild!) + expect(screen.queryByText('Single task')).toBeNull() + expect(view.container.querySelector('[data-slot="composer-subagent-detail"]')).toBeNull() + expect(header.getAttribute('aria-expanded')).toBe('false') + fireEvent.click(header) + expect(view.container.querySelector('[data-slot="composer-subagent-detail"]')).toBeTruthy() + expect((screen.getByRole('textbox') as HTMLInputElement).value).toBe('Keep this draft') +}) + it('retires the live frame only after every child settles, without depending on the parent busy state', () => { upsertSubagent('owner', { subagent_id: 'child', goal: 'Live task' }) render( diff --git a/apps/desktop/src/app/chat/composer/status-stack/subagent-section.tsx b/apps/desktop/src/app/chat/composer/status-stack/subagent-section.tsx index e9c3410ab79a..7b08ce55deca 100644 --- a/apps/desktop/src/app/chat/composer/status-stack/subagent-section.tsx +++ b/apps/desktop/src/app/chat/composer/status-stack/subagent-section.tsx @@ -64,32 +64,35 @@ export function SubagentSection({ sessionId }: SubagentSectionProps) { return (
item.status === 'running') ? t.agents.running : t.agents.queued} + className="text-(--ui-purple)" + spinner="braille" + /> + } + defaultCollapsed={false} icon={} label={t.statusStack.subagents(live.length)} - preview={ - <> - {live.slice(0, 3).map(row)} - {live.length > 3 && ( -

{t.agents.moreAgents(live.length - 3)}

- )} - - } >
{live.map(row)}
+ {detail && ( +
+ setDrafts(previous => ({ ...previous, [detail.id]: text }))} + subagentId={detail.id} + text={drafts[detail.id] ?? ''} + /> + + +
+ )} - {detail && ( -
- setDrafts(previous => ({ ...previous, [detail.id]: text }))} - subagentId={detail.id} - text={drafts[detail.id] ?? ''} - /> - - -
- )}
) } From 7777f8c350fcfaa5c81a2a24d6e7ecbb62789a31 Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Tue, 8 Sep 2026 12:21:41 -0700 Subject: [PATCH 221/227] feat: add GPT Image 2.5 generation and editing to OpenAI provider --- plugins/image_gen/openai/__init__.py | 32 ++++++++++--- plugins/image_gen/openai/plugin.yaml | 2 +- .../plugins/image_gen/test_openai_provider.py | 45 +++++++++++++------ .../user-guide/features/built-in-plugins.md | 2 +- .../user-guide/features/image-generation.md | 32 ++++++++++++- 5 files changed, 90 insertions(+), 23 deletions(-) diff --git a/plugins/image_gen/openai/__init__.py b/plugins/image_gen/openai/__init__.py index d54f109cfeb7..80e495d4094a 100644 --- a/plugins/image_gen/openai/__init__.py +++ b/plugins/image_gen/openai/__init__.py @@ -1,4 +1,4 @@ -"""OpenAI ``gpt-image-2`` at three quality tiers (virtual ids ``gpt-image-2-low/-medium/-high``); +"""OpenAI GPT Image 2 and 2.5 Flare/Sunburst quality tiers; base64 output → image cache. Selection: ``OPENAI_IMAGE_MODEL`` → ``image_gen.openai.model`` → ``image_gen.model`` → :data:`DEFAULT_MODEL`.""" @@ -18,9 +18,29 @@ logger = logging.getLogger(__name__) +# Keep subscription routing independent: Codex does not verify explicit image model selection. +MODELS = { + **{key: {**meta, "api_model": API_MODEL} for key, meta in GPT_IMAGE_2_TIERS.items()}, + **{ + model if quality == "auto" else f"{model}-{quality}": { + "display": f"GPT Image 2.5 {name} ({quality.title()})", + "speed": speed, + "strengths": strengths, + "api_model": model, + "quality": quality, + } + for model, name, speed, strengths in ( + ("gpt-image-2.5-flare", "Flare", "Fast", "Everyday image generation and editing"), + ("gpt-image-2.5-sunburst", "Sunburst", "Slower", "Precision generation and editing"), + ) + for quality in ("auto", "low", "medium", "high", "xhigh", "max") + }, +} + + def _resolve_model() -> Tuple[str, Dict[str, Any]]: return resolve_static_model( - GPT_IMAGE_2_TIERS, DEFAULT_MODEL, env_var="OPENAI_IMAGE_MODEL", config_key="openai") + MODELS, DEFAULT_MODEL, env_var="OPENAI_IMAGE_MODEL", config_key="openai") def _load_image_bytes(ref: str) -> Tuple[bytes, str]: @@ -57,16 +77,16 @@ def _named_bytes_io(ref: str) -> io.BytesIO: class OpenAIImageGenProvider(StaticImageGenProvider): - """OpenAI ``images.generate`` / ``images.edit`` backend — gpt-image-2.""" + """OpenAI ``images.generate`` / ``images.edit`` backend with selectable API models.""" provider_id = "openai" label = "OpenAI" - models = GPT_IMAGE_2_TIERS + models = MODELS default_model_id = DEFAULT_MODEL price = "varies" setup = dict( name="OpenAI", badge="paid", - tag="gpt-image-2 at low/medium/high quality tiers — text-to-image & image editing", + tag="GPT Image 2 / 2.5 Flare / 2.5 Sunburst — text-to-image & image editing", key="OPENAI_API_KEY", prompt="OpenAI API key", url="https://platform.openai.com/api-keys") def is_available(self) -> bool: @@ -106,7 +126,7 @@ def generate( # gpt-image-2 returns b64_json unconditionally and REJECTS # ``response_format`` as an unknown parameter. Don't send it. request: Dict[str, Any] = dict( - model=API_MODEL, prompt=prompt, size=size, n=1, quality=meta["quality"]) + model=meta["api_model"], prompt=prompt, size=size, n=1, quality=meta["quality"]) if is_edit: try: files = [_named_bytes_io(ref) for ref in sources] diff --git a/plugins/image_gen/openai/plugin.yaml b/plugins/image_gen/openai/plugin.yaml index 18e4d86390db..5e3db55b4858 100644 --- a/plugins/image_gen/openai/plugin.yaml +++ b/plugins/image_gen/openai/plugin.yaml @@ -1,6 +1,6 @@ name: openai version: 1.0.0 -description: "OpenAI image generation backend (gpt-image-2). Saves generated images to $HERMES_HOME/cache/images/." +description: "OpenAI image generation backend (GPT Image 2 and GPT Image 2.5 Flare/Sunburst). Saves generated images to $HERMES_HOME/cache/images/." author: NousResearch kind: backend requires_env: diff --git a/tests/plugins/image_gen/test_openai_provider.py b/tests/plugins/image_gen/test_openai_provider.py index a3306f4372ec..93ad4013213f 100644 --- a/tests/plugins/image_gen/test_openai_provider.py +++ b/tests/plugins/image_gen/test_openai_provider.py @@ -57,9 +57,10 @@ def test_name(self, provider): def test_default_model(self, provider): assert provider.default_model() == "gpt-image-2-medium" - def test_list_models_three_tiers(self, provider): + def test_picker_matches_resolvable_catalog(self, provider): ids = [m["id"] for m in provider.list_models()] - assert ids == ["gpt-image-2-low", "gpt-image-2-medium", "gpt-image-2-high"] + assert set(ids) == set(provider.models) + assert provider.default_model() in ids def test_catalog_entries_have_display_speed_strengths(self, provider): for entry in provider.list_models(): @@ -171,24 +172,40 @@ def test_b64_saves_to_cache(self, provider, tmp_path): # gpt-image-2 rejects response_format — we must NOT send it. assert "response_format" not in call_kwargs - @pytest.mark.parametrize("tier,expected_quality", [ - ("gpt-image-2-low", "low"), - ("gpt-image-2-medium", "medium"), - ("gpt-image-2-high", "high"), + @pytest.mark.parametrize("api_model,quality", [ + ("gpt-image-2", quality) for quality in ("low", "medium", "high") + ] + [ + (model, quality) + for model in ("gpt-image-2.5-flare", "gpt-image-2.5-sunburst") + for quality in ("auto", "low", "medium", "high", "xhigh", "max") ]) - def test_tier_maps_to_quality(self, provider, monkeypatch, tier, expected_quality): - monkeypatch.setenv("OPENAI_IMAGE_MODEL", tier) + @pytest.mark.parametrize("editing", [False, True]) + def test_selection_reaches_image_request( + self, provider, monkeypatch, tmp_path, api_model, quality, editing + ): + import yaml + + tier = api_model if quality == "auto" else f"{api_model}-{quality}" + monkeypatch.delenv("OPENAI_IMAGE_MODEL", raising=False) + (tmp_path / "config.yaml").write_text(yaml.safe_dump({ + "image_gen": {"openai": {"model": tier}} + })) + source = tmp_path / "source.png" + source.write_bytes(bytes.fromhex(_PNG_HEX)) fake_client = MagicMock() - fake_client.images.generate.return_value = _fake_response(b64=_b64_png()) + call = fake_client.images.edit if editing else fake_client.images.generate + call.return_value = _fake_response(b64=_b64_png()) with _patched_openai(fake_client): - result = provider.generate("a cat") + result = provider.generate("a cat", image_url=str(source) if editing else None) + assert result["success"] is True assert result["model"] == tier - assert result["quality"] == expected_quality - assert fake_client.images.generate.call_args.kwargs["quality"] == expected_quality - # Always the same underlying API model regardless of tier. - assert fake_client.images.generate.call_args.kwargs["model"] == "gpt-image-2" + assert result["quality"] == quality + assert call.call_args.kwargs["quality"] == quality + assert call.call_args.kwargs["model"] == api_model + assert "response_format" not in call.call_args.kwargs + assert Path(result["image"]).read_bytes() == bytes.fromhex(_PNG_HEX) @pytest.mark.parametrize("aspect,expected_size", [ ("landscape", "1536x1024"), diff --git a/website/docs/user-guide/features/built-in-plugins.md b/website/docs/user-guide/features/built-in-plugins.md index ac18d4c92535..211496340277 100644 --- a/website/docs/user-guide/features/built-in-plugins.md +++ b/website/docs/user-guide/features/built-in-plugins.md @@ -61,7 +61,7 @@ The repo ships these bundled plugins under `plugins/`. All are opt-in — enable | `teams_pipeline` | standalone | Microsoft Teams meeting pipeline — Graph-backed, transcript-first meeting summaries | | `spotify` | backend (7 tools) | Native Spotify playback, queue, search, playlists, albums, library | | `google_meet` | standalone | Join Meet calls, live-caption transcription, optional realtime duplex audio | -| `image_gen/openai` | image backend | OpenAI `gpt-image-2` image generation backend (alternative to FAL) | +| `image_gen/openai` | image backend | OpenAI GPT Image 2 and 2.5 Flare/Sunburst generation and editing (API key) | | `image_gen/openai-codex` | image backend | OpenAI image generation via Codex OAuth | | `image_gen/xai` | image backend | xAI `grok-2-image` backend | | `hermes-achievements` | dashboard tab | Steam-style collectible badges generated from your real Hermes session history | diff --git a/website/docs/user-guide/features/image-generation.md b/website/docs/user-guide/features/image-generation.md index f9ad54552493..b9904cc07444 100644 --- a/website/docs/user-guide/features/image-generation.md +++ b/website/docs/user-guide/features/image-generation.md @@ -126,6 +126,36 @@ Auth reuses the same env vars as the Meta chat provider — `MODEL_API_KEY` as aliases. Set `META_BASE_URL` to point at a proxy or alternate host. Text-to-image only for now; responses are saved to `$HERMES_HOME/cache/images/`. +## OpenAI API: GPT Image 2.5 + +The **OpenAI** provider supports GPT Image 2.5 Flare (fast everyday creation) +and Sunburst (precision generation and editing), using `OPENAI_API_KEY`. +Select them through `hermes tools` → Image Generation → OpenAI, or set: + +```bash +hermes config set image_gen.provider openai +hermes config set image_gen.openai.model gpt-image-2.5-flare +``` + +`gpt-image-2.5-flare` and `gpt-image-2.5-sunburst` use automatic quality. +Append `-low`, `-medium`, `-high`, `-xhigh`, or `-max` to select a fixed quality, +for example `gpt-image-2.5-sunburst-high`. Both support generation and editing +with up to 16 reference images. Existing GPT Image 2 selections and the +`gpt-image-2-medium` default are unchanged. + +This is paid API usage, separate from a ChatGPT/Codex subscription. Both models +cost $5 per million text-input tokens, $8 per million image-input tokens, and +$30 per million image-output tokens (cached input rates are $1.25 and $2, +respectively). Per-image cost varies with usage; the GPT Image 2 calculator +does not estimate 2.5 token consumption. See the official +[Flare](https://developers.openai.com/api/docs/models/gpt-image-2.5-flare) and +[Sunburst](https://developers.openai.com/api/docs/models/gpt-image-2.5-sunburst) docs. + +The **OpenAI (Codex auth)** provider remains separate: its backend can accept +an image-model value without honoring that selection, so a successful image +alone does not verify Flare or Sunburst routing. These selections are offered +only through the direct OpenAI API provider, not Codex auth or FAL. + ## Usage The agent-facing schema is intentionally minimal — the model picks up whatever you've configured: @@ -167,7 +197,7 @@ Two inputs drive the edit: | Backend | Image-to-image | Reference cap | How | |---|---|---|---| | **FAL.ai** (edit-capable models below) | ✓ | up to 9 | routes to the model's `/edit` endpoint | -| **OpenAI** (`gpt-image-2`) | ✓ | up to 16 | `images.edit()` | +| **OpenAI** (GPT Image 2 / 2.5 Flare / Sunburst) | ✓ | up to 16 | `images.edit()` | | **xAI** (Grok Imagine) | ✓ | 1 | `/v1/images/edits` (`grok-imagine-image-quality`) | | **Krea** (`Krea 2`) | ✓ | up to 10 | reference-guided generation (`image_style_references`) | | **OpenAI (Codex auth)** | ✓ | up to 16 | Codex Responses `image_generation` tool with `input_image` content parts | From b1f003e18633298d549668b8e186af84cca45b76 Mon Sep 17 00:00:00 2001 From: Teknium <127238744+teknium1@users.noreply.github.com> Date: Tue, 8 Sep 2026 13:34:50 -0700 Subject: [PATCH 222/227] feat: add FAL GPT Image 2.5 generation and editing selections --- tests/tools/test_image_generation.py | 23 ++++++++++++ tools/image_generation_catalog.py | 25 +++++++++++++ .../user-guide/features/image-generation.md | 37 +++++++++++++++++-- 3 files changed, 82 insertions(+), 3 deletions(-) diff --git a/tests/tools/test_image_generation.py b/tests/tools/test_image_generation.py index 66313884911f..0fdf95210123 100644 --- a/tests/tools/test_image_generation.py +++ b/tests/tools/test_image_generation.py @@ -30,6 +30,29 @@ def image_tool(): # Catalog integrity # --------------------------------------------------------------------------- +@pytest.mark.parametrize("variant", ["flare", "sunburst"]) +@pytest.mark.parametrize("aspect,size", [ + ("landscape", "landscape_4_3"), ("square", "square_hd"), ("portrait", "portrait_4_3"), +]) +def test_image_25_selection_routes_generation_and_edits(image_tool, monkeypatch, variant, aspect, size): + model = f"openai/gpt-image-2.5/{variant}/text-to-image" + monkeypatch.setenv("FAL_IMAGE_MODEL", model) + monkeypatch.setenv("FAL_KEY", "test-key") + selected, meta = image_tool._resolve_fal_model() + assert selected == model + refs = [f"https://example.com/{i}.png" for i in range(17)] + for sources, endpoint in (([], model), (refs, f"openai/gpt-image-2.5/{variant}/edit")): + actual, payload = image_tool._prepare_fal_request( + selected, meta, "a cup", aspect, 42, {"guidance_scale": 9}, sources, + ) + assert actual == endpoint + assert payload["quality"] == "medium" + assert payload["image_size"] == size + assert "seed" not in payload and "guidance_scale" not in payload + assert payload.get("image_urls", []) == sources[:16] + assert meta["upscale"] is False + + class TestFalCatalog: """Every FAL_MODELS entry must have a consistent shape.""" diff --git a/tools/image_generation_catalog.py b/tools/image_generation_catalog.py index fc9e3ddfbe2b..fa81b9a179c4 100644 --- a/tools/image_generation_catalog.py +++ b/tools/image_generation_catalog.py @@ -153,6 +153,31 @@ def _model( }, max_reference_images=16, ), + # Same minimum pixel count as GPT Image 2; keep medium quality explicit + # rather than inheriting FAL's higher-cost high default. + **{ + f"openai/gpt-image-2.5/{variant}/text-to-image": _model( + f"GPT Image 2.5 {variant.title()}", speed, strengths, "Token-based pricing", + sizes={ + "landscape": "landscape_4_3", "square": "square_hd", "portrait": "portrait_4_3", + }, + defaults={"quality": "medium", "num_images": 1, "output_format": "png"}, + supports={ + "prompt", "image_size", "quality", "num_images", "output_format", "background", + "output_compression", "sync_mode", + }, + edit_endpoint=f"openai/gpt-image-2.5/{variant}/edit", + edit_supports={ + "prompt", "image_urls", "image_size", "quality", "num_images", "output_format", + "background", "output_compression", "sync_mode", "mask_url", "input_fidelity", + }, + max_reference_images=16, + ) + for variant, speed, strengths in ( + ("flare", "Fast", "Everyday creation, natural lighting and textures"), + ("sunburst", "Slower", "Precision editing, subject and composition consistency"), + ) + }, "fal-ai/ideogram/v3": _model( "Ideogram V3", "~5s", "Best typography", "$0.03-0.09/image", defaults={"rendering_speed": "BALANCED", "expand_prompt": True, "style": "AUTO"}, diff --git a/website/docs/user-guide/features/image-generation.md b/website/docs/user-guide/features/image-generation.md index b9904cc07444..862ee885d19f 100644 --- a/website/docs/user-guide/features/image-generation.md +++ b/website/docs/user-guide/features/image-generation.md @@ -126,6 +126,37 @@ Auth reuses the same env vars as the Meta chat provider — `MODEL_API_KEY` as aliases. Set `META_BASE_URL` to point at a proxy or alternate host. Text-to-image only for now; responses are saved to `$HERMES_HOME/cache/images/`. +## FAL: GPT Image 2.5 + +Select **GPT Image 2.5 Flare** or **GPT Image 2.5 Sunburst** under +`hermes tools` → Image Generation → FAL.ai. The model IDs are: + +- `openai/gpt-image-2.5/flare/text-to-image` +- `openai/gpt-image-2.5/sunburst/text-to-image` + +For example: + +```bash +hermes config set image_gen.provider fal +hermes config set image_gen.model openai/gpt-image-2.5/flare/text-to-image +``` + +Providing `image_url` or reference images automatically selects the corresponding +`openai/gpt-image-2.5/flare/edit` or `openai/gpt-image-2.5/sunburst/edit` endpoint. +Both accept up to 16 source images. Hermes pins quality to `medium`, matching its +existing FAL GPT Image policy rather than FAL's higher-cost `high` default. +Landscape and portrait use 4:3 presets to satisfy the minimum pixel count; +square uses `square_hd`. Upscaling remains off unless requested. + +FAL bills by tokens, not a fixed image price: $5/M text input, $1.25/M cached +text input, $10/M text output, $8/M image input, $2/M cached image input, and +$30/M image output, rounded up to $0.0001 per request. See the +[Flare](https://fal.ai/models/openai/gpt-image-2.5/flare/text-to-image) and +[Sunburst](https://fal.ai/models/openai/gpt-image-2.5/sunburst/text-to-image) +pages. Direct FAL requires a funded `FAL_KEY`; managed-gateway availability +depends on that gateway's endpoint allowlist and is not implied by FAL availability. +Existing provider and model defaults are unchanged. + ## OpenAI API: GPT Image 2.5 The **OpenAI** provider supports GPT Image 2.5 Flare (fast everyday creation) @@ -154,7 +185,7 @@ does not estimate 2.5 token consumption. See the official The **OpenAI (Codex auth)** provider remains separate: its backend can accept an image-model value without honoring that selection, so a successful image alone does not verify Flare or Sunburst routing. These selections are offered -only through the direct OpenAI API provider, not Codex auth or FAL. +through the direct OpenAI API provider and FAL, not as verified Codex-auth selections. ## Usage @@ -196,7 +227,7 @@ Two inputs drive the edit: | Backend | Image-to-image | Reference cap | How | |---|---|---|---| -| **FAL.ai** (edit-capable models below) | ✓ | up to 9 | routes to the model's `/edit` endpoint | +| **FAL.ai** (edit-capable models below) | ✓ | up to 16 (per model) | routes to the model's `/edit` endpoint | | **OpenAI** (GPT Image 2 / 2.5 Flare / Sunburst) | ✓ | up to 16 | `images.edit()` | | **xAI** (Grok Imagine) | ✓ | 1 | `/v1/images/edits` (`grok-imagine-image-quality`) | | **Krea** (`Krea 2`) | ✓ | up to 10 | reference-guided generation (`image_style_references`) | @@ -205,7 +236,7 @@ Two inputs drive the edit: FAL models with an editing endpoint: `flux-2/klein/9b`, `flux-2-pro`, `nano-banana-pro`, `gpt-image-1.5`, `gpt-image-2`, `ideogram/v3`, and -`qwen-image`. Pure text-to-image FAL models (`z-image/turbo`, `recraft`, +`qwen-image`, plus GPT Image 2.5 Flare and Sunburst above. Pure text-to-image FAL models (`z-image/turbo`, `recraft`, `krea/*`) reject image inputs with a clear error pointing you at an edit-capable model. From 1175ac4bb931097dc286542bf52f0e3f92aff916 Mon Sep 17 00:00:00 2001 From: brooklyn! Date: Tue, 8 Sep 2026 18:41:41 -0500 Subject: [PATCH 223/227] feat(desktop): default glass to 29% tint on the sidebar --- apps/desktop/electron/translucency.test.ts | 64 +++++++++---------- .../src/store/translucency.win10.test.ts | 19 ++---- apps/shared/src/translucency.ts | 42 +++--------- 3 files changed, 46 insertions(+), 79 deletions(-) diff --git a/apps/desktop/electron/translucency.test.ts b/apps/desktop/electron/translucency.test.ts index e9c9ed0c61ff..d525c7c864af 100644 --- a/apps/desktop/electron/translucency.test.ts +++ b/apps/desktop/electron/translucency.test.ts @@ -603,9 +603,7 @@ describe('what an update actually changes natively', () => { }) it('leaves a window alone when glass is selected but off', () => { - // The light default carries one point of fade. Someone who dragged the - // tint to zero asked for an opaque window, and that point must not follow - // them there — off has to mean exactly 1, not 0.9999. + // A saved fade must not follow the tint to zero: off means opaque. expect(windowOpacityFor({ ...glass(0), fade: 1 })).toBe(1) expect(windowOpacityFor({ ...glass(0), fade: 40 })).toBe(1) }) @@ -615,12 +613,7 @@ describe('what an update actually changes natively', () => { }) }) -/** - * The shipped defaults, per platform. These are the numbers a fresh profile - * gets before anyone opens Settings, so they are the ones most people will - * ever see — and they differ by platform because the lever means different - * things behind macOS vibrancy and Windows acrylic. - */ +/** Fresh profiles share the sidebar treatment, with native frost per platform. */ describe('the defaults a fresh profile lands on', () => { const mac = (appearance: 'dark' | 'light') => defaultTranslucencyValues(appearance, false) const win = (appearance: 'dark' | 'light') => defaultTranslucencyValues(appearance, true) @@ -641,21 +634,16 @@ describe('the defaults a fresh profile lands on', () => { expect(defaultTranslucencyState('dark', false, false).mode).toBe('clear') }) - it('tints light more heavily than dark, on both platforms', () => { - // A dark field already separates from what is behind it; a bright one - // needs real thinning before the desktop reads as a layer underneath. - expect(mac('light').intensity).toBeGreaterThan(mac('dark').intensity) - expect(win('light').intensity).toBeGreaterThan(win('dark').intensity) - }) - - it('asks far less of Windows, which composites its own tint in DWM', () => { - expect(win('light').intensity).toBeLessThan(mac('light').intensity) - expect(win('dark').intensity).toBeLessThan(mac('dark').intensity) + it('keeps tint consistent across appearances and platforms', () => { + for (const values of [mac('light'), mac('dark'), win('light'), win('dark')]) { + expect(values.intensity).toBe(mac('light').intensity) + } }) - it('never fades a Windows window — setOpacity dims the composited backdrop', () => { - expect(win('light').fade).toBe(0) - expect(win('dark').fade).toBe(0) + it('keeps the content column opaque at the native level', () => { + for (const values of [mac('light'), mac('dark'), win('light'), win('dark')]) { + expect(windowOpacityFor({ ...values, mode: 'glass' })).toBe(1) + } }) it('defaults each platform onto a frost that platform can actually render', () => { @@ -665,9 +653,9 @@ describe('the defaults a fresh profile lands on', () => { } }) - it('opens the whole window, not just the sidebar rail', () => { + it('uses the normalized scope default for every appearance and platform', () => { for (const values of [mac('light'), mac('dark'), win('light'), win('dark')]) { - expect(values.scope).toBe('window') + expect(values.scope).toBe(normalizeScope(undefined)) } }) }) @@ -680,9 +668,14 @@ describe('the defaults a fresh profile lands on', () => { describe('resolving the book for the painted appearance', () => { const empty = normalizeBook(null, true) - it('falls all the way through to the platform default', () => { - expect(resolveTranslucency(empty, 'dark', false).intensity).toBe(defaultTranslucencyValues('dark', false).intensity) - expect(resolveTranslucency(empty, 'dark', true).intensity).toBe(defaultTranslucencyValues('dark', true).intensity) + it('agrees with the native first-window defaults in either appearance', () => { + for (const appearance of ['light', 'dark'] as const) { + for (const isWindows of [false, true]) { + expect(resolveTranslucency(empty, appearance, isWindows)).toEqual( + defaultTranslucencyState(appearance, true, isWindows) + ) + } + } }) it('scopes an edit to the appearance it was made in', () => { @@ -692,14 +685,17 @@ describe('resolving the book for the painted appearance', () => { expect(resolveTranslucency(book, 'dark', false).intensity).toBe(defaultTranslucencyValues('dark', false).intensity) }) - it('carries a v1 state into BOTH appearances via base', () => { - // Someone who tuned a window before appearances were split keeps exactly - // what was on screen, in either appearance, until they edit one of them. - const migrated = normalizeBook({ intensity: 40, mode: 'glass' }, true) + it('preserves a saved whole-window treatment in both appearances', () => { + const saved = { intensity: 40, scope: 'window', mode: 'glass' } as const + const migrated = normalizeBook(saved, true) + + expect(migrated.base).toEqual({ intensity: saved.intensity, scope: saved.scope }) - expect(migrated.base.intensity).toBe(40) - expect(resolveTranslucency(migrated, 'light', false).intensity).toBe(40) - expect(resolveTranslucency(migrated, 'dark', false).intensity).toBe(40) + for (const appearance of ['light', 'dark'] as const) { + for (const isWindows of [false, true]) { + expect(resolveTranslucency(migrated, appearance, isWindows)).toMatchObject(saved) + } + } }) it('lets an appearance override base without disturbing the other', () => { diff --git a/apps/desktop/src/store/translucency.win10.test.ts b/apps/desktop/src/store/translucency.win10.test.ts index ac48eb3e8ca6..99862bbf75e1 100644 --- a/apps/desktop/src/store/translucency.win10.test.ts +++ b/apps/desktop/src/store/translucency.win10.test.ts @@ -10,10 +10,9 @@ import { describe, expect, it, vi } from 'vitest' // That is the exact shape of the bug this test guards: the $translucency // computed used to pass GLASS_IS_WINDOWS as the "isWindows" argument to // resolveTranslucency. On Win10 glass is unsupported, so GLASS_IS_WINDOWS is -// false and the fallback defaults came from the MAC table (light intensity 66, -// dark intensity 22) instead of the WINDOWS table (light intensity 20, dark -// intensity 5) — an untouched profile rendered at ~70% opacity. The fix passes -// isWindowsPlatform() instead, which is true on Win32 regardless of glass. +// false and the fallback defaults came from the MAC table instead of the +// WINDOWS table. The fix passes isWindowsPlatform() instead, which is true +// on Win32 regardless of glass support. vi.hoisted(() => { Object.defineProperty(globalThis.navigator, 'platform', { configurable: true, value: 'Win32' }) Object.defineProperty(globalThis.window, 'hermesDesktop', { @@ -26,9 +25,7 @@ import { defaultTranslucencyValues } from '@hermes/shared/translucency' import { $translucency, $translucencyBook, GLASS_SUPPORTED, setAppearance } from './translucency' -// The windows table is the one that must win on Win10. These are the numbers -// the issue calls out: mac light 66 / mac dark 22 vs windows light 20 / -// windows dark 5. +// The Windows table must win on Win10, even when tint matches macOS. const WINDOWS_DARK = defaultTranslucencyValues('dark', true) const WINDOWS_LIGHT = defaultTranslucencyValues('light', true) @@ -45,15 +42,12 @@ describe('Win10 translucency defaults (regression for #90824)', () => { // platform defaults. The store's initial appearance is dark. expect($translucency.get()).toEqual({ ...WINDOWS_DARK, mode: 'clear' }) - // The bug's signature: mac dark intensity is 22, windows dark is 5. - expect($translucency.get().intensity).toBe(5) - expect($translucency.get().intensity).not.toBe(22) + expect($translucency.get().material).not.toBe(defaultTranslucencyValues('dark', false).material) // Light appearance must resolve the windows light table too. setAppearance('light') expect($translucency.get()).toEqual({ ...WINDOWS_LIGHT, mode: 'clear' }) - expect($translucency.get().intensity).toBe(20) - expect($translucency.get().intensity).not.toBe(66) + expect($translucency.get().material).not.toBe(defaultTranslucencyValues('light', false).material) }) it('keeps the mode clear when glass is unsupported', () => { @@ -69,6 +63,5 @@ describe('Win10 translucency defaults (regression for #90824)', () => { setAppearance('dark') expect($translucency.get()).toEqual({ ...WINDOWS_DARK, mode: 'clear' }) - expect($translucency.get().intensity).toBe(5) }) }) diff --git a/apps/shared/src/translucency.ts b/apps/shared/src/translucency.ts index 75fc2139d957..ed3af92abe0e 100644 --- a/apps/shared/src/translucency.ts +++ b/apps/shared/src/translucency.ts @@ -52,7 +52,7 @@ export const GLASS_SCOPES = ['window', 'sidebar'] as const export type GlassScope = (typeof GLASS_SCOPES)[number] -export const DEFAULT_GLASS_SCOPE: GlassScope = 'window' +export const DEFAULT_GLASS_SCOPE: GlassScope = 'sidebar' /** * Electron `setBackgroundMaterial` values. `'auto'` is deliberately absent — @@ -125,41 +125,19 @@ export type TranslucencyValues = Omit export type Appearance = 'light' | 'dark' /** - * Per-appearance defaults, per platform family. Glass ships ON: it is the - * better-looking half of the feature, and a lever that starts at zero is a - * feature nobody finds. - * - * The two platforms need different numbers because the lever means different - * things behind them. `intensity` is how much of the theme tint the renderer - * REMOVES (see `glassSurfaceKeep`), and what shows through underneath is a - * native material with its own weight: - * - * - macOS vibrancy is genuinely sheer, so the tint has to come most of the way - * off before the desktop reads at all. Light leans heavy — a bright desktop - * behind a bright window needs real thinning before the field separates — - * with a single point of fade so the window edge reads as glass rather than - * as paint. Dark takes far less: a dark field already separates, and the - * tint that flatters light would smother it. - * - Windows acrylic composites its OWN tint in DWM before the page is drawn, - * so the renderer's tint stacks on top of a backdrop that is already doing - * the work. The same numbers that read as frost on a Mac read as a washed - * sheet here; these stay low and let DWM carry it. Fade stays at zero — - * `setOpacity` over a system backdrop dims the composited result rather than - * deepening it. - * - * Both sit on the frost each platform renders best: 'header' and 'titlebar' - * are macOS-only rungs (on Windows they collapse onto mica — see - * `glassMaterialsFor`), while 'under-window' is the acrylic rung, the live - * blur closest to what macOS calls under-window. + * Glass starts at 29% tint, confined to the sidebar in both appearances. + * Fade stays off so the content column and text remain fully opaque. + * Frost keeps its platform/appearance tuning: macOS uses header/titlebar, + * while Windows uses under-window, the live acrylic backdrop. */ const DEFAULT_VALUES: Record<'mac' | 'windows', Record> = { mac: { - light: { intensity: 66, fade: 1, material: 'header', scope: 'window' }, - dark: { intensity: 22, fade: 0, material: 'titlebar', scope: 'window' } + light: { intensity: 29, fade: 0, material: 'header', scope: DEFAULT_GLASS_SCOPE }, + dark: { intensity: 29, fade: 0, material: 'titlebar', scope: DEFAULT_GLASS_SCOPE } }, windows: { - light: { intensity: 20, fade: 0, material: 'under-window', scope: 'window' }, - dark: { intensity: 5, fade: 0, material: 'under-window', scope: 'window' } + light: { intensity: 29, fade: 0, material: 'under-window', scope: DEFAULT_GLASS_SCOPE }, + dark: { intensity: 29, fade: 0, material: 'under-window', scope: DEFAULT_GLASS_SCOPE } } } @@ -274,7 +252,7 @@ export function normalizeMaterial(value: unknown): GlassMaterial { return GLASS_MATERIALS.includes(value as GlassMaterial) ? (value as GlassMaterial) : DEFAULT_GLASS_MATERIAL } -/** Unknown or unsupported values fall back to whole-window glass. */ +/** Unknown or unsupported values fall back to sidebar glass. */ export function normalizeScope(value: unknown): GlassScope { return GLASS_SCOPES.includes(value as GlassScope) ? (value as GlassScope) : DEFAULT_GLASS_SCOPE } From 608b97c3055ce6779ba9b6d9282b79d246dfaea3 Mon Sep 17 00:00:00 2001 From: brooklyn! Date: Tue, 8 Sep 2026 18:41:41 -0500 Subject: [PATCH 224/227] docs(desktop): document sidebar glass defaults --- apps/desktop/DESIGN.md | 9 +++++++++ 1 file changed, 9 insertions(+) diff --git a/apps/desktop/DESIGN.md b/apps/desktop/DESIGN.md index ed72f83c81a1..8f3103392137 100644 --- a/apps/desktop/DESIGN.md +++ b/apps/desktop/DESIGN.md @@ -86,6 +86,15 @@ Menus and popovers use their own shared `shadow-md` + dashed targets and local blur. These are semantic surface classes, not licenses for call-site shadow or border inventions. +## Window glass + +Glass defaults to **29% Tint, Sidebar only** in both light and dark appearances. +Fade defaults to zero so the content column and text stay opaque. Native frost +keeps its platform/appearance defaults. Explicitly saved settings take precedence; +changing defaults must not overwrite a user's existing choices. The shared +`apps/shared/src/translucency.ts` resolver owns these defaults for both the +renderer and Electron's first window paint. + ## Stroke & color tokens | Token | Use | From 258b351fc1507623bf8dc9f363b815c0fa54b049 Mon Sep 17 00:00:00 2001 From: brooklyn! Date: Tue, 8 Sep 2026 18:45:15 -0500 Subject: [PATCH 225/227] fix(desktop): keep background reports behind bounded disclosures --- .../thread/system-message.test.tsx | 26 +++++++++++++++-- .../assistant-ui/thread/system-message.tsx | 28 +++++++++++++++---- 2 files changed, 45 insertions(+), 9 deletions(-) diff --git a/apps/desktop/src/components/assistant-ui/thread/system-message.test.tsx b/apps/desktop/src/components/assistant-ui/thread/system-message.test.tsx index 14ea92d7ac2b..5ee1206fa66e 100644 --- a/apps/desktop/src/components/assistant-ui/thread/system-message.test.tsx +++ b/apps/desktop/src/components/assistant-ui/thread/system-message.test.tsx @@ -1,5 +1,5 @@ import { AssistantRuntimeProvider, type ThreadMessage, useExternalStoreRuntime } from '@assistant-ui/react' -import { cleanup, render } from '@testing-library/react' +import { cleanup, fireEvent, render } from '@testing-library/react' import { afterEach, describe, expect, it } from 'vitest' import { $displayTimestamps } from '@/store/display-timestamps' @@ -14,13 +14,13 @@ $displayTimestamps.set(true) const timestamp = new Date('2026-05-01T00:00:00.000Z') stubThreadEnvironment() -function Harness({ text }: { text: string }) { +function Harness({ text, asyncResult }: { text: string; asyncResult?: string }) { const message = { id: 'system-1', role: 'system', content: [{ type: 'text', text }], createdAt: timestamp, - metadata: { custom: { timelineTimestamp: timestamp.getTime() / 1000 } } + metadata: { custom: { timelineTimestamp: timestamp.getTime() / 1000, asyncResult } } } as unknown as ThreadMessage const runtime = useExternalStoreRuntime({ @@ -46,6 +46,26 @@ function expectTimestampSeparated(container: HTMLElement, precedingText: string) afterEach(cleanup) +describe('background report disclosure', () => { + it('keeps result bodies out of the transcript until opened and removes them when collapsed', () => { + const report = '{"blockers":[{"title":"Local-model readiness uses the wrong endpoint"}]}' + const { container, getByRole } = render() + + expect(container.textContent).not.toContain('blockers') + expectTimestampSeparated(container, '2 background agents finished') + const toggle = getByRole('button', { name: '2 background agents finished' }) + expect(toggle.getAttribute('aria-expanded')).toBe('false') + + fireEvent.click(toggle) + expect(toggle.getAttribute('aria-expanded')).toBe('true') + expect(container.textContent).toContain(report) + + fireEvent.click(toggle) + expect(toggle.getAttribute('aria-expanded')).toBe('false') + expect(container.textContent).not.toContain('blockers') + }) +}) + describe('system message timestamp text separation', () => { it('separates an ordinary system row timestamp in accessible and copied text', () => { const { container } = render() diff --git a/apps/desktop/src/components/assistant-ui/thread/system-message.tsx b/apps/desktop/src/components/assistant-ui/thread/system-message.tsx index 08a919e9de24..1bdf62754860 100644 --- a/apps/desktop/src/components/assistant-ui/thread/system-message.tsx +++ b/apps/desktop/src/components/assistant-ui/thread/system-message.tsx @@ -1,10 +1,10 @@ import { MessagePrimitive, useAuiState } from '@assistant-ui/react' -import { type FC } from 'react' +import { type FC, useState } from 'react' import { MarkdownTextContent } from '@/components/assistant-ui/markdown-text' import { messageContentText } from '@/components/assistant-ui/thread/content' import { MessageTimelineTimestamp } from '@/components/assistant-ui/thread/timeline-timestamp' -import { SCAFFOLD_LABEL_CLASS } from '@/components/chat/scaffold-row' +import { SCAFFOLD_LABEL_CLASS, ScaffoldRow } from '@/components/chat/scaffold-row' import { Codicon } from '@/components/ui/codicon' import { ToolIcon } from '@/components/ui/tool-icon' import { LinkifiedText } from '@/lib/external-link' @@ -17,6 +17,7 @@ const REVIEW_NOTE_RE = /^review:(?