runtime: cut the fleet to v0.20.5 (2026.8.19) with all fleet patches carried

.coderabbit.yaml

@@ -0,0 +1,18 @@
+# CodeRabbit configuration.
+#
+# Auto-review is DISABLED repo-wide: the app was enabled at the org level on
+# 2026-08-14 and immediately began reviewing every opened PR. This repo's
+# merge gate is CI ("All required checks pass") plus maintainer review —
+# CodeRabbit reviews carry no merge weight (protect-main ruleset requires
+# 0 approving reviews), so auto-firing on the full PR firehose adds comment
+# noise without gating value.
+#
+# The bot stays installed and summonable on demand: comment
+# `@coderabbitai review` on any PR to request a one-off review, or
+# `@coderabbitai ignore` to mute it on a PR it has already joined.
+#
+# To re-enable auto-review, flip `enabled: true` below (or delete this file —
+# the app default is on).
+reviews:
+  auto_review:
+    enabled: false

.dockerignore

@@ -40,6 +40,10 @@ ui-tui/packages/hermes-ink/dist/
 # Environment files
 .env
 .env.*
+# ...but keep the template: docker/stage2-hook.sh seeds $HERMES_HOME/.env from
+# /opt/hermes/.env.example on first boot (OOF-285 — excluding it silently broke
+# first-boot .env seeding and the API_SERVER_KEY generation that depends on it).
+!.env.example
 
 # IDE
 .vscode/
@@ -98,7 +102,6 @@ plans/
 .plans/
 
 # Repo-level dotfiles that are git-only or dev-tooling config
-.env.example
 .envrc
 .gitattributes
 .hadolint.yaml

.env.example

@@ -104,6 +104,14 @@
 # $10/month subscription. Get your key at: https://opencode.ai/auth
 # OPENCODE_GO_API_KEY=
 
+# =============================================================================
+# LLM PROVIDER (OpenCode Free)
+# =============================================================================
+# OpenCode Free provides keyless free models (Ox Alpha / x-preview-f-free,
+# big-pickle, etc.). NO env var and NO account needed — requests are sent
+# anonymously (the free tier rejects any unrecognized Authorization header).
+# Select it with `hermes model` or `/model free`.
+
 # =============================================================================
 # LLM PROVIDER (Hugging Face Inference Providers)
 # =============================================================================

.gitattributes

@@ -8,3 +8,27 @@ web/package-lock.json linguist-generated=true
 Dockerfile  text eol=lf
 *.dockerfile text eol=lf
 docker/entrypoint.sh text eol=lf
+
+# Enforce LF for all source/text files. Windows editors and tools default to
+# CRLF; without normalization a Windows contributor's edit turns into a
+# whole-file phantom diff (every line "changed" by its ending), breaks
+# string-match patch tooling, and pollutes review. `text` normalizes to LF
+# at check-in; `eol=lf` also checks out as LF so the working tree matches
+# the index on every platform. PowerShell files are the deliberate
+# exception (PS 5.1 tooling expects CRLF).
+*.py    text eol=lf
+*.ts    text eol=lf
+*.tsx   text eol=lf
+*.js    text eol=lf
+*.mjs   text eol=lf
+*.cjs   text eol=lf
+*.jsx   text eol=lf
+*.json  text eol=lf
+*.yaml  text eol=lf
+*.yml   text eol=lf
+*.toml  text eol=lf
+*.md    text eol=lf
+*.css   text eol=lf
+*.html  text eol=lf
+*.svg   text eol=lf
+*.ps1   text eol=crlf

.github/actions/detect-changes/action.yml

@@ -24,6 +24,12 @@ outputs:
   docker_meta:
     description: Docker setup and meta files have changed.
     value: ${{ steps.classify.outputs.docker_meta }}
+  docker:
+    description: Files included in the docker image have changed.
+    value: ${{ steps.classify.outputs.docker }}
+  nix:
+    description: Run `nix flake check` (flake inputs, or any product Python change).
+    value: ${{ steps.classify.outputs.nix }}
   site:
     description: Build the Docusaurus docs site.
     value: ${{ steps.classify.outputs.site }}
@@ -33,9 +39,18 @@ outputs:
   deps:
     description: Check pyproject.toml dependency upper bounds.
     value: ${{ steps.classify.outputs.deps }}
+  uv_lock:
+    description: Run `uv lock --check` (pyproject.toml / uv.lock changes only).
+    value: ${{ steps.classify.outputs.uv_lock }}
   npm_lock:
     description: Post/update the semantic package-lock.json diff PR comment.
     value: ${{ steps.classify.outputs.npm_lock }}
+  installer:
+    description: Run the PowerShell installer tests on a Windows runner.
+    value: ${{ steps.classify.outputs.installer }}
+  rust:
+    description: Run `cargo test` for the Tauri bootstrap installer.
+    value: ${{ steps.classify.outputs.rust }}
   mcp_catalog:
     description: Require MCP catalog security review label.
     value: ${{ steps.classify.outputs.mcp_catalog }}

.github/workflows/ci-review-comment.yml

@@ -0,0 +1,80 @@
+name: CI review comment
+
+# Live-updating PR review comment.
+#
+# The poller runs for up to 40 minutes.
+# This run lives in its own workflow.
+# A run stays in progress until its last job ends, and GitHub refuses
+# ``gh run rerun`` on a run that is in progress.
+#
+# ``workflow_run`` starts this when CI starts. It always reads the workflow
+# and the scripts from the default branch, never from the PR head.
+# It makes a write token safe here.
+#
+# The poller reads job results through the API. Thus it watches the CI run
+# and the separate docker run, and it depends on neither.
+
+on:
+  workflow_run:
+    workflows: [CI]
+    # ``in_progress``, not ``requested``: a first-time contributor's run
+    # sits in ``action_required`` until a maintainer approves it, and
+    # ``requested`` fires at creation — the poller would wait out its
+    # whole timeout on a run that never starts. ``in_progress`` fires
+    # when the run actually starts, and it also fires on re-runs, which
+    # ``requested`` does not.
+    types: [in_progress]
+
+permissions:
+  contents: read
+  actions: read
+  pull-requests: write
+
+# One poller per CI run. A new push starts a new CI run, and its poller
+# cancels the poller of the run that GitHub superseded. The group keys
+# on the head repository too: fork PRs often share a branch name
+# (``main``, ``patch-1``), and two PRs must not cancel each other.
+concurrency:
+  group: ci-review-comment-${{ github.event.workflow_run.head_repository.full_name }}-${{ github.event.workflow_run.head_branch }}
+  cancel-in-progress: true
+
+jobs:
+  comment:
+    name: CI review comment (live)
+    # Fork PRs get no comment: the poller needs a write token, and the
+    # ``pull_requests`` payload is empty for a fork run.
+    if: >-
+      github.event.workflow_run.event == 'pull_request' &&
+      github.event.workflow_run.head_repository.full_name == github.repository
+    runs-on: ubuntu-latest
+    timeout-minutes: 60
+    steps:
+      - name: Checkout trusted default branch
+        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
+        with:
+          ref: ${{ github.event.repository.default_branch }}
+          persist-credentials: false
+
+      - name: Run live comment poller
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          # The CI run to report on — not this run. The name must not be
+          # GITHUB_RUN_ID: the runner sets the GITHUB_* defaults itself and
+          # ignores an env: override, so that name silently resolves to THIS
+          # run. The poller then watches itself, which stays in_progress for
+          # as long as the poller runs, and it waits out its whole timeout.
+          CI_RUN_ID: ${{ github.event.workflow_run.id }}
+          # Sibling runs for the same commit that the comment also covers,
+          # one workflow name per line (a name can contain a comma).
+          # The poller resolves each name to its runs through the API.
+          WATCH_WORKFLOWS: |
+            Docker Build, Test, and Publish
+          PR_NUMBER: ${{ github.event.workflow_run.pull_requests[0].number }}
+          RUN_URL: ${{ github.event.workflow_run.html_url }}
+          # Commit info for the review comment header.
+          COMMIT_SHA: ${{ github.event.workflow_run.head_sha }}
+          COMMIT_MESSAGE: ${{ github.event.workflow_run.head_commit.message }}
+        run: |
+          python3 -u scripts/ci/live_comment.py \
+            --interval 15 \
+            --timeout 3000