From 54f03722f91deeb13be55f84305ac64f989f01ae Mon Sep 17 00:00:00 2001 From: Agent Bot Date: Fri, 21 Aug 2026 23:04:38 -0400 Subject: [PATCH 1/2] docs: record Windows installer fix verified via real CI, Linux still open Confirms PR #172's cuda_v13 prune fixed Windows Tauri Desktop Build (verified via manual workflow_dispatch, real makensis run producing a 957MB installer artifact). Linux's linuxdeploy failure persists but fails in ~20s (not the ~2min of a real large-payload build), suggesting a possibly distinct root cause worth investigating separately - not a blocker since desktop-linux is continue-on-error. Co-Authored-By: Claude Sonnet 5 Claude-Session: https://claude.ai/code/session_01WexPHS5v6ZRnwzQGR4JG4A --- docs/DEPENDENCY_BUNDLING_PLAN.md | 15 +++++++++----- docs/governance/DEFERRED_WORK.md | 34 ++++++++++++++++++++++---------- 2 files changed, 34 insertions(+), 15 deletions(-) diff --git a/docs/DEPENDENCY_BUNDLING_PLAN.md b/docs/DEPENDENCY_BUNDLING_PLAN.md index c8aeaa41..a416689f 100644 --- a/docs/DEPENDENCY_BUNDLING_PLAN.md +++ b/docs/DEPENDENCY_BUNDLING_PLAN.md @@ -273,11 +273,16 @@ Video generation stays fully out per the acceptance criterion. script). Verified for real: ran the fetch script end-to-end against the live `windows-amd64` v0.32.13 asset before and after the change — `cuda_v13` confirmed absent post-fetch, `cuda_v12` (~1.1GB) and the small - `vulkan` fallback dir (~50MB) still present. **Not yet verified:** whether - this is sufficient to actually clear CI's `Tauri Desktop Build`/ - `Tauri Desktop Build (Linux)` failures — needs a real CI run, not local - extraction alone. Linux's `linuxdeploy` failure hasn't been independently - confirmed to share the same root cause as Windows's `makensis` failure. + `vulkan` fallback dir (~50MB) still present. **Verified against real CI + 2026-08-22** (manual `workflow_dispatch` run against the fix branch): + Windows `Tauri Desktop Build` now succeeds, produced a real + `Alphonso_2.6.2_x64-setup.exe` (~957MB). Merged via PR #172 (`96354f9`). + **Linux `Tauri Desktop Build (Linux)` still fails** — different failure + shape (fails in ~20s vs. the prior successful run's ~2 minutes, suggesting + a possibly distinct root cause, not necessarily the same size issue) — not + a blocker since that job is `continue-on-error: true`, but still open; see + `docs/governance/DEFERRED_WORK.md`'s 2026-08-21/22 entry for the resume + hint. ### PY — Python / Voice OS diff --git a/docs/governance/DEFERRED_WORK.md b/docs/governance/DEFERRED_WORK.md index cc574e00..899529ac 100644 --- a/docs/governance/DEFERRED_WORK.md +++ b/docs/governance/DEFERRED_WORK.md @@ -36,16 +36,30 @@ Rule 12 / Rule 11. This register survives the session. Future agents resume from binaries run on both older and newer driver installs, while dropping v12 instead would have saved more space (~1.1GB vs. ~630MB) at the cost of silently losing GPU acceleration for anyone without the newest driver. - **Not yet verified:** whether this cut is enough to actually clear the - NSIS/linuxdeploy failures — that needs a real CI Tauri Desktop Build run, - not local extraction alone. If CI still fails, the agreed fallback order is - Option 3 (split lite installer + separate runtime downloader), then as a - last resort Option 1 (download Ollama at first run instead of bundling it, - which reverses `DEPENDENCY_BUNDLING_PLAN.md`'s explicit "works offline on - first launch" decision from 2026-08-16 — only accept that regression if - 2 and 3 both fail). Resume hint: watch the next `main` push's `Tauri - Desktop Build`/`Tauri Desktop Build (Linux)` job results; if still red, - move to Option 3 before Option 1. + **Verified 2026-08-22 via a real CI run** (manually dispatched against the + fix branch with `gh workflow run ci.yml --ref ...`, since these jobs only + trigger on push-to-main/workflow_dispatch, not on `pull_request`): **Windows + `Tauri Desktop Build` now succeeds** — `makensis` completed in ~9 minutes + and produced a real `Alphonso_2.6.2_x64-setup.exe` artifact (~957MB, + confirmed via `gh api .../artifacts`), no more datablock error. Merged to + `main` at `96354f9` (PR #172). + **Linux `Tauri Desktop Build (Linux)` still fails** — same + `failed to run linuxdeploy` error, now in ~20 seconds (vs. ~2 minutes for + the last known-good run on 2026-08-16), with zero diagnostic output between + "Bundling ... .AppImage" and the failure — Tauri swallows linuxdeploy's own + stderr. The near-instant failure time suggests this may NOT be the same + size-driven root cause as Windows's NSIS bug (a real size problem would + likely fail partway through processing ~1GB+ of payload, not in 20s flat) — + worth investigating as a possibly distinct issue (FUSE/AppImage execution + environment on the runner, a linuxdeploy/plugin version regression, disk + space) rather than assuming the same CUDA-trim fix will resolve it. + **This is not currently blocking anything**: `desktop-linux` has + `continue-on-error: true` in `ci.yml` (same as macOS), so it was never a + required check — Windows was the only gating job, and it's fixed. Resume + hint if picked up: get real stderr out of linuxdeploy directly (run + `linuxdeploy-x86_64.AppImage` by hand against the built `app` binary + outside of Tauri's bundler, or add `RUST_LOG=debug`/`--verbose` to + `tauri build` if supported) before assuming it's the same size issue. - [2026-08-18] Hermes agent-backend delegation (per-agent Ollama/NVIDIA/Gemini/Hermes provider picker, wiring 9 in-app agents to a separate live Hermes Agent From 50444db940d573a31e75c788268eab1bdfcd1e18 Mon Sep 17 00:00:00 2001 From: Agent Bot Date: Fri, 21 Aug 2026 23:28:06 -0400 Subject: [PATCH 2/2] chore: gitignore local Q&A E2E Test.md working notes Machine-local QA report pasted in from Slack (Viktor agent) - findings get triaged into DEFERRED_WORK.md / real issues, not tracked here. Co-Authored-By: Claude Sonnet 5 Claude-Session: https://claude.ai/code/session_01WexPHS5v6ZRnwzQGR4JG4A --- .gitignore | 5 +++++ 1 file changed, 5 insertions(+) diff --git a/.gitignore b/.gitignore index cea65242..c1f02abb 100644 --- a/.gitignore +++ b/.gitignore @@ -86,6 +86,11 @@ desktop.ini *credential* audits/private/ +# External/informal QA reports pasted in from other channels (e.g. Slack) — +# machine-local working notes, not a repo-governed doc; findings get triaged +# into DEFERRED_WORK.md / real issues instead of living here. +Q&A E2E Test.md + # Fetched at build time by scripts/fetch-ollama-runtime.mjs — never commit # vendored runtime binaries (see docs/DEPENDENCY_BUNDLING_PLAN.md). The # directory itself must stay tracked (src-tauri/vendor/ollama/.gitkeep) —