Repository navigation
Conversation
Design Review (Fable 5, fork) — ✅ PASSDesign-level review of Design-Verdict: PASS Verify the reapplied single commit matches the reviewed content of #14043, since this PR inherits that review rather than restating it. All index claims checked against base: Autopilot's stage loop, prompt, and spec are absent; [DESIGN-REVIEWED] 84aa116 Description read[DESCRIPTION-READ] c83873d96fc72ab98b15e72ddd4f8cf970d3f7275c20f9f73a8efc5ee0a7d2ff That sha256 names everything this verdict read: the PR description, plus the 1 evidence file(s) the workflow collected from it and handed this review. It covers that evidence because the media strip replaces every attachment URL with the same placeholder, so an edit swapping one attachment for another would leave the description's captured text identical while what this review judged changed. Recompute it with |
GPT 5.6 Review (fork) — ✅ no blocking findingsReviewed Review detailsNo findings. |
Opus 5 Review (fork) — ✅ no blocking findingsReviewed Review detailsThe retirement RFC is flipped to FINDING — docs/request-for-change/rfc-retire-chat-autopilot.md:3 — [OPUS-REVIEWED] 84aa116 |
|
Superseded by #16146, rebuilt from current main as one commit with the Opus partial-status correction and current RFC index preserved. |
Problem / Motivation
Goal: Preserve the Autopilot timeout incident evidence while recording that the stage-budget proposal was superseded before acceptance.
Six distinct 30-minute stage timeouts occurred in nine days. Two native ACP records proved that work continued after the dashboard said the stage had stopped. Since the original RFC opened, backend removal #15362 and frontend removal #15359 have both merged, so the stage-budget decision no longer applies.
This replaces #14043. Its conflict-resolution merge left four commits, while repository hygiene permits at most two. The published branch is preserved; this PR reapplies the five-file documentation result as one commit on current
main.Why it matters
The incident record should remain reviewable without misleading maintainers into approving behavior for a deleted feature. Issue #14058 owns the narrower lifecycle defect that remains at the ordinary chat-turn ceiling:
_bounded_turncancels_run_chatwithout first sending nativesession/cancel.Not a goal
What changed (motivation → approach → change)
The superseded stage-budget RFC and sanitized evidence appendix are applied to current
mainas one commit. They now link issue #14058 as the owner of the remaining cancellation defect. The retirement RFC and index record both backend and frontend removal as implemented. The assets README explicitly indexes Markdown evidence only, avoiding an incomplete inventory of unrelated binary assets. Six deduplicated incidents, two confirmed dashboard/native continuations, provenance hashes, and reproduction details remain preserved without raw transcripts.Backwards compatibility
Compatible: documentation and audit metadata only. No runtime, configuration, stored session, or public interface changes.
Tests
python3.12 scripts/docs_lint.py --test— passed../scripts/docs-lint.sh— passed across 391 Markdown files.git diff --check origin/main...HEAD— passed.Manual verification
main."orchestrator"is a retired mode and restores as plain chat.mainstill has the ordinary-turn cancellation path recorded in issue dashboard: Stop is a no-op after an Autopilot stage-timeout wedge -- lanes show [Stopped] but the session stays "In progress" and never escalates to the documented hard-kill / Force-reset #14058.Screenshots / video
N/A — this PR changes documents only. The source screenshots contained unrelated session content; the appendix preserves only sanitized lifecycle facts.
Related Issues
Checklist
main.Contribution License Agreement