Skip to content

docs: add self-review checklist + Stop phase operating mode (A4+A5+W6)#904

Merged
s977043 merged 1 commit into
mainfrom
feat/agent-self-review-and-stop-mode
May 25, 2026
Merged

docs: add self-review checklist + Stop phase operating mode (A4+A5+W6)#904
s977043 merged 1 commit into
mainfrom
feat/agent-self-review-and-stop-mode

Conversation

@s977043

@s977043 s977043 commented May 25, 2026

Copy link
Copy Markdown
Owner

Summary

P0 process-hardening from the multi-perspective review (3 of 4 items). Companion to PR #902 which fixed the immediate defects; this codifies the patterns so the same defects don't recur.

A4. Codify-then-validate

Before merging a new rule, answer 3 questions: failure scenarios / reopen condition / gray zone. The Stop rule in #898 lacked answers to all three → #902 補修. This is the same-session example referenced.

A5. Code-gen self-review

Before merging workflow / script automation, verify: concurrency / default coupling / failure mode. These were the convergent gaps in #900#902.

W6. Stop phase operating mode table

Gray-zone matrix: status report (single OK / repeated NG), retrospective (推奨), tidy-ups (NG), retrospective-derived meta PR (reopen #1 — 新情報 で正当, ただし A4/A5 必須).

Files

  • AGENTS.md — new "Self-Review Checklist" section
  • docs/development/retrospectives/2026-05-21-25.md — operating mode table under existing Stop-on-Stop-Judge

Test plan

  • node scripts/fix-dashes.mjs --check clean
  • lint-staged pre-commit
  • CI green

S4 (community skill fixtures backfill, 2-3 skill slice) is the remaining P0 — separate PR.

P0 process-hardening from retrospective review (3 of 4):
- A4 Codify-then-validate: 3-question gate for new rules
- A5 Code-gen self-review: concurrency / default-coupling / failure-mode
- W6 Stop phase operating mode: explicit gray-zone table

S4 (community skill fixtures backfill) is the remaining P0; tracked separately
as it is 2-3 skills × 4 files.
@gemini-code-assist

Copy link
Copy Markdown
Contributor

Warning

You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again!

@vercel

vercel Bot commented May 25, 2026

Copy link
Copy Markdown

The latest updates on your projects. Learn more about Vercel for GitHub.

Project Deployment Actions Updated (UTC)
ai-review-kit Ready Ready Preview, Comment May 25, 2026 10:19pm

@github-actions

Copy link
Copy Markdown
Contributor

No changes to review compared to main.

@github-actions

Copy link
Copy Markdown
Contributor

PlanGate Review

  • decision: pass
  • plan status: skipped-by-label
  • exec status: skipped-by-label

PlanGate review decision: pass

severity count policy
critical 0 fail
major 0 fail-if-required / warn
minor 0 comment-only
info 0 skipped

ポリシー: critical=fail / major=fail-if-required (warn) / minor=comment-only / info=skipped — spec: pages/reference/cli-review-plan-spec.md

@s977043 s977043 merged commit c83e872 into main May 25, 2026
19 checks passed
@s977043 s977043 deleted the feat/agent-self-review-and-stop-mode branch May 25, 2026 22:31
s977043 added a commit that referenced this pull request May 26, 2026
…906) (#907)

Defect discovered during v0.57.0 release: kick workflow's GITHUB_TOKEN-authored
push cannot trigger downstream pull_request workflows by GitHub design
(no-recursion safety), so the release PR stays BLOCKED despite the kick
appearing successful.

Fix applies "Code-gen self-review" from AGENTS.md (#904):

- Failure mode (loud): added a verification step that polls check-runs for 90s
  and fails the workflow if no non-Vercel checks start. The silent success was
  the #906 failure pattern.
- Default coupling: prefer RELEASE_KICK_PAT secret, fall back to GITHUB_TOKEN
  with a warning and an immediate exit 1 in the verify step.
- Concurrency / auto-detect: preserved from #902.

Runbook updated with PAT setup instructions.

Local script (scripts/release-please-kick.sh) remains the unconditional
fallback — it pushes via user PAT (gh OAuth), bypassing the no-recursion limit.

Closes #906.
s977043 added a commit that referenced this pull request May 27, 2026
Per Codex slice judgment: investigate + design, defer impl. This doc resolves
the 3 open questions named in execution-context-contract.md replay section.

- Option A/B/C analysis (recommend B: additive debug.execution.snapshot)
- Drift handling (report, not block)
- 3-PR slice plan (pre / test / impl)
- Failing integration test spec

Codify-then-validate applied to the design itself (AGENTS.md from #904).
s977043 added a commit that referenced this pull request May 27, 2026
Per Codex slice judgment: investigate + design, defer impl. This doc resolves
the 3 open questions named in execution-context-contract.md replay section.

- Option A/B/C analysis (recommend B: additive debug.execution.snapshot)
- Drift handling (report, not block)
- 3-PR slice plan (pre / test / impl)
- Failing integration test spec

Codify-then-validate applied to the design itself (AGENTS.md from #904).
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant