メニュー
Lifecycle contract for opening PRs, driving them green, and fixing CI. Triggers on: opening a PR, driving a PR to green, fixing CI on a PR, /green, /es, /er, /advice on a PR.
デフォルトでは、最初にソースを確認する Prompt が選択されています。直接コマンドに切り替えるか、ローカルコピーをダウンロードすることもできます。
インストールを決める前に、SKILL.md と SkillsMP に表示されている付属ファイルをお読みください。
SOC 職業分類に基づく
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
直接コマンドでは確認用 Prompt が省略されます。実行前にソースを確認してください。
npx skills add https://github.com/jleechanorg/claude-commands --skill draft-first-prコマンドは1行のまま表示されます。コピー前に横へスクロールして全体を確認してください。
ローカルで確認しますか？SkillsMP が現在取得できるファイルをダウンロードできます。
SKILL.md を表示中
Token-efficient second opinion slash command /advice. Extracts decision point + artifact (≤150 lines), then fans out in parallel: (1) Opus subagent reviewer with fallback chain codex→agy→cursor, (2) /research on the decision topic, (3) /secondo multi-model opinion. Use instead of advisor() which ships the full conversation uncached.
Use this skill when working in repositories managed by Agent Orchestrator or when the user asks how to use `ao` properly. Covers the default AO workflow: bootstrap with `ao start`, dispatch work with `ao spawn`, inspect progress with `ao status` or `ao session ls`, steer sessions with `ao send`, and recover or clean up sessions safely. Includes strict parameter fidelity, pre-spawn cap cleanup, quota-wall fallback, and post-spawn verification.
Generate a full agento PR status report — draft readiness, canonical /green, zero-touch rate, inline display, and Slack summary.
Every new PR is opened as DRAFT (gh pr create --draft) and stays draft until quality is proven. Flipping to ready-for-review is a deliberate, gated action — not a side effect of pushing code.
This is the full state machine — every other file (pr-green-definition, /green, CLAUDE.md, AGENTS.md) points here rather than restating it:
DRAFT
→ /es PASS @ SHA
→ /er PASS @ SHA
→ /advice APPROVED @ SHA
→ mark ready (gh pr ready <N>)
→ /green: CI green + no merge conflicts, BOTH verified @ current HEAD SHA
→ separate merge authorization: explicit human "MERGE APPROVED" (case-insensitive)
in the most recent live user message
Each arrow is a gate, not a formality — do not skip ahead, and do not treat an earlier gate's pass as still valid once HEAD has moved (see "SHA-binding rule" below).
While a PR is draft, run these in sequence — do not skip ahead:
/es — evidence bundle passes (real evidence per ~/.claude/skills/evidence-standards/SKILL.md + repo-specific extensions), verified at the PR's current HEAD SHA./er — evidence review verdict is PASS (not PARTIAL/FAIL/INCONCLUSIVE), verified at the same current HEAD SHA — re-run if /es was earned at an older SHA./advice — second-opinion approval on the change itself (binary APPROVED at <SHA> / NOT APPROVED at <SHA> — see ~/.claude/skills/advice/SKILL.md), bound to the same current HEAD SHA.Only after all three pass at the same current SHA: flip the PR from draft to ready-for-review (gh pr ready <N>).
/es, /er, /advice, and /green are each earned at a specific commit SHA — none of them carry forward across a HEAD move. If a new commit lands after a verdict (a nit fix, a rebase, a CI-requested change), that verdict is STALE and must be re-earned — or explicitly re-affirmed at the new SHA — before it counts toward the next gate in the chain. This applies uniformly:
/es PASS does not justify running /er against it./er PASS does not justify running /advice against it./advice APPROVED does not justify marking the PR ready./green does not justify reporting merge-readiness.Before trusting any prior verdict, compare the SHA it was stamped with against the PR's live head: gh pr view <N> --json headRefOid --jq '.headRefOid'. Any mismatch means re-run that gate — never carry a verdict forward on memory.
/greenOnce ready-for-review, drive /green per ~/.claude/skills/pr-green-definition/SKILL.md (2-gate CI+mergeable definition, SHA-bound) — do not restate the gate mechanics here.
Do NOT burn CI cycles chasing green on an unproven draft — that's what starved capacity before this policy. Get through the 3 draft gates first, then spend CI budget on the 2-gate /green loop.
If CI is running more than ~10 minutes past its normal runtime for the check in question: run the equivalent tests locally NOW and post proof labeled "local run — CI still pending" (exact command, output, timestamp, git SHA). Do not treat polling/waiting as the primary strategy; Gate 1 remains pending until remote CI itself resolves successfully.
Long-open PRs that skipped straight to chasing CI green (e.g. the level-up auto-PR class) were starved by CI contention — CI capacity is a shared resource, not a private queue. The ci-value-audit-v2 findings (green-gate workflow: 341 hr/wk consumed, 50.7% cancel rate pre-#8637) show that driving unproven work through full CI repeatedly is the dominant cost driver. Gating quality (/es//er//advice) in draft — before CI spend — front-loads correctness and back-loads CI cost only onto PRs already known-good.
Neither draft-phase gate above nor /green includes CodeRabbit or Bugbot approval. CodeRabbit/Bugbot: optional advisory reviewers — read their feedback, take what's useful; never a gate, never a wait, at any phase.
Merge still requires the literal phrase MERGE APPROVED (case-insensitive) from the human in the most recent live message, scoped per-repo per existing policy (e.g. $GITHUB_REPOSITORY). Nothing in this skill changes that gate.