원클릭으로
retro
Review the current session and propose improvements to CLAUDE.md, hooks, agents, and skills.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
메뉴
Review the current session and propose improvements to CLAUDE.md, hooks, agents, and skills.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
| name | retro |
| description | Review the current session and propose improvements to CLAUDE.md, hooks, agents, and skills. |
| user-invocable | true |
Analyze the current session's patterns and propose improvements to the Claude Code configuration (CLAUDE.md, hooks, agents, skills, rules).
Before launching the agent, compile a session summary to pass as context:
git log --oneline for recent commits)Format as a concise bullet list of session events.
Before launching the agent, gather quantitative metrics:
git log --oneline — count commits in this sessionInclude these metrics in the agent prompt as structured data.
Before launching the agent, sweep the session for review findings that were surfaced but never acted on, so they are not silently lost:
security-guidance PreToolUse warnings that were acknowledged but not fixed (the edit went through anyway)code-reviewer / silent-failure-hunter / security-review findings marked out-of-scope, deferred, or "later"ReportFindings items the user deferred, and any "I'll note this for follow-up" the session made but never wrote downFor each surviving finding, capture it into TODO.md (project memory, or ~/.claude/pending-cognify/TODO.md if no project context) with: what the finding was, which reviewer/hook surfaced it, why it was deflected, and a concrete first step to close it. Include these captured items in the agent prompt so the retrospective can decide whether any warrant a rule/hook change.
If the sweep finds nothing, note "no deflected findings this session" and continue.
Launch the session-retrospective agent in the background using the Agent tool:
When the agent returns, present its findings in two sections:
Section A: Patterns to Reinforce — workflows, rules, or agent invocations that worked well this session. These should be preserved and not accidentally removed in future config changes.
Section B: Improvement Proposals — numbered list where each proposal includes:
~/.claude/rules/foo.md, settings.json)Use AskUserQuestion to let the user select which proposals to implement (multiSelect).
For each approved proposal, implement the change. Follow existing patterns:
cookbooks/claude-code/files/hooks/, registered in settings.jsoncookbooks/claude-code/files/rules/ with appropriate frontmattercookbooks/claude-code/files/agents/SKILL.md in cookbooks/claude-code/files/skills/<name>/cookbooks/claude-code/files/CLAUDE.mdAfter implementation, sync deploy targets and commit.
config(~/.claude/todo/sources.yaml + sources.local.yaml)に列挙された外部 capture ソース — Slack saved(後で見る)/ 📌 リアクション / self-DM、Google Tasks、 Apple Reminders(remind --dump)、Claude Code transcript の deferral 発話、 カレンダー連携 Notion 議事録 — を sweep し、 TODO 形に正規化(完了条件の起草 + provenance 付与)して機密性 routing どおりの memory store(ai-memory / memory-local)に tags:["todo"] で書き込む毎日ループ skill。 explicit ソース(明示マーク)は自動保存 + 領収書 1 行、inferred ソース(推論抽出)は 候補提案のみ(承認後に保存)。dedup は provenance キー。結果は ~/.claude/todo/ledger.md に追記。設計の真実源は ~/.claude/docs/todo-management.md。 「todo collect」「TODO 集めて」「capture 集めて」「TODO 収集」でトリガー。 注: ソース側の状態は絶対に変更しない(Task の完了化・saved 解除等をしない)。 例外は連携ストアの Apple Reminders のみ(リンクトークン annotate と close 伝播 complete)。 仕事ソースの sink は memory-local 固定 — 個人 ai-memory へ流すのは egress 違反。
全 TODO ストア(各 repo の TODO.md、ai-memory / memory-local の tags:["todo"]、 federated な GitHub issues)を横断列挙し、項目ごとに 2 秒 probe で stale を検出して 済み / 前提消滅 / 継続 / 要判断 に分類する、TODO 管理パイプラインの週次棚卸しループ skill。 probe なしの即答列挙(list モード)も担当。結果は ~/.claude/todo/ledger.md に書き出す。 ストア設計・capture routing は ~/.claude/docs/todo-management.md が真実源。 デフォルトは dry-run(列挙 + probe + 分類提案のみ。forget も TODO.md 編集もしない)。 「todo reconcile」「TODO 棚卸し」「TODO 掃除」で dry-run、 「TODO 見せて」「open な TODO は」で list モード、「LIVE mode」明記で適用モード。 注: forget / TODO.md 編集は機械的証拠か明示承認がある場合のみ。TODO.md 編集は PR 経由(main 直 push 禁止)。issues と TODO.md は列挙するだけで複製しない。
Check the connection and authentication status of every MCP server and re-authenticate any that need it, one at a time. Use when the SessionStart MCP health hook reports "NEEDS-AUTH" / "MCP connectors need attention", when an mcp__* tool call fails with an auth or connection error, or when the user asks to check MCP status, reconnect MCP, "MCP 認証", "check mcp", "mcp 繋がってない", "認証されていない MCP を認証して".
self-heal observability ループの Layer 2-B(issue 解決)。shin1ohno/setup の open な self-heal ラベル issue を 1 件選び、fleet を調査して根本原因を特定し、修正する自律ループ。 allowlist 内の remediation class(既知サービスの再収束/cookbook drift 修正)は 調査→修正→PR→CI green→merge→auto-mitamae 適用→機能検証→issue close まで完全自律。 allowlist 外(新規設計・破壊的変更・auth/secret・原因不明・3 回失敗)は PR/診断を残して self-heal-needs-human を付け停止。merge は CI green 必須、破壊的操作禁止、1 回 1 issue。 owner(shin1ohno) が issue/PR に付けたコメント・更新は GO 承認として拾い、needs-human でも 再着手する(第三者のコメント・更新は無視。bot コメントは <!-- self-heal-bot --> で識別)。 「self-heal resolve」「self-heal issue を直す」「fleet alert を解決」でトリガー。 対は self-heal-create(ES 状態を issue 化する loop)。
Scaffold a new documentation-hub repository for a project: README.md, CLAUDE.md (AI agent instructions), .claude/rules/ discipline files, bundled skills/agents (interview, retro, writing, notion-cli, claude-docs-researcher, session-retrospective, sync-bundled-tools), .gitignore, and an initial git commit. Use when user asks to "set up a docs hub", "create a documentation repository", "新しいプロジェクトのドキュメントハブを作る", "ドキュメント管理リポジトリを scaffold", "init docs project", or starts a fresh repo intended to manage external doc links (Notion, Slack, etc.) rather than source code.
Progressive load testing and performance tuning for Docker services. Identifies hard constraints, finds breaking points, and proposes tuning within safe limits.