メニュー
This skill should be used during /maigo:review when fetching or reusing PR context (title / body / diff / CI status / linked issues), caching the first fetch into review-rubric.md so subsequent re-review rounds skip re-fetching.
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。
SOC 職業分類に基づく
This skill should be used when assembling git commits — deciding what to stage, how to scope a bash/agent command's working directory, whether to amend an unreleased commit or stack a new one, and how to size a diff against the correct baseline. Applies to the commit-assembly step of /maigo:go, /maigo:quick, /maigo:team, and /maigo:address-comments, and to the orchestrator generally.
This skill should be used when the maigo orchestrator, running in the Claude Code harness, decides whether to delegate work to a subagent instead of doing it on the main thread, needs to keep a long task's acceptance criteria from drifting after context compaction, or needs to keep verification independent from whoever produced the work. Scope note: 只在 Claude Code harness 下才適用——這裡的 orchestrator context 會被計費、可能被壓縮、且可 spawn subagent 分攤負載；沒有 subagent 能力的 harness 不適用此 skill。
This skill should be used by all maigo agents at startup, before beginning work, to load relevant cross-project memory entries with relevance-based ordering and a 10-entry cap. Consumers: Raana, Tomori, Soyo, the orchestrator itself, and any future agent that reads ~/.config/maigo/memory/.
This skill should be used by the maigo orchestrator on every /maigo command, alongside narration, to govern the conversational conduct of the main dialogue — AskUserQuestion widget discipline and Taiwanese Mandarin word-choice norms.
This skill should be used when performing code review on a diff (whether implemented by another agent or an external PR), enforcing a strict reviewer stance, applying a mandatory 9-item checklist, demanding evidence, and giving specific改法 instead of vague critique.
This skill should be used when orchestrating the full MyGO!!!!! teammate flow — Raana explores, Tomori plans, Anon implements, Soyo reviews, Taki validates. Applies to /maigo:go and /maigo:team (both sequential and parallel variants).
| name | pr-context-cache |
| description | This skill should be used during /maigo:review when fetching or reusing PR context (title / body / diff / CI status / linked issues), caching the first fetch into review-rubric.md so subsequent re-review rounds skip re-fetching. |
Owner Agent: Raana
Consumers: /maigo:review step 1
/maigo:review 的第一步是 Raana 抓 PR context。「re-review」（同一個 PR 改完再跑一次）
時這些資料幾乎沒變——重抓只是浪費時間。第一次 fetch 後 cache 到
.maigo/review-rubric.md 開頭的機讀區段，re-review 偵測同 source 且 diff sha 未變
→ 直接還原，跳過全部 gh / git 重抓。
機械流程由 script 代勞（cache 偵測 / 驗證 / fetch / truncate / 寫檔一條龍）：
python3 "${CLAUDE_PLUGIN_ROOT:-.}/scripts/pr_context_cache.py" <source> \
[--rubric .maigo/review-rubric.md] [--base main]
<source>：GitHub PR URL / PR 編號（需要 gh CLI）、本地 branch 名、或 commit rangepython3 scripts/pr_context_cache.py 即可stdout 第一行是 cache_hit: true|false，其後是 cache 區段全文——
含 Source / PR number / Title / Body（截 500 行）/ Linked issues / CI status /
Diff stat / Review threads（inline review thread，含 resolve 狀態）/ Review
summaries（gh pr view --json reviews）/ Conversation comments（gh pr view --json comments） / Diff sha / Full diff（截 2000 行）。
Review threads / summaries / comments 只在 <source> 是 PR 時抓（branch / range
diff 沒有對應的 GitHub review thread 可抓）。未解決（[OPEN]）的 thread 在輸出
裡會被特別標出——下結論前先檢查這些 thread 對照目前 diff 是否已經處理，
避免漏看 reviewer 留下但尚未收斂的架構意見（真實事故：只抓 diff + PR body +
reviewDecision，漏看某 PR 上一位 reviewer 對 isinstance-based type-switch 設計
留下的 OPEN thread，直到使用者問「有沒有看 TP 說了什麼」才補回）。
<!-- pr-context-cache:start v1 --> … <!-- pr-context-cache:end --> 區段
（無檔案 → 建立；無區段 → prepend；有舊區段 → 整段取代）script 跑不起來（找不到路徑、無 gh CLI、git 失敗 → exit 1 + stderr）→
Raana 回退手動抓：依 /maigo:review step 1 列的 gh pr view / gh pr diff / gh pr checks（或 git diff / git log）指令直接 fetch，不寫 cache。
review-rubric.md 其餘內容（rubric 本身由 Tomori 撰寫）/maigo:describe-pr 不走這個 skill）