agentleFS
HsuanYuLeeHsuanYuLee/polaris/.claude/skills/review-pr/SKILL.md

review-pr

Skill5 starsChanged 5 days agoMITView on GitHub

Review someone else's PR as a code reviewer: read the diff and the repo's own rules, leave inline comments, and submit APPROVE or COMMENT.

185 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use diagnose-ci-failures, review-delta and 10 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • zerx-labSkill

    diagnose-ci-failures

    Diagnose CI failures for a PR using the GitHub CLI, extract error logs, and generate a plan to fix them. Use when the user asks to check CI status, pull CI issues, triage test failures, or investigate PR build failures.

    Also in 3 projects that use this

  • n24q02mSkill

    review-delta

    Review only changes since last commit using impact analysis. Token-efficient delta review with automatic blast-radius detection.

    Also in 3 projects that use this

  • GeiserXSkill

    docs-loop

    Use when the user asks to audit docs, check documentation for stale claims, or verify docs against the code.

    Also in 3 projects that use this

  • GeiserXSkill

    full-loop

    Drive a repository to done with sergio-loop, then keep improving it with refine-loop at a descending theta. Use only when the user explicitly invokes full-loop.

    Also in 3 projects that use this

  • n24q02mSkill

    build-graph

    Build or update the code review knowledge graph. Run this first to initialize, or let hooks keep it updated automatically.

    Also in 3 projects that use this

  • +7 more from GeiserX

The file

Files
SKILL.md

review-pr

以 reviewer 角色審查別人的單一 PR,依 repo rules / handbook / diff context 留 inline comments,並送出 GitHub review。

Contract

此 skill 只處理單一 PR review。多 PR discovery 與 batch orchestration 交給 review-inbox;修自己的 PR review comments 就直接修(那不需要重簽成功的定義)。

Reviewer stance:prioritize bugs、behavior regressions、security、type safety、project rule violations、missing tests。不要用 personal style preference 擋 merge。 reviewer-side 屬 read-only lane,但對 changes_requested、active unresolved comments、 awaiting_re_review、mergeable_ready 的語義必須與 author-side mutable lane 一致;不得自行重寫。 本 skill 可輸出 reviewer 結論(送哪一種見〈Flow〉第 6 步), 但不得把 reviewer 結論寫成 author-side stage authority;「可 merge / 可 release / 已完成」仍必須沿用 shared PR / workflow state,而不是 reviewer prose。

這支 skill 有沒有被叫對:evals/evals.json

上面那條界線(什麼走這支、什麼走 review-inbox / request-pr-review)不是只寫在散文裡, 它有一份具名的案例集:evals/evals.json,13 句真的會被打出來的話,7 句該觸發、6 句 不該。每一條帶著它為什麼在那裡(notes)。

它是給人讀的,不是給腳本跑的——這裡沒有 runner,也刻意不要有一支。它的用途是:改 frontmatter 的 description 之前先讀那 13 句,問「改完之後這 13 句的答案還一樣嗎」。負向 那 6 句尤其重要,因為觸發詞放寬的代價從來不出現在正向案例上。

什麼時候要更新它:這支 skill 被叫錯、或該叫沒叫到的那一刻——把那句原話補成第 14 條, 標好它該不該觸發。一句在真實對話裡走錯的話,比十句想像出來的案例有用。同一趟摩擦也記進 你手上那張單可以改的那部分(SKILL-UTILITY,見 driving-work-to-done),兩者不重複:那裡記 「這一趟它幫到還是擋到」,這裡記「這句話該路由到哪」。

Reference Loading

SituationLoad
Any runreview-pr-entry-fetch-flow.md, pr-input-resolver.md, workspace-config.yaml
Analysisreview-pr-analysis-flow.md, library-change-protocol.md as needed
Writing findings upreview-comment-form.md
Submit and notifyreview-pr-submit-flow.md, scripts/validate-language-policy.sh, external-write-gate.md, github-slack-user-mapping.md
Re-reviewreview-pr-rereview-learning-flow.md, review-lesson-extraction.md

Large PR 分批 review 可派 sub-agent。派工的 prompt 要帶哪些東西(含它回傳什麼的 Completion Envelope), 在 review-pr-analysis-flow.md〈Batch Review〉。Sub-agent 只做 analysis,不送出 review、不改檔。

Flow

  1. 從使用者輸入或 Slack context 解析 PR URL;找不到單一 PR 時停止或轉 review-inbox。
  2. 依 pr-input-resolver.md 解析 owner、repo、number、本地 project path;找不到本地 repo 時使用 remote read mode。
  3. 用 scripts/fetch-pr-info.sh <owner/repo> <pr_number> [--my-user <username>] 取得 metadata、 files、review strategy、existing reviews、approval state、re-review signal。它只放行 open 且非 draft 的 PR——已合併、已關閉、還在 draft 的一律拒絕並說出是哪一種狀態。 打 --help 問得到用法。
  4. 先讀那個 repo 自己的 CLAUDE.md、AGENTS.md 與 .claude/rules/,讀 PR base branch 上的版本:有就讀, 三份都沒有就在 review 摘要裡說出沒有;以其中一條規矩為依據的 finding 附上它的檔名與行號(使用者 2026-10-30:「各 repo 可能有自己的 claude.md 規則,施作或 review 前務必去讀過」)。 再讀 workspace handbook、PR description、changed files、diff、既有 review comments,建立去重清單。
  5. Review changed files;large PR 依 reference 分組派 sub-agent fan-out。
  6. 合併 findings,依 severity 決定 APPROVE 或 COMMENT——分級照舊,而 must-fix 送的是 COMMENT,不擋對方的分支。要擋要有使用者的原話,見 references/review-pr-submit-flow.md〈Review Action〉。
  7. Review body、inline comments、Slack notification 送出前跑 language gate。
  8. Submit GitHub review,查詢 approve status,輸出摘要。
  9. 若有 validated repo-specific pattern,依 standard-first rule 更新 handbook。
  10. Slack source 時回覆原始 thread。

Severity Boundary

must-fix 必須是可從 code / diff / rules 直接證明會造成 bug、安全風險、型別錯誤、 或違反關鍵規範。外部 API 行為、language/library behavior、或僅基於慣例的推論,在未驗證前 最多是 should-fix。

擋人的門檻是「這份 diff 讓系統變壞」,不是「我發現了一件真的事」。 這兩句常常同時成立 而答案相反:一顆 PR 把事情往好的方向推,同時它旁邊躺著一個真的問題——那個問題是真的,但它 不是這顆 PR 造成的,擋掉這顆 PR 也修不好它。

所以這顆 PR 一個 hunk 都沒動的檔案上的既有問題,不是 must-fix。它仍然是一個發現, 要有地方去:留一則 COMMENT,或開一張單。不要讓它沒有落點——一個沒有出口的發現,下一 次會回到 must-fix 那一格,因為那是唯一保證有人看的地方。

同一句話的另外兩個面:跨 repo 的上線順序、別人手上還沒做完的決定,這一份 diff 都改不了, 所以它們也不擋。

Write Rules

  • GitHub review、inline comments、Slack replies 都是 external write。
  • 使用 scripts/validate-language-policy.sh 或 external write gate 驗證 final text。
  • 不重複留言已由其他 reviewer 指出的同語意問題。
  • Suggested change 只在能精準替換 diff range 時使用。

下面這一行是機器讀的:往別人看得到的地方送文字的其他 skill,從這裡問出「送出去之前要過 哪一道檢查」,不各自寫死一條路徑。這道關卡住在這裡,因為 review 這件事本身就是對外寫入, 它是這支 skill 的原生需求,不是為了別人才存在的。

<!-- POLARIS-EXTERNAL-WRITE-GATE: bash .claude/skills/review-pr/scripts/polaris-external-write-gate.sh -->

Completion

輸出 PR、review result、must-fix / should-fix / nit counts、approve status、Slack notification status,以及 handbook updates if any。

<!-- PROSE-EXTERNAL-PATHS: docs-manager/ — 動手對象:那是 specs 站台自己的 repo,這支 skill 往它寫東西、讀它的結構,不是我們抄一份放著的知識 -->

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.