SKILL / SEED
公开中
review-pr
Review a pull request for correctness, regressions, and missing tests. Use when the user asks to review a PR or /review-pr.
GrokClaude CodeCursorCodex
一键装到本地 agent
和 SkillsMP 一样，先看源码再安装。Grok 用分享链接；Claude / Cursor / Codex 下载 zip 解到技能目录。
META
NAMEreview-pr
SLUGreview-pr
SOURCEseed
BYTES630
AI 怎么用
- 复制上面的 AI URL
- 粘贴到 Claude / ChatGPT / Grok / Cursor
- 再说你要做什么，AI 会先读 skill 再动手
SKILL.MD
AGENT READABLEname: review-pr
description: Review a pull request for correctness, regressions, and missing tests. Use when the user asks to review a PR or /review-pr.
Review PR
Review the current branch or a given PR as a careful reviewer.
Steps
- Identify the diff (
git diff main...HEADor the provided PR). - Restate the intended change in 2-3 sentences before commenting.
- Hunt for: broken existing behavior, missing tests, unsafe defaults, leaked secrets, and unclear names.
- Group findings by severity: blocker, should-fix, nit.
- End with a short verdict: approve, request changes, or ask a clarifying question.
FULL BUNDLE (1 files)
# Agent Skill Package: review-pr You are loading a published Agent Skill. Follow SKILL.md exactly. Supporting files from the original zip are inlined below. When SKILL.md says to read `references/...` or `scripts/...`, use the matching FILE section here — do not say the file is missing. Canonical URL: https://skill.hk/s/review-pr.md Human page: https://skill.hk/s/review-pr Files (1): - SKILL.md ======================================================================== FILE: SKILL.md ======================================================================== --- name: review-pr description: Review a pull request for correctness, regressions, and missing tests. Use when the user asks to review a PR or /review-pr. --- # Review PR Review the current branch or a given PR as a careful reviewer. ## Steps 1. Identify the diff (`git diff main...HEAD` or the provided PR). 2. Restate the intended change in 2-3 sentences before commenting. 3. Hunt for: broken existing behavior, missing tests, unsafe defaults, leaked secrets, and unclear names. 4. Group findings by severity: blocker, should-fix, nit. 5. End with a short verdict: approve, request changes, or ask a clarifying question.