菜单
Use when reviewing pull requests, diffs, patches, or repository files to identify merge-blocking issues, important risks, and evidence gaps without directly fixing the reviewed artifact.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Use when handling local Git CLI daily operations such as checking status or diff, staging files, creating conventional commits, restoring local changes, reverting commits, or resolving merge/rebase/cherry-pick conflicts.
Use when explaining code, answering technical questions, comparing approaches, or providing repository-grounded analysis without entering an implementation or remediation workflow.
Use when evaluating system architecture, platform design, or high-level technical structure to identify production risks, evidence gaps, and prioritized remediation without directly redesigning or implementing the system.
Use when improving internal code structure without intentionally changing externally observable behavior, such as module splits, simplification, dependency cleanup, or maintainability and performance-oriented restructuring.
Use when feature ideas, issues, stakeholder notes, or partial specs need to be reviewed and turned into a concise, testable PRD with clear scope, user stories, acceptance criteria, and explicit open decisions.
Use when a PRD, feature spec, or acceptance criteria already exist and need to be turned into a concise test plan and traceable test cases with clear priorities, coverage, and execution focus.
| name | review-code |
| description | Use when reviewing pull requests, diffs, patches, or repository files to identify merge-blocking issues, important risks, and evidence gaps without directly fixing the reviewed artifact. |
| argument-hint | [PR-url | diff | patch | file...] |
review-code 是纯评审 lane。它的职责不是直接修代码，而是基于最小必要证据识别阻断问题、重要风险与证据缺口，并给出清晰、可执行、可追溯的评审结论。
它默认是代码优先的通用评审 skill，可用于未来安装到用户全局家目录后的普通仓库场景；当评审对象命中脚本、文档、规则、skills、prompt 或其他 agent 行为资产时，再追加专项检查视角。
适用于：
以下情况不要直接进入本 skill：
遇到这些情况时，优先转交：
superpowers:receiving-code-reviewsuperpowers:requesting-code-reviewskills/fix-bug/SKILL.mdskills/develop-feature/SKILL.mdskills/refactor/SKILL.mdskills/answer/SKILL.mdskills/architecture-review/SKILL.md评审对象 / 范围 / 评审目标 / 不做 / 当前证据。review-code 退化成代修、重写方案或顺手实现的 lane。Align
对象 / 范围 / 目标 / 不做 / 证据PR / diff / patch / fileClassify
code / script / doc / mixedpre-merge / post-change audit / targeted file reviewlow / standard / high-riskrules / skills / prompts / agent behavior 时，显式追加对应视角Gather Evidence
evidence gapReview
目标是否达成 / 实现是否正确 / 验证是否充分 / 交付是否可用code 默认检查：正确性、安全、边界条件、回归风险、复杂度、可维护性script 默认检查：参数与路径安全、危险操作保护、失败退出、恢复路径、幂等性doc 默认检查：事实一致性、结构清晰度、步骤可执行性、术语一致性、过时信息Decide
Review Action: Approve | Comment | Request ChangesQuality Result: PASS | CONDITIONAL PASS | FAILApprove + PASSComment + CONDITIONAL PASSRequest Changes + FAILReport
Findings -> Open Questions -> Review Action -> Quality Result -> Evidence -> Risks / Gaps -> Done / Partial / SkippedCritical
Important
Suggestion
每条 finding 至少应包含：
LocationSeverityIssueWhy it mattersEvidenceRecommended fixReview Action 面向协作动作：
Approve
Comment
Request Changes
Quality Result 面向治理判定：
PASS
MUST 已满足，且证据与结论一致CONDITIONAL PASS
FAIL
若要偏离默认三种组合，必须显式说明原因、牺牲项与残余风险。
按需读取配套规则与参考文件：
rules/deliverable-quality-gate.md
PASS / CONDITIONAL PASS / FAIL 判定时读取rules/output-style.md
rules/roles.md
skills/review-code/references/review-check-matrix.md
skills/review-code/references/review-verdict-examples-and-anti-patterns.md
不要把所有规则与参考文件一次性压进上下文。
出现以下信号时，说明评审正在退化：
对用户的交付至少应包含：
FindingsOpen QuestionsReview ActionQuality ResultEvidenceRisks / GapsDone / Partial / Skipped输出 findings 时，默认按严重度排序；若改动范围很小，默认最多先给 Top 5 个最高价值问题。
若未发现阻断问题，必须明确写出：未发现阻断问题，并补充证据边界、测试缺口或残余风险。
若当前路径没有更优替代方案，应明确写出：当前路径合理，暂无更优替代建议。
rules/output-style.md 与 rules/deliverable-quality-gate.md