code-review

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Code Review

代码审查

Overview

概述

Comprehensive code review against the original plan, coding standards, and learned project patterns. This skill dispatches a dedicated code-reviewer agent for thorough analysis, ensuring every change is evidence-based, plan-aligned, and convention-aware before it reaches the main branch.

Announce at start: "I'm using the code-review skill to review the implementation."

对照原始计划、编码标准和已沉淀的项目模式开展全面代码审查。本技能会调度专属的code-reviewer agent执行深度分析，确保所有变更在并入主分支前均有据可依、符合计划且遵循约定规范。

启动时声明： "我正在使用code-review技能评审本次实现。"

Phase 1: Gather Context

阶段1：收集上下文

Goal: Identify what changed, what the plan required, and what conventions apply.

目标： 明确变更内容、计划要求和适用规范。

Actions

执行动作

Retrieve the changes to review:

bash

git diff HEAD~N..HEAD          # or specific commit range
git log --oneline HEAD~N..HEAD # what was done

Locate the plan document:

bash

ls docs/plans/*.md | tail -1

Load project conventions from
```
memory/learned-patterns.md
```
Identify:
- What files were changed
- What the plan/spec required
- What conventions apply

拉取待评审的变更内容：

bash

git diff HEAD~N..HEAD          # 或指定提交范围
git log --oneline HEAD~N..HEAD # 查看提交说明

查找计划文档：

bash

ls docs/plans/*.md | tail -1

从
```
memory/learned-patterns.md
```
加载项目约定规范
明确以下信息：
- 哪些文件发生了变更
- 计划/需求文档的要求是什么
- 哪些约定规范适用于本次变更

STOP — Do NOT proceed to Phase 2 until:

暂停 — 满足以下条件前请勿进入阶段2：

All changed files are identified
The plan or spec requirements are loaded
Relevant conventions from memory are loaded
You can state what was supposed to be built

已识别所有变更文件
已加载计划或需求要求
已加载知识库中相关的约定规范
你能清晰说明本次需要交付的功能内容

Phase 2: Dispatch Code Reviewer

阶段2：调度代码审查者

Goal: Send structured review request to the

code-reviewer

agent.

目标： 向

code-reviewer

agent发送结构化的评审请求。

Review Prompt Template

评审提示词模板

Review the following changes against:
1. Plan: [plan document or requirements]
2. Conventions: [learned patterns from memory]
3. Standards: [CLAUDE.md rules]

Changes:
[git diff output or file list]

Check for:
- Plan alignment (did we build what was specified?)
- Code quality (DRY, YAGNI, naming, structure)
- Error handling (edge cases, failure modes)
- Security (injection, XSS, auth issues)
- Test coverage (are changes tested?)
- Performance (obvious bottlenecks)
- Documentation (are public APIs documented?)

对照以下要求评审变更内容：
1. 计划：[计划文档或需求内容]
2. 约定：[知识库中的沉淀模式]
3. 标准：[CLAUDE.md规则]

变更内容：
[git diff输出或变更文件列表]

检查维度：
- 计划对齐度：是否实现了需求指定的功能？
- 代码质量：是否符合DRY、YAGNI原则，命名、结构是否规范？
- 错误处理：是否覆盖了边缘场景、故障模式？
- 安全性：是否存在注入、XSS、权限相关问题？
- 测试覆盖：变更内容是否有对应的测试？
- 性能：是否存在明显的性能瓶颈？
- 文档：公共API是否有对应的文档说明？

STOP — Do NOT proceed to Phase 3 until:

暂停 — 满足以下条件前请勿进入阶段3：

Review request has been dispatched
Reviewer agent has returned findings

已发送评审请求
评审agent已返回评审结果

Phase 3: Categorize and Resolve Issues

阶段3：问题分类与修复

Goal: Classify findings and fix all Critical issues.

目标： 对评审发现的问题分类，修复所有严重问题。

Issue Categorization Table

问题分类表

Category	Definition	Action Required
Critical	Bugs, security issues, data loss risk, plan violations	Must fix before merge
Important	Code quality, missing tests, convention violations	Should fix before merge
Suggestions	Style, naming, minor improvements	Nice to have, fix if time allows

分类	定义	要求动作
严重（Critical）	Bug、安全问题、数据丢失风险、违反计划要求	合并前必须修复
重要（Important）	代码质量问题、缺失测试、违反约定规范	合并前应当修复
建议（Suggestions）	风格、命名、微小优化	非必需，时间允许时修复

Fix Loop

修复循环

For Critical and Important issues:

Fix the issue
Run tests to verify the fix
Re-dispatch code-reviewer agent for the specific fix
Repeat until no Critical issues remain

针对严重和重要问题：

修复问题
运行测试验证修复效果
针对修复内容重新调度code-reviewer agent评审
重复上述步骤直到无严重问题残留

STOP — Do NOT proceed to Phase 4 until:

暂停 — 满足以下条件前请勿进入阶段4：

All Critical issues are resolved
All Important issues are resolved or explicitly deferred with justification
Test suite passes after all fixes

所有严重问题已解决
所有重要问题已解决或已明确说明延期理由
所有修复完成后测试集全部通过

Phase 4: Self-Learning Integration

阶段4：自学习集成

Goal: Persist patterns discovered during review for future sessions.

目标： 存储评审过程中发现的新模式，供后续会话使用。

Actions

执行动作

If new patterns were identified, update
```
memory/learned-patterns.md
```
If a common mistake was found, note it for future reference
If the plan needed adjustment, update
```
memory/decisions-log.md
```

如果识别到新模式，更新
```
memory/learned-patterns.md
```
如果发现了共性错误，记录下来供后续参考
如果计划需要调整，更新
```
memory/decisions-log.md
```

Review Output Format

评审输出格式

markdown

undefined

markdown

undefined

Code Review Summary

代码审查总结

Scope: [files/components reviewed] Plan alignment: [aligned / minor deviations / major deviations]

评审范围： [评审的文件/组件] 计划对齐度： [完全对齐/少量偏差/严重偏差]

Critical Issues (N)

严重问题（N个）

[Issue title] —
```
file:line
```
Problem: [description] Fix: [specific recommendation]

[问题标题] —
```
文件:行号
```
问题：[问题描述] 修复方案：[具体修复建议]

Important Issues (N)

重要问题（N个）

[Issue title] —
```
file:line
```
Problem: [description] Fix: [specific recommendation]

[问题标题] —
```
文件:行号
```
问题：[问题描述] 修复方案：[具体修复建议]

Suggestions (N)

优化建议（N个）

[Suggestion] —
```
file:line
```

[建议内容] —
```
文件:行号
```

What Was Done Well

做得好的地方

[Positive observations]

---

[正面评价内容]

---

Decision Table: Review Depth

决策表：评审深度

Change Type	Review Depth	Reviewer
New feature (>100 lines)	Full review: plan alignment + quality + security + tests	code-reviewer agent
Bug fix (<50 lines)	Focused review: regression test + root cause + fix correctness	code-reviewer agent
Refactor (no behavior change)	Behavior preservation: all tests pass + no regressions	code-reviewer agent
Config/infra change	Security + correctness: no secrets exposed, valid syntax	code-reviewer agent
Documentation only	Accuracy + completeness: matches current code behavior	Inline review

变更类型	评审深度	评审方
新功能（>100行）	全量评审：计划对齐+质量+安全+测试	code-reviewer agent
Bug修复（<50行）	聚焦评审：回归测试+根因+修复正确性	code-reviewer agent
重构（无行为变更）	行为一致性：所有测试通过+无回归	code-reviewer agent
配置/基础设施变更	安全+正确性：无密钥泄露、语法有效	code-reviewer agent
仅文档变更	准确性+完整性：与当前代码行为一致	直接评审

Anti-Patterns / Common Mistakes

反模式/常见错误

Anti-Pattern	Why It Is Wrong	Correct Approach
Skipping review for "small fixes"	Small changes cause production outages	Review everything
Reviewing without the plan	Cannot verify correctness without requirements	Always load the plan first
Fixing issues without re-running tests	Fixes can introduce new bugs	Run full test suite after every fix
Generic feedback ("looks good")	Not actionable, misses real issues	Cite specific code lines with fix recommendations
Reviewing your own code alone	Author blindness misses defects	Always dispatch code-reviewer agent
Deferring Critical issues	Critical issues become production incidents	Must fix before merge, no exceptions

反模式	错误原因	正确做法
认为是“小修复”跳过评审	小变更也可能导致生产故障	所有变更都要评审
不对照计划开展评审	没有需求依据无法验证正确性	始终先加载计划再评审
修复问题后不重新运行测试	修复可能引入新Bug	每次修复后运行完整测试集
泛泛的反馈（“看起来不错”）	无法落地，会遗漏真实问题	引用具体代码行并给出修复建议
仅自己评审自己的代码	作者盲区会遗漏缺陷	始终调度code-reviewer agent评审
延期处理严重问题	严重问题会演变为生产事故	合并前必须修复，无例外

Rationalizations — STOP If You Think These

常见借口 — 如果你有这些想法请立刻停止

Excuse	Reality
"It's just a typo fix"	Typo fixes can break APIs. Review it.
"I'm confident in this code"	Confidence does not equal correctness. Review it.
"The tests pass"	Tests can miss bugs. Review it.
"It's just styling/formatting"	Style changes can introduce bugs. Review it.
"Nobody will notice"	That is exactly when bugs ship. Review it.
"I'll review it later"	Later never comes. Review it now.
"The deadline is tight"	Shipping bugs costs more than reviewing. Review it.

借口	现实
“只是改了个 typo”	Typo修复也可能破坏API，必须评审
“我对这段代码很有信心”	信心不等于正确性，必须评审
“测试都通过了”	测试也可能遗漏Bug，必须评审
“只是样式/格式调整”	样式变更也可能引入Bug，必须评审
“没人会注意到”	这正是Bug发布的原因，必须评审
“我之后再评审”	之后永远不会来，现在就评审
“ deadline 很紧”	线上Bug的修复成本远高于评审成本，必须评审

Subagent Dispatch Opportunities

子代理调度场景

Task Pattern	Dispatch To	When
Reviewing multiple independent files/modules	`Agent` tool with `subagent_type="Explore"`	When review scope spans multiple unrelated modules
Security-focused review pass	`Agent` tool invoking `security-review` skill	When changes touch auth, input handling, or external APIs
Performance impact assessment	`Agent` tool invoking `performance-optimization` skill	When changes affect hot paths or data-heavy operations

Follow the

dispatching-parallel-agents

skill protocol when dispatching.

任务模式	调度对象	触发时机
评审多个独立文件/模块	`Agent` 工具， `subagent_type="Explore"`	评审范围覆盖多个无关模块时
聚焦安全的评审环节	调用 `security-review` 技能的 `Agent` 工具	变更涉及权限、输入处理或外部API时
性能影响评估	调用 `performance-optimization` 技能的 `Agent` 工具	变更影响热点路径或大数据量操作时

调度时遵循

dispatching-parallel-agents

技能协议。

Integration Points

集成点

Skill	Relationship
`planning`	Review checks implementation against the approved plan
`test-driven-development`	Review verifies test coverage and TDD compliance
`verification-before-completion`	Review is a prerequisite for verification
`self-learning`	Review findings feed into learned patterns
`acceptance-testing`	Review checks that acceptance tests exist for all criteria
`systematic-debugging`	If review reveals a bug, switch to debugging skill
`security-review`	Security findings during review trigger deeper security analysis

技能	关联关系
`planning`	评审对照已审批的计划校验实现
`test-driven-development`	评审验证测试覆盖度和TDD合规性
`verification-before-completion`	评审是验证环节的前置条件
`self-learning`	评审结果沉淀为已学习模式
`acceptance-testing`	评审检查所有验收标准是否有对应的验收测试
`systematic-debugging`	如果评审发现Bug，切换到调试技能
`security-review`	评审中发现的安全问题会触发深度安全分析

Iron Law

铁律

┌─────────────────────────────────────────────────────────────────┐
│  HARD-GATE: NO MERGE WITHOUT REVIEW                            │
│                                                                 │
│  Every change gets reviewed. No exceptions for "small fixes"   │
│  or "obvious changes." If you are about to merge without       │
│  review, STOP immediately.                                     │
└─────────────────────────────────────────────────────────────────┘

┌─────────────────────────────────────────────────────────────────┐
│  硬性门禁：无评审不合并                                          │
│                                                                 │
│  所有变更都必须评审。即便是“小修复”或“明显变更”也不例外。如果你  │
│  打算不经评审就合并，请立刻停止。                                 │
└─────────────────────────────────────────────────────────────────┘

Skill Type

技能类型

RIGID — The four-phase process is mandatory. Every change must be reviewed by the code-reviewer agent. No merge without review. No exceptions.

严格执行（RIGID） — 四阶段流程为强制要求。所有变更必须经过code-reviewer agent评审，无评审不合并，无任何例外。