Code Review / 代码审查与质量

Multi-dimensional code review skill for AI agents — five axes, severity labels, and an approval standard. / 面向 AI 智能体的多维代码审查技能——五维审查、严重程度标签、明确的通过标准。

English | 中文说明

What it is

An AI agent skill that enforces disciplined, multi-dimensional review of every change before merge. No change ships unreviewed. It evaluates code across five axes — correctness, readability, architecture, security, and performance — and outputs a structured review report with severity-labeled findings so authors know exactly what blocks the merge and what is optional.

Key features

• Five-axis review / 五维审查 — correctness, readability & simplicity, architecture, security, performance
• Severity labels / 严重程度标签 — Critical / Required / Optional / Nit / FYI, so blocking issues never hide behind cosmetic nits
• Approval standard / 通过标准 — approve when the change definitively improves overall code health, even if imperfect
• Change sizing / 变更规模 — guidance on reviewable change sizes (~100 / ~300 / ~1000 lines) and four splitting strategies (stack, by file group, horizontal, vertical)
• Structural remedies / 结构修复 — every structural problem comes with a named refactoring move, not just a complaint
• Reference checklists / 深度检查表 — OWASP Top 10 security checklist and a deep performance checklist (N+1 queries, network waterfalls, memory, frontend, caching, concurrency)
• Anti-rationalization table / 反借口表 — calls out "it works, that's good enough", "AI-generated code is probably fine", and other classic review-evading excuses

How it works

Review runs in five steps: understand context → review tests first → review implementation → categorize findings by severity → verify the author's verification story. Output follows a fixed report template ending in an Approve / Request changes verdict.

Works with any diff source: git diff, GitHub PR, GitLab MR, or pasted code. No tool dependencies.

Usage

Drop the folder into your agent's skills directory. The agent triggers on phrases like "review this", "code review", "审查代码", "审查这个 PR".

Repository structure

├── SKILL.md                            # Main skill instructions / 主指令
└── references/
    ├── security-checklist.md           # OWASP Top 10 checklist / 安全检查表
    └── performance-checklist.md         # Performance deep dive / 性能检查表

License

MIT

---

中文说明

这是什么

一个 AI 智能体技能：合并前对任何变更做多维度审查，没有例外。从正确性、可读性、架构、安全、性能五个维度评估代码，输出带严重程度标签的结构化审查报告，让作者清楚知道哪些是阻塞项、哪些可选。

核心特性

• 五维审查：正确性、可读性与简洁性、架构、安全、性能
• 严重程度标签：严重 / 必改 / 建议 / 细节 / 备忘，真问题不会被格式细节淹没
• 通过标准：变更确实提升整体代码健康度即通过，不追求完美
• 变更规模：~100 行好审、~300 行可接受、~1000 行必须拆分，附四种拆分策略
• 结构修复：每个结构问题都配一个命名重构动作，不只是指出问题
• 参考检查表：OWASP Top 10 安全检查表 + 性能深入检查表（N+1 查询、网络瀑布流、内存、前端、缓存、并发）
• 反借口表：戳破「能跑就行」「AI 生成的应该没问题」等经典合理化借口

使用方式

将文件夹放入智能体的 skills 目录即可。用户说 "review this"、"code review"、"审查代码"、"审查这个 PR" 时自动触发。

许可

MIT