multi-agent-code-review is an intelligent multi-agent collaborative code review system designed for AI coding assistants and development teams. It leverages a parallel agent architecture where specialized reviewers work simultaneously to provide comprehensive, multi-dimensional code analysis.
Unlike traditional single-reviewer code review, this system coordinates multiple expert agents—each focused on a specific domain—to deliver thorough coverage across architecture, security, performance, code quality, and testing.
- Review Commander: Task coordination and report aggregation
- Architecture Reviewer: System design, module organization, dependency analysis
- Security Reviewer: Vulnerability detection, injection prevention, auth/authorization
- Performance Reviewer: Hotspot identification, complexity analysis, N+1 detection
- Code Quality Reviewer: Style consistency, readability, best practices
- Testing Reviewer: Coverage assessment, edge cases, mock quality
- Context Gathering - Understand business context and scope
- High-Level Review - Architecture and design assessment
- Line-by-Line Review - Deep analysis of every code change
- Summary & Decision - Structured report with verdict
- Markdown: GitHub PR comments, documentation, easy to share
- HTML: Embedded reports, email-friendly, print-ready
| Tag | Meaning | Action |
|---|---|---|
| 🔴 [blocking] | Critical issue | Must fix before merge |
| 🟡 [important] | Significant issue | Should fix |
| 🟢 [nit] | Minor issue | Non-blocking suggestion |
| 💡 [suggestion] | Alternative approach | Consider |
| 📚 [learning] | Educational comment | No action needed |
| 🎉 [praise] | Well done | Keep it up |
| Language | Framework Coverage |
|---|---|
| TypeScript | React, Vue, Angular, NestJS, Next.js |
| Python | Django, FastAPI, Flask, AsyncIO |
| Go | Standard library, Gin, Echo, gRPC |
Activate via command:
/review <files or PR URL>
Activate via description:
"帮我审查这些代码"
"review the PR for security issues"
"审查这个函数的性能和安全性"
Output format options:
review src/**/*.ts --format markdown # Default
review src/**/*.ts --format html # HTML report
┌─────────────────────────────────────────────────────────────┐
│ Review Commander │
│ (Task Coordination & Reporting) │
└───────────────────────────┬─────────────────────────────────┘
│
┌──────────────────┼──────────────────┐
▼ ▼ ▼
┌────────────┐ ┌────────────┐ ┌────────────┐
│Architecture│ │ Security │ │Performance │
│ Reviewer │ │ Reviewer │ │ Reviewer │
└────────────┘ └────────────┘ └────────────┘
│ │ │
▼ ▼ ▼
┌────────────┐ ┌────────────┐ ┌────────────┐
│ Code │ │ Testing │ │ API │
│ Quality │ │ Reviewer │ │ Reviewer │
└────────────┘ └────────────┘ └────────────┘
multi-agent-code-review/
├── SKILL.md # Main entry point
├── README.md # This file
├── test_pool.md # Test cases
└── references/
├── agents.md # Agent definitions
├── pipeline.md # Collaboration workflow
├── review-phases.md # 4-phase review process
├── severity-labels.md # Tagging system
├── output-template.md # Markdown & HTML templates
└── languages/
├── typescript.md # TypeScript guidelines
├── python.md # Python guidelines
└── go.md # Go guidelines
| Condition | Decision |
|---|---|
| Has 🔴 blocking issues | Request Changes |
| Has 🟡 important issues | Comment |
| Only 🟢 nit or 💡 suggestion | Approve |
| Exceptional quality | Approve with Praise |
multi-agent-code-review 是一款智能多智能体协作代码审查系统，专为 AI 编程助手和开发团队设计。它采用并行智能体架构，让专业审查员同时工作，从架构、安全、性能、代码质量和测试等多个维度提供全面的代码分析。
不同于传统的单一审查员代码审查，本系统协调多个专家智能体——每个专注于特定领域——提供彻底、全面的代码审查覆盖。
- 主编 (Review Commander): 任务协调与报告汇总
- 架构审查员 (Architecture Reviewer): 系统设计、模块组织、依赖分析
- 安全审查员 (Security Reviewer): 漏洞检测、注入防护、认证授权
- 性能审查员 (Performance Reviewer): 热点识别、复杂度分析、N+1 检测
- 代码质量审查员 (Code Quality Reviewer): 风格一致性、可读性、最佳实践
- 测试审查员 (Testing Reviewer): 覆盖率评估、边界用例、Mock 质量
- 上下文收集 - 理解业务背景和审查范围
- 高层审查 - 架构和设计评估
- 逐行审查 - 深度分析每一处代码变更
- 总结决策 - 生成结构化报告并给出 verdict
- Markdown: GitHub PR 评论、文档、易于分享
- HTML: 内嵌报告、邮件友好、打印就绪
| 标签 | 含义 | 行动 |
|---|---|---|
| 🔴 [blocking] | 严重问题 | 合并前必须修复 |
| 🟡 [important] | 重要问题 | 应当修复 |
| 🟢 [nit] | 小问题 | 非阻塞建议 |
| 💡 [suggestion] | 替代方案 | 考虑采纳 |
| 📚 [learning] | 教育评论 | 无需操作 |
| 🎉 [praise] | 优秀表现 | 继续保持 |
| 语言 | 框架覆盖 |
|---|---|
| TypeScript | React, Vue, Angular, NestJS, Next.js |
| Python | Django, FastAPI, Flask, AsyncIO |
| Go | 标准库, Gin, Echo, gRPC |
通过命令激活:
/review <文件或 PR 地址>
通过描述激活:
"帮我审查这些代码"
"审查这个 PR 的安全问题"
"代码审查"
输出格式选项:
review src/**/*.ts --format markdown # 默认 Markdown
review src/**/*.ts --format html # HTML 报告
┌─────────────────────────────────────────────────────────────┐
│ 主编 (Review Commander) │
│ (任务协调与报告汇总) │
└───────────────────────────┬─────────────────────────────────┘
│
┌──────────────────┼──────────────────┐
▼ ▼ ▼
┌────────────┐ ┌────────────┐ ┌────────────┐
│ 架构 │ │ 安全 │ │ 性能 │
│ 审查员 │ │ 审查员 │ │ 审查员 │
└────────────┘ └────────────┘ └────────────┘
│ │ │
▼ ▼ ▼
┌────────────┐ ┌────────────┐ ┌────────────┐
│ 代码 │ │ 测试 │ │ API │
│ 质量 │ │ 审查员 │ │ 审查员 │
└────────────┘ └────────────┘ └────────────┘
multi-agent-code-review/
├── SKILL.md # 主入口
├── README.md # 项目文档
├── test_pool.md # 测试用例池
└── references/
├── agents.md # 智能体定义
├── pipeline.md # 协作流程
├── review-phases.md # 四阶段审查
├── severity-labels.md # 标签系统
├── output-template.md # 输出模板
└── languages/
├── typescript.md # TypeScript 指南
├── python.md # Python 指南
└── go.md # Go 指南
| 条件 | 决策 |
|---|---|
| 存在 🔴 blocking 问题 | Request Changes - 必须修改后才能合并 |
| 存在 🟡 important 问题 | Comment - 建议修改 |
| 只有 🟢 nit 或 💡 suggestion | Approve - 可以合并 |
| 代码质量优秀 | Approve with Praise - 表扬 |
MIT