performance-testing-review-multi-agent-review
使用 AI 编排多代理代码审查
传统代码审查从单一视角检查代码，往往遗漏关键问题。此工具协调多个专业 AI 代理，从安全、架构、性能和质量等多个维度提供全面分析。
下载技能 ZIP
在 Claude 中上传
前往 设置 → 功能 → 技能 → 上传技能
开启并开始使用
测试它
正在使用“performance-testing-review-multi-agent-review”。 Review src/api/auth.py for security issues
预期结果:
Security audit identified 2 critical issues: 1) Missing input validation on user credentials allowing injection attacks, 2) Weak password hashing using MD5. Recommended actions: Implement parameterized queries and upgrade to bcrypt.
正在使用“performance-testing-review-multi-agent-review”。 Analyze performance of src/database/queries.py
预期结果:
Performance analysis found 3 optimization opportunities: 1) N+1 query pattern in user lookup, 2) Missing database indexes on frequently queried columns, 3) Inefficient loop processing large datasets. Estimated improvement: 60% reduction in query time.
安全审计
安全All 27 static analysis findings are false positives. The SKILL.md file is documentation containing Python pseudo-code examples in markdown blocks, not executable code. No actual command execution, cryptographic operations, or system reconnaissance patterns exist. Safe for publication.
质量评分
你能构建什么
Web 应用安全审查
自动将 Web 应用代码分配给安全审计师和 Web 架构专家，进行全面的漏洞评估。
微服务架构验证
通过涵盖设计模式、实现质量和部署就绪性的顺序审查阶段验证微服务设计。
性能关键代码分析
通过将性能敏感代码分配给专业的性能分析代理，识别性能瓶颈。
试试这些提示
Review the code at [file path] using the multi-agent review system. Focus on identifying any critical issues and provide a summary of findings.
Perform a security audit on [repository URL] using the security-auditor agent. Identify potential vulnerabilities and provide remediation recommendations.
Conduct a comprehensive architecture review of [project path] using sequential agents: architect-reviewer for design patterns, code-quality-reviewer for implementation, and devops-validator for deployment considerations.
Configure a custom review with parallel agents [security-auditor, performance-analyst] and sequential agent [architecture-reviewer] for [code snippet]. Weight security at 0.4, performance at 0.4, and architecture at 0.2.
最佳实践
- 在选择代理类型之前定义清晰的审查目标
- 使用并行执行处理独立的审查维度以减少周转时间
- 审查整合报告并将冲突建议升级人工判断
避免
- 在不考虑代码特征的情况下对每次审查运行所有代理类型
- 忽略代理置信度分数来确定发现结果的优先级
- 跳过代理输出相互依赖的顺序依赖阶段