AI 编码 Agent 团队自动化代码评审工作流生成器

You are an expert in AI-powered code review automation and multi-agent development workflows.

Generate a complete automated code review workflow for a team of AI coding agents:

Team Composition

• Primary coder: [e.g., Claude Code / Codex / Cursor]
• Reviewer agent: [e.g., dedicated review agent with specific expertise]
• Security scanner: [e.g., agent focused on vulnerability detection]
• Test generator: [e.g., agent that writes tests for new code]

Workflow Requirements

Pre-Review Checks

1. Lint and format validation
2. Type checking pass
3. Dependency vulnerability scan
4. Diff size guard (flag PRs > 500 lines for split)

Review Stages

1. Architecture Review: Does the change align with existing patterns?
2. Logic Review: Are there edge cases, race conditions, or logic errors?
3. Security Review: SQL injection, XSS, auth bypass, secret leaks
4. Performance Review: N+1 queries, unnecessary allocations, missing indexes
5. Test Coverage: Are critical paths tested? Mutation testing score?

Review Output Format

For each finding:

• Severity: critical / warning / suggestion / nitpick
• File and line range
• Description of the issue
• Suggested fix (code snippet)
• Confidence score (0-100%)

Quality Gates

• Block merge if: any critical finding, test coverage < 80%, or security scan fails
• Auto-approve if: only nitpicks, all tests pass, and diff < 50 lines

Generate:

1. GitHub Actions YAML workflow
2. Review agent system prompts for each role
3. PR comment template with findings summary
4. Metrics dashboard config for tracking review quality over time