AgentSkillsCN

review

对近期变更或指定文件进行多代理代码审查。根据文件类型,精准挑选专业评审人员。

SKILL.md
--- frontmatter
name: review
description: Run a multi-agent code review on recent changes or specified files. Selects specialist reviewers based on file types.

/review

Run a multi-agent code review on recent changes or specified files.

Usage

code
/review              # Review uncommitted changes
/review <file/dir>   # Review specific files or directory
/review --full       # Full project architecture review

Workflow

  1. Identify scope: Determine what to review:

    • No args: git diff and git diff --cached for uncommitted changes
    • File/dir arg: the specified paths
    • --full: entire project structure
  2. Select reviewers: Based on what changed, spin up relevant review agents as a team:

    • .go files (handlers/services/database) → code-reviewer
    • .go files (handlers) + templates → htmx-debugger
    • Auth, middleware, secrets → security-reviewer
    • Docker, Makefile, CI → devops-reviewer
    • Cross-cutting changes → systems-architect
    • Template/CSS/UX changes → ux-reviewer
    • SQL, queries, performance-sensitive code → performance-analyst
    • ML/LLM/embedding code → ml-engineer
  3. Run review team: Launch selected agents in parallel using TeamCreate. Each agent reviews the changed files through their specialist lens.

  4. Compile report: Collect findings from all agents and produce a unified, deduplicated report:

    • Critical — Must fix before merge
    • Important — Should fix, creates tech debt if deferred
    • Suggestion — Nice to have, style/optimization
  5. Create issues: For any Critical or Important findings, create beads issues to track the fixes (bd create --title="[Bug] ..." --type=bug --priority=1 for Critical, --priority=2 for Important). Do not just report findings without tracking them.