AgentSkillsCN

review-checklist

为 Open Artel 多智能体项目制定代码审查标准。明确审查清单、评审结论类型、边界违规检测、回归问题核查,以及审查文件格式。

SKILL.md
--- frontmatter
name: review-checklist
description: Code review standards for Open Artel multi-agent projects. Defines the review checklist, verdict types, boundary violation detection, regression checking, and review file format.

Review Checklist

When reviewing an agent's submitted work, check every item:

Required Checks

  • Acceptance criteria met — Every criterion from the task brief is satisfied
  • Files within agent boundary — Agent only modified files in their domain (per AGENTS.md)
  • No boundary violations — No changes to files owned by other agents
  • Build passes — Project builds without errors (if applicable)
  • No regressions — Existing functionality still works
  • Consistent with conventions — Code follows project style and patterns
  • Commit message format — Uses [AGENT:x] [ACTION:y] [TASK:z] header
  • Task brief updated — Task status reflects current state

Optional Checks (when applicable)

  • Tests pass — All existing tests still pass
  • New tests added — New functionality has test coverage
  • Types correct — TypeScript types are accurate, no any usage
  • Error handling — Async operations have proper error handling
  • Documentation updated — Relevant docs reflect changes

Verdict Types

VerdictMeaningNext Action
APPROVEDAll required checks passMerge to pre-mortal, assign next task
CHANGES REQUESTEDMinor issues found, fixableAgent addresses feedback, re-submits
REJECTEDFundamental problems, wrong approachTask re-scoped or reassigned

Decision Criteria

  • APPROVED: All required checks pass. Minor style issues can be noted but don't block.
  • CHANGES REQUESTED: One or more required checks fail, but the approach is sound. Provide specific feedback on what to fix.
  • REJECTED: Approach is fundamentally wrong, boundary violations are severe, or work doesn't match the task brief at all.

Boundary Violation Detection

How to Check

  1. Read the task brief — Check which files are in scope
  2. Read the diffgit diff HEAD~1 or git diff <branch>
  3. Compare against AGENTS.md — Verify each modified file is in the agent's domain
  4. Check .ai/boundaries.md — For detailed ownership rules

Common Violations

ViolationExampleSeverity
Cursor modifies UI componentssrc/components/ui/Button.tsxHigh — Lovable's domain
Lovable modifies hooks with logicsrc/hooks/use-gateway-api.tsHigh — Cursor's domain
Any agent modifies configpackage.json, tsconfig.jsonMedium — Claude Code's domain
Any agent modifies .ai/ files.ai/tasks/TASK-001.mdLow — usually Claude Code's domain

Response to Violations

  • High severity: REJECT — agent must not modify other agents' files
  • Medium severity: CHANGES REQUESTED — remove unauthorized changes
  • Low severity: Note in review, approve if changes are minor and correct

Regression Checking

What to Check

  1. Build status — Does the project still build?
  2. Test suite — Do existing tests still pass?
  3. Related features — Do features that depend on changed code still work?
  4. UI consistency — Do visual components still render correctly?

How to Check

bash
# Build check
npm run build

# Test check
npm run test

# Type check
npx tsc --noEmit

# Lint check
npm run lint

Regression Response

  • If build breaks: REJECT immediately
  • If tests fail: CHANGES REQUESTED with specific test failures
  • If related features break: CHANGES REQUESTED with reproduction steps
  • If UI inconsistency: CHANGES REQUESTED with screenshots or descriptions

Review File Format

Reviews are written to .ai/reviews/<task-id>-review.md.

See .ai/templates/review.md for the complete template.

Key sections:

  • Checklist — All items checked/unchecked
  • Files Reviewed — Table of files with assessments
  • Findings — Numbered list of observations
  • Feedback — Specific guidance for the agent
  • Decision — Verdict and next action

Review Commit Format

After completing a review, commit with:

code
[AGENT:kimi] [ACTION:approve] [TASK:TASK-XXX] Brief summary

or

code
[AGENT:kimi] [ACTION:reject] [TASK:TASK-XXX] Brief reason

See .ai/reviews/TASK-XXX-review.md for details.