AgentSkillsCN

review-merge-readiness

请求并执行结构化代码审查：在完成重要任务后、每轮执行结束时，或在合并前进行。根据git diff范围，对照计划与需求，按严重程度（关键/重要/次要）列出问题清单，并就合并准备情况给出清晰的结论。触发关键词包括：请求代码审查、PR审查、合并准备、生产就绪。

SKILL.md
--- frontmatter
name: review-merge-readiness
description: "Request/execute structured code review: use after completing important tasks, at end of each execution batch, or before merge. Based on git diff range, compare against plan and requirements, output issue list by Critical/Important/Minor severity, and provide clear verdict on merge readiness. Trigger words: request code review, PR review, merge readiness, production readiness."

Requesting Code Review (Structured Review: Requirements Alignment + Production Readiness)

Goal: Make code review a repeatable process, not random comments.

Core principle: Review early, review often.

When Review is Required

Mandatory:

  • •After completing important features
  • •After each batch in workflow-execute-plans ends (default 3 tasks per batch)
  • •Before merging to main branch

Optional but valuable:

  • •When stuck (get a fresh perspective)
  • •Before major refactoring (establish baseline first)
  • •After fixing complex bugs (verify no new regressions)

How to Initiate (Minimum Information Set)

1) Determine Review Scope (git SHA)

Prefer using "baseline before plan/task started" as BASE_SHA:

bash
# Common approach: use main as baseline
BASE_SHA=$(git merge-base HEAD main 2>/dev/null || git merge-base HEAD master)
HEAD_SHA=$(git rev-parse HEAD)

If you want to review just the most recent commit within a small task:

bash
BASE_SHA=$(git rev-parse HEAD~1)
HEAD_SHA=$(git rev-parse HEAD)

2) Prepare Review Basis (Plan / Requirements)

Review must provide:

  • •WHAT_WAS_IMPLEMENTED: What you just completed (1-3 bullet points)
  • •PLAN_OR_REQUIREMENTS: Corresponding plan excerpt or requirements document path (e.g., run_dir/03-plans/feature-plan.md)

3) Get diff (Evidence)

bash
git diff --stat "$BASE_SHA..$HEAD_SHA"
git diff "$BASE_SHA..$HEAD_SHA"

Review Execution Methods (Choose One)

Method A: Sub-Agent Review (if your system supports it)

Use template: review-merge-readiness/code-reviewer.md, fill in placeholders and execute.

Method B: Self Review (works without sub-agent)

Check each item per template checklist and output same structured result (Strengths + Issues by severity + Verdict).

Review Output Format (Must Be Structured)

Output must include:

  • •Strengths: Specific positives (at least 1)
  • •Issues: Categorized by severity:
    • •Critical (Must Fix): Security/data loss/functionality bugs/would cause production incidents
    • •Important (Should Fix): Obviously poor architecture/missing error handling/test gaps/requirements misalignment
    • •Minor (Nice to Have): Style/small optimizations/documentation polish
  • •Assessment: Merge readiness (Yes/No/With fixes) + 1-2 sentence technical reasoning

Each Issue must include:

  • •Location: file:line
  • •What's wrong
  • •Why it matters
  • •How to fix (provide minimal change suggestion)

Relationship with Other Review Skills

  • •review-clean-code: More focused on "maintainability/cleanliness", suitable for deep code smell investigation
  • •review-react-best-practices: More focused on React/Next.js performance patterns (waterfalls/bundle/re-renders). Use when the diff touches React UI, data fetching, or performance-sensitive areas.
  • •This skill: More focused on "requirements alignment/production readiness/merge verdict" conclusion-oriented review

If you just need a "can we merge?" verdict: use this skill. If you want a deep health check: additionally invoke review-clean-code.