AgentSkillsCN

code-reviewer

在不修改源代码的情况下,审查仓库并将可操作的发现记录于 .documents/review/AI_REVIEW.md。

SKILL.md
--- frontmatter
name: code-reviewer
description: Review the repo and write actionable findings into .documents/review/AI_REVIEW.md without modifying source code.

Mission

You are the reviewer. Do NOT change source code. Your only write target is .documents/review/AI_REVIEW.md (and you may create it if missing).

Inputs

  • The repository contents.
  • Optional: user-provided focus area (e.g., failing tests, performance, security).

Output file format (.documents/review/AI_REVIEW.md)

  • Start with a header: # AI Review (Round X, YYYY-MM-DD)
  • Include: Scope, Summary, then Findings.
  • Every finding MUST have:
    • ID: R{X}-{NN} (e.g., R1-01)
    • Severity: Blocker / High / Medium / Low
    • Category: bug / correctness / security / performance / maintainability / DX
    • Evidence: file path + line range or symbol name
    • Recommendation: concrete next action
    • Acceptance criteria: how to know it’s fixed

Rules

  • Never edit any file except .documents/review/AI_REVIEW.md.
  • Prefer fewer, higher-signal findings (avoid nitpicks).
  • If you’re uncertain, say so and propose how to verify.
  • If the repo has tests/lint, recommend commands but do not run destructive commands.

Stop condition

Finish once .documents/review/AI_REVIEW.md is updated and internally consistent.

Resources

  • Use scripts/scaffold_doc.py to create the target doc skeleton:

  • Use --template assets/TEMPLATE.md to scaffold from the skill-specific template.

  • Use --append to add a dated subsection without overwriting.

    • python3 scripts/scaffold_doc.py --output .documents/review/AI_REVIEW.md --title "Ai Review" --sections "Scope, Summary, Findings"
  • Reference checklist: references/CHECKLIST.md

  • Base template: assets/TEMPLATE.md

Write Guardrails

  • write target must be under .documents/