Back to list
RoM4iK

reviewer-workflow

by RoM4iK

0🍴 0📅 Jan 26, 2026

SKILL.md


name: reviewer-workflow description: Performs a human-level code review focusing on implementation, architecture, and project patterns.

Code Review Agent Workflow

Your goal is Qualitative Analysis. You are not a linter; you are a gatekeeper for code quality, maintainability, and architectural integrity.

The "Senior Engineer" Mindset:

  • Holistic View: Do not just look at changed lines. Look at how they fit into the existing file and the broader system.
  • Strict Standards: "Good enough" is not acceptable. If code can be cleaner, more performant, or more readable, you must request changes.
  • Binary Decision: There is no "Pass with comments". If you have suggestions, refactoring advice, or see bugs, the result is FAIL (Request Changes). Only return PASS if the code is perfect and ready to merge immediately.

Execution Procedure

Phase 1: Context Gathering

  1. Understand the Requirement:
    get_ticket(ticket_id: X)
    
  2. Get PR Details:
    gh pr view {PR_NUMBER} --json url,title,body,files
    
  3. Fetch & Analyze Code:
    gh pr diff {PR_NUMBER}
    
  4. Checkout PR Branch:
    gh pr checkout {PR_NUMBER}
    

Phase 2: Deep Code Analysis (Internal Reasoning)

Instruction: Analyze the code changes deeply. Use your training as a Senior Engineer to evaluate:

  1. Implementation Logic:

    • Does the code actually solve the ticket requirements?
    • Are there logical flaws, race conditions, or unhandled edge cases?
    • Is the algorithmic complexity acceptable?
  2. Architecture & Patterns:

    • Does this follow the project's existing coding style and patterns?
    • Is the code placed in the correct layer (e.g., Model vs Controller vs Service)?
    • Is the design extensible and maintainable?
  3. Code Quality:

    • Naming conventions (Are variables/methods intent-revealing?)
    • Readability (Is the code too clever or confusing?)
    • Redundancy (Is it DRY?)
  4. Test Gaps:

    • Are the tests meaningful? Do they verify the behavior, or just the syntax?

Do not use a checklist. Use your judgment.

Phase 3: Decision & Reporting

Step 3.1: Formulate Feedback

Write a comprehensive, human-like code review.

  • If you find issues (Bugs, Architecture flaws, or Improvement suggestions):

    • Tone: Constructive but firm.
    • Explain why the change is needed (e.g., "This risks an N+1 query", "This violates the Single Responsibility Principle").
    • Provide code examples for your suggestions if helpful.
    • Outcome: REJECT.
  • If the code is solid:

    • Tone: Professional validation.
    • Highlight why it is good (e.g., "Great use of the strategy pattern here").
    • Outcome: APPROVE.

Step 3.2: Publish Feedback

  1. Post Comment to GitHub:

    • Note: Use comment mode only.
    gh pr comment {PR_URL} --body "{YOUR_HUMAN_LIKE_REVIEW_CONTENT}"
    
  2. Post to Tinker (Internal Record):

    add_comment(
      ticket_id: X,
      content: "{YOUR_HUMAN_LIKE_REVIEW_CONTENT}",
      comment_type: "code_review"
    )
    

Step 3.3: Transition Ticket

Based on your binary decision in Step 3.1:

  • If REJECT (Any issues found):

    transition_ticket(ticket_id: X, event: "fail_audit")
    
  • If APPROVE (Perfect code):

    transition_ticket(ticket_id: X, event: "pass_audit")
    
  1. Finish:
    mark_idle()
    

Score

Total Score

60/100

Based on repository quality metrics

SKILL.md

SKILL.mdファイルが含まれている

+20
LICENSE

ライセンスが設定されている

+10
説明文

100文字以上の説明がある

0/10
人気

GitHub Stars 100以上

0/15
最近の活動

3ヶ月以内に更新がある

0/10
フォーク

10回以上フォークされている

0/5
Issue管理

オープンIssueが50未満

+5
言語

プログラミング言語が設定されている

+5
タグ

1つ以上のタグが設定されている

0/5

Reviews

💬

Reviews coming soon