---

name: triage-review description: Fetch PR review comments from AI reviewers, verify accuracy, and apply valid fixes. Use when the user wants to process AI code review comments, apply AI reviewer suggestions, or verify and implement automated code review feedback. Works with GitHub PRs and uses web search to verify AI claims before applying changes. allowed-tools: [Bash, Read, TodoWrite, Grep, Glob, Task, WebSearch]

Apply Verified Fixes from AI Review Comments

Context

Current State

• Current branch: git branch --show-current
• Uncommitted changes: git status --porcelain
• PR number (if exists): gh pr view --json number --jq '.number' 2>/dev/null || echo "No PR found"
• PR URL: gh pr view --json url --jq '.url' 2>/dev/null || echo ""

PR Information (if PR exists)

• PR title: gh pr view --json title --jq '.title' 2>/dev/null || echo ""
• PR state: gh pr view --json state --jq '.state' 2>/dev/null || echo ""
• Changed files in PR: gh pr view --json files --jq '.files[].path' 2>/dev/null || echo ""

Review Comments

• All PR comments: gh pr view --json comments --jq '.comments[] | "[\(.author.login)] \(.body)"' 2>/dev/null || echo "No comments"
• Review threads: gh pr view --json reviews --jq '.reviews[] | "[\(.author.login)] \(.state): \(.body)"' 2>/dev/null || echo "No reviews"

Your task

1. Identify AI Reviewers

   • Look for comments from: google-code-assist, github-copilot, copilot[bot], google-code-assist[bot]
   • Extract all suggestions from AI reviewers

2. Verify Each AI Suggestion For each AI comment, perform fact-checking:

   a) Extract Technical Claims

   • API usage recommendations
   • Security vulnerability claims
   • Performance optimization suggestions
   • Best practice recommendations

   b) Verify Using Web Search Use BOTH mcp__gemini__google-search and mcp__openai__openai-search with the SAME query to cross-verify information:

   • Search Strategy:

     1. Send identical queries to both search tools
     2. Compare results from both sources
     3. Look for consensus or discrepancies
     4. Prioritize information that appears in both results

   • Verification Process:

     • API/Method validation: Search "[language] [method_name] documentation" in both tools
     • Security claims: Search "[vulnerability] CVE [year]" and compare findings
     • Performance claims: Search "[technique] benchmark comparison" across both engines
     • Best practices: Search "[technology] official best practices" and verify consistency

   • Cross-Verification Examples:

     CopyQuery: "React useEffect cleanup function best practice"
     → Compare if both tools return similar official React documentation
     
     Query: "Log4j CVE-2021-44228 vulnerability details"
     → Verify if both tools confirm the same security issue
     
     Query: "Python list comprehension vs for loop performance"
     → Check if benchmarks from both searches align
     

   • Decision Criteria:

     • ✅ Apply if: Both tools confirm the same information
     • ⚠️ Review carefully if: Results differ between tools
     • ❌ Skip if: Neither tool can verify the claim

3. Categorize Verified Suggestions After verification, classify each suggestion:

   ✅ Verified & Apply:

   • Confirmed security vulnerabilities
   • Documented API misuse
   • Proven performance issues
   • Official best practices

   ⚠️ Partially Verified:

   • Mixed opinions in community
   • Context-dependent improvements
   • Style preferences with some backing

   ❌ Incorrect/Unverified:

   • Outdated recommendations
   • Non-existent APIs/methods
   • False security claims
   • Subjective preferences without backing

4. Apply Only Verified Fixes For each verified suggestion:

   CopyFile: [filename]
   Line: [line number or range]
   Issue: [Brief description in Japanese]
   Verification: [What was confirmed via search]
   Fix: [Exact change to apply]
   Source: [Documentation/CVE/Benchmark URL]
   

5. Implementation Process

   • Read current file content
   • Apply verified changes ONE topic at a time
   • After each topic fix:
     1. Save the modified files
     2. Create a focused commit

   Commit Guidelines:

   • One commit per logical fix topic
   • Group related changes in same file(s) together
   • Keep commits atomic and reversible
   • Use Conventional Commits format

6. Final Report in Japanese

   Copy## AI レビュー検証・修正レポート
   
   ### ✅ 検証済み・適用した修正
   1. **[File]**: [What was fixed]
      - 根拠: [Verification source/URL]
      - 修正内容: [Brief description]
   
   ### ⚠️ 部分的に検証・条件付き適用
   1. **[File]**: [What was conditionally fixed]
      - 検証結果: [Mixed findings]
      - 判断理由: [Why applied or not]
   
   ### ❌ 誤り・未適用の提案
   1. **[Suggestion]**: [Why it was incorrect]
      - 検証結果: [What search revealed]
   
   ### 検証サマリー
   - AI提案総数: X件
   - 検証済み修正: Y件
   - 誤った提案: Z件
   

Important Notes

• ALWAYS verify AI suggestions before applying
• Prioritize official documentation over blog posts
• Check publication dates to avoid outdated advice
• If unable to verify, mark as "unverified" and skip
• Document all verification sources for transparency
• Be especially careful with security-related changes
• When in doubt, search multiple sources
• Use AskUserQuestionTool when you need clarification on:
  • Whether to apply a partially verified suggestion
  • How to prioritize conflicting recommendations
  • Whether a security-related change should be applied