emasoft/amia-ai-pr-review-methodology

AI PR Review Methodology Skill

Overview

Evidence-based PR review using 4 phases and 5 analysis dimensions to catch false positives, redundant code, and unverified assumptions.

Prerequisites

• PR workflow knowledge and ability to read diffs
• Shell access and codebase access for verification

Instructions

1. Gather Context (Phase 1): Read phase-1-context-gathering. Read complete files (not just diffs), search for duplicates, understand root cause, verify all claims.

2. Structured Analysis (Phase 2): Read phase-2-structured-analysis, then apply dimensions D1-D5 in order:

   • dimension-1-problem-verification
   • dimension-2-redundancy-check
   • dimension-3-system-integration
   • dimension-4-senior-review
   • dimension-5-false-positive-detection

3. Evidence Requirements (Phase 3): Compile missing evidence: Problem Demonstration, Solution Validation, Assumption Verification, Cross-Platform Testing.

4. Scenario Protocol (if applicable): Match PR type to the appropriate scenario:

   • scenario-bug-fixes
   • scenario-dependency-updates
   • scenario-path-changes
   • scenario-performance

5. Generate Review (Phase 4): Use review-output-template and run quick-reference-checklist before submitting.

Checklist

Copy this checklist and track your progress:

• [ ] Read complete files, not just the diff
• [ ] Search for existing solutions and duplicates
• [ ] Verify all claims and understand root cause
• [ ] Apply all 5 analysis dimensions (D1-D5)
• [ ] Compile missing evidence list
• [ ] Apply scenario-specific protocol if applicable
• [ ] Generate structured review using the template

Output

| Section | Content | |---------|---------| | Summary | PR overview and overall assessment | | Findings | Strengths, questions, blocking red flags | | Recommendation | APPROVE / REQUEST CHANGES / COMMENT |

Output discipline: All scripts support --output-file <path>.

Error Handling

Non-zero exit codes on failure. See detailed guide in Resources.

Resources

• review-output-template — Review output format

  • T.1 When to generate the review output
  • T.2 The complete review output template
  • T.3 How to fill each section
  • T.4 Choosing the final recommendation
  • T.5 Setting the confidence level
  • T.6 Writing the author note
  • T.7 Example: A completed review output

• detailed-guide — Full methodology reference

  • 4 Phases Overview
  • 5 Analysis Dimensions
  • Evidence Requirements
  • Scenario-Specific Protocols
  • Error Handling
  • Reference Documents Index

Examples

gh pr diff 42 > /tmp/pr42.diff
gh pr view 42 --json title,body,files
# Apply D1-D5 dimensions, compile evidence, generate review
gh pr review 42 --body "## Summary\nReviewed with AI PR methodology.\n### Recommendation: APPROVE"