ai-code-review-collaboration（AI 代码审查协作）

Multi-AI Code Review Collaboration Framework

A systematic approach to getting diverse AI perspectives on code, then synthesizing insights into actionable improvements.

When to Use This

• Complex codebases where blind spots are likely
• Production/live trading code where reliability is critical
• Architecture decisions with trade-offs
• When you want to validate Claude's recommendations
• Before major refactoring or deployment

The Process

Phase 1: Initial Review (Claude)

1. Claude reviews code and documents findings
2. Creates context prompt for external AI
3. Identifies platform-specific constraints external AI must understand

Phase 2: External AI Review

1. User pastes prompt to external AI (Gemini, DeepSeek, GPT, etc.)
2. External AI provides structured review
3. User brings response back to Claude

Phase 3: Synthesis & Debate

1. Claude evaluates external AI's points
2. Categorizes into: Valid, Partially Valid, Invalid
3. Explains reasoning for each categorization
4. Creates response prompt for user to continue dialogue

Phase 4: Consensus

1. Continue rounds until agreement reached
2. Document final action plan
3. Prioritize fixes by risk and effort

Prompt Templates

Template A: Initial External AI Request

**[AI NAME] CODE REVIEW REQUEST - [PROJECT TYPE]**

I need a comprehensive code review. This code [CRITICAL CONTEXT - e.g., "runs on live funded accounts"]. Please review thoroughly and respond in a format I can share with another AI for collaborative discussion.

---

## CRITICAL PLATFORM CONTEXT (Read First)

[List platform-specific constraints that might not be obvious]
[List what IS and ISN'T possible on this platform]
[Explain why certain "standard" patterns don't apply]

---

## REVIEW SCOPE

Please analyze:
1. **Logic & Correctness** - [specific concerns]
2. **Risk Management** - [specific concerns]
3. **Performance** - [specific concerns]
4. **Reliability** - [specific concerns]
5. **Code Quality** - [specific concerns]
6. **Scalability** - [planned expansion, multi-instance needs, performance at scale]
7. **Future Updateability** - [extension points, configuration extensibility, technical debt, breaking change risks]

---

## FUTURE ROADMAP (if applicable)

[Describe planned features, scaling needs, and future requirements so the reviewer can assess how well the current architecture supports them]

---

## THE CODE

[Include full code or key sections]

---

## RESPONSE FORMAT

Structure your response as:

**[AI NAME] CODE REVIEW - ROUND 1**

## 🔴 CRITICAL ISSUES (Must Fix)
## 🟡 IMPORTANT CONCERNS (Should Fix)
## 🟢 MINOR SUGGESTIONS (Nice to Have)
## ✅ WELL IMPLEMENTED
## 🔮 SCALABILITY ASSESSMENT
## 🔧 FUTURE UPDATEABILITY ASSESSMENT
## ❓ QUESTIONS / CLARIFICATIONS NEEDED
## 📋 PRIORITIZED ACTION PLAN

---

## EXISTING FINDINGS (if any)

[Include prior AI findings so new AI can confirm/challenge]

Template B: Response to External AI

**CLAUDE'S RESPONSE TO [AI NAME] - ROUND [N]**

## ✅ FULL AGREEMENT
[Points we agree on completely]

## 🤝 CONCESSIONS & MODIFICATIONS
[Points where Claude adjusts position with explanation]

## 🛡️ POINTS I STILL MAINTAIN
[Disagreements with detailed reasoning]

## 🔍 NEW OBSERVATIONS
[Anything new Claude notices based on discussion]

## 📋 UPDATED ACTION PLAN
[Current consensus on what to fix]

## 🤝 CLOSING QUESTION
[Ask if they agree or have remaining concerns]

Template C: Final Consensus Summary

## MULTI-AI REVIEW CONSENSUS

**Participants:** [List AIs involved]
**Code Reviewed:** [File/project name]
**Date:** [Date]

### AGREED FIXES (In Priority Order)
| # | Issue | Fix | Effort | Risk |
|---|-------|-----|--------|------|
| 1 | [Issue] | [Solution] | [Low/Med/High] | [Low/Med/High] |

### EXPLICITLY REJECTED SUGGESTIONS
| Suggestion | Rejected Because |
|------------|------------------|
| [Suggestion] | [Platform constraint / Not applicable / etc.] |

### VERIFIED AS CORRECT
- [Item 1 that was reviewed and confirmed good]
- [Item 2]

### OPEN QUESTIONS FOR FUTURE
- [Any unresolved items to revisit later]

Best Practices

1. Always provide platform context - External AIs apply generic patterns without knowing constraints
2. Be specific about what CAN'T be done - Prevents suggestions for impossible approaches
3. Request structured responses - Makes synthesis easier
4. Track rounds - Label each exchange for clarity
5. Document consensus - Final agreement should be explicit
6. Implement incrementally - Test each fix before moving to next

Reference Files

• references/prompt-templates.md - Copy-paste ready templates
• references/platform-contexts.md - Pre-written context blocks for common platforms
• references/synthesis-checklist.md - How to evaluate external AI suggestions