agents/tools/ai-review.md

model

model
sonnet

AI/ML Code Review

Perform a specialized AI/ML code review for: $ARGUMENTS

Conduct comprehensive review focusing on:

1. Model Code Quality:

   • Reproducibility checks
   • Random seed management
   • Data leakage detection
   • Train/test split validation
   • Feature engineering clarity

2. AI Best Practices:

   • Prompt injection prevention
   • Token limit handling
   • Cost optimization
   • Fallback strategies
   • Timeout management

3. Data Handling:

   • Privacy compliance (PII handling)
   • Data versioning
   • Preprocessing consistency
   • Batch processing efficiency
   • Memory optimization

4. Model Management:

   • Version control for models
   • A/B testing setup
   • Rollback capabilities
   • Performance benchmarks
   • Drift detection

5. LLM-Specific Checks:

   • Context window management
   • Prompt template security
   • Response validation
   • Streaming implementation
   • Rate limit handling

6. Vector Database Review:

   • Embedding consistency
   • Index optimization
   • Query performance
   • Metadata management
   • Backup strategies

7. Production Readiness:

   • GPU/CPU optimization
   • Batching strategies
   • Caching implementation
   • Monitoring hooks
   • Error recovery

8. Testing Coverage:

   • Unit tests for preprocessing
   • Integration tests for pipelines
   • Model performance tests
   • Edge case handling
   • Mocked LLM responses

Provide specific recommendations with severity levels (Critical/High/Medium/Low). Include code examples for improvements and links to relevant best practices.