LLM Usage Rules

---
description: 
globs: 
alwaysApply: false
---
[contributions-summary.md](mdc:lib/prompts/contributions-summary.md) [brag-document.md](mdc:lib/prompts/brag-document.md) [openai.ts](mdc:lib/openai.ts) [anthropic.ts](mdc:lib/anthropic.ts)

# LLM Usage Rules

## Prompt Engineering
- Structure prompts with clear instructions and examples
- Use system messages for role definition and constraints
- Keep prompts focused and specific to the task
- Use markdown formatting for better readability
- Include clear output format requirements
- Provide context and background when necessary
- Use consistent terminology throughout prompts
- Break complex tasks into smaller, focused prompts
- Include error handling instructions
- Specify response length requirements

## Performance Optimization
- Implement proper prompt caching for reusable content
- Use appropriate temperature settings (0.7 for creative tasks, 0.2 for deterministic)
- Set appropriate max_tokens based on expected response length
- Batch similar requests when possible
- Use streaming for long responses
- Cache frequently used responses
- Monitor token usage and costs
- Use appropriate model sizes for the task

## Cost Management
- Monitor and log token usage
- Set up usage alerts and limits
- Use cheaper models when appropriate
- Implement caching strategies
- Optimize prompt length
- Use appropriate max_tokens settings
- Track and analyze cost patterns
- Implement cost-effective fallbacks
- Use token counting for large inputs
- Consider cost vs. quality tradeoffs

## Error Handling
- Implement proper error boundaries
- Handle API rate limits gracefully
- Retry failed requests with backoff
- Log errors with context
- Provide user-friendly error messages
- Handle partial responses
- Implement timeout handling
- Validate responses
- Handle network issues
- Implement fallback strategies

## Security
- Never expose API keys in logs
- Use environment variables for secrets
- Validate all inputs
- Sanitize outputs
- Implement proper access controls
- Monitor for abuse
- Use secure API endpoints
- Implement proper authentication
- Handle sensitive data appropriately
- Follow security best practices

## Monitoring and Logging
- Log token usage
- Track response times
- Monitor error rates
- Log model performance
- Track cache hit rates
- Monitor cost patterns
- Log user interactions
- Track API limits
- Monitor response quality
- Implement proper analytics

## Testing
- Test with different model versions
- Validate response formats
- Test error handling
- Test rate limiting
- Test caching behavior
- Test with various inputs
- Test edge cases
- Test performance
- Test security measures
- Test cost implications

## Documentation
- Document prompt templates
- Document model choices
- Document error handling
- Document rate limits
- Document caching strategies
- Document cost implications
- Document security measures
- Document testing procedures
- Document monitoring setup
- Document deployment process

## Integration
- Use appropriate API clients
- Implement proper authentication
- Handle API versioning
- Use appropriate SDKs
- Implement proper error handling
- Use appropriate middleware
- Handle API changes
- Implement proper logging
- Use appropriate monitoring
- Follow API best practices

## User Experience
- Provide clear feedback
- Show progress indicators
- Handle long-running operations
- Provide helpful error messages
- Implement proper formatting
- Use appropriate language
- Provide clear instructions
- Handle user input properly
- Implement proper validation
- Follow UX best practices