Thank you for your interest in contributing!
- Check existing issues first
- Provide clear reproduction steps
- Include relevant logs or screenshots
-
Create a new directory under
prompts/:prompts/ └── your-agent/ ├── core.md # Main system prompt ├── core-cli.md # CLI-specific instructions └── instructions.md # Detailed task instructions -
Define the agent's role and responsibilities in
core.md -
Add task-specific instructions in
instructions.md -
Update the README.md with the new agent type
Add validated code patterns to knowledge/patterns.jsonl:
{
"id": "P-XXX",
"name": "Pattern Name",
"description": "When and why to use this pattern",
"code": "Example code snippet",
"confidence": 0.9,
"uses": 0
}Add recorded mistakes and fixes to knowledge/lessons.jsonl:
{
"id": "L-XXX",
"trigger": "Context when this applies",
"mistake": "What went wrong",
"fix": "How to fix it",
"success_rate": 1.0
}Add known anti-patterns to knowledge/anti-patterns.jsonl:
{
"id": "AP-XXX",
"name": "Anti-Pattern Name",
"description": "Why this is problematic",
"example": "Bad code example",
"fix": "Correct approach"
}- Be specific and actionable
- Include examples where possible
- Define clear success criteria
- Keep token count reasonable (< 4000 tokens per prompt)
- Use unique IDs (check existing entries)
- Provide concrete code examples
- Set realistic confidence scores (0.0-1.0)
- Test patterns before adding
- Keep README.md updated
- Document architectural decisions
- Include diagrams where helpful
- Fork the repository
- Create a feature branch:
git checkout -b feature/your-feature - Make your changes
- Test prompts with Claude API if possible
- Submit a pull request with:
- Clear description of changes
- Rationale for new patterns/lessons
- Any test results
- Be respectful and constructive
- Focus on improving the system
- Share learnings with the community
Open an issue or reach out to the maintainers.