Contributing

Join the CatalystRL community and help build the future of self-improving AI agents.

Repository Structure

catalystrl-claude/
├── skills/                    # Skill definitions
│   ├── commit/               # Example: commit skill
│   │   ├── SKILL.md          # Main prompt
│   │   ├── config.json       # Configuration
│   │   ├── rules.json        # Validation rules
│   │   └── scripts/          # Deterministic scripts
│   └── ...
├── golden-blueprint/          # Blueprint documentation
│   ├── 01_ABES_Integration.md
│   ├── ...
│   └── Templates/
└── lib/                       # Shared utilities
    └── discovery/             # Dynamic discovery scripts

Development Setup

1. Fork and clone

git clone https://github.com/YOUR_USERNAME/catalystrl-claude.git
cd catalystrl-claude

2. Install to Claude Code

# Link skills to Claude Code directory
ln -s $(pwd)/skills/* ~/.claude/skills/

# Verify installation
ls ~/.claude/skills/

3. Test your changes

# Run the compliance validator
./golden-blueprint/validate-golden-compliance.sh your-skill

Creating a New Skill

1Create directory structure
mkdir -p skills/my-skill/scripts
cd skills/my-skill
2Copy templates
cp ../../golden-blueprint/Templates/golden-config.json config.json
cp ../../golden-blueprint/Templates/golden-prompt.md SKILL.md
3Customize and implement

Replace placeholders in config.json and SKILL.md with your skill's configuration and logic.

4Validate compliance
../../golden-blueprint/validate-golden-compliance.sh my-skill

Skill Requirements by Tier

TierDQRequired ComponentsTesting
Tier 160-90%All 15Required
Tier 230-60%1-8, 11-12Recommended
Tier 310-30%1-8, 12Optional

DQ (Deterministic Quotient) = Scripted Steps / Total Steps. Higher DQ means more automation, less LLM reasoning.

Pull Request Guidelines

1.Run validation - Ensure skill passes compliance check
2.Test locally - Verify skill works in Claude Code
3.Document changes - Update SKILL.md with any new features
4.Write clear PR description - Explain what and why
5.Add tests - Required for Tier 1 skills

Code Style

SKILL.md

  • • YAML frontmatter required
  • • Step-graph syntax for dependencies
  • • Clear step descriptions
  • • Error handling section

config.json

  • • Use $discover: for dynamic data
  • • Include agent metadata
  • • Define gates for safety
  • • Configure memory context

Scripts

  • • Bash or Python
  • • Exit codes for success/failure
  • • Structured output (JSON preferred)
  • • Include in scripts/ directory

Naming

  • • Lowercase with hyphens
  • • 2-3 words maximum
  • • Family prefix for related skills
  • • No "agent" suffix

Questions?

The repository is coming soon. Once public, you can open issues on GitHub or join the community discussions.

View repository progress →