Skills System¶

Skills are reusable workflow instructions that all supported coding agents (Claude Code, GitHub Copilot CLI, OpenCode) can use. A single skill definition propagates automatically to every agent.

Adding a New Skill¶

1. Create a .md file in .claude/commands/ with YAML frontmatter:

   ---
   name: my-skill
   description: >
     Brief description of when this skill should be triggered.
     This text is used in the skill catalog for all agents.
   ---
   
   # My Skill
   
   Instructions the agent should follow when this skill is activated.
   

2. Run the propagation script:

   ./scripts/generate-agent-instructions.sh
   

That's it. The skill is now available in all three agents.

How It Works Per Agent¶

Claude Code¶

Skills in ~/.claude/commands/ become slash commands invokable via /skill-name. Claude reads these natively.

GitHub Copilot CLI¶

Copilot reads .github/copilot-instructions.md for project-level instructions. The generator creates this file by combining all rules from CLAUDE.md (with machine-specific paths sanitized) and a catalog of available skills.

OpenCode¶

OpenCode reads CLAUDE.md natively but has no slash command system. The generator appends an Available Skills (Auto-Generated) section listing all skills. This section is idempotent — re-running replaces it.

Automatic Sync Points¶

The script runs automatically at two points:

• ./install.sh — install_skills() during installation
• coding --<agent> — ensure_agent_instructions() at every launch

Skills stay current even if you add new ones between installs.

Path Sanitization¶

Generated files never contain machine-specific paths:

This makes .github/copilot-instructions.md safe to commit.

Current Skills¶

Script Reference¶

# Sync all skills to all agents
./scripts/generate-agent-instructions.sh

# With explicit paths (used internally)
./scripts/generate-agent-instructions.sh <project_dir> <coding_repo>