Repository Structure

AGENTS.md

Instructions for AI agents working on this repository.

Repository Structure

agent-skills/
├── skills/      # Skills with tooling, scripts, or workflows
├── prompts/     # Pure instruction prompts (text only)
├── clawdbot/    # Clawdbot-specific skills
├── scripts/     # Repo maintenance scripts
├── README.md    # Main documentation (auto-generated sections)
└── AGENTS.md    # This file

Adding Content

Adding a Skill (with tooling)

1. Create folder in /skills:

   mkdir -p skills/my-skill
   

2. Create SKILL.md with frontmatter:

   ---
   name: my-skill
   description: What this skill does
   ---
   
   # My Skill
   
   Instructions, scripts, templates...
   

3. Add supporting files:

   • scripts/ — Helper scripts
   • templates/ — Template files
   • examples/ — Example inputs/outputs

Adding a Prompt (pure guidance)

1. Create folder in /prompts:

   mkdir -p prompts/my-prompt
   

2. Create SKILL.md with frontmatter:

   ---
   name: my-prompt
   description: Guidance for X
   ---
   
   # My Prompt
   
   Instructions and guidelines...
   

Prompts should be self-contained — no external dependencies or scripts.

Adding Clawdbot Skills

Same as skills, but in /clawdbot:

mkdir -p clawdbot/my-clawdbot-skill

SKILL.md Requirements

Required frontmatter:

• name: Identifier (lowercase, hyphens)
• description: One-line description

Good SKILL.md structure:

1. Quick usage examples at the top
2. Detailed instructions
3. Edge cases / troubleshooting
4. References to supporting files

Validation

Before committing:

./scripts/validate-skills.sh

Sync Rule (Required)

If you add or update a skill locally, you must also:

1. Add/update the corresponding skill folder in this repo
2. Update the README skill list (Skills + Clawdbot-specific)
3. Run ./scripts/validate-skills.sh
4. Open a draft PR with a clear What/Why/How/Testing description

Use clawdbot/skill-sync if you want a helper, but the steps above are still required.

Conventions

• Skill names: lowercase with hyphens (elegant-reports)
• Scripts should be executable (chmod +x)
• No hardcoded machine-specific paths
• Don't commit credentials or API keys