Clawdi Store

Template store for agents and skills on the Clawdi platform.

Structure

agents/           Agent templates (agent.json + workspace files)
skills/           Standalone skill templates (skill.json + SKILL.md)
scripts/          CI validation scripts

Adding an Agent

1. Create agents/your-agent-name/ directory
2. Add agent.json manifest (see schema below)
3. Add IDENTITY.md with name + emoji (must match agent.json)
4. Add SOUL.md with agent personality
5. Add skills under skills/your-skill/SKILL.md
6. Submit a PR — CI validates automatically

agent.json Schema

{
  "version": 1,
  "name": "My Agent",
  "emoji": "🤖",
  "description": "What this agent does (shown in catalog card)",
  "headline": "Short one-liner (shown in catalog list)",
  "category": "productivity",
  "author": "Your Name",
  "tags": ["tag1", "tag2"],
  "featured": false,
  "trust_level": "community",
  "languages": [],
  "suggested_prompts": [
    "How can you help me?",
    "Show me what you can do"
  ],
  "model": "claude-sonnet-4-6",
  "skills": {
    "inline": ["my-bundled-skill"],
    "remote": [
      {
        "slug": "external-skill",
        "source": "git",
        "url": "https://github.com/owner/repo",
        "subdir": "skills/external-skill"
      }
    ]
  }
}

Field	Required	Type	Description
version	yes	int	Increment on updates. Frontend detects available upgrades.
name	yes	string	Display name (must match IDENTITY.md)
emoji	yes	string	Display emoji (must match IDENTITY.md)
description	yes	string	Full description for catalog detail page
headline	recommended	string	Short one-liner for catalog card
category	yes	string	e.g. productivity, finance, marketing
author	yes	string	Creator name
tags	yes	string[]	Searchable tags
suggested_prompts	recommended	string[]	Chat welcome prompts (max 4)
featured	recommended	bool	Show in featured section
trust_level	recommended	string	community, verified, or official
languages	recommended	string[]	Language codes, empty = all
model	optional	string	Suggested LLM model
skills.inline	optional	string[]	Skill directories under skills/
skills.remote	optional	object[]	External skill references

IDENTITY.md

Must contain name and emoji matching agent.json:

- Name: My Agent
- Emoji: 🤖

Directory Layout

agents/my-agent/
├── agent.json          # Manifest (required)
├── IDENTITY.md         # Name + emoji (required, must match agent.json)
├── SOUL.md             # Agent personality (recommended)
├── USER.md             # User context
├── README.md           # Shown on catalog detail page
├── skills/
│   └── my-skill/
│       ├── SKILL.md    # Skill definition (required per skill)
│       └── scripts/    # Implementation files
└── ...

Adding a Skill (Standalone)

1. Create skills/your-skill-name/ directory
2. Add skill.json manifest
3. Add SKILL.md with skill definition
4. Submit a PR

skill.json Schema

{
  "version": 1,
  "name": "Web Search",
  "emoji": "🔍",
  "description": "Search the web",
  "category": "productivity",
  "author": "Clawdi"
}

How It Works

The Clawdi backend maintains a local shallow clone of this repo. Every 5 minutes:

1. git pull to fetch latest changes
2. Scan agents/*/agent.json → upsert agent catalog DB
3. Scan skills/*/skill.json → upsert skill catalog DB

Agent installation tars the entire directory and extracts it to the CVM workspace in a single RPC call.

CI Validation

Every push and PR runs scripts/validate.py which checks:

• Required fields in agent.json / skill.json
• Type correctness (version=int, tags=array, etc.)
• IDENTITY.md exists and matches agent.json name/emoji
• Inline skills have SKILL.md
• No orphan skill directories

See CONTRIBUTING.md for contribution requirements and OPEN_SOURCE_RELEASE.md for the public release checklist.

Security and Publishing

• Keep committed USER.md files as placeholders only. Do not commit populated runtime profiles, credentials, wallet data, session files, local databases, or generated logs.
• Review NOTICE when adding vendored third-party assets.
• See SECURITY.md for vulnerability reporting, secret handling, and financial workflow guidance.

License

MIT. See LICENSE and NOTICE.