HomeDocumentation

Documentation

Everything you need to know about LarryBrain skills and OpenClaw agents.

What is OpenClaw?

OpenClaw is an AI agent platform built on top of Claude by Anthropic. It gives you a personalized AI agent that remembers things, uses tools, integrates with your services, and works alongside you - over chat, voice, or API.

Unlike ChatGPT or Claude.ai, OpenClaw agents are fully customizable with persistent memory, configurable personalities, and can be extended with skills.

✨ Agent capabilities

✓ Persistent memory across sessions
✓ Tool use (web, calendar, email, code, images)
✓ Skill installation from LarryBrain marketplace
✓ Custom personality & instructions
✓ Multi-channel (WhatsApp, Discord, Telegram, CLI)

What are Skills?

Skills are knowledge packs that teach your OpenClaw agent new abilities. Think of them like apps for your AI - each skill bundles everything the agent needs to become an expert in a specific domain.

A skill can include any combination of:

→SKILL.md - the main instruction file that tells the agent what to do, how to behave, and what tools to use
→Reference files - docs, examples, templates, best practices the agent can read
→Scripts - shell scripts or tools the agent can execute
→Configuration - API endpoints, default settings, environment requirements

Skills are composable - install as many as you want and they stack together, giving your agent multi-domain expertise. An agent with a marketing skill, a GitHub skill, and a weather skill can do all three.

Skill Structure

Every skill lives in a folder with a SKILL.md file at the root. This is the only required file - everything else is optional.

my-awesome-skill/
├── SKILL.md          # Required - main instructions
├── reference/        # Optional - docs, examples
│   ├── api-docs.md
│   └── examples.md
├── scripts/          # Optional - executable tools
│   └── fetch-data.sh
└── templates/        # Optional - reusable templates
    └── report.md

Writing SKILL.md

The SKILL.md file is what OpenClaw reads to understand your skill. Write it as if you're explaining to a smart colleague what to do. Include:

Name and description - What the skill does in 1-2 sentences. This appears in the marketplace.
When to use - Tell the agent when this skill is relevant. E.g. "Use when the user asks about TikTok marketing or wants to create slideshow posts."
How it works - Step-by-step instructions. Be specific - the agent follows these literally. Include tool usage, API calls, file paths, and expected outputs.
Configuration - Any API keys, credentials, or settings the user needs to provide. Use environment variables or a config file.
Examples - Show concrete examples of user requests and how the agent should respond. The more examples, the better.
Constraints - Things the agent should NOT do. Rate limits, safety rules, platform-specific rules.

💡 Pro tip

Test your skill on your own OpenClaw agent before submitting. Drop the folder into your workspace skills/ directory and try it. If your agent can follow the instructions, users' agents will too.

Example SKILL.md

# Weather Skill

Get current weather and forecasts for any city worldwide.

## When to use
Use when the user asks about weather, temperature, forecasts,
or whether they need an umbrella.

## How it works
1. User asks about weather for a location
2. Fetch data from wttr.in (no API key needed):
   `curl -s "wttr.in/{city}?format=j1"`
3. Parse the JSON response
4. Present: current temp, conditions, wind, humidity
5. If they ask for forecast, show next 3 days

## Examples
User: "What's the weather in London?"
→ Fetch wttr.in/London, present current conditions

User: "Do I need a jacket tomorrow?"
→ Fetch forecast, check tomorrow's temp + rain chance

## Constraints
- wttr.in rate limit: ~100 requests/hour
- Always include the data source
- Use metric units unless user specifies otherwise

The LarryBrain Skill

The LarryBrain skill is the gateway to the entire marketplace. Install it and your agent gains the ability to search, discover, and install skills automatically.

When your agent doesn't know how to do something, it searches LarryBrain for a skill that can help. It explains what each skill does and installs the best fit - all through natural conversation.

How it works

1. You ask your agent for help with something it doesn't know
2. The LarryBrain skill searches the marketplace API for relevant skills
3. Your agent explains what's available, shows ratings and install counts
4. If you're subscribed, it installs the skill instantly
5. If you're not subscribed, it shows you how to sign up

Skills are served on-demand via API - they're not downloaded permanently to your machine. Your agent checks subscription status every time it uses a premium skill. This ensures creators get paid for every active user and prevents piracy.

Installing Skills

Two ways to install skills:

Option 1: Ask your agent

If you have the LarryBrain skill installed, just ask naturally:

"I need help with TikTok marketing, find me a skill for that"

Option 2: Ask your agent directly

"Install the weather oracle skill from LarryBrain"

Your agent checks LarryBrain first when you ask it to install any skill.

📋 Free vs Premium

Some skills are free and available to everyone. Premium skills require a Pro subscription ($29.99/mo).

Publishing Skills

Share your skills with the community and earn 50% of subscription revenue from every user who joins through your referral link - forever.

Requirements

→GitHub account - must be at least 2 months old (spam prevention)
→Active subscription - creators must be active subscribers
→Public GitHub repo - with a SKILL.md file in the root directory
→1-3 categories - from the fixed list
→Pass security review - automated scan + team review

Step-by-step submission

1
Build your skill
Create a folder with a SKILL.md file. Add any reference files, scripts, or templates your agent needs. Test it on your own OpenClaw agent first.
2
Push to GitHub
Create a public repository and push your skill folder. The SKILL.md must be at the root of the repo.
3
Go to the Publish page
Sign in with Google or GitHub, connect your GitHub account, and make sure you have an active subscription.
4
Fill in the details
Name your skill, write a clear one-line description, pick an emoji icon, and choose 1-3 categories. Paste your GitHub repo URL.
5
Submit for review
Hit submit. We'll pull your SKILL.md, run an automated security scan, and queue it for team review.
6
Wait for approval
Reviews take 1-3 business days. You'll get an email when approved (or if changes are needed).
7
Go live
Once approved, your skill appears in the marketplace. Share your referral link to earn 50% of new subscriptions.

💰 Earnings

You get a unique referral link when your skill is published. When someone signs up for LarryBrain Pro through your link, you earn 50% of their subscription fee every month, forever. No cap. Payouts on the 15th of each month.

Security Review

Every submitted skill goes through a two-stage security review before it appears in the marketplace. This protects users from malicious or poorly-written skills.

Stage 1: Automated Scan

Our automated scanner checks the skill repo for:

✓
Data exfiltration patterns - Attempts to send user data to external servers - curl/wget to unknown URLs, encoded payloads, base64-encoded URLs
✓
Destructive commands - rm -rf, format, dd, or anything that could damage the user's system
✓
Credential harvesting - Instructions that trick the agent into reading and sending API keys, passwords, or tokens
✓
Privilege escalation - Attempts to modify system files, install packages, change permissions, or escape the workspace
✓
Hidden instructions - Prompt injection attempts - instructions hidden in comments, invisible characters, or encoded text designed to override the agent's safety rules
✓
Network abuse - Scripts that make excessive API calls, run DDoS patterns, or abuse external services
✓
Obfuscated code - Minified scripts, encoded payloads, or deliberately obscured logic that hides what the skill actually does

Stage 2: Team Review

A human reviewer then checks:

✓
Does it do what it claims? - The description must honestly match what the SKILL.md instructs the agent to do
✓
Quality bar - Instructions must be clear enough for an agent to follow. Vague or broken skills get sent back with feedback
✓
No ToS violations - Skills can't enable spam, harassment, fraud, impersonation, or violate any platform's terms of service
✓
Appropriate categories - Categories must accurately reflect the skill's functionality

Additional safeguards

✓GitHub account age - must be 2+ months old to publish (blocks throwaway spam accounts)
✓Subscription required - creators have skin in the game
✓Version tracking - every update goes through review. We track git commits so skills can't change after approval
✓Community reports - users can flag skills. Flagged skills are suspended pending re-review

If your skill is rejected

You'll receive an email explaining what failed and why. Fix the issues, push to your repo, and resubmit. Most rejections are for quality issues (unclear instructions), not security - and are easy to fix.

Guidelines

✅ Allowed

Productivity and automation skills
Skills wrapping public APIs (weather, GitHub, etc.)
Educational and reference skills
Creative tools (writing, design, media)
Developer tools and DevOps automation
Fun, entertainment, and games

❌ Not allowed

Skills that enable spam or manipulation
Skills that bypass security systems or safety rules
Misleading descriptions (bait and switch)
Skills that violate any platform's Terms of Service
Credential harvesting or data exfiltration
Skills that impersonate other people or agents