Out of the box, OpenClaw can chat, search the web, read files, and run shell commands. That's useful — but the real power comes from the skill system. Skills are modular capability packs that teach your AI assistant how to interact with specific tools, APIs, and workflows. Think of them as plugins, except they're just markdown files with instructions.

What Are OpenClaw Skills?

A skill in OpenClaw is a directory containing a SKILL.md file — a structured document that tells the AI how to use a specific tool or perform a specific task. When you ask your assistant to do something, OpenClaw checks the available skills, finds the relevant one, reads its instructions, and follows them.

Skills aren't code in the traditional sense. They're closer to detailed runbooks. A GitHub skill might explain how to use the gh CLI to create issues, review PRs, and check CI status. A weather skill tells the assistant which API to call and how to format the response. The AI reads the skill, understands the tool, and executes accordingly.

This design means anyone can create a skill — you don't need to write a plugin in a specific language or conform to a framework. Write clear instructions in markdown, and your AI can follow them.

How Skills Get Loaded

OpenClaw scans skill directories on startup and builds a registry of available capabilities. When a request comes in, the system matches it against skill descriptions using semantic understanding — not just keyword matching. If you ask "check my GitHub issues," OpenClaw knows to load the GitHub skill even if you didn't say the word "skill."

Skills are loaded lazily. The system reads only the SKILL.md file for the matched skill at the time it's needed, keeping token usage efficient. If a skill references other files (scripts, templates, configs), those are resolved relative to the skill's directory.

The skill registry is visible in your session context. You can ask your assistant "what skills do you have?" and it will list every loaded capability with a short description.

Popular Skills Worth Installing

Here are the skills most users install first:

GitHub — Full GitHub integration via the gh CLI. Create and manage issues, open PRs, review code, check CI runs, and query repos. If you're a developer, this is non-negotiable. Your assistant becomes a code review partner that can triage issues and monitor your repos.

Weather — Current conditions, forecasts, and rain alerts for any location. Pairs well with the heartbeat system — your assistant can check the weather before you leave the house and send a heads-up via Telegram.

Exa Web Search — Neural web search that understands semantic meaning, not just keywords. Better than basic web search for research, finding similar pages, and extracting clean content from URLs. Free tier available.

n8n Integration — Connect your AI assistant to n8n workflows for complex automations. Trigger workflows from chat, pass data between systems, and build multi-step processes that your assistant can kick off on demand.

Brave Search — A lightweight alternative for web search and content extraction. Good for quick lookups without the overhead of a full research tool.

mcporter — Manage MCP (Model Context Protocol) servers. If you're connecting OpenClaw to external tools via MCP — databases, APIs, custom services — this skill handles the configuration and communication.

Installing Skills with ClawHub

ClawHub is the skill registry — think of it as npm for OpenClaw skills. The CLI makes installation straightforward:

Skills install to ~/.openclaw/workspace/skills/ by default. Each skill gets its own directory with the SKILL.md file and any supporting assets. You can also install skills manually by placing the directory in the right location — no build step required.

Creating Your Own Skills

The real unlock is building custom skills for your specific workflows. The format is simple:

A good SKILL.md includes a clear description (so the AI knows when to activate it), a list of what tools or APIs it uses, detailed instructions for each action, and error handling guidance. The more specific your instructions, the better the AI performs.

For example, a skill for managing your Shopify store might include instructions for authenticating with the Admin API, a list of common operations (inventory checks, order lookups, product updates), and templates for how to format responses. You're essentially writing a manual that an AI can follow.

MCP: The Connection Layer

Many skills use the Model Context Protocol (MCP) under the hood. MCP is a standard way for AI assistants to communicate with external tools and data sources. An MCP server exposes a set of tools (functions the AI can call), and the skill tells the AI how and when to use them.

The mcporter skill handles MCP server management — connecting, authenticating, listing available tools, and inspecting results. If you're building complex integrations with databases, APIs, or custom services, understanding MCP is worth the investment.

Skill Security Considerations

Skills can include shell commands and API calls, so trust matters. A few guidelines:

✓ Key Points

The skill system is what turns OpenClaw from a general-purpose chatbot into a tool that actually knows your stack. Start with the built-in skills, add a few from ClawHub, and build custom ones for the workflows that matter most to you.

Ready to install? → Get OpenClaw running on your server