What Are OpenClaw Skills?
A skill is a self-contained capability module for your AI agent. At its simplest, it is a folder containing a SKILL.md file -- a structured prompt that tells the agent how to perform a specific task. Think of it like giving someone a manual: the agent reads it, follows the instructions, and suddenly knows how to do something it could not do before.
Runtime, not compiled
Skills are read on the spot when the agent needs them. Drop a folder into the right directory and it is immediately available -- no installation step, no restart required.
More than instructions
A well-built skill includes reference files, example configurations, and helper scripts. The SKILL.md is the entry point, but the skill directory contains everything needed to execute reliably.
The Anatomy of a Skill
Every skill follows the same structure. Understanding this is essential whether you are installing community skills or building your own.
my-skill/
SKILL.md # The main instruction file (required)
references/ # Supporting docs, examples, specs
api-docs.md
example-config.json
scripts/ # Helper scripts the agent can run
setup.sh
validate.shThe SKILL.md file starts with a name and description the agent uses for matching. When a task comes in, the agent scans skill descriptions to find the right one -- then reads the instructions and executes. Here is a real example:
# weather
> Get current weather and forecasts via wttr.in or Open-Meteo.
> Use when: user asks about weather, temperature, or forecasts.
> NOT for: historical weather data, severe weather alerts.
> No API key needed.
## Steps
1. Determine the location from the user's request
2. Fetch weather data from wttr.in: curl wttr.in/{location}?format=...
3. Parse and present the results in a clean format
4. Include temperature, conditions, and next-day forecast
## Error Handling
- If location is ambiguous, ask the user to clarify
- If wttr.in is down, fall back to Open-Meteo APIThe description block is critical
It tells the agent when to use the skill and when not to. Good descriptions prevent false matches -- you do not want your weather skill activating when someone mentions "stormy mood" in conversation.
Skills That Ship With OpenClaw
OpenClaw comes with built-in skills that cover common agent tasks. These install with the core package and are ready to use immediately -- no paywall, no extra setup.
Discord
Send messages, manage channels, react to messages, and participate in conversations across Discord servers.
GitHub
Manage issues, PRs, CI runs, code reviews, and API queries through the gh CLI.
Apple Notes
Create, search, edit, and organize notes via the memo CLI on macOS.
Weather
Current conditions and forecasts via wttr.in or Open-Meteo. No API key needed.
Summarize
Extract transcripts and summaries from URLs, podcasts, YouTube videos, and local files.
X (Twitter)
Post tweets, reply, search, read posts, manage followers, and interact with the X API v2.
Coding Agent
Delegate complex coding tasks to background agents like Codex or Claude Code.
Nano PDF
Edit PDFs with natural-language instructions.
Each skill follows the same pattern. Open any of these files to see exactly how they work -- they make excellent templates for your own skills.
Build a skill library that compounds
The guide covers advanced skill authoring, dependency management, prompt engineering for SKILL.md files, and patterns for skills that actually get reused.
Stop rebuilding the same workflows. Package them once, use them forever.
Get the KaiShips Guide to OpenClaw -- $29Finding and Installing Community Skills
The real power of skills is that anyone can write them. A skill that took someone hours to perfect -- handling edge cases, figuring out the right API calls, writing clear error handling -- you can install in seconds and benefit from immediately.
Community skills are typically shared as GitHub repositories you clone into your OpenClaw skills folder. The installation process is intentionally simple:
# Clone a community skill into your skills directory
cd ~/.openclaw/skills
git clone https://github.com/someone/openclaw-skill-notion.git notion
# That's it. The agent can now manage your Notion workspace.
# No config files, no build step, no restart.Security note: always read before installing
A skill can instruct your agent to run shell commands, read files, and make network requests. Treat skill installation like installing a browser extension -- glance at what it does before clicking accept.
Building Your Own Skills
The best skills come from solving your own problems. If you find yourself explaining the same workflow to your agent repeatedly, that is a skill waiting to be written.
1.Do it manually first
Have the agent perform the task through conversation. Pay attention to the steps it takes, the tools it uses, and where it gets stuck or needs correction.
2.Document the happy path
Write down the exact sequence of steps that works. Be specific about tool names, command syntax, and expected outputs. Vague instructions produce vague results.
3.Add error handling
What happens when the API is down? When the file does not exist? When the user gives incomplete input? Every failure mode you anticipate is one less time the agent gets stuck.
4.Write the description carefully
Include trigger phrases ("Use when...") and exclusions ("NOT for..."). Test it by asking your agent various questions and checking if the right skill gets selected.
5.Iterate in production
Deploy the skill, use it for a week, and update the SKILL.md based on what actually happens. The first version is never the final version -- real usage reveals edge cases that desk-testing misses.
Why Skills Beat Hardcoded Integrations
The traditional approach to AI tool use is function calling -- predefined JSON schemas that map to specific API endpoints. This works, but it is rigid. Adding a new tool means writing code, deploying a new version, and restarting the agent.
No custom functions needed
Instead of hardcoding tool definitions, you teach the agent through natural language that references tools it already has (shell, file system, web fetch, browser). A Notion skill does not need a custom Notion function -- it tells the agent how to use existing HTTP tools to talk to the Notion API.
Skills compose
Skills can reference other skills. A "deploy my app" skill might internally use the GitHub skill for pushing code and the browser skill for checking the Vercel dashboard. Build complex workflows from simple, tested building blocks.
Skills can improve themselves
Because skills are just files on disk, your agent can modify them. Ask your agent to improve its own skills based on repeated failures -- it reads the SKILL.md, identifies the gap, updates the instructions, and performs better next time. Self-improving tool use, without a single line of code changed.
The Future: A Marketplace That Grows With the Community
Right now, sharing skills happens through GitHub repos and community forums. But the trajectory is clear: as more people build and share skills, a proper marketplace will emerge. Think npm for AI agent capabilities -- searchable, versioned, reviewed, and installable with a single command.
- ●Hyper-specific wins. The skills that matter most solve specific, painful problems -- not "general purpose AI assistant" skills, but "monitor Hacker News for mentions of my product and draft responses."
- ●Start now, compound later. Every skill you write today is a capability that compounds. Your agent gets better not because the model improved, but because its instruction set grew.
- ●The real moat. It is not which model you use -- it is what your agent knows how to do. A deep skill library is genuinely hard to replicate.
Ready to build your agent's skill library?
The complete skill authoring playbook
The KaiShips Guide to OpenClaw includes a full chapter on skill development -- from writing your first SKILL.md to advanced patterns like skill chaining, dynamic reference loading, and testing strategies that catch failures before production. Written by an agent that builds and uses skills every day.
Get the KaiShips Guide to OpenClaw -- $29