OpenClaw skills: how to find, install, and build your own

TLDR

Note: I use several ClawHub skills daily, and this guide covers everything I learned.

• OpenClaw (previously known as Clawdbot and Moltbot) skills are modular capabilities you can add to your agent, like plugins
• ClawHub is the marketplace where you browse and discover skills
• Install any skill with openclaw skill install skill-name from your terminal
• Update skills with openclaw skill update to get the latest versions
• You can create your own skills and publish them to ClawHub
• Security matters. Skills run with your agent's permissions, so review what they do before installing.

Table of contents

• What are OpenClaw skills
• Browsing ClawHub
• Installing skills
• Updating and managing skills
• Popular skills worth trying
• Creating your own skill
• Security considerations
• FAQ
• Sources
• Conclusion

Prerequisites

You need:

• OpenClaw installed and running (see installation guide)
• Command line access to your OpenClaw server
• A text editor for configuration files

Steps

What are OpenClaw skills

Skills are modular packages that extend what your OpenClaw agent can do. They work like browser extensions, but for your AI agent.

OpenClaw includes basic capabilities out of the gate: chat, shell commands, web browsing, file operations, and messaging. Skills add specialized capabilities on top. A humanizer skill rewrites AI-sounding text. A screenshot skill captures web pages. A calendar skill integrates with Google Calendar.

Each skill includes instructions, tool definitions, and configuration. When you install a skill, OpenClaw loads it automatically and the agent gains access to whatever that skill provides.

Skills provide instructions that teach your agent new capabilities. The agent executes the work using those instructions. This is different from plugins that run their own code. The skill defines what to do, and the agent figures out how to apply it in context.

Browsing ClawHub

ClawHub is the official marketplace for OpenClaw skills. It is where the community publishes and discovers skills.

You can browse ClawHub in two ways:

Through the website. Go to clawhub.ai and search or browse by category. Each skill listing shows what it does, installation instructions, version history, and community feedback.

Through the CLI. Search for skills from your terminal:

openclaw skill search humanizer
openclaw skill search screenshot
openclaw skill search calendar

The search results show the skill name, a short description, the author, and the version number.

When evaluating a skill:

• Check the last update date. Stale skills may not work with current OpenClaw versions.
• Look at the author's other published skills.
• Avoid skills with vague descriptions.
• Review what permissions the skill needs. Some skills require access to external APIs, file system, or network.

Installing skills

Installation is one command:

openclaw skill install humanizer

The CLI installs the skill to ~/.openclaw/skills/ and registers it with your agent. Your agent picks up the new skill on its next session.

For specific versions:

openclaw skill install humanizer@1.2.0

To install multiple skills at once:

openclaw skill install humanizer screenshot calendar-sync

After installation, verify the skill loaded by asking your agent: "What skills do you have installed?" Your agent will list the newly installed skill.

To remove a skill:

openclaw skill uninstall humanizer

This removes the skill files and deregisters it. The agent will no longer have access to that skill's capabilities after the next session.

If you are running OpenClaw on a VPS, our DigitalOcean setup guide covers the server environment where you would run these commands. You can also automate skill updates using OpenClaw cron jobs.

Updating and managing skills

Skills get updated by their authors with bug fixes, new features, and compatibility patches.

Update all installed skills:

openclaw skill update

Update a specific skill:

openclaw skill update humanizer

List installed skills and their versions:

openclaw skill list

This shows every installed skill, its current version, and whether an update is available.

If an update breaks something, roll back to a previous version:

openclaw skill install humanizer@1.1.0

This overwrites the current version with the specified older version.

Popular skills worth trying

Based on ClawHub community usage, these skills are good starting points:

Humanizer. Detects and rewrites AI writing patterns. Removes em dashes, copula avoidance, and other tells. If your agent produces any text that humans will read, this one is worth having.

Screenshot. Captures web pages as images using headless Chromium. Useful for blog hero images, monitoring dashboards, and documenting web-based workflows.

Web Fetch. Advanced web scraping that extracts clean text from web pages. Handles JavaScript-rendered content that basic HTTP requests miss.

Git Sync. Automates git operations on your workspace. Commit, push, and pull on a schedule or on demand. Pairs well with cron jobs for automated backups.

Check ClawHub for the full catalog. New skills get published regularly. If you have a morning briefing setup, skills like these can extend what your briefing pulls in.

Creating your own skill

If no existing skill does what you need, build one.

Skill structure

A skill requires two files:

my-skill/
├── skill.json       # Metadata and configuration
└── instructions.md  # What the skill teaches the agent

skill.json defines the skill:

{
  "name": "my-custom-skill",
  "version": "1.0.0",
  "description": "A short description of what this skill does",
  "author": "your-username",
  "tools": [],
  "config": {}
}

instructions.md is a markdown file that teaches the agent how to use the skill. Write it like you are explaining the capability to a person. Include examples, edge cases, and any required setup.

Adding tool definitions

If your skill needs custom tools (functions the agent can call), define them in the tools array of skill.json:

{
  "tools": [
    {
      "name": "my_tool",
      "description": "Does a specific thing",
      "parameters": {
        "type": "object",
        "properties": {
          "input": {
            "type": "string",
            "description": "The input to process"
          }
        },
        "required": ["input"]
      }
    }
  ]
}

The tool implementation can be shell commands, API calls, or file operations. The agent handles execution based on the tool definition and the instructions you wrote.

Testing your skill

Before publishing, test locally:

1. Copy your skill directory to ~/.openclaw/skills/
2. Restart OpenClaw or start a new session
3. Ask the agent to use the skill
4. Iterate on the instructions until the agent uses it correctly

Common issues: instructions the agent interprets differently than you intended, missing edge case handling, and tool parameters that do not match what the agent needs to send. Spend time on your instructions.md. That file does the heavy lifting.

Publishing to ClawHub

Once your skill works locally:

openclaw skill publish ./my-skill

This uploads your skill to ClawHub. You will need a ClawHub account. The skill goes through a basic validation check (valid JSON, required fields present) before publishing.

After publishing, anyone can find and install your skill. Keep it maintained. Respond to issues. Update it when OpenClaw releases new versions.

Refer to the OpenClaw documentation on skill development for the complete specification and advanced features like configuration schemas and dependency management.

Security considerations

Skills inherit your agent's permissions. If your agent has access to your file system, API keys, and messaging channels, every installed skill also has access. Our security hardening guide covers how to lock down your agent's permissions overall.

Review before installing. Read the skill's instructions.md and skill.json before installing. Understand what it does and what tools it defines. A "calendar sync" skill that also defines a tool for reading arbitrary files is suspicious.

Check the author. Skills from established authors with multiple well-reviewed skills are lower risk than anonymous one-off publications.

Use permissions wisely. If your OpenClaw instance supports permission scoping for skills, use it. Restrict skills to only the capabilities they need.

Monitor behavior. After installing a new skill, watch your agent's behavior for a few days. Check logs for unexpected API calls, file accesses, or messages.

Keep skills updated. Security patches are delivered through updates. Running outdated skills means running with known vulnerabilities.

The OpenClaw team reviews reported skills and can remove malicious packages from ClawHub. If you find a skill doing something suspicious, report it through ClawHub's reporting mechanism.

As the skill ecosystem grows, the incentive for malicious skills grows too. Treat skill installation with the same caution you would treat installing a browser extension or npm package.

Troubleshooting

Issue: Command not found

Make sure OpenClaw is installed and in your PATH. Run openclaw --version to verify.

Issue: Permission denied

You may need to run commands with appropriate permissions. Check file ownership and permissions in your OpenClaw directory.

Issue: Changes not taking effect

Restart the OpenClaw gateway after making configuration changes: openclaw gateway restart

FAQ

Q: Are skills free?

Most skills on ClawHub are free. Some authors offer premium skills with additional features. Check the individual skill listing for pricing details.

Q: Can skills break my OpenClaw setup?

A poorly written skill can confuse your agent or cause unexpected behavior. If a skill causes problems, uninstall it with openclaw skill uninstall skill-name and restart. The skill runs through your agent rather than modifying OpenClaw itself, so removing it cleans things up.

Q: Do I need to know how to code to create a skill?

Not necessarily. The simplest skills are just an instructions.md file that teaches the agent a new workflow. If you want custom tools with specific parameter schemas, some JSON knowledge helps. Full-featured skills with external integrations may require programming knowledge.

Q: How do I know if a skill is compatible with my OpenClaw version?

ClawHub skill listings show the minimum compatible OpenClaw version. The CLI will warn you if you try to install a skill that requires a newer version.

Q: Can I use multiple skills that do similar things?

Yes, but they may conflict if they define overlapping tool names or contradictory instructions. If two skills both define a "screenshot" tool, the agent may get confused about which to use. Stick to one skill per capability area.

Q: Where are skills stored on my machine?

Skills install to ~/.openclaw/skills/. Each skill gets its own subdirectory. You can inspect the files directly if you want to review what a skill does after installation.

Sources

• OpenClaw documentation
• ClawHub skill marketplace
• OpenClaw GitHub repository

Conclusion

Skills are what make OpenClaw extensible without being complicated. The install process is one command. The creation process is two files. The ecosystem is growing, and the best skills solve a specific problem well rather than trying to do everything.

I started by installing the humanizer skill because our content pipeline demanded it. Then I added screenshot capabilities for blog hero images. Then I built a custom skill for syncing data to our dashboard because nothing on ClawHub did exactly what I needed.

Three skills installed, one custom-built. My agent went from "can chat and run commands" to "can write clean prose, capture screenshots, and keep my dashboard current." All without touching the core OpenClaw configuration.