OpenClaw Skills Documentation
OpenClaw Skills are local automation modules that extend your AI gateway with real actions. This documentation explains how to install OpenClaw Skills, understand permissions, and design skills that are safe, observable, and easy to reuse.
Quick start
Install the OpenClaw CLI and add your first skill
The fastest way to get started is to install the CLI, then add a skill by owner and repo. Every OpenClaw Skills entry provides its permissions and documentation before you run anything locally.
Install command
curl -fsSL https://openclawskills.best/install.sh | sh1. Install the CLI
Run the install command and confirm it completes without errors.
2. Add a skill
Use openclaw add @owner/repo to register a skill locally.
3. Review permissions
Read skill.md and verify required permissions before you run it.
What are OpenClaw Skills?
OpenClaw Skills are local automation modules that let your AI gateway run real actions on your machine. Each OpenClaw Skills entry ships with documentation, a permission manifest, and metadata so teams can verify what will run before execution.
The OpenClaw Skills directory is designed to make discovery and review easier, while keeping execution local and auditable. You can use skills for workflow automation, infrastructure tasks, or connecting APIs to your agent stack.
How OpenClaw Skills work
Discovery
Browse the directory, confirm the maintainer, and review skill.md for the scope of the automation.
Registration
Add a skill locally and store its metadata, documentation, and permissions for future runs.
Execution
Run the skill with explicit permissions and log outputs to keep automation observable.
skill.md format (recommended fields)
The skill.md file is the primary source of truth for an OpenClaw Skills entry. Use it to describe what the skill does, the permissions it needs, and how to install or run it. Keeping skill.md complete helps users validate a skill before executing any automation.
Common fields
- name
- description
- version
- permissions
- inputs
- outputs
- requirements
- examples
Permissions & security
OpenClaw Skills are reviewed through explicit permissions, so every skill lists the actions it needs before running. Teams can use security tiers (such as L1–L3) to summarize risk and approval requirements for each OpenClaw Skills entry.
L1 — Low risk
Read-only or minimal side effects. Good for quick automation.
L2 — Moderate
Writes data or calls external APIs. Requires review.
L3 — Sensitive
Touches production data or credentials. Strong approval needed.
Review the OpenClaw Skills Security Policy to understand disclosure and reporting guidelines.
Build an OpenClaw Skill
When you build a skill, focus on clear inputs, deterministic outputs, and minimal permissions. Treat skill.md as the contract between your automation and the person who runs it.
- Define a single purpose and list required inputs.
- Document outputs and side effects in skill.md.
- Limit permissions to the smallest safe set.
- Provide at least one runnable example.
Marketplace & discovery
The OpenClaw Skills directory helps you find automation by category, maintainer, or task. Use tags to discover integrations for databases, developer tools, or AI workflows.
FAQ
How do I install OpenClaw Skills?
Install the OpenClaw CLI, then add skills by owner and repo. Always review permissions listed in skill.md before running.
Are OpenClaw Skills safe to run locally?
Skills are designed to be permissioned. Review required permissions and security tier labels before executing any automation.
What should I include in skill.md?
Document what the skill does, inputs, outputs, and permissions. Include examples so users can validate behavior quickly.
Related links
Keep exploring the OpenClaw Skills ecosystem: