Skills System

Out of the box, OpenClaw handles a wide range of tasks. But every workflow has its own patterns, tools, and conventions. Skills let you extend OpenClaw with specialized instructions that load exactly when needed.

Think of skills as curated knowledge packs. They contain instructions, scripts, references, and sometimes assets that help the agent work in a specific domain. Weather, web search, video processing, health checks, SEO content creation. Each skill is a self-contained module that the agent reads when relevant.

TL;DR: Skills are specialized instruction packs that auto-load based on context. They live in the skills directory with a SKILL.md file and supporting scripts. You can use existing skills, create new ones, and chain them together for complex workflows.

How Skills Work

When you ask OpenClaw to do something, it checks if a skill matches the request. If it does, it reads the skill's SKILL.md file and follows those instructions for the task.

Skills live in the skills directory. Each skill is a folder with at minimum a SKILL.md file. The folder name is the skill identifier. Inside SKILL.md, you describe what the skill does, when to use it, and how to use it.

The agent scans available skills and matches them against your request. If your prompt mentions "weather", the weather skill loads. If you ask about "GitHub PRs", the GitHub skill loads. Matching is based on description and trigger phrases defined in the skill.

Skills can include supporting files: scripts, reference data, assets. A video processing skill might include an ffmpeg command reference. A web search skill might include API documentation. The SKILL.md file points to these resources.

Skill Directory Structure

A skill has a specific directory structure.

~/.openclaw/workspace/skills/ weather/ SKILL.md assets/ github/ SKILL.md scripts/ seo-content-brief/ SKILL.md templates/ examples/

SKILL.md is the entry point. It describes the skill, its triggers, and how to use it. Supporting files live in subdirectories within the skill folder.

When OpenClaw loads a skill, it reads SKILL.md and makes the supporting files available. The agent can reference scripts, read templates, or execute commands as described in the skill.

When Skills Auto-Load

Skills auto-load when the agent detects a match in your request. The matching logic looks at the skill description and trigger phrases.

If you say "get the weather in Singapore", the weather skill loads. If you say "create a PR for this branch", the github skill loads. If you say "write an SEO content brief for this keyword", the seo-content-brief skill loads.

You can also explicitly request a skill. "Use the weather skill to check the forecast" is explicit and always works.

Some skills are loaded proactively by the agent. During session startup, the agent might load relevant skills based on your workspace setup or recent context. TOOLS.md references specific skills, which gives the agent a hint about what might be needed.

The goal is that the right skill loads without you having to think about it. The agent should pick up the weather skill when you ask about weather, and the github skill when you ask about PRs.

Using Existing Skills

OpenClaw ships with a set of built-in skills. These cover common developer tasks: web search, weather, GitHub operations, tmux control, health checks, and more.

To use a built-in skill, just ask for what you want. The agent detects the match and loads the skill. You do not need to know which skill handles it.

For workspace-specific skills, make sure they are in the skills directory. If you have a custom skill for your company's deployment pipeline, put it in ~/.openclaw/workspace/skills/deploy/ and describe it in SKILL.md. The agent will pick it up when relevant.

Built-in skills are read-only. You can use them but not modify them. Custom skills in the workspace are fully editable. You own them and can change them to fit your workflow.

Creating New Skills

Create a skill when you have a repetitive workflow that would benefit from specialized instructions.

The process is straightforward. Create a folder in the skills directory. Write a SKILL.md that describes what the skill does, when to use it, and how. Add any supporting files the skill needs.

A simple SKILL.md structure:

# Skill Name

Description of what this skill does and when to use it.

## Usage

How to use this skill. Include examples.

## Tools

What tools or commands this skill uses.

## Examples

Example prompts or workflows.

The description field is important. It is what the agent uses to match your request to the skill. Make it clear and include common trigger phrases.

If your skill requires specific scripts, put them in a scripts/ subdirectory. If it needs templates, put those in templates/. Keep the skill self-contained.

Skill Chaining for Complex Tasks

Some tasks need more than one skill. Skill chaining is when you sequence skills to handle a multi-step workflow.

Example: "Research this competitor, create an SEO content brief, and draft the article." This involves web research, SEO brief creation, and content writing. Three skills, chained together.

The main agent orchestrates the chain. It detects that multiple skills are needed, loads them in sequence, and passes context between them. The research output feeds into the SEO brief, which feeds into the draft.

You do not need to explicitly name each skill in the chain. You describe the end goal and let the agent figure out which skills to use and in what order. "Write a technical blog post about this API" might chain the web search skill (for research) into the technical-blog-writing skill.

Skill chaining works best when each skill produces output that the next skill can consume. Clear interfaces between skills make this smooth. If a skill outputs structured data, the next skill can read that structure. If it outputs freeform text, the next skill extracts what it needs.

Maintaining Skills

Skills age. Your stack changes, your workflow evolves, tools get deprecated. Periodically review your custom skills and update them.

Check if the skill instructions still match how you actually work. If you changed your deployment process, update the deploy skill. If you switched from CircleCI to GitHub Actions, update the CI skill.

The skill maintenance loop is simple. Every few weeks, skim your custom skills. Ask: does this still reflect how I work? If not, update it.

Built-in skills are maintained by OpenClaw updates. Watch release notes for changes to built-in skills.

A well-maintained skill library makes OpenClaw progressively more valuable. Each skill you create is a permanent improvement to how the agent works for you.

Sources: OpenClaw Skills Documentation