OpenClaw Skills: Find, Install and Manage Them

What an OpenClaw skill is

A skill is a folder containing a SKILL.md file: a short markdown guide with a name and description at the top. When a skill is relevant, the agent reads it and follows its instructions using the tools it already has. OpenClaw follows the open AgentSkills format.

---
name: weather-brief
description: Short weather summary for a city, for morning briefings
metadata: { "openclaw": { "requires": { "bins": ["curl"] } } }
---
When the user asks for the weather or a morning briefing:
1. Fetch the forecast for their city.
2. Reply in 2 lines: temperature range and rain chance.
3. Warn if rain is above 60%.

Where skills come from

Want a gate for everyone on your machine? security.installPolicy runs your own policy command before any skill install continues, and blocks installs if the check fails. See the security guide.

Control which agent gets which skills

Where a skill is installed decides who can see it. An allowlist decides who does:

{
  agents: {
    defaults: { skills: ["github", "weather"] },  // shared baseline
    entries: {
      writer: { default: true },                  // inherits github, weather
      docs: { skills: ["docs-search"] },          // replaces the baseline
      "locked-down": { skills: [] },              // no skills
    },
  },
}

When two skills share a name, the higher source wins:

1. <workspace>/skillsWorkspace skills, the highest priority
2. <workspace>/.agents/skillsProject agent skills
3. ~/.agents/skillsPersonal agent skills
4. ~/.openclaw/skillsShared managed skills (--global)
5. …/workshop-skillsSkills learned in Skill Workshop
6. bundledShipped with OpenClaw
7. skills.load.extraDirsExtra folders and plugin skills, the lowest priority

An allowlist controls which skills an agent sees. It isn't a security boundary for shell access, so limit exec separately.

Create your own skill

The fastest way: ask your agent. "Every time I send a receipt photo, add it to my expenses sheet. Save that as a skill." Skill Workshop turns it into a proposal you review before it goes live:

openclaw skills workshop list
openclaw skills workshop inspect <proposal-id>
openclaw skills workshop apply <proposal-id>
openclaw skills workshop reject <proposal-id> --reason "Not reusable"

Or write one by hand with the generator:

Use {baseDir} in the body to point at files in the skill's folder. Give a skill its API key with skills.entries.<name>.apiKey. Note that these keys are injected on the host, not inside a sandbox.

How much do skills cost in tokens?

Every eligible skill adds its name, description and location to the system prompt. That's about 24 tokens per skill, plus the length of those fields, on every model call. Full instructions are only loaded when used.

An estimate: (97 characters + description + about 40 characters for name and location) ÷ 4 per skill. Too many skills? OpenClaw shortens descriptions to fit skills.limits.maxSkillsPromptChars. Trim unused skills with allowlists to save money. See pricing.

Fix common skill problems

OpenClaw skills questions