Agent Skills

Skills extend what agents can do. A skill is a SKILL.mdfile that the agent reads before starting tasks that match the skill's description. Think of them as specialised instruction sets.

Installing from ClawHub

ClawHub is the public registry for agent skills. Install any skill with one command:

openclaw skills install <slug>
# e.g.
openclaw skills install github
openclaw skills install summarize
openclaw skills install weather

Browsing installed skills

openclaw skills list

Skills install to ~/.openclaw/workspace/skills/<name>/SKILL.md.

Writing your own skill

A minimal skill has two parts:

skill.json — manifest

{
  "name": "my-skill",
  "version": "1.0.0",
  "description": "One sentence: what tasks does this skill handle?",
  "author": "you"
}

SKILL.md — instructions

# My Skill

## When to use this skill
Describe the trigger conditions clearly. The agent pattern-matches 
your task description against all skill descriptions to decide which 
SKILL.md to load.

## Steps
1. Step one
2. Step two
3. Verify and report

## Tools available
- `exec`: shell commands
- `web_fetch`: fetch a URL
- `message`: send a Telegram/Discord message

Keeping skills up to date

openclaw skills update --all

Asistry runs a weekly auto-update cron that upgrades all installed skills and vets any changes for security issues before applying them.

Security vetting

Before installing any skill from an external source, the built-in skill-vetter skill checks for red flags: subprocess calls, eval, network exfiltration, suspicious patterns. Skills from ClawHub are pre-vetted, but you can re-vet any skill at any time:

# Ask J.A.R.V.I.S to vet a skill before installing
"Vet the github skill from ClawHub before I install it"