Building Custom OpenClaw Skills

What Are Skills?

Skills are modular capability packages for OpenClaw agents. Each skill contains:

• SKILL.md — instructions the agent follows
• Scripts — helper scripts and tools
• References — documentation and examples
• Assets — any supporting files

Your First Skill

Let's create a simple skill that checks website uptime.

Directory Structure

skills/
  uptime-checker/
    SKILL.md
    check.sh

SKILL.md

# Uptime Checker

Check if a website is responding.

## Usage

When the user asks to check a website's status, use the `check.sh`
script in this skill's directory.

## Commands

- `bash check.sh <url>` — Returns HTTP status code and response time

check.sh

#!/bin/bash
URL="${1:?Usage: check.sh <url>}"
STATUS=$(curl -s -o /dev/null -w "%{http_code}" --max-time 10 "$URL")
TIME=$(curl -s -o /dev/null -w "%{time_total}" --max-time 10 "$URL")
echo "Status: $STATUS | Response time: ${TIME}s | URL: $URL"

Skill Best Practices

Keep Instructions Clear

Your SKILL.md is what the agent reads. Be explicit about: - When to use the skill - What commands are available - Expected inputs and outputs - Error handling

Use Relative Paths

Skills reference files relative to the skill directory. The agent resolves these automatically.

Make Skills Self-Contained

A good skill works without external dependencies. If it needs something, document it in the prerequisites section.

Advanced: Skills with Configuration

Some skills need per-user configuration. Store this in TOOLS.md:

### Uptime Checker
- Monitored sites: copypastelearn.com, lucaberton.com
- Check interval: 30 minutes
- Alert via: Discord #alerts channel

Sharing Skills

Package your skill and share it on ClawhHub. The community benefits from reusable, well-documented skills.

Built-in Skills

OpenClaw comes with several skills out of the box: - Weather — forecasts via wttr.in or Open-Meteo - Discord — channel management and messaging - Health Check — system security auditing - Skill Creator — meta-skill for building new skills

---

Ready to go deeper? Check out our hands-on course: OpenClaw Agent — practical exercises you can follow along on your own machine.

Further reading

To go deeper, the Ansible intelligent assistant and MCP server expands on these patterns in production.