OpenClaw Skills Explained: Teaching Your AI New Abilities

TL;DR: Skills are plug-and-play ability packs for your AI assistant. Each skill bundles instructions, tools, and scripts into a folder. Drop in a weather skill and your AI can check forecasts. Drop in a GitHub skill and it can manage your repos. No code required to use them — but you can build your own.

What Are Skills?

Out of the box, your AI assistant can do a lot: chat, answer questions, write, analyze, reason. But it can't check the weather, manage your GitHub issues, or send a Slack message. Those require skills.

A skill is a self-contained package that teaches your AI how to do something new. It's not a plugin or an API integration in the traditional sense — it's more like giving your assistant a manual and a toolkit for a specific job.

Each skill is just a folder containing:

• SKILL.md — Instructions that tell the AI what this skill does and how to use it
• Scripts (optional) — Shell scripts, Node.js scripts, or Python scripts that the AI can execute
• Config (optional) — Settings specific to this skill

That's it. No complex plugin architecture, no compilation, no dependency hell. A skill can be as simple as a single markdown file with instructions, or as sophisticated as a full automation suite with multiple scripts.

How Skills Work Under the Hood

When OpenClaw starts, it reads all the SKILL.md files from your skills directory. These instructions get loaded into the AI's system prompt — the set of instructions that shapes every response.

So when you ask "What's the weather in Berlin?", the AI sees:

1. Its personality (SOUL.md)
2. Your context (MEMORY.md, USER.md)
3. The weather skill instructions (SKILL.md) — which say something like: "When the user asks about weather, run the weather.sh script with the location as an argument"
4. Your message

The AI reads the skill instructions, understands it has a weather tool available, decides to use it, runs the script, and formats the result for you.

It's the same process a human would follow if you gave them a manual: read the instructions, use the tools, deliver the result.

Built-in Skills

OpenClaw comes with several skills ready to use:

Weather

Ask about weather anywhere in the world. Current conditions, forecasts, severe weather alerts.

You: What's the weather like in Amsterdam this weekend?

AI: Amsterdam this weekend:
Saturday: 14°C, partly cloudy, 20% chance of rain
Sunday: 16°C, sunny, light wind from the west
Good weekend for being outside — maybe bring a light jacket Saturday morning.

The weather skill calls a weather API, parses the response, and gives you a natural-language summary instead of raw data.

GitHub

Manage repositories, issues, pull requests, and actions — all through chat.

You: What PRs are open on the api-server repo?

AI: 3 open PRs on api-server:
1. #142 "Add rate limiting middleware" by sarah — ready for review
2. #139 "Fix connection pool leak" by alex — 2 approvals, ready to merge
3. #137 "Update dependencies" by dependabot — failing CI

Want me to merge #139?

Browser

Browse the web, read pages, fill forms, take screenshots. The AI can research topics, check websites, and interact with web applications.

Tmux

Manage terminal sessions. Run long-running commands, check on processes, manage development environments — without leaving your chat.

Notion

Read and write to your Notion workspace. Create pages, update databases, search your knowledge base.

Slack

Send messages, read channels, manage threads in your Slack workspace.

Calendar

Check your schedule, create events, get reminders about upcoming meetings.

Email

Read, search, draft, and send emails. Triage your inbox without opening it.

Community Skills: ClawdHub

Beyond the built-in skills, there's a growing community creating and sharing skills at clawdhub.com.

Community skills cover everything from:

• Home automation (control smart lights, thermostats)
• Finance (check stock prices, crypto portfolios)
• Productivity (Todoist, Linear, Jira integration)
• Media (music control, podcast management)
• DevOps (Docker management, server monitoring)
• Social media (Twitter posting, analytics)

Installing a community skill is typically as simple as cloning a repo into your skills directory:

cd ~/.openclaw/skills/
git clone https://github.com/clawdhub/skill-todoist

Restart OpenClaw and the new skill is active.

Anatomy of a Skill

Let's look at what a skill actually contains. Here's the weather skill structure:

skills/
└── weather/
    ├── SKILL.md
    └── weather.sh

SKILL.md

# Weather Skill

You can check the weather for any location.

## Usage

When the user asks about weather, run:

    ./skills/weather/weather.sh <location>

The script returns JSON with current conditions and forecast.
Format the response naturally — don't dump raw JSON.

## Examples

- "What's the weather?" → Use their location from USER.md
- "Weather in Tokyo" → Use "Tokyo" as location
- "Will it rain tomorrow?" → Get forecast, focus on precipitation

weather.sh

#!/bin/bash
LOCATION="${1// /+}"
curl -s "wttr.in/${LOCATION}?format=j1"

That's a complete skill. A markdown file with instructions and a shell script that does the work. The AI reads the instructions, knows when and how to use the script, and handles the rest.

Example: The Weather Skill in Action

Here's a more detailed look at how a skill interaction flows:

1. You say: "Should I bring an umbrella today?"
2. AI reads: SKILL.md for the weather skill, sees it can check weather
3. AI decides: This is a weather question, I should use the weather skill
4. AI checks: USER.md says you're in Berlin
5. AI runs: ./skills/weather/weather.sh Berlin
6. Script returns: JSON with current conditions showing 40% chance of rain in the afternoon
7. AI responds: "Yeah, bring one. It's clear this morning but there's a 40% chance of rain after 2pm. Nothing heavy, but enough to be annoying if you're caught without one."

The AI doesn't just relay data — it interprets it in context (it knows you're asking because you're going out) and gives a practical answer.

Creating Your Own Skills

This is where it gets interesting. You don't need to be a programmer to create useful skills — many skills are just instruction files.

A Pure-Instruction Skill

Want your AI to always format meeting notes a specific way? Create a skill:

# Meeting Notes Skill

When the user says "meeting notes" or asks you to take notes on a meeting:

1. Use this format:
   - **Date**: [date]
   - **Attendees**: [names]
   - **Key Decisions**: [bullet list]
   - **Action Items**: [who → what → by when]
   - **Open Questions**: [unresolved items]

2. Always end with action items
3. Keep it concise — no fluff
4. Ask for clarification if the meeting topic is unclear

No scripts needed. The AI reads the instructions and follows the format. This is a skill.

A Script-Based Skill

For skills that need to interact with external services, add a script:

# Hetzner Server Status Skill

Check the status of Hetzner cloud servers.

## Usage

Run: `./skills/hetzner/status.sh [server-name]`

Returns server status, IP, and resource usage.
If no server name given, list all servers.

#!/bin/bash
# status.sh
curl -s -H "Authorization: Bearer $HETZNER_API_TOKEN" \
  "https://api.hetzner.cloud/v1/servers" | jq '.servers[] | {name, status, public_net}'

Requesting Custom Skills

Don't want to build it yourself? You can:

1. Check ClawdHub — someone might have already built what you need
2. Ask your AI — OpenClaw can often help you write the SKILL.md and scripts
3. Request on GitHub — open an issue on the OpenClaw repo describing the skill you want
4. Ask lobsterfarm — managed users can request custom skills as part of the service

Skills vs Plugins vs Integrations

How are skills different from what other platforms offer?

The fundamental difference: skills run on your machine with full access to your environment. Plugins and integrations are sandboxed third-party services. Skills can do anything your computer can do. Plugins can only do what the platform allows.

The Bottom Line

Skills are what transform a chatbot into a personal assistant. Without them, your AI can only talk. With them, it can actually do things — check weather, manage code, browse the web, control your infrastructure, and anything else you can script.

The skill system is deliberately simple: markdown instructions + optional scripts. No complex APIs, no approval processes, no vendor lock-in. If you can write a bash script, you can extend your AI assistant.

Don't want to manage server infrastructure? lobsterfarm provides managed OpenClaw hosting — deployment, updates, and support handled for you.

Get started with lobsterfarm → · Browse community skills →