OpenClaw Skills are reusable capability bundles that extend what the OpenClaw AI assistant can do. Each Skill packages instructions, prompt templates, and optional tool definitions that the AI loads on demand. Think of them as the OpenClaw equivalent of Claude Skills or ChatGPT plugins — but they work with whichever AI model you have configured, including local Ollama models.

This guide covers what Skills are, how they work under the hood, how to install and manage them in the OpenClaw Desktop App, the recommended starter Skills for common use cases, and how to author your own.

What is an OpenClaw Skill?

An OpenClaw Skill is a small package — usually a single directory — containing:

  • A SKILL.md manifest with frontmatter (name, description, trigger conditions) and the natural-language instructions the AI follows when the Skill is active.
  • Optional asset files — templates, scripts, reference documents, or other resources the Skill needs.
  • Optional tool declarations — references to MCP servers or built-in tools the Skill can invoke.

When the AI assistant decides a Skill is relevant to the current message, it loads the Skill's instructions into the active context. The user does not need to manually toggle Skills on or off for each conversation — the assistant's description field acts as a trigger, similar to how Claude Code skills work.

Why use Skills?

Three reasons:

  1. Consistency — instead of pasting the same long prompt into every conversation, a Skill encodes that prompt once and the assistant uses it whenever the trigger matches.
  2. Composability — Skills can be combined. Install three Skills and the assistant picks whichever is relevant per message.
  3. Portability — a Skill is just a directory. You can share it on GitHub, install it in any OpenClaw instance, or version-control it alongside your project.

How to install a Skill in OpenClaw Easy

Skills are managed from the OpenClaw Desktop App's Skills section:

  1. Open the OpenClaw Easy desktop app.
  2. Click the Skills tab in the left sidebar.
  3. Browse the catalog of available Skills, or click Add from URL to install a Skill from a public Git repository.
  4. Click Install on the Skill you want. The app downloads the Skill, registers it with the local gateway, and the next message you send will see the Skill available.

No terminal, no npm install, no manual file editing. The Skill becomes available across every channel you have connected — WhatsApp, Telegram, Slack, Discord, Feishu, Line, and the built-in desktop chat.

Recommended starter Skills

If you are new to OpenClaw Skills, these are good starting points. The exact catalog evolves over time — check the in-app Skills browser for the current list.

Web search

Lets the assistant fetch live web pages and search results when a question requires current information. Useful for "what is happening in X today" or "latest version of Y". Typically wraps a public search API or a self-hosted search service.

File operations

Gives the assistant tools to read, write, and search files on your local machine. Pairs well with the Codex / coding Skills for repository work.

Code execution

Runs short Python or JavaScript snippets in a sandbox. Useful for math, data transformation, or quick prototyping inside a conversation.

Calendar / scheduling

Connects the assistant to a calendar (Google Calendar, iCal) so you can ask "what is on my calendar this week" or "schedule a meeting for Thursday at 3pm".

Channel-specific personalities

Custom system prompts for specific channels. For example, a Skill that makes the assistant terse and professional on the work Slack, casual on the personal WhatsApp.

Domain experts

Skills that prime the assistant with deep knowledge of a specific domain — legal contract review, medical terminology, recipe authoring, etc. Activated by trigger keywords in the user message.

How to create your own Skill

The minimum Skill is a single SKILL.md file:

---
name: my-skill
description: When to use this skill — be specific so the AI auto-matches it correctly.
---

# My Skill

Instructions for the assistant when this skill is active.

You can reference files, prescribe formats, set tone, or
explain domain context here.

Steps:

  1. Create a directory named after your Skill (use kebab-case, e.g. my-skill).
  2. Add SKILL.md with the frontmatter and instructions.
  3. In OpenClaw Easy, go to Skills → Add Local Skill → point at the directory.
  4. The Skill is now active. Send a message that matches the description trigger and the assistant will use your instructions.

Tip: The description field is the most important part of a Skill. It is how the AI decides whether to invoke your Skill for a given message. Write it as a clear "when to use this" sentence — specific enough that the AI does not invoke it for unrelated queries, but broad enough that legitimate triggers actually match.

Skills vs. MCP servers

OpenClaw supports two complementary extension mechanisms:

  • Skills — packaged instructions and prompt templates. Lightweight, no separate process, easy to author in markdown. Best for: domain expertise, tone/format guidelines, workflow templates.
  • MCP servers — separate processes that expose tools to the AI over the Model Context Protocol. More powerful: stateful, can do I/O, can call external APIs. Best for: filesystem access, database queries, custom integrations.

You can use both. A Skill can declare that it depends on an MCP server, and the assistant will load the MCP tools when the Skill is active. For most users, Skills are the right starting point — they cover 80% of customization needs with 10% of the complexity.

Sharing and discovering Skills

OpenClaw Skills are designed to be shareable. Common distribution patterns:

  • GitHub repos — publish a Skill as a public repo, share the URL, anyone can install via "Add from URL"
  • In-app catalog — curated Skills available directly in the Skills browser
  • Team-internal — share a Skill directory via internal Git or shared folder for team-specific workflows

The Skills format is open. There is no proprietary marketplace lock-in — a Skill that works in OpenClaw Easy works in any OpenClaw-compatible runtime.

Frequently asked questions

Do OpenClaw Skills cost money?

The Skills framework is free and open-source (Apache-2.0). Individual Skills published by third parties are usually free, but a Skill could in principle be commercial — the framework does not prevent it. Skills do not call any paid service on their own; they just shape the prompt the AI sees, so the only AI cost is whatever your configured provider charges per token.

Can a Skill access my files or data without permission?

No. A pure-instructions Skill cannot read files or make network calls — it only changes how the AI responds. If a Skill needs file or network access, it must declare a dependency on an MCP server (or built-in tool), and OpenClaw Easy will prompt for permission the first time the tool is invoked.

How many Skills can I install?

There is no hard limit. The assistant only loads the Skill's instructions into context when its description matches the current message — so installing 50 Skills does not bloat every conversation. Only the relevant ones are activated per message.

Do Skills work offline?

Yes, if you are using a local model via Ollama. The Skill itself runs inside the OpenClaw runtime on your machine — no external service needed. With cloud AI providers (Claude / GPT / Gemini), you need an internet connection for the AI call itself, but the Skill loading and activation happen locally.

Can I disable a Skill without uninstalling it?

Yes. Skills have a per-Skill enabled toggle in the Skills section. Toggle it off and the Skill stays installed but the assistant ignores it until you re-enable.

Next steps