How to Install OpenClaw Skills: The Complete Guide

What Are OpenClaw Skills?

Skills are the building blocks that determine what your OpenClaw AI assistant can actually do. Out of the box, OpenClaw knows how to send and receive messages, route requests, and call AI models. Skills are what teach it to control Spotify, search GitHub, manage your Obsidian vault, book calendar events, transcribe audio, or run a cron-triggered daily briefing.

Each skill is a folder containing a SKILL.md file with YAML frontmatter that declares the skill's name, description, required binaries, and installation instructions — plus natural language instructions that tell the AI agent how and when to use the skill's tools.

As of v2026.2.21, OpenClaw ships with 53 bundled skills and hundreds more are available through ClawHub, the community skill registry. You can also build and load your own.

The Three Skill Locations (and Why Order Matters)

OpenClaw resolves skills from three locations, each with a defined priority. When two skills share the same name, the highest-priority location wins:

1. Workspace skills — <your-workspace>/skills/ — highest priority, project-specific
2. Managed/local skills — ~/.openclaw/skills/ — user-level, installed via CLI or ClawHub
3. Bundled skills — shipped inside the openclaw package — lowest priority, always available

You can also register additional directories via the skills.load.extraDirs config key if your workflow requires it — for example, pointing at a shared team skills library on a network mount or a monorepo subfolder.

Understanding this order is important: if you install a custom version of a bundled skill (say, a modified github skill) into your workspace, it silently overrides the bundled one. This is a feature, not a bug — but it can cause confusion if you forget you have a local override in place.

Bundled Skills: What You Already Have

The 53 bundled skills cover the most common automation surfaces. They are pre-loaded and available immediately — no installation required — but many depend on external binaries or API keys that you need to supply:

Productivity & Notes

• apple-notes, apple-reminders — macOS-only; uses AppleScript
• bear-notes — Bear app via x-callback-url scheme (macOS/iOS)
• notion — Notion API; requires NOTION_TOKEN env var
• obsidian — Obsidian vault via Local REST API plugin
• things-mac — Things 3 via URL scheme (macOS only)
• trello — Trello API; requires API key + token

Developer Tools

• github — GitHub operations via gh CLI (issues, PRs, CI runs, code review)
• gh-issues — GitHub Issues focused skill
• coding-agent — Spawn a sub-agent for code generation tasks
• skill-creator — Meta-skill: helps you build new skills interactively
• tmux — Terminal session management via tmux
• session-logs — Access and search conversation session logs

Search & Media

• spotify-player — Playback control via spogo or spotify_player CLI
• songsee — Song identification from audio
• gifgrep — GIF search via Tenor API
• goplaces — Location search and mapping
• video-frames — Extract and analyse frames from video files
• nano-pdf — Read and summarise PDF files
• summarize — Summarise long text, articles, or documents

AI & Voice

• openai-image-gen — DALL·E image generation via OpenAI API
• openai-whisper — Local Whisper transcription (requires Python + model)
• openai-whisper-api — Whisper via OpenAI API (no local model needed)
• sherpa-onnx-tts — Fully local text-to-speech using Sherpa ONNX (no cloud required)
• gemini — Direct Gemini model calls for multimodal tasks
• voice-call — Realtime voice conversation over phone/VoIP

Communication & Integrations

• discord, slack, imsg — Channel-specific actions (send, react, search)
• himalaya — Email management via the Himalaya CLI
• 1password — Credential lookup via op CLI (read-only by default)
• oracle — Oracle Database query skill
• healthcheck — HTTP endpoint health monitoring
• model-usage — Local AI model usage and cost reporting

Method 1 — Enable a Bundled Skill

Bundled skills load automatically, but many need external dependencies. Here is the workflow:

1. Check what a skill needs — ask your agent: "What does the spotify-player skill require?" or open skills/spotify-player/SKILL.md and read the requires block in the frontmatter.
2. Install the dependency. Most skills list Homebrew, apt, or npm install commands. For Spotify control for example: brew install spogo.
3. Add the required API key to your .env or openclaw.json config if the skill calls an external API.
4. Reload or restart OpenClaw. Skills are loaded at startup. Run /reload in your chat interface if your deployment supports hot reload, or restart the process.
5. Test it. Send a message that would invoke the skill — for example, "Play Radiohead on Spotify." If the skill loads correctly, the agent will use it automatically.

No YAML config change is required to use bundled skills — they are on by default. You only need to disable one explicitly if it conflicts with a custom skill you are building.

Method 2 — Install a Community Skill from ClawHub

ClawHub (clawhub.ai) is the official community registry for OpenClaw skills — the place where the project now directs all new community-contributed skills to keep the core lean. Installing from ClawHub requires the clawhub CLI:

# Install the ClawHub CLI (one-time)
npm install -g clawhub

# Search for skills
clawhub search <keyword>

# Install a skill into ~/.openclaw/skills/
clawhub install <skill-name>

# Update a specific skill
clawhub update <skill-name>

# Update all installed ClawHub skills
clawhub update --all

# List your installed skills
clawhub list

Installed skills land in ~/.openclaw/skills/ by default and are immediately visible to all your OpenClaw workspaces. If you want a skill scoped to a single project only, pass --dir ./skills to install it into your workspace folder instead.

The clawhub skill itself is a bundled meta-skill — meaning once OpenClaw is running, you can also ask your agent directly: "Search ClawHub for a weather skill and install it." The agent will use the clawhub CLI on your behalf.

Method 3 — Load a Workspace Skill (Custom or Team)

Workspace skills live in a skills/ subfolder inside your OpenClaw workspace directory. This is the right approach for:

• Skills you build yourself for internal tooling
• Modified versions of bundled skills you want to override
• Team skills checked into a shared repository

The minimum viable skill structure is:

skills/
└── my-custom-skill/
    └── SKILL.md

A minimal SKILL.md looks like this:

---
name: my-custom-skill
description: "Fetches live inventory data from our internal API."
metadata:
  {
    "openclaw": {
      "emoji": "📦",
      "requires": { "bins": ["curl"] }
    }
  }
---

# My Custom Skill

Call our inventory API:

```bash
curl -H "Authorization: Bearer $INVENTORY_TOKEN" https://api.internal/inventory
```

Use this skill when the user asks about stock levels, product availability,
or inventory status. Always show the result as a structured table.

OpenClaw discovers workspace skills at startup. No config registration is needed — it scans the skills/ directory automatically. Use the skill-creator bundled skill to interactively scaffold new skills from within your chat interface.

Skill Dependencies: Automatic vs Manual Installation

When OpenClaw loads a skill, it reads the requires block in the skill's YAML frontmatter and checks whether the listed binaries exist on your system. If they do not, it can attempt automatic installation depending on your platform and config:

• Homebrew (kind: brew): Installs via brew install <formula> — macOS/Linux with Homebrew
• npm/pnpm/yarn/bun (kind: node): Installs the package globally
• apt (kind: apt): Runs apt-get install — Debian/Ubuntu Linux
• Manual (kind: manual): Displays a URL or instructions; no auto-install

Auto-install requires that OpenClaw has permission to run the relevant package manager. In Docker deployments, this usually means the container image includes the package manager and the process has sufficient privileges. In production setups it is often cleaner to pre-install all skill dependencies in your Dockerfile or server provisioning script and disable auto-install.

Checking Which Skills Are Active

Ask your agent: "List all active skills" or run /status in a terminal session. The status output includes loaded skills, their source location (bundled/managed/workspace), and whether all required dependencies were found. Skills with missing dependencies load with a warning and their tools are unavailable until the dependency is installed.

Disabling a Skill

To disable a bundled or managed skill, add it to the skills.load.exclude list in your openclaw.json:

{
  "skills": {
    "load": {
      "exclude": ["food-order", "nano-banana-pro"]
    }
  }
}

This is useful when you want to prevent the agent from using a skill in a specific deployment — for example, disabling voice-call in a text-only workspace, or disabling openai-image-gen to control API spend.

Common Setup Issues

• Skill not triggering: The agent will only invoke a skill if the description in SKILL.md matches the user's intent. Vague or overly broad descriptions cause the agent to skip the skill. Make the description specific about when to use it and when not to.
• Binary not found: If the auto-install step fails silently, manually run the install command from the install block in the SKILL.md frontmatter and verify which <binary> returns a path.
• Wrong skill version overriding bundled: Run /status and check the source path next to the skill name. If you see a workspace path instead of the bundled path, you have a local override — intentional or not.
• API key not being picked up: Skills that call external APIs read keys from environment variables. Make sure you have set the variable in your .env file and that OpenClaw loaded it at startup (check OPENCLAW_LOAD_SHELL_ENV=1 if you're using shell env).