How to Install OpenClaw Skills

What Are OpenClaw Skills?

A skill gives your OpenClaw AI agent a capability beyond plain conversation. Under the hood a skill is a folder containing a SKILL.md file — a Markdown document with YAML frontmatter that tells the agent what the skill is for and walks it through the procedure step by step. A skill can also bundle helper files such as scripts, templates, or a setup script.

That is worth being precise about, because skills and plugins are different things in OpenClaw. A plugin is executable code registered under plugins.entries in your config, and it is how channels like Telegram or WhatsApp are wired in. A skill is instructions plus optional files that the agent reads and follows. Installing a skill does not mean editing plugins.entries.

OpenClaw ships with a large set of bundled skills already in the box, and you add more from ClawHub, the official skill registry at clawhub.ai. Skills cover web search, image generation, GitHub and Notion access, browser automation, transcript reading, writing assistance, and much more.

Think of skills like apps on your phone — your agent starts with the basics, and you install exactly the capabilities you need.

Where to Find Skills

There are three main ways to discover OpenClaw skills:

• ClawHub — The official skill registry at clawhub.ai. Browse by owner and slug, read what each skill does, and install it by reference. Skill refs look like @owner/slug.
• openclaw skills search — Search ClawHub straight from your agent's terminal. Run openclaw skills search web search to find search-related skills, or add --limit 10 to cap the results.
• OpenClaw Launch Skills page — Browse our curated marketplace of 4,000+ skills at openclawlaunch.com/skills, with one-click install onto a managed instance.

How to Install a Skill

There are three methods to install a skill, depending on your setup:

Method 1: One-Click on OpenClaw Launch

If you're using OpenClaw Launch, this is the shortest path and needs no command line. Deploy an instance first, then open the Skills page in your dashboard, pick the instance you want from the selector, and click Install. The skill's files are written into that container's skills directory and the agent picks them up. The same card flips to Uninstall when you want it gone.

Skills that talk to an external service still need their own credential — the install button does not ask for one. Follow the setup steps inside the skill itself, which usually means adding a key under skills.entries in your config or signing in through a CLI the skill wraps. See Skill Configuration below.

Method 2: openclaw skills install

From your agent's terminal, install any ClawHub skill by its @owner/slug reference:

openclaw skills install @owner/weather

The same command also takes a git repository or a local directory, which is how you install something that was never published to ClawHub:

openclaw skills install git:https://github.com/you/my-skill --as my-skill
openclaw skills install ./path/to/my-skill

Useful flags: --version pins a specific release, --global installs into the shared managed skills directory instead of the current agent's workspace, --force overwrites an existing skill of the same name, and --agent <id> targets a particular agent when you run more than one.

Two flags cover two different ClawHub safety states, and it is worth keeping them apart. --acknowledge-clawhub-risk accepts a trust warning on a release ClawHub has already flagged, without stopping for an interactive prompt. --force-install is the other case: a GitHub-backed release whose ClawHub scan simply hasn't finished yet, which the flag installs anyway rather than waiting. Either way you are choosing to skip a check, so only use them when you already trust the publisher.

Method 3: Drop In a SKILL.md Folder

Because a skill is just a folder with a SKILL.md in it, you can write one by hand. To make it available to every agent, create a directory in the shared managed skills directory — ~/.openclaw/skills/<slug>/ — and put your SKILL.md inside it, along with any scripts it needs. To scope it to one agent instead, use that agent's own workspace skills/ directory.

~/.openclaw/skills/my-skill/
├── SKILL.md          # frontmatter + instructions
└── scripts/
    └── run.sh        # optional helper files

The frontmatter needs at minimum a name and a description. The description is what the agent reads when deciding whether the skill applies, so write it as a trigger — say when to use the skill, not just what it is. A skill file with no frontmatter is never loaded at all.

With skills.load.watch enabled in your config, the gateway notices the new folder and loads it without a restart. If you would rather be certain, restarting the gateway always picks it up cleanly.

Popular Skills to Start With

These are the most-installed skills across managed OpenClaw Launch instances, so they are a reasonable place to begin:

For a curated walk through the rest, see the best OpenClaw skills, or the GitHub skill guide for the most-installed one on its own.

Managing Installed Skills

Skill management lives in the openclaw skills command group. Run these from your agent's terminal:

• openclaw skills list — Every skill the agent can see, with its status and where it came from. Add --eligible for only the ones ready to use, or --json for machine-readable output.
• openclaw skills check — The troubleshooting command. It reports how many skills are eligible, visible to the model, available as a command, and how many are blocked by missing requirements.
• openclaw skills info <name> — Details for one skill.
• openclaw skills update --all — Updates every tracked ClawHub-installed skill. Pass a single @owner/slug instead of --all to update just one, and add --global to target the shared managed directory.
• openclaw skills verify <name> — Verifies a ClawHub skill against ClawHub.

There is no uninstall subcommand. To remove a skill you installed by hand, delete its folder from the skills directory; on OpenClaw Launch, click Uninstall on the skill's card in your dashboard.

If you want an agent to see only a chosen subset of what is installed, set an allowlist with agents.list[].skills (or agents.defaults.skills) in your config. The skills stay installed; the agent simply loads only the ones you name.

Skill Configuration

Most skills need no configuration at all — the instructions in SKILL.md are the whole skill. What varies is credentials: a skill that reaches an external service needs a way to authenticate.

Per-Skill Settings in openclaw.json

Credentials belong in the skills.entries section of your config, keyed by the skill's slug. Note the nesting — it is skills.entries.<slug>, not skills.<slug>:

{
  "skills": {
    "entries": {
      "nano-banana-pro": {
        "apiKey": "your-gemini-key-here"
      },
      "some-skill": {
        "env": {
          "SOME_SERVICE_TOKEN": "your-token-here"
        }
      },
      "youtube-watcher": {
        "enabled": false
      }
    }
  }
}

The two credential forms do slightly different jobs. env is a map of environment-variable names to values, and satisfies a skill whatever variable it declares it needs. apiKey is the shorthand, and it works for skills that name a primary environment variable in their own metadata — OpenClaw maps your apiKey onto that variable for you. If a skill stays ineligible after you set apiKey, name the variable explicitly under env instead.

The same block is how you disable a skill without deleting it: set "enabled": false and the skill stays installed, with its settings intact, for whenever you want it back.

Why Exporting a Variable in the Terminal Isn't Enough

This trips people up, so it is worth stating plainly. Running export GEMINI_API_KEY=… in a web terminal sets that variable for that shell session only. The agent is served by a gateway process that is already running, and it cannot see a variable exported into a shell that started after it. The skill will still report a missing requirement.

So put the credential in skills.entries as above, or set it in the environment the gateway itself starts with. An ambient variable does work — it just has to be there before the gateway comes up, not typed into a terminal afterwards.

Some skills sidestep all of this by authenticating through a CLI they wrap: the github skill uses whatever gh auth login already established, for example, and that credential lives on disk rather than in the environment. Each skill's own SKILL.md states what it needs, and openclaw skills check lists the skills currently held back by a missing requirement.

Where Skills Are Loaded From

The skills.load section controls discovery. skills.load.watch makes the gateway pick up changes to skill files without a restart, and skills.load.extraDirs adds directories beyond the managed skills path — useful when you keep a set of skills in your own repo and mount it in.

Troubleshooting

Skill Not Found

If openclaw skills install returns a not-found error:

• Check the ref format — it is @owner/slug, and both halves matter
• Run openclaw skills search <keyword> to confirm the exact slug on ClawHub
• For a skill that lives only in a repo, use the git: form rather than a bare name

Installation Errors

If a skill fails to install:

• Check connectivity — ClawHub skills are downloaded at install time
• Verify the agent can write to its skills directory
• If the install stops on a ClawHub trust warning, the release has been flagged — proceed only if you trust the publisher. A GitHub-backed release still mid-scan is the separate case, and waiting for the scan is the safe option there
• Update OpenClaw if the skill declares a newer minimum version

Skill Not Responding

If a skill is installed but your agent doesn't use it, run openclaw skills check first — it separates the four reasons this happens, which saves guessing:

• Missing requirements is the usual answer. The skill needs a binary or a credential the container doesn't have yet, so it is loaded but not eligible.
• Excluded by agent allowlist means an agents.*.skills allowlist is filtering it out.
• No frontmatter in a hand-written SKILL.md means it was never loaded at all — it won't appear in the list.
• Otherwise, ask for it by name in chat: “use the web-search skill to look up…”, and check the gateway logs for skill-related errors.

One more case worth knowing: past roughly 100 registered commands, OpenClaw drops the per-skill slash commands on Telegram and Discord and keeps only the general /skill command. The skills themselves keep working — the agent still loads and uses them when you ask in plain language. Only the named shortcuts go away, and an agent skill allowlist is the way to bring the set back under the ceiling.

Rate Limits on ClawHub

ClawHub applies a per-IP rate limit on skill downloads, so installing many skills back to back can return a 429. Stop and wait roughly 10–15 minutes rather than retrying in a loop, which extends the cooldown. Authenticated ClawHub requests get a much higher limit. See the ClawHub rate limit guide for the full fix.

Skills on OpenClaw Launch

OpenClaw Launch makes skill management easier in two ways. Our curated marketplace of 4,000+ skills is available directly from your dashboard with one-click install — no command line and no config editing — and every managed OpenClaw instance also has a web terminal, so the full openclaw skills command group is there when you want it.

Deployed instances arrive with OpenClaw's bundled skills already in place, which covers a lot before you install anything. To add more, browse the Skills page, choose the instance, and click install.

Skills do not update themselves by default. Run openclaw skills update --all when you want the latest ClawHub versions, or install the auto-updater skill, which does that once a day on its own.

Frequently Asked Questions

How many skills can I install?

There is no hard limit on the number of skills you can install. However, each active skill consumes some memory and processing resources. Most users run between 5 and 15 skills without any performance issues. If you notice slower responses, try disabling skills you don't use frequently.

Are OpenClaw skills free?

The vast majority of skills on ClawHub are free to install and use. Some skills connect to external APIs (like Tavily or OpenAI) that have their own pricing — in those cases, you need an API key from the provider. The skill itself is free; the external service may have costs.

How do I find the best skills?

Start with the Skills page on OpenClaw Launch, which curates popular skills and shows how many instances have installed each one. From the terminal, use openclaw skills search with descriptive keywords like “web scraping” or “PDF reader”.

Can I create my own skill?

Yes, and it is deliberately low-ceremony: a skill is a folder with a SKILL.md inside it. Write the frontmatter and the instructions, drop it in the skills directory, and the agent can use it. The bundled skill-creator skill will walk your agent through building and testing one with you. When it is ready, you can publish it to ClawHub for others to install.

What is ClawHub?

ClawHub is the official skill registry for OpenClaw, at clawhub.ai. You install from it by reference — openclaw skills install @owner/slug — and search it with openclaw skills search. ClawHub scans published releases and OpenClaw surfaces a trust warning for a release it has flagged, so treat an unverified skill the way you would any other third-party code.

What's Next?

• Browse all skills — Explore 4,000+ skills for your AI agent
• Best OpenClaw skills — Curated list of the top skills for every use case
• ClawHub guide — Learn how to navigate the official skill marketplace
• Tavily search setup — Set up AI-powered web search for your agent
• Connect GitHub to your agent — Give your agent push, PR, and issue access via the GitHub skill
• See pricing — Deploy with skills ready to go, from $3 for the first month, then $6/mo