ClawHub: The Skill Registry for OpenClaw Agents

Every OpenClaw agent starts the same way: a gateway, a channel, and a blank slate. It can chat. It can reason. But it can't do much that's specific to your stack, your workflows, or the tools you actually use — not until you teach it.

That's what skills are for. And ClawHub is where you find them.

ClawHub is the public registry for OpenClaw skills and plugins. Think of it as the npm of AI agent capabilities: a versioned, searchable, community-driven library of things your agent can learn to do. One command to install, one command to update, and your agent gains a new ability without you writing a line of code.

This guide covers everything — what ClawHub is, how to use it from both OpenClaw's native commands and the standalone CLI, how to publish your own skills, and the real-world patterns that make the registry useful rather than just decorative.

What Is a Skill, Exactly?

Before getting into ClawHub specifically, it helps to be precise about what a skill is in OpenClaw terms.

A skill is a versioned bundle of files that teaches your OpenClaw agent how to perform a specific task. At minimum, a skill contains a SKILL.md file — a markdown document that describes what the skill does, when to use it, and how to execute it. That file becomes part of the agent's system context, giving it structured instructions it can follow.

Beyond SKILL.md, a skill can include:

• Supporting scripts (shell, Python, Node) the agent can invoke
• Config templates and reference files
• Metadata that controls when the skill gets loaded

When OpenClaw loads your workspace, it scans the skills/ directory and makes every installed skill available to the agent. The agent reads the skill descriptions, knows what it's capable of, and applies the right one when the task matches.

The power is in composability. A single agent with five well-chosen skills is dramatically more useful than one with zero. And because skills are text files, they're auditable, versionable, and easy to inspect before you trust them.

Finding Skills on ClawHub

The ClawHub site lives at clawhub.ai. You can browse, search by keyword or tag, and read the full SKILL.md content for any skill before installing it.

From the terminal, OpenClaw ships with native search built in:

openclaw skills search "calendar"
openclaw skills search "github pr review"
openclaw skills search "postgres"

The search is powered by vector embeddings, not keyword matching — so "email triage" will surface skills tagged for inbox management even if they don't use that exact phrase. Practically, this means you should search by what you want to do, not by what you think the skill is called.

You can also use the standalone ClawHub CLI if you want more control:

clawhub search "postgres backups"
clawhub search "web scraping" --limit 10

Both paths return the same registry data. Use whichever fits your workflow.

Installing Skills

Once you find something you want, installation is one command:

openclaw skills install <skill-slug>

This fetches the latest version of the skill from ClawHub and installs it into your active workspace's skills/ directory. OpenClaw picks it up on the next session — no restart needed for subsequent installs once you've already started a session, but a fresh session ensures the agent's context includes the new skill cleanly.

The ClawHub CLI gives you more options when you need them:

# Install a specific version
clawhub install my-skill-pack --version 1.2.0

# Overwrite if already installed
clawhub install my-skill-pack --force

Installed skills are recorded in .clawhub/lock.json in your workdir, similar to a package-lock file. This makes it easy to audit what's installed, replicate your setup on another machine, or roll back to a known state.

Keeping Skills Updated

Skill authors push updates when they add features, fix bugs, or improve instructions. Staying current is straightforward:

# Update a single skill
openclaw skills update <skill-slug>

# Update everything at once
openclaw skills update --all

Or via the ClawHub CLI:

clawhub update my-skill-pack
clawhub update --all

The update process compares your local skill files against the registry using content hashes. If your local files have been modified since install, the CLI asks before overwriting — or errors in non-interactive mode unless you pass --force. This prevents you from accidentally wiping local customizations you've made to a skill.

Installing Plugins from ClawHub

ClawHub isn't just for skills — it also hosts plugins. Plugins are code packages that extend OpenClaw's runtime, adding new tools, channel integrations, or capabilities that go beyond what a skill's text files can do.

# Install a plugin directly from ClawHub
openclaw plugins install clawhub:<package-name>

# OpenClaw also tries ClawHub automatically for bare package names
openclaw plugins install openclaw-codex-app-server

Plugins installed this way get ClawHub source metadata recorded alongside them, so openclaw plugins update --all can pull new versions from the registry automatically.

The distinction between skills and plugins matters: skills run as agent instructions (text-based), while plugins run as code (Node modules that hook into the gateway). Skills are lower risk to install and inspect; plugins require more trust because they execute code in the gateway process.

The ClawHub CLI: When You Need More

OpenClaw's native openclaw skills commands cover the day-to-day install/update flow. The standalone clawhub CLI is the tool you reach for when you need registry-authenticated workflows — publishing, syncing, managing versions, or running automation in CI.

Install it:

npm install -g clawhub
# or
pnpm add -g clawhub

Authenticate:

clawhub login

This opens a browser auth flow and stores your token in the CLI config. From there, all authenticated commands work:

clawhub whoami
clawhub list       # skills you've published
clawhub sync       # scan local skills and publish new/updated ones

Publishing Your Own Skills

This is where ClawHub gets genuinely interesting. Any skill you build locally can be published to the registry and shared with the OpenClaw community — or just backed up to the cloud for your own use.

Publishing a single skill:

clawhub skill publish ./my-skill \
  --slug my-skill \
  --name "My Skill" \
  --version 1.0.0 \
  --tags latest

Publishing multiple skills at once with sync:

# Dry run first — see what would be uploaded
clawhub sync --dry-run

# Publish everything
clawhub sync --all

The sync command scans your workdir for skill folders, compares them against the registry by content hash, and uploads new or changed skills. You can control how versions are bumped:

clawhub sync --all --bump minor --changelog "Added support for X"

Each publish creates a new semver version. Tags like latest point to a version and can be moved — useful for rolling back to a previous version if an update breaks something.

Security Requirements for Publishing

To publish, your GitHub account must be at least one week old. This isn't a hard KYC requirement — it's a basic spam throttle that keeps the registry clean without blocking real contributors. Once you're authenticated, you can publish as often as you want.

Skills with more than 3 unique reports from other users are auto-hidden pending moderation. If you're publishing skills, keeping your SKILL.md accurate and your behavior honest is both good practice and required to stay in good standing.

Publishing Plugins to ClawHub

If you've built a code plugin and want to distribute it through ClawHub, the publish path goes through your GitHub repository:

# Dry run to verify the build plan
clawhub package publish your-org/your-plugin --dry-run

# Publish it
clawhub package publish your-org/your-plugin

# Publish a specific version
clawhub package publish your-org/your-plugin@v1.0.0

Your plugin's package.json needs specific OpenClaw metadata for the registry to accept it:

{"{"}
  "name": "@myorg/openclaw-my-plugin",
  "version": "1.0.0",
  "type": "module",
  "openclaw": {"{"} 
    "extensions": ["./index.ts"],
    "compat": {"{"} 
      "pluginApi": ">=2026.3.24-beta.2",
      "minGatewayVersion": "2026.3.24-beta.2"
    {"}"},
    "build": {"{"} 
      "openclawVersion": "2026.3.24-beta.2",
      "pluginSdkVersion": "2026.3.24-beta.2"
    {"}"}
  {"}"}
{"}"}

The compat block ensures users installing your plugin know what gateway version they need. Don't skip it — plugins that silently fail on mismatched versions get reported and pulled.

How Skills Load in OpenClaw

Understanding the load order prevents surprises. When OpenClaw starts a session, it loads skills from these locations in priority order:

1. Workspace skills — <workspace>/skills/ — highest priority
2. Bundled skills — skills shipped with OpenClaw itself

Workspace skills win when there's a name conflict. This means you can override a bundled skill or a ClawHub-installed skill by putting a local version in your workspace's skills/ folder with the same name. It's a clean escape hatch for customization without forking.

The ClawHub CLI installs into the current working directory's skills/ folder by default, falling back to your configured OpenClaw workspace if one is set. You can override this:

clawhub install my-skill --workdir /path/to/workspace

Or via environment variable:

export CLAWHUB_WORKDIR=/path/to/workspace
clawhub install my-skill

Practical Patterns That Work

A few things I've found actually matter in practice:

Search by intent, not by name. The vector search means "automate my morning standup" will surface relevant skills even without exact keyword matches. Keyword searches are fine but often miss adjacent options.

Inspect before you trust. Every skill on ClawHub has a public SKILL.md page. Read it. A skill that's vague about what it does or asks for unusual permissions should make you pause. Skills are text, so you can always open the installed files in your workspace and audit them directly.

Use clawhub sync as a backup strategy. If you've built custom skills for your own workflows, clawhub sync --all is the fastest way to back them up to the registry. They can be private-ish (not promoted, but discoverable via direct URL) or fully public. Either way, you get versioned cloud backup with one command.

Pin versions in production. The lockfile at .clawhub/lock.json records what version each skill is at. For production agent setups, check this file into version control. It gives you reproducible installs and a clear audit trail of what changed when behavior changes.

Don't skip --dry-run on sync. Before running clawhub sync --all on a workspace with many skills, run with --dry-run first. You'll see exactly what would be published and can catch accidental inclusions before they go public.

Environment Variables Worth Knowing

For CI pipelines or automated setups, the ClawHub CLI respects these environment variables:

CLAWHUB_SITE        # Override the registry site URL
CLAWHUB_REGISTRY    # Override the registry API URL
CLAWHUB_CONFIG_PATH # Where the CLI stores your auth token
CLAWHUB_WORKDIR     # Default workdir for installs
CLAWHUB_DISABLE_TELEMETRY=1  # Disable install count telemetry on sync

The telemetry note: when you run clawhub sync logged in, the CLI sends a minimal usage snapshot to compute install counts for the registry's ranking. It's opt-out, not opt-in — set the env var if you'd rather not participate.

The Registry Is a Distribution Channel

Here's the framing I find most useful: ClawHub isn't just a library of tools, it's a distribution channel for agent capabilities.

If you're building OpenClaw setups for clients, for your team, or for yourself across multiple machines, ClawHub gives you a managed, versioned way to distribute exactly the skills each agent needs. You publish once, install everywhere, update from a single source of truth.

That's a meaningful shift from the "copy these files manually" approach that most teams default to. It also means the OpenClaw ecosystem as a whole benefits: every skill published to ClawHub is a capability any agent can gain with one command. The registry gets more useful as it grows, and it grows every time someone publishes something they built for themselves.

The best time to publish a skill you've built is before you forget you built it.

The OpenClaw Playbook

Want a full agent setup — skills, crons, memory, and all?

ClawKit is the complete guide to running OpenClaw as a real business operator. Covers skill architecture, workspace setup, cron automation, multi-agent coordination, and the exact patterns I use to run autonomous systems every day.

Getting Started Today

If you're not already using ClawHub, the entry point is simple:

# Search for something useful
openclaw skills search "your use case here"

# Install it
openclaw skills install <skill-slug>

# Start a new session — your agent now knows how to do this

If you've built anything custom, consider publishing it. The bar is low (a GitHub account, one command), the upside is version control and discoverability, and the registry benefits from every real-world skill that lands in it.

The best OpenClaw setups I've seen aren't the ones with the most powerful models or the longest prompts — they're the ones with the right skills loaded for the work at hand. ClawHub makes finding and keeping those skills manageable.

Want the complete guide to building, configuring, and operating an OpenClaw agent at scale? Get ClawKit — $9.99. It covers skill architecture, workspace design, cron automation, memory systems, and every pattern I use to run autonomous agents that actually ship work.

Originally published at openclawplaybook.ai. Get The OpenClaw Playbook — $9.99