OpenClaw ClawdHub Guide: Install, Manage, and Build Community Skills (2026)

There’s a moment every OpenClaw user hits about two weeks in. You’ve got crons running, Telegram connected, maybe a custom persona. And then you realize: I need it to do something it doesn’t do by default.

Maybe it’s deep research across multiple sources. Maybe it’s pulling Google Search Console data automatically. Maybe it’s connecting to Obsidian, or running a morning briefing with Bitcoin price plus weather plus calendar — all in one message.

You could build that from scratch. Or you could open ClawdHub, type one command, and have it running in 90 seconds.

ClawdHub is OpenClaw’s community skill marketplace — the place where pre-built automations live, get updated, and get installed with a single CLI command. We use it constantly. This guide covers everything: what ClawdHub is, how to find skills worth installing, the exact commands to install and manage them, and how to know when a skill is actually doing what it promises.

What Is ClawdHub?

ClawdHub (clawdhub.com) is the official skill registry for OpenClaw. Think of it like npm for AI agent behaviors — a central place to publish, discover, and install packaged automations called “skills.”

Each skill is a directory containing:

• A SKILL.md file that tells OpenClaw exactly what the skill does and how to trigger it
• Optional scripts, config files, and reference data the agent needs to execute the skill
• A version number so you can stay current or pin to a specific release

When OpenClaw processes a request, it scans its installed skills. If your request matches a skill’s description, the agent reads that SKILL.md and follows the specialized instructions — rather than winging it from general knowledge.

That’s the key insight most people miss. Skills aren’t plugins that add features to the software. They’re context injections that give the AI agent domain-specific knowledge and step-by-step procedures for specific tasks. The difference in output quality is significant.

We’ve compared the same task with and without the relevant skill loaded. With a skill: structured output, correct tool sequence, proper error handling. Without: a reasonable attempt that misses half the steps and requires follow-up corrections. Skills close that gap.

How OpenClaw Skills Actually Work

Before installing anything, it helps to understand the mechanics. This isn’t essential knowledge for basic use — but if you’re going to manage 10+ skills, you want to understand what’s happening under the hood.

The Skill Directory Structure

Skills live in your OpenClaw workspace under a skills/ folder. Each skill gets its own subdirectory:

~/.openclaw/workspace/skills/
├── deep-research/
│   └── SKILL.md
├── obsidian-daily/
│   ├── SKILL.md
│   └── templates/
├── gsc/
│   ├── SKILL.md
│   └── scripts/
│       └── fetch_data.py
└── morning-routine/
    └── SKILL.md

The workspace directory is typically at ~/.openclaw/workspace/ on Mac/Linux or C:\Users\[username]\.openclaw\workspace\ on Windows.

How OpenClaw Detects and Uses Skills

At the start of every conversation, OpenClaw reads the <description> entry from every installed skill’s SKILL.md and includes them in your system prompt under an <available_skills> block. When you send a message, the agent checks whether any skill clearly matches your request.

If it finds one, it runs the read tool to load the full SKILL.md before responding — giving itself the specialized instructions for that task. If no skill matches, it proceeds from general knowledge.

This means skill descriptions matter enormously. A well-written skill description that covers multiple trigger phrases (“SEO audit”, “why am I not ranking”, “meta tags review”) gets activated reliably. A vague description gets missed.

The ClawdHub team maintains this quality. Community-submitted skills go through a review process before they appear in the registry — which is why the install-to-works ratio is much higher than random GitHub scripts.

Installing the ClawdHub CLI

ClawdHub has its own CLI tool for managing skills. If you don’t have it installed already:

npm install -g clawdhub

Verify it’s working:

clawdhub --version

That’s it. No API keys, no accounts, no auth for basic usage. The registry is public for reads.

On Windows, run this in PowerShell as your regular user (not Administrator). On Mac/Linux, you might need sudo npm install -g clawdhub depending on your Node setup — or use a version manager like nvm to avoid permission issues.

If you’re on the Windows OpenClaw setup and already have Node.js from the OpenClaw install, this just works.

Browsing Available Skills

Three ways to find skills worth installing:

Method 1: ClawdHub Website

Head to clawdhub.com and browse the directory. You can filter by category (research, content, development, productivity, integrations) and see install counts, last updated dates, and skill descriptions before committing to anything.

The “Most Installed” list is a reasonable starting point. It shows what’s actually being used — not just what got published.

Method 2: CLI Search

clawdhub search "research"
clawdhub search "obsidian"
clawdhub search "seo"

Returns a list of matching skills with their descriptions and current versions. Fast when you know what category you’re looking for.

Method 3: List All Available Skills

clawdhub list

Dumps the full registry. Long, but useful if you want to spot skills you didn’t know existed. Pipe it through grep for specific tools:

clawdhub list | grep -i "twitter\|x.com\|social"

Installing Skills: The Exact Commands

Installing a skill is one command:

clawdhub install deep-research

This downloads the skill to your OpenClaw workspace’s skills/ directory. No restart required — OpenClaw picks up new skills at the start of the next conversation.

To install multiple skills at once:

clawdhub install deep-research obsidian-daily gsc brave-search

To install a specific version (useful for pinning in production setups):

clawdhub install deep-research@1.2.0

To install into a specific workspace path (if you have multiple OpenClaw setups):

clawdhub install deep-research --workspace /path/to/workspace

The Skills We Actually Have Running

After about 18 months of running OpenClaw as our primary AI operating system, here’s what’s installed on our production setup and what each one actually does for us:

Skill	What It Does For Us	Frequency of Use
deep-research	Multi-source research with source synthesis. Better than raw web search for anything requiring comparison or analysis.	Daily
obsidian-daily	Manages daily notes in Obsidian vault — append entries, read past notes, search content.	Daily
gsc	Pulls Google Search Console data — queries, clicks, impressions, CTR — without manually logging in.	Weekly
brave-search	Direct Brave Search API calls for research sub-agents that don’t need full browser automation.	Daily
pinch-to-post	Full WordPress automation via REST API — create, update, manage posts, WooCommerce, SEO meta.	Daily (via crons)
github	GitHub CLI integration — issues, PRs, workflow runs via gh commands.	Weekly
coding-agent	Structured instructions for running Codex CLI, Claude Code, or OpenCode as sub-agents for complex builds.	Several times/week
morning-routine	Habit checklist and streak tracking for morning workflow.	Daily
pre-mortem-analyst	Structured failure-analysis — assumes a project failed and works backward. Genuinely useful before major decisions.	Occasionally
seo-optimizer	Analyzes HTML/CSS pages for SEO issues. Good for technical audits.	Weekly

Most of these we discovered on ClawdHub, not from the OpenClaw docs. The marketplace surface area is larger than what the official documentation highlights.

Managing Installed Skills

See What’s Currently Installed

clawdhub installed

Lists all skills in your workspace with their current version and the latest available version — so you can immediately see what’s outdated.

Update a Specific Skill

clawdhub update deep-research

Pulls the latest version and overwrites the skill directory. If the skill has scripts or config files you’ve customized, check the changelog first — some updates replace files rather than merging.

Update All Skills At Once

clawdhub update --all

This is what our auto-updater skill runs once daily via cron. We covered the cron job setup in detail here — but the short version is: schedule this command to run nightly and you never think about skill freshness again.

Remove a Skill

clawdhub remove obsidian-daily

Deletes the skill directory from your workspace. The agent will no longer see it in the available skills list. Good for cleaning up skills you installed to test but don’t use.

There’s a real reason to prune skills you don’t use: every installed skill’s description gets included in your system prompt. That’s context tokens. If you have 40 skills installed, you’re burning meaningful context budget just on skill descriptions — context that could go toward your actual task. Keep it to what you use.

Sync Skills (Declarative Management)

For teams or multi-machine setups, you can define your skill list in a config file and sync to it:

# clawdhub.config.json
{
  "skills": [
    "deep-research",
    "obsidian-daily",
    "gsc",
    "brave-search",
    "pinch-to-post",
    "github",
    "coding-agent"
  ]
}

clawdhub sync

This installs missing skills and removes ones not in the list. Useful if you’re version-controlling your OpenClaw workspace config — which we’d recommend for anyone running this seriously.

Reading a Skill Before You Install It

Here’s what most guides skip: before installing any skill, preview what it actually does.

clawdhub info deep-research

This shows you the full skill metadata including the SKILL.md content — the exact instructions the AI will follow when the skill triggers. It’s worth reading, especially for skills that:

• Require API keys (the skill will tell you)
• Make external API calls you might want to be aware of
• Have configuration options you need to set up
• Depend on other tools being installed (like a CLI tool or Python package)

The gsc skill, for example, requires setting up a Google Cloud service account and placing the credentials file at a specific path. You won’t discover that at install time — you’ll discover it when the skill activates and fails. Read the info first.

Skills That Require Setup After Install

Some skills are plug-and-play. Others need a bit of configuration before they work. Here’s how to tell the difference and what to do about it.

API Key Skills

Skills like brave-search, exa, tavily, and apollo need API keys in your environment. The standard pattern is adding them to your system environment or OpenClaw’s config:

# Add to your shell profile or Windows environment variables
BRAVE_API_KEY=your_key_here
EXA_API_KEY=your_key_here
TAVILY_API_KEY=your_key_here

The skill’s SKILL.md will tell you the exact variable name it expects. If you’re managing multiple API keys, we covered OpenClaw configuration best practices including environment management in our advanced config guide.

CLI Dependency Skills

The github skill requires the gh CLI installed and authenticated. The obsidian-daily skill requires obsidian-cli. The coding-agent skill requires your coding CLI of choice (Codex, Claude Code, etc.).

These skills don’t install their dependencies automatically — they assume the dependency exists. Check the clawdhub info [skill-name] output for a “Prerequisites” or “Requirements” section.

Skills With Scripts

Some skills include Python or shell scripts that run as part of their automation. Check that you have the required interpreter and any Python packages listed in the skill’s README.

For Python dependencies, a quick install usually handles it:

pip install paramiko google-auth requests

The skill will typically call these scripts via OpenClaw’s exec tool — so if Python isn’t in your PATH, the script won’t run. On Windows specifically, verify python --version works from PowerShell before debugging why a skill isn’t executing.

Troubleshooting: When a Skill Isn’t Triggering

You installed a skill and it’s not activating when you expect. Here’s the diagnostic process we use.

Step 1: Verify the Skill Is Actually Installed

clawdhub installed | grep deep-research

If it’s not in the list, it’s not installed. Re-run the install command and check for errors.

Step 2: Check the Skill Directory Exists

Navigate to your workspace skills directory and confirm the folder is there with a SKILL.md file:

# Mac/Linux
ls ~/.openclaw/workspace/skills/

# Windows PowerShell
dir $env:USERPROFILE\.openclaw\workspace\skills\

Step 3: Check Your Request Phrasing

Skills trigger based on description matching. If your request doesn’t match the skill’s description, the agent won’t select it. Try phrasing your request using words directly from the skill description.

For example, the pre-mortem-analyst skill triggers on: “pre-mortem”, “premortem”, “imagine this failed”, “what could go wrong”, “risk analysis”, “stress test”, “what would kill this”, “project risks”.

Asking “what are the risks of this approach” might work. Asking “is this a good idea” probably won’t trigger it — too vague, too many possible interpretations.

Step 4: Too Many Skills Installed?

There is a practical limit. If you have 50+ skills installed, you might hit context window issues where the system prompt with all skill descriptions gets truncated. The agent never sees the full skill list and can’t select properly.

Prune inactive skills. Run clawdhub installed, go through the list honestly, and remove anything you haven’t used in 30 days.

Building Your Own Skill and Publishing It

The ClawdHub community grows through user contributions. If you’ve built an automation that works well, packaging it as a skill and publishing it takes about 20 minutes.

The Minimal Skill Structure

Every skill needs at minimum:

my-skill/
└── SKILL.md

The SKILL.md format:

# My Skill Name

## Description (Used for skill matching)
Brief description of what this skill does. Include trigger phrases: "specific phrase 1", "specific phrase 2", "related task keyword". Use when user wants to [accomplish X]. Also triggers on [Y] and [Z].

## Instructions
Step-by-step instructions for the AI agent.

1. First, do this
2. Then do that
3. Handle errors like this

## Configuration Required
- ENV_VAR_NAME: Description of what this is and where to get it

## Prerequisites
- tool-name >= version

We covered writing custom skills in detail in our Custom OpenClaw Skills guide — including advanced patterns like multi-step workflows and skills that spawn sub-agents.

Publishing to ClawdHub

# From your skill directory
cd my-skill/

# Login to ClawdHub (one-time)
clawdhub login

# Publish
clawdhub publish

The CLI will prompt for a skill name (must be unique in the registry), description, and category. After submission, there’s a brief review period — typically a few days — before it appears publicly.

For versioned updates to an existing skill:

clawdhub publish --version 1.1.0

Use semantic versioning. Breaking changes = major version bump. New features = minor. Bug fixes = patch.

ClawdHub vs. Installing Skills Manually from GitHub

You can also install skills directly from GitHub repositories rather than ClawdHub:

clawdhub install github:username/repo-name

Or just clone the repo directly into your skills directory:

git clone https://github.com/username/some-skill ~/.openclaw/workspace/skills/some-skill

When should you do this versus using ClawdHub?

• Use ClawdHub for anything where you want automatic update tracking, community vetting, and easy management via the CLI. This is the right default for 90% of skills.
• Use GitHub direct for skills you’re actively developing yourself, skills from private repos, or experimental skills that haven’t been submitted to the registry yet.
• Manual creation for skills that are highly specific to your setup and wouldn’t make sense for public use — like a skill that knows your specific server infrastructure or internal tooling.

Mixing is fine. We have ClawdHub-managed skills alongside custom skills we wrote ourselves. The clawdhub installed command only shows ClawdHub-tracked skills — manually created skills won’t appear there, but they’ll still work in OpenClaw as long as they’re in the skills directory with a valid SKILL.md.

A Realistic Starter Pack

If you’re setting up ClawdHub from scratch and want a reasonable starting set without bloat, here’s what we’d install on day one:

clawdhub install deep-research brave-search coding-agent github

That gives you:

• Serious research capability (deep-research)
• Fast web search without burning browser overhead (brave-search)
• Structured coding workflows with sub-agents (coding-agent)
• GitHub operations via CLI (github)

Then add based on your actual workflow over the next week. Don’t install 20 skills at once — you won’t know what’s helping and what’s just consuming context. Add one or two, use them for a few days, then decide whether they stay.

If you’re running OpenClaw as a content operation like we do, the second wave would be:

clawdhub install pinch-to-post gsc seo-optimizer obsidian-daily

And if you’re using OpenClaw as an operations system for a business:

clawdhub install morning-routine auto-updater pre-mortem-analyst first-principles-decomposer

The auto-updater skill in particular is worth installing early — it sets up a cron that keeps all your other skills current. You’ll never manually run clawdhub update --all again.

What ClawdHub Doesn’t Do

One honest note before you over-rely on the marketplace: ClawdHub skills are instructions, not integrations. They tell the AI agent how to accomplish a task — but the agent still needs the underlying tools, credentials, and infrastructure to actually execute.

A skill that manages your WordPress site requires WordPress to be running and the REST API to be accessible. A skill that queries your database requires the database credentials. A skill that controls your browser requires the browser tool to be enabled in OpenClaw.

The skill is the playbook. You still have to build the field.

This is occasionally frustrating when a skill description sounds like it does everything and you install it expecting magic — only to find three configuration steps and a missing API key standing between you and the automation working. That’s not a ClawdHub problem; that’s the nature of connecting AI to real infrastructure. Read the info page before installing anything complex.

Integrating ClawdHub Into Your OpenClaw Setup

The cleanest way to manage skills across a serious OpenClaw deployment:

1. Keep a clawdhub.config.json in your workspace root with your canonical skill list
2. Version control that config file (along with your SOUL.md, AGENTS.md, and TOOLS.md)
3. Use the auto-updater skill to run clawdhub update --all nightly
4. Periodically audit your installed list and prune what you don’t use
5. For custom skills you write, store them in skills/[skill-name]/ and track them separately from ClawdHub-managed skills

This way, rebuilding your OpenClaw setup on a new machine is a 5-minute operation: install OpenClaw, copy your workspace directory (or pull it from git), run clawdhub sync, and you’re back to full capability.

We tested this last month when moving between machines. Without workspace files and ClawdHub sync, rebuilding a mature OpenClaw setup would take hours. With them: 8 minutes including waiting for npm installs.

If you haven’t set up version control on your OpenClaw workspace yet, our workspace files guide covers exactly how to structure and protect your configuration.

Frequently Asked Questions

Is ClawdHub free to use?

Yes. Browsing, installing, and updating skills from ClawdHub is completely free. Publishing skills to the registry is also free. There is no paid tier for basic skill management.

Do skills work with all AI models in OpenClaw?

Skills are model-agnostic — they inject instructions into the system prompt and work with whatever model OpenClaw is using, whether that’s Claude, GPT-4, Gemini, or any other supported model. Some skills are optimized for specific model capabilities (e.g., skills that use extended thinking work best with Claude’s reasoning models), but most work universally.

Can I use ClawdHub skills if I’m self-hosting OpenClaw?

Yes. ClawdHub skills are just files in your workspace. Whether you’re running OpenClaw on a cloud server, a local Windows machine, or a Raspberry Pi, the skill installation process is identical. The skills directory path is the same regardless of host.

How do I know if a skill is trustworthy?

ClawdHub reviews submissions before publishing. That said, skills execute code via OpenClaw’s exec tool — which means a malicious skill could theoretically run commands on your system. Always read the SKILL.md and any included scripts before installing. Stick to skills from verified publishers or with high install counts and recent update activity. The same common sense you’d apply to any npm package applies here.

What’s the difference between a skill and an MCP server?

MCP servers are external tool providers that give OpenClaw new capabilities via the Model Context Protocol — things like database access, web search APIs, or custom tool endpoints. Skills are instructions that tell the agent how to use existing tools more effectively. They complement each other: an MCP server might add a database query tool, and a skill might tell the agent the exact workflow for using that tool to generate weekly reports. We covered MCP integration in our OpenClaw MCP guide.

Can skills call other skills?

Not directly — only one skill activates per request. But skills can instruct the agent to spawn sub-agents that have access to their own skill sets, or can reference procedures that effectively chain tasks. Complex multi-skill workflows are usually better handled through sub-agent orchestration rather than trying to chain skills directly.

How often are skills on ClawdHub updated?

Varies by skill and publisher. Well-maintained skills get updated when OpenClaw adds new capabilities or when the underlying tools change their API. The clawdhub installed command shows you the last-updated date alongside each skill — anything updated in the last 60 days is actively maintained. Skills not updated in 6+ months might be abandoned; check the ClawdHub page for any maintenance notes before relying on them in production.

Does installing lots of skills slow down OpenClaw?

The performance impact is on context window usage, not on processing speed. Each skill description adds tokens to your system prompt. With 10-15 skills, the overhead is negligible. With 40+ skills, you’re using a meaningful chunk of available context before the conversation even starts. Practical advice: keep your installed count under 25, and ruthlessly remove skills you’re not actively using.