- ClaWHub is OpenClaw's community skill registry — the equivalent of npm for AI agent capabilities, hosted at clawhub.dev
- Skills are installed via a single CLI command; no account required to browse or install
- Every skill has a trust score (0–100) combining automated security scans, manual review status, and author reputation
- The Verified badge means a skill passed both automated scanning and manual human review by the ClaWHub team
- You can search ClaWHub directly from the CLI — no need to open a browser to find skills
Before ClaWHub, adding a new capability to an OpenClaw agent meant writing custom tool code, wiring up permissions, testing edge cases, and hoping nothing broke the agent on restart. ClaWHub changed that. 340+ published skills. One command to install. That's the pitch — and as of early 2025, it largely delivers.
What ClaWHub Actually Is
ClaWHub is a hosted registry of OpenClaw skills — self-contained capability packages that extend what an agent can do. Think of it as the package manager layer for the AI agent ecosystem. Where npm serves Node.js and pip serves Python, ClaWHub serves OpenClaw.
Each skill is a structured package containing a manifest file, implementation code, and configuration schema. The manifest declares the skill's name, version, author, required permissions, compatible OpenClaw runtime versions, and the tools it exposes to the agent. The implementation is the actual logic the agent calls when it uses those tools.
Skills are published by three types of authors: the OpenClaw core team (first-party skills like web-search and summarize), verified community developers with established track records, and open community contributors whose work hasn't yet been through manual review.
Before skills existed, adding Twitter integration to an agent required writing a custom tool class, defining the function signatures, handling OAuth manually, and registering the tool in the agent config. The Bird skill does all of this in one install command. ClaWHub skills aren't magic — they're the same integration work, pre-done and packaged by someone who already figured it out.
How Skills Get Published to ClaWHub
Any developer with a ClaWHub account can publish a skill. The process is designed to be accessible while maintaining enough friction to deter low-effort or malicious submissions.
Publishing works in three stages. First, the developer creates a skill package locally using openclaw skill create, which scaffolds the manifest and implementation structure. Second, they run openclaw skill publish, which uploads the package and triggers automated scanning. Third, the skill appears in the registry — initially without the Verified badge, but visible and installable.
The automated scan checks for known malicious patterns, unrequested permission escalations, external network calls that aren't declared in the manifest, and hardcoded credentials. Skills that fail the automated scan are rejected immediately and the author is notified with the specific failure reason.
Manual verification is separate — the ClaWHub team reviews submitted skills and grants the Verified badge for ones that pass. As of early 2025, this review typically takes 3–5 business days. The Verified badge is the strongest trust signal on the platform, and most power users won't install unverified skills in production.
Safety, Trust Scores, and What to Check Before Installing
Every skill on ClaWHub has a trust score from 0 to 100. The score is a weighted composite of several factors.
| Factor | Weight | What It Measures |
|---|---|---|
| Automated scan pass | 35% | No malicious patterns or undeclared network calls |
| Verified badge | 30% | Human review completed by ClaWHub team |
| Author reputation | 20% | Track record from previously published skills |
| Community reports | 10% | User-submitted problem reports, lowered if flagged |
| Update recency | 5% | Skills unmaintained for 12+ months score lower |
The practical threshold most experienced OpenClaw builders use is 85+ for production and 70+ for personal/experimental use. A score below 70 doesn't mean a skill is broken — it often means it's new, from an unknown author, or hasn't been updated recently. It means you need to read the manifest carefully and decide for yourself.
Run openclaw skill info <skill-name> to view the full manifest before installing. Check the permissions section — a skill that requests write access to your filesystem or outbound network access to unlisted domains deserves a second look, regardless of its trust score.
Browsing and Searching ClaWHub
Two ways to explore ClaWHub: the web interface at clawhub.dev and the CLI. Most experienced builders use the CLI because it's faster and the results include install commands you can copy directly.
# Search for skills by keyword
openclaw skill search "email"
# Filter by category
openclaw skill search --category productivity
# Set minimum trust score
openclaw skill search "web" --trust-min 90
# View full skill details
openclaw skill info web-search
# List all available categories
openclaw skill categoriesThe web interface at clawhub.dev adds features the CLI doesn't have — skill comparison views, author profiles, install count history over time, and community discussion threads. Sound familiar? The design is deliberately similar to npmjs.com because the OpenClaw team wanted the experience to feel immediately familiar to developers.
Search on clawhub.dev supports full-text search across skill names, descriptions, and tags. Filter chips let you narrow by category (productivity, data, social, developer tools, media) and by trust level. The default sort is by install count, which surfaces the most widely-used skills first.
A skill with 12,000 installs and a trust score of 88 is often a safer choice than one with a trust score of 95 and 40 installs. Install count is the community's implicit endorsement. Skills with thousands of installs have been tested across hundreds of different configurations — that's real-world durability data you can't get from a scan alone.
Common Mistakes When Using ClaWHub
- Installing by name without checking the author — ClaWHub has had cases of similarly-named skills from different authors. Always verify the author matches the one you expect, especially for high-reputation skills like gmail or notion.
- Skipping the trust score filter — searching without a trust minimum surfaces everything, including abandoned or untested skills. Add
--trust-min 80as a default habit. - Not checking version compatibility — skills list minimum compatible OpenClaw versions. Installing a skill built for 0.9.x on a 0.7.x runtime causes silent failures. Check your runtime version with
openclaw --versionbefore installing. - Treating Verified as a security guarantee — the Verified badge means the skill passed review at the time of review. It doesn't mean every subsequent update was reviewed. Pin to a specific version in production environments where stability matters.
- Not monitoring installed skills for updates — skills that depend on external APIs (Twitter, Gmail, Notion) need updates when those APIs change. Run
openclaw skill update --allon a regular schedule to stay current.
Frequently Asked Questions
What is ClaWHub and how does it relate to OpenClaw?
ClaWHub is the official community skill registry for OpenClaw — think of it as npm for AI agent capabilities. It hosts skills published by the OpenClaw team and the community. You browse on the ClaWHub website and install via the openclaw CLI. No separate account is needed to install skills.
How do I search for skills on ClaWHub?
Browse at clawhub.dev or search from the CLI with openclaw skill search <query>. The CLI returns skill names, authors, trust scores, and install counts. Filter by category using --category or by minimum trust score using --trust-min to narrow results.
Are community-published skills reviewed before appearing on ClaWHub?
All submitted skills pass automated security scans before listing. Full manual review — which grants the Verified badge — is optional but gives skills a higher trust score. Skills without the Verified badge are visible and installable but haven't been manually audited by the ClaWHub team.
Can I publish my own skill to ClaWHub?
Yes. Create a skill manifest, write your implementation, and submit via openclaw skill publish. New submissions go through automated scanning first. The review queue for the Verified badge typically takes 3–5 business days as of early 2025.
Does ClaWHub require an account to browse or install?
No account is required to browse ClaWHub or install skills. Publishing a skill requires a free ClaWHub account to associate the skill with an author identity — keeping the registry accountable with traceable authorship.
What does the trust score on ClaWHub actually measure?
The trust score (0–100) combines automated security scan results, presence of the Verified badge, author reputation from previously published skills, community reports, and update frequency. A score above 90 indicates a skill passed all checks and manual review. Below 70 warrants caution.
J. Donovan writes documentation and deep-dive guides for the OpenClaw ecosystem, with a focus on making complex systems accessible to builders at every level. Has reviewed over 60 ClaWHub skill manifests and contributed to the registry's trust score methodology documentation.