ClawHub vs SkillzBench: Which Agent Skills Platform Should You Trust?
ClawHub is the skill registry for OpenClaw agents: it hosts 3,286+ community skills with vector search, SemVer versioning, and a free tier, with Pro at $20 per human seat per month for private repos and governance. SkillzBench is a free, independent alternative that verifies, security-scans, and benchmarks agent skills, then connects them to any agent through a free MCP server. If you run OpenClaw and want the biggest catalog, use ClawHub; if you want to know which skills are safe and actually improve your agent before installing anything, SkillzBench is the safer starting point.
TL;DR
- ClawHub is the official skill registry for OpenClaw, hosting 3,286+ community skills with 1.5M+ downloads as of February 2026.
- ClawHub's pricing starts at $0 forever for public repos with unlimited agents; Pro costs $20 per human seat per month and adds private repos, SSO/SAML, and a larger AI review pool.
- SkillzBench is entirely free: the skills directory, the MCP connector, sandbox benchmarks, and change alerts all cost nothing, with revenue coming from blog advertising only.
- Security is the deciding difference. Koi Security's February 2026 audit found 341 malicious skills among 2,857 audited (a 12% infection rate), and Unit 42 later found five more evasive malicious skills that survived VirusTotal and ClawScan screening.
- ClawHub skills run in-process with the OpenClaw Gateway with no sandboxing, meaning a malicious skill gets the same filesystem and network access as the agent itself.
- SkillzBench benchmarks every skill in a sandbox with the same model, with and without the skill, and publishes the Lift in percentage points so you can see whether a skill actually helps.
- SkillzBench works across agents (Claude Code, Codex, Cursor, Gemini CLI, and chat-app agents) through one free MCP server, while ClawHub is tied to the OpenClaw ecosystem.
Head to head
| Dimension | ClawHub | SkillzBench |
|---|---|---|
| Pricing | Free forever for public repos and unlimited agents; Pro $20 per human seat per month; Enterprise custom annual | Free. Everything: directory, MCP connector, benchmarks, alerts 🏆 |
| Catalog size | 3,286+ skills, 1.5M+ downloads, 2,999 stars, 11 categories (Feb 2026) 🏆 | Continuously crawls and dedupes skills from GitHub; check skillzbench.com for current counts |
| Security vetting | VirusTotal + ClawScan automated screening; no human review; no sandboxing at runtime | Commit-SHA pinning, validation, license check, security scan, and quality judgment on every version 🏆 |
| Benchmarking | Community star ratings, downloads, and comments; no performance benchmarks | Sandbox tests with the same model, with and without the skill, publishing Lift in percentage points 🏆 |
| Agent compatibility | OpenClaw ecosystem agents (OpenClaw Core, Moltbot, Clawdbot, Moltbook) | Claude Code, Codex, Cursor, Gemini CLI, and chat-app agents via one free MCP server 🏆 |
| Publishing | Publish skills and plugins with changelogs, tags, rename, merge, and soft-delete; GitHub account 1+ week old required 🏆 | Not a publishing registry; it indexes and verifies existing skills |
| Ongoing monitoring | Community reporting (3-strike auto-hide), moderator review, and appeals | Tracks upstream changes, re-scans, re-benchmarks, and notifies you automatically 🏆 |
| Independence | Official OpenClaw registry; hosts and distributes the skills it rates via community signals | Independent directory; ad-funded, and ad money never influences scores or listings 🏆 |
The contenders
ClawHub is the public skill and plugin registry for OpenClaw, often described as "npm for AI agents." It hosts text-based agent skills (a SKILL.md plus supporting files) alongside a native package catalog for code plugins, bundle plugins, and experimental whole-agent Claw packages. The platform is CLI-first: you search, install, pin, update, and publish skills through the clawhub CLI, and a web app built on TanStack Start and Convex handles browsing and rendering SKILL.md files. Search uses OpenAI embeddings (text-embedding-3-small) for vector search, so natural-language queries surface conceptually relevant skills instead of relying on brittle keywords. Publishing is open to anyone with a GitHub account at least one week old, with changelogs, tags, rename-with-redirect, and duplicate merging built in. Since the February 2026 ClawHavoc campaign, published skills go through VirusTotal scanning and ClawScan pattern analysis, and OpenClaw has partnered with NVIDIA on skill analysis. The registry tracks 3,286+ skills and 1.5M+ downloads as of February 2026. Pricing is per human seat, never per agent.
pricing 0/month
- + Largest catalog in the OpenClaw ecosystem: 3,286+ skills and 1.5M+ downloads as of February 2026
- + Free tier is genuinely usable, with unlimited public repos and unlimited agents forever
- + Vector search makes discovery far easier than keyword matching
- + CLI-first design fits developer workflows, including inspect-before-install and local pinning
- + Post-ClawHavoc, every published skill is screened through VirusTotal and ClawScan before download
- − No sandboxing: skills run in-process with the Gateway with full filesystem, network, and credential access
- − No human code review before publishing; the bar is a GitHub account one week old
- − Koi Security found 341 malicious skills in a 2,857-skill audit (12% infection rate) in February 2026
- − Unit 42 documented five evasive malicious skills that survived VirusTotal and ClawScan screening between February and May 2026, including infostealers and novel agentic threats
- − Automated scanners catch known malware but cannot reliably catch social engineering or prompt injection, which was the primary ClawHavoc attack method
SkillzBench is AIQ Labs' free, independent directory of AI agent skills plus a free MCP tool that connects any agent to the skills you choose. Agent skills (SKILL.md instruction packages for Claude Code, Codex, Cursor, Gemini CLI, and chat-app agents) are everywhere, but published research found roughly 26% of them carry at least one vulnerability, and until now nobody could tell which ones actually help. SkillzBench answers that with five pillars. Discover continuously crawls and dedupes skills from GitHub. Verify pins every version to a commit SHA, validates it, checks its license, security-scans it, and judges its quality. Benchmark tests every skill in a sandbox with the same model, with and without the skill, and publishes the Lift in percentage points. Equip gives you one free MCP server plus per-agent loadouts: you toggle skills and tools per agent, and the agent only sees what you chose. Stay current tracks upstream changes, re-scans, re-benchmarks, and notifies you. A 0-100 trust score combines security, quality, license, maintenance, and community signals, and Lift is always shown separately from ratings. Everything on the platform is free; revenue comes from advertising around blog articles, and ad money never touches scores or listings.
pricing 0
- + Everything is free: directory, MCP connector, benchmarks, and alerts, with no seat pricing at all
- + Independent verification: commit-SHA pinning, license checks, and security scans happen before you ever install a skill
- + Sandbox benchmarking shows the Lift in percentage points, so you know whether a skill actually improves your agent's output
- + Works across agents (Claude Code, Codex, Cursor, Gemini CLI, chat-app agents) rather than locking you into one ecosystem
- + Per-agent loadouts enforce least privilege: an agent only sees the skills you explicitly enabled
- + Ongoing monitoring re-scans and re-benchmarks when upstream skills change, catching drift after you install
- − It is a verification and connection layer, not a publishing registry, so you will not publish your own skills through it
- − It does not host ClawHub's full catalog, so OpenClaw-specific skills may require going to ClawHub directly
- − The MCP connector adds one more component to manage if you already have a fully native OpenClaw setup
- − Trust scores and Lift figures are SkillzBench's own methodology, so results may differ from your production workloads
Verdict
Winner: none
These two platforms solve different halves of the same problem: ClawHub is a distribution registry with scale, while SkillzBench is a verification and benchmarking layer that works across agents. Many teams will realistically use both, so neither wins outright.
| Option | Overall | Breakdown |
|---|---|---|
| ClawHub | 7.5 | : 9: 8: 8: 5: 9: 6 |
| SkillzBench | 8 | : 10: 7: 8: 9: 6: 10 |
Pricing compared
Security: the difference that matters most
What each platform actually does day to day
Which should you pick?
- :
- :
- :
- :
- :
- :
FAQ
Is ClawHub safe to use for business agents?
It carries real risk that you should manage actively. Koi Security's February 2026 audit found 341 malicious skills out of 2,857 audited, a 12% infection rate, and Unit 42 documented five more evasive malicious skills that survived VirusTotal and ClawScan screening between February and May 2026. ClawHub skills also run in-process with the Gateway with no sandboxing, giving them full access to credentials, sessions, and the filesystem. If you use ClawHub, inspect skills before installing, pin trusted versions, and treat any skill asking you to run terminal commands as hostile until verified.
What does the ClawHavoc campaign mean for small businesses?
It showed that agent skill marketplaces are a real supply chain attack surface, not a theoretical one. The 341 malicious skills used polished documentation and fake "Prerequisites" sections to trick users into running commands that installed Atomic Stealer, which harvests crypto wallets, browser passwords, SSH keys, .env files, and Keychain data. For a small business, a single infected machine can expose customer records, API keys, and payment credentials. The lesson is to verify skills independently before installation, which is exactly the problem SkillzBench's scanning and benchmarking layer addresses.
Is SkillzBench really free, and how does it make money?
Yes, everything on SkillzBench is free: the skills directory, the trust scores, the sandbox benchmarks with published Lift figures, the free MCP connector, per-agent loadouts, and change alerts. The platform earns revenue from advertising shown in and around its blog articles, never by charging users, and ad money never influences trust scores or listings. Lift is always shown separately from ratings so the two signals cannot be conflated.
Can I use ClawHub and SkillzBench together?
Yes, and for OpenClaw teams that is arguably the right setup. ClawHub gives you the largest catalog and native CLI distribution for OpenClaw agents. SkillzBench gives you independent verification: commit-SHA pinning, security scans, license checks, and sandbox benchmarks showing each skill's Lift in percentage points. A practical workflow is to shortlist skills on ClawHub, check their verification status and Lift on SkillzBench, and only then install, pinning the version you tested.
What is "Lift" and why does it matter?
Lift is SkillzBench's benchmarking metric. Every skill is tested in a sandbox with the same model, once with the skill and once without, and the difference in performance is published in percentage points. This matters because popularity is not the same as usefulness: a skill can have hundreds of stars and still make your agent's output worse, or add risk without adding capability. Lift is always shown separately from trust ratings, so you can weigh safety and performance independently.
How does ClawHub's per-seat pricing work in practice?
ClawHub bills per human seat, never per agent. The Free tier is $0 forever for public repos with unlimited agents. Pro costs $20 per human seat per month and includes a pool of 250 platform AI reviews and 10 verify credits per seat monthly, with metered overage at $0.10 per review and $2.00 per verify. Agents run free; the costs scale with how many humans need private repos, governance, and standing 24/7 agents (up to 10 on Pro).