OpenClaw Skills in 2026: 11 Verified ClawHub Picks (and the Ones to Skip)
Most OpenClaw skill lists are padded with skills that do not exist. I checked each on ClawHub, so here are the ones that install cleanly and are safe to run.
OpenClaw skills are markdown instruction files (SKILL.md) that teach your agent how to use a tool, published on the ClawHub registry. The most useful verified ones in 2026 include GitHub, Slack, Discord and Google Workspace, many written by OpenClaw's own creator, plus Agent Browser, ElevenLabs and a skill security auditor. Install them with "openclaw skills install @author/slug", not the "claw install" command the copycat lists give you.
On this page
- TL;DR
- What are OpenClaw skills, and how are they different from plugins?
- How do you actually install an OpenClaw skill?
- Where do OpenClaw skills live?
- The 11 best OpenClaw skills at a glance
- Which skills should a developer install first?
- Which skills handle Slack, Discord and voice?
- Which skills run your Gmail, Calendar and smart home?
- Research and self-improvement
- Security
- Which OpenClaw skills should you skip?
- Are ClawHub skills safe to install?
- Can you skip the registry entirely?
- How do you install and vet a skill, step by step?
- Final take
- Common questions
Search "best OpenClaw skills" and you get the same list twenty times over: twenty skills, a download count next to each, and an install command that looks official.
So I installed OpenClaw, opened the ClawHub registry, and checked them one by one.
Two things fell apart fast. The install command nearly every list prints is wrong, and the download numbers do not match anything on the actual skill pages.
Almost every "best skills" list was copied from another list, not from the tool.
So here is the version I would trust: eleven skills I confirmed are real, each with the command that actually works, and an honest note on the ones to skip. No invented numbers.
What are OpenClaw skills, and how are they different from plugins?
An OpenClaw skill is not code. In the docs' own words, "skills are markdown instruction files that teach the agent how and when to use tools." Each one is a folder with a SKILL.md file: a bit of YAML frontmatter (a name and a description) and a markdown body telling the agent how to do something.
That is worth knowing before you install anything, because it means you can read a skill in full before you trust it.
It is prose, not a binary. You can vet every word before you enable it.
Plugins are the other half. A plugin is the actual code package, published to ClawHub separately. Skills are the instructions; plugins are the tools those instructions drive. A lot of guides blur the two. They are not the same thing.
How do you actually install an OpenClaw skill?
With openclaw skills install @author/slug, and nothing else. Almost every "best skills" article tells you to run claw install github, but there is no claw binary, and skill names are not global: the real command uses openclaw (or the clawhub CLI), and every skill is namespaced to its author.
So the GitHub skill is not claw install github. It is this:
openclaw skills install @steipete/github
You can see it straight off the live page above. The other forms are real too, and worth knowing:
openclaw skills install git:owner/repo@refinstalls straight from a Git repo.openclaw skills install ./path/to/skill --as my-toolinstalls a local folder.- Add
--globalto make a skill available to every local agent. - The separate
clawhubCLI (clawhub install @author/slug) is for registry-authenticated work.
If a guide hands you claw install <name> with no author, it was written from other guides, not from the tool.
Where do OpenClaw skills live?
Two places, both open: the ClawHub registry at clawhub.ai, and the community monorepo on GitHub. Both let you read a skill's full SKILL.md before it ever touches your machine.
Read the SKILL.md before you install it. That one habit is the whole safety story.
The registry is ClawHub. It is browsable, with Top, Trending and New filters, and each skill has a page showing its SKILL.md, its author, its version and a security-scan badge.
Community skills also live in the open github.com/openclaw/skills monorepo, where they are submitted as author/skill folders. So "OpenClaw skills on GitHub" is a real thing, not just marketing.
One number I will not repeat: some lists claim ClawHub hosts "13,700+ skills". I could not confirm any total from the registry itself, and the figures floating around contradict each other wildly. So treat the catalog as "large and growing", and judge each skill on its own page, not on a headline count.
The 11 best OpenClaw skills at a glance
Every skill below resolves to a real ClawHub or monorepo page. I have left download counts out on purpose, they are not verifiable, and the slug is what you actually need.
| Skill | Category | By | Install slug |
|---|---|---|---|
| GitHub | Dev & code | @steipete | @steipete/github |
| Agent Browser | Dev & code | @thesethrose | @thesethrose/agent-browser |
| Vercel | Dev & code | @leonaaardob | @leonaaardob/lb-vercel-skill |
| Slack | Communication | @steipete | @steipete/slack |
| Discord | Communication | @steipete | @steipete/discord |
| ElevenLabs | Communication | @odrobnik | @odrobnik/elevenlabs |
| Google Workspace (GOG) | Productivity | @steipete | @steipete/gog |
| Home Assistant | Productivity | @iahmadzain | @iahmadzain/home-assistant |
| Summarize | Research | OpenClaw core | bundled |
| Self-Improving Agent | Self-improvement | @pskoett | @pskoett/self-improving-agent |
| Skill Security Auditor | Security | @akhmittra | @akhmittra/skill-security-auditor |
Which skills should a developer install first?
Method 1
GitHub
Best for: Running your dev workflow by chat
The most useful first skill for a developer. Written by @steipete, OpenClaw's own creator, it wraps the gh CLI so the agent can read issues, check CI, review pull requests and query the API.
It is MIT-0 and passes ClawHub's security scan. That author signal matters: a skill by the person who built OpenClaw is about as trustworthy as the registry gets.
openclaw skills install @steipete/githubMethod 2
Agent Browser
Best for: Autonomous web browsing
When your agent needs to actually use a website, not just call an API, Agent Browser is the skill. It is a headless browser-automation tool by @thesethrose, living in the official openclaw/skills monorepo so you can read its source in full.
Good for scraping, form-filling and any task that only exists behind a UI. Give it a scoped profile, not your logged-in browser.
openclaw skills install @thesethrose/agent-browserMethod 3
Vercel
Best for: Shipping deployments from chat
If you deploy on Vercel, this skill by @leonaaardob wraps the Vercel CLI so you can trigger and check deployments in plain language. "Ship the latest to preview" becomes one instruction.
Handy, but this is exactly the kind of skill to scope carefully, a deploy tool with your token can do real things. Use a project-scoped key.
openclaw skills install @leonaaardob/lb-vercel-skillWhich skills handle Slack, Discord and voice?
Method 4
Slack
Best for: Team messaging automation
Another @steipete skill. It drives Slack through a bot token, so the agent can post updates, read channels and react. This is the one that turns an agent into something your team actually notices, a daily standup summary, an alert when a build breaks.
openclaw skills install @steipete/slackMethod 5
Discord
Best for: Running a community
The @steipete Discord skill handles messages, reactions, threads, polls and pins. Discord is also a native OpenClaw messaging channel, so you can both talk to your agent there and have it manage a server.
Useful if you run a community and want an assistant that lives where the members already are.
openclaw skills install @steipete/discordMethod 6
ElevenLabs
Best for: Giving your agent a voice
For text-to-speech and voice, the @odrobnik ElevenLabs skill connects the agent to ElevenLabs' API so it can speak, or place voice calls. There are a few ElevenLabs skills on ClawHub; this is the cleanest single-author one I found.
Worth it if you want spoken briefings or a phone-style agent rather than a text bot.
openclaw skills install @odrobnik/elevenlabsWhich skills run your Gmail, Calendar and smart home?
Method 7
Google Workspace (GOG)
Best for: Gmail, Calendar, Drive and Docs
GOG is the productivity anchor, and it is another @steipete skill. It is a Google Workspace CLI: Gmail, Calendar, Drive, Contacts, Sheets and Docs, all reachable in plain language.
This is the one that makes an agent feel like an assistant rather than a toy, "what's on my calendar tomorrow and who hasn't replied" becomes a single ask.
openclaw skills install @steipete/gogMethod 8
Home Assistant
Best for: Local smart-home control
If you run Home Assistant, this skill by @iahmadzain lets the agent control your local smart home, lights, scenes, sensors, without routing anything through a cloud service. Everything stays on your own network.
The privacy-friendly pick for anyone whose reason for self-hosting is keeping data at home in the first place.
openclaw skills install @iahmadzain/home-assistantResearch and self-improvement
Method 9
Summarize
Best for: Condensing long content
You do not install this one. Summarize ships bundled with OpenClaw in the core repo's skills folder, so it is there the moment you set the agent up. Point it at an article, a thread or a document and it gives you the short version.
It is a good reminder that the base agent already does a lot before you touch the registry at all.
Bundled with OpenClaw, no separate install needed.
See OpenClaw's bundled skills →Method 10
Self-Improving Agent
Best for: An agent that learns from its mistakes
The @pskoett Self-Improving Agent skill captures learnings, errors and corrections over time, so the agent gets a little sharper at your recurring tasks instead of repeating the same mistakes.
It is actively maintained and MIT-0. Of the "self-improvement" skills on ClawHub, this is the clean one, I skipped a more popular fork over an unresolved data-exfiltration report (more on that below).
openclaw skills install @pskoett/self-improving-agentSecurity
Method 11
Skill Security Auditor
Best for: Vetting a skill before you install it
The most important skill on this list, and the one no copycat guide leads with. The @akhmittra Skill Security Auditor audits another ClawHub skill for risky patterns before you enable it, exactly the check you want given anyone can publish.
Install this first, then run it against everything else. Note the slug is skill-security-auditor, not the bare security-auditor the lists print.
openclaw skills install @akhmittra/skill-security-auditorWhich OpenClaw skills should you skip?
Mission Control, Frontend Design and Capability Evolver. All three show up on the recycled lists, and none survived a check. Cutting is half the job, so here is what is wrong with each:
- Mission Control is not a skill you can install. The "morning briefing dashboard" is actually a separate third-party dashboard app you run with Docker, not something you
openclaw skills install. - Frontend Design does not exist under that slug. There are real UI-generation skills (one is
anti-slop-design), but the entry every list copies points at nothing. - Capability Evolver is a real skill, but it is a contested community fork with an unresolved data-exfiltration report filed against it. Real is not the same as safe.
The pattern to avoid in general: a skill with almost no history, a closed or vague source, or one that asks for broad Bash or root-level access it does not obviously need. That is the profile a bad skill hides behind.
Are ClawHub skills safe to install?
Short answer: treat every one as untrusted until you have read it.
That is not me being cautious, it is OpenClaw's own guidance. The docs say plainly: "Treat third-party skills as untrusted code. Read them before enabling." Because a skill is just a SKILL.md file, you genuinely can read it first, so do.
On the registry side, ClawHub runs automated security scans, shows a scan-state badge on each skill, and gates publishing behind an aged GitHub account. That is real, but it is automated, not human-curated, and publishing is open to anyone.
You will also find posts throwing around scary, specific breach numbers, "X% of the registry is malicious", a named mass-poisoning attack. I could not verify any of those against a primary source, so I am not going to repeat them as fact. The honest, verifiable position is simpler: read the SKILL.md, prefer skills by known authors like @steipete, run the security auditor, avoid broad-permission skills, and sandbox the agent.
Can you skip the registry entirely?
Yes: run an agent that already has the tools you need built in, so there is no registry to vet. That sidesteps the whole "is this skill safe" question in one move.
That is why I built Grip AI, a lightweight OpenClaw alternative written in Python. Full disclosure, it is mine, and it is free and MIT-licensed. It ships browser automation, scheduling and messaging as core features, and it sandboxes the agent to its own working directory by default. Nothing to install from an open marketplace, nothing unvetted running on your machine.
It is not a replacement for OpenClaw's huge ecosystem. It is the answer when you want a small, readable agent you fully control. If that is the trade you want, the full comparison is in the OpenClaw alternatives roundup.
How do you install and vet a skill, step by step?
- Find it on clawhub.ai and open its page. Note the author, the version, and the security-scan badge.
- Read the
SKILL.md. It is right there on the page. If it wants broadBashor root access it does not need, stop. - Prefer known authors. A skill by @steipete or in the official monorepo is a safer bet than an anonymous one with no history.
- Install with the real command:
openclaw skills install @author/slug. Add--globalonly if you want every local agent to have it. - Sandbox the agent and give it scoped API keys, never your main credentials.
Want an agent with the tools already built in?
Grip AI is the lightweight OpenClaw alternative I wrote and use daily. Browser, scheduling and messaging are baked in, the agent stays sandboxed, and there is no registry to vet. Free and MIT-licensed.
Get Grip AI on GitHubFinal take
OpenClaw's skills are genuinely useful, and the best of them are written by the people who built OpenClaw itself. But the guides ranking for "best OpenClaw skills" mostly copied each other, wrong command, invented download counts, a couple of skills that do not exist. I ran the same verification on the best Codex skills, and the copy-paste problem is identical there.
Use the real command, openclaw skills install @author/slug. Read the SKILL.md before you enable anything. Start with the security auditor.
And if vetting a marketplace is not how you want to spend your time, run a lean agent that already has the tools built in.
Common questions
What are OpenClaw skills, and how are they different from plugins?
A skill is a markdown file (SKILL.md) that teaches the agent how and when to use a tool it already has. A plugin is the code side, a package published to ClawHub. In short: skills are instructions, plugins are the tools those instructions drive.
How do I install an OpenClaw skill from ClawHub?
Use "openclaw skills install @author/slug", for example "openclaw skills install @steipete/github". Slugs are namespaced to their author, so you always need the @author prefix. The old "claw install name" command many lists show does not exist; the binary is openclaw (or clawhub).
Are ClawHub skills safe to install?
Treat them as untrusted code, in OpenClaw's own words. ClawHub runs automated security scans and shows a scan badge, but anyone can publish. Read the SKILL.md before enabling it, avoid skills that ask for broad Bash or root access, and sandbox the agent.
What are the most useful OpenClaw skills right now?
For most people: GitHub, Slack and Google Workspace for daily work, Agent Browser for web automation, and a search skill wrapping Tavily or Exa. Many of the strongest are written by @steipete, OpenClaw's creator, which is a good authenticity signal.
Where do OpenClaw community skills live on GitHub?
On the ClawHub registry at clawhub.ai, and in the open github.com/openclaw/skills monorepo where community skills are submitted. Each skill is a folder with a SKILL.md file, so you can read exactly what it does before you install it.
Can I use OpenClaw's abilities without installing skills?
Yes. The lighter route is an agent that ships its core tools built in, so there is no registry to vet. Grip AI, a lightweight OpenClaw alternative, sandboxes the agent to its working directory and includes browser, scheduling and messaging tools out of the box.
SEO Specialist and product builder with 10+ years in search. The notes come from the work, not the theory.