Know what an AI skill does before your assistant runs it.
Skills give your AI assistant new abilities, and the same access to your computer that you have. Skill Scanner reads every line first and tells you, in plain words, whether it's safe to install.
Free, with no account. Every check runs on your own device.
Red lines are bad, green ones are safe. They only show their color under the shield, and some bad lines are invisible until it passes over them.
A skill is a set of instructions your AI follows.
You install it once. From then on your assistant reads it and does what it says, often without asking you again.
It runs with your permissions.
Whatever you can open on your computer, it can open too: documents, photos, projects.
That includes your passwords and keys.
The files that sign you in to GitHub, your cloud accounts and your servers sit right next to everything else.
And it can reach the internet.
One line is enough to send what it read to someone else's server.
It can even change your AI for later.
Some skills write standing orders into your assistant's memory, so the harm outlasts the skill.
Skill Scanner reads every line first.
It shows which of these a skill really does, then tells you what to do about it. For this backup skill: reading files is expected, the rest is not.
See it in 44 seconds.
Check a skill now.
This is the Skill Scanner extension, running right here. Paste a GitHub link or the install command you were about to run, or paste the skill's text. You get the same plain answer you'd get in Chrome.
- Works with GitHub links, ClawHub, and commands like npx skills add.
- Pasted text is checked inside this page and never sent anywhere.
- The extension adds one-click checks on GitHub, the full report, and a warning if a skill you use changes.
What it looks for.
68 checks across 12 kinds of risk, plus a lookup of the software packages a skill relies on in a database of known security holes. These are the risks people run into most. Each example is checked by the real scanner, right now, in your browser.
It instructs the AI to keep its actions secret from you.
Text you can't see on the page, but your AI reads and obeys: comments, invisible characters, look-alike letters.
What we found.
In our tests, Skill Scanner checked 7,728 public AI skills that anyone can install.
The most common risks in skills confirmed dangerous
- Reads your private keys or saved logins13 skills
- Downloads and runs code from the internet7 skills
- Downloads a program and runs it5 skills
- Tries to take over your AI assistant5 skills
- Sends your secrets to the internet4 skills
Checked with Skill Scanner 2.1.0 in October 2026. Suspicious means the skill was graded C, D or F: worth a closer look before installing, or not safe to install. The risks above come from the 30 dangerous skills we confirmed by hand; false alarms are left out. Automated checks can be wrong, so we don't name any skill.
Your skills stay yours.
- Checks run on your device.Pasted text and uploaded files are scanned inside your browser and never sent anywhere.
- No account, no cookies.No sign-up, no ads, no cookies, and no analytics in the extension. This website only counts visits and a few clicks, like starting the game, without knowing who you are. The video only loads from YouTube if you press play.
- Every connection is listed.It only contacts GitHub or ClawHub when you check a link, and OSV.dev to look up software packages with known security holes. The extension keeps a log of each one.
- An AI second opinion, only if you want one.Connect your own Claude, ChatGPT or Gemini account, or a free model on your computer. You see exactly what would be sent before anything is.
Questions people ask.
- What is an AI skill?
- A skill is a small package, usually a SKILL.md file with instructions and sometimes a few scripts, that teaches an AI assistant a new job. Claude Code, OpenAI Codex, OpenClaw and other agents install them from GitHub or from registries like ClawHub. Once installed, the assistant follows the skill with the same access to your computer that you have.
- Can an AI skill be dangerous?
- Yes. A skill can tell your assistant to read your passwords and keys, send your files to someone else's server, download and run a program, or leave standing orders in your AI's memory. Most skills are fine, though. Of the 7,728 public skills we checked, 170 (2.2%) looked suspicious, and we confirmed 30 as dangerous by hand.
- How do I check if a skill is safe before I install it?
- Paste its GitHub link, its install command or its text into the checker on this page. Or add the Chrome extension and click it on the skill's GitHub page. In a few seconds you get a verdict, the reasons in everyday words, and the exact lines behind them.
- Which skills can it check?
- Skills in the SKILL.md format, along with their scripts, and MCP tool definitions. You can give it a GitHub link, a ClawHub skill, an install command such as npx skills add, or the skill's text. The extension can also check files you upload.
- What do the grades mean?
- A means we found nothing risky. B means a few small notes and nothing harmful. C means some parts deserve a closer look before you install. D means risky: only install it if you trust the author and understand each item. F means don't install it.
- Does Skill Scanner send my skills anywhere?
- Text you paste and files you upload are checked inside your browser and never leave it. When you check a link, it downloads that skill from GitHub or ClawHub, the same as opening it yourself. If the skill lists software packages, their names and versions are looked up on OSV.dev, Google's public database of known security holes.
- Is it free? Do I need an account?
- It's free and there's no account. The AI second opinion is optional: it uses your own Claude, ChatGPT or Gemini key, or a free model running on your computer, and it shows you exactly what would be sent before anything is.
- Can it miss something?
- Yes. It's an automated check, so a new trick can get past it, and it sometimes flags something harmless, like a security tool that quotes attack code in order to block it. Treat a good grade as one reason to trust a skill, next to knowing who wrote it.
Check skills where you find them.
Open a skill on GitHub and click Skill Scanner, or paste the install command you were about to run. You get a plain answer in seconds, and a warning if a skill you trust changes later.
Add to Chrome, it's freeTalk to us.
Write to us with a question, or tell us about your organization.
- Questions and feedbackHow a check works, a result you don't understand, or a skill we missed or flagged by mistake.
- OrganizationsRolling Skill Scanner out to a team, checking the skills your company uses, or working together.