Free skill security scanner

Is this skill safe? Check before you install it.

Get a clear answer: safe, use with caution, or do not install.

  • Never run
  • Scanned by NVIDIA SkillSpector
  • Nothing stored
01

Your skill

Add the skill to scan

Free · nothing is run or stored

Check a skill before you install it

Free security scan for AI agent skills that finds prompt injection, credential theft and data leaks without running anything.

Why it matters

Installing an agent skill is more like running a script than reading a doc. Your agent will open the files it names, visit the links it lists and run the commands it gives. The risky line is rarely hidden. It sits in plain English among normal setup steps, right where a quick skim stops paying attention.

The skill evaluator asks whether a skill is any good. This tool only asks whether it is safe. A polished skill that quietly tells your agent to read your SSH key is more dangerous than a sloppy one, and a quality score would only make it look better.

What the tool does

Paste SKILL.md text, give a public HTTPS or GitHub link, or upload a SKILL.md or ZIP up to 2 MB. GitHub folders are fetched up to 10 MB, 100 files and 5 MB unpacked. Links must be HTTPS, can't carry passwords and can't point at a private network, and every redirect is checked again. The skill is read as text in a temporary folder. Nothing is installed or run.

Two checks run. One matches known risky patterns. An AI check reads for intent, which catches lines that look harmless but aren't. If the AI check can't run, the scan uses patterns only and says so. You get a score out of 100, a verdict, counts by severity, and each finding with its file, line, excerpt, confidence and a suggested fix. Nothing is saved, so every run is a fresh scan.

Terms worth knowing

SkillSpector
NVIDIA's open scanner for agent skills. It flags each risky line with a rule, a severity, a location and a confidence.
Security score
100 minus the risk score. No findings scores 100, and each finding lowers it by its severity and confidence.
Verdict
The call next to the score: Safe to try, Use with caution, or Do not install. A risk score of 51 or more is do not install; 21 to 50 is caution.
Prompt injection
Text that tries to take over the agent instead of doing the task, like "ignore earlier rules" or a fake system message.
Data exfiltration
An instruction that sends your files, secrets or chats somewhere else, like an upload or webhook the task never needed.
Supply-chain risk
Setup steps that download and run outside code, often a pipe-to-shell install or a package with no pinned version.
Pattern and AI checks
Pattern checks find known shapes; the AI check reads for intent. The result always says which ones ran.

Frequently asked questions

Quick answers to common questions.

Is the skill installed or executed?

No. Files are read as text in a temporary folder. Nothing is unpacked into a workspace, no script is run and no instruction is carried out.

How is this different from the skill evaluator?

The evaluator adds a quality score, saves the result, and gives you a share page and a README badge. This tool only checks security and saves nothing. Use it when the only question is whether a skill is safe to install.

What does the security score mean?

It is 100 minus the risk score. Findings count by severity (critical 50, high 25, medium 10, low 5), scaled by confidence, with extra weight in script files and less for repeats of the same rule. A risk score of 51 or more reads Do not install, 21 to 50 reads Use with caution, and below that reads Safe to try.

Does a clean scan mean the skill is safe?

No, and the result says so. The scan only reads the text in front of it, so it can't follow scripts the skill downloads later or know what the author intends. Read anything you didn't write and test it in a sandbox first.

What does it actually look for?

Four kinds of risk: prompt injection, access to passwords and keys (SSH keys, cloud credentials, tokens), data leaks, and unsafe installs. Every finding shows the file and line so you can check it yourself.

Is my skill stored, logged, or used for training?

No. It sits in a temporary folder during the scan and is deleted after. That is also why there is no share link: nothing is kept.

Can I scan a repository with several skills in it?

Yes, use the GitHub link. Each SKILL.md is scored on its own and the verdict reflects the riskiest one, so many safe skills don't add up to a bad score. The result shows how many skills and files were read.

How many scans can I run?

Three every 30 minutes without an account, or ten on a paid plan when signed in. Your IP address is only kept as a hash. A large scan with the AI check can take up to four minutes.