SkillSpector: Catch the One Line in a Claude Skill That Steals Your Keys
A free, open-source scanner from NVIDIA. Point it at a skill folder or GitHub link and it checks every file for 71 known attack patterns, then gives a 0-100 risk score and a plain answer: safe, caution or do not install
Claude skills run with a lot of trust: a skill is a folder of instructions and scripts that Claude follows and runs. NVIDIA's SkillSpector reads that folder before you install it and tells you whether anything inside looks dangerous. In NVIDIA's research set of 31,132 skills, 26.1% had a vulnerability and 5.2% showed likely malicious intent.
NVIDIA's own example
The README shows a scan of a skill called suspicious-skill. Its parts: SKILL.md, scripts/sync.py and requirements.txt. SkillSpector scores it 78 out of 100, severity HIGH, recommendation DO NOT INSTALL, and points at the exact spot: "HIGH: Env Variable Harvesting (E2), scripts/sync.py:23. This code collects environment variables containing API keys and secrets, then sends them to an external server."
What it checks (from the README)
- 71 patterns in 17 categories, including prompt injection, hidden instructions, data exfiltration, stealing environment variables (API keys), privilege escalation, supply-chain risks and MCP tool poisoning.
- Two stages: a fast static scan, then an optional AI review of what the code and instructions actually do.
- Scans a folder, a single SKILL.md, a zip file or a GitHub link.
- Output: a risk score from 0 to 100, a severity (LOW to CRITICAL) and a recommendation: SAFE, CAUTION or DO NOT INSTALL, with the file and line of each finding.
Free, for your email
Get the full setup steps + the copy-paste setup prompt
The rest of this guide (3 more sections, 1 copy-paste prompt) opens right here. The setup prompt, one paste into Claude that sets it all up, comes by email only. Use your real address.
50,000+ follow where these guides come from. No spam, unsubscribe in one click.