AI Compass / Quick reads
Third-party skills are third-party code
Skills are handy, but they run with your permissions. A scan before installation takes minutes. It is not an all-clear.
In depth: Vet third-party skills before they run
A skill is a folder with instructions and often scripts that an agent such as Claude Code loads when needed. The agent treats what is in it like an instruction from you. NVIDIA has published a free scanner, SkillSpector (Apache 2.0). According to the project README, of 31,132 skills examined, 26.1% contained vulnerabilities and 5.2% probably showed malicious intent (externally checked, the study itself not read). AI LABS shows the workflow: first scan the deliberately dangerous test skills, then your own candidate. The standard scan looks for patterns, the AI scan checks whether a skill does what it claims. A test skill that lies about its purpose scored 0 in the standard scan and 100 in the AI scan.
Tool and alternative
SkillSpector is free; the AI scan uses up quota or runs locally via Ollama. You no longer need the conversion to Claude from the video: according to the README (11 October 2026), the Claude command line is now built in as a provider. A side note on irony: in the same breath, Jack Roberts recommends “just giving” repo links to the agent (15 September). I advise against that.
Sources
- AI LABS, “They Finally Fixed Claude Code’s Biggest Problem”, 17 June 2026 (0:33–6:35)
- Jack Roberts, “7 New Github Repos You Need NOW”, 15 September 2026 (2:39–3:13)
- NVIDIA SkillSpector, README, retrieved 11 October 2026
Advertising in the source: AI LABS sponsored by Nimbalyst (approx. 8:05–9:00), the finished skill is in the paid community. Jack Roberts with affiliate links (Floot, Glaido) and a masterclass.
Want to try this with your team? That’s what the AI Lab is for. See the AI Lab