SkillSpector
Review AI agent skills for prompt injection, data theft, and unsafe code before installing them, using NVIDIA's open-source security scanner.
What is SkillSpector?
SkillSpector helps you inspect an agent skill before giving it access to your workflow. The NVIDIA project scans repositories, URLs, archives, directories, and individual files for suspicious instructions and security risks. Use static checks for a fast first pass, then add optional language-model analysis for deeper review. Reports include findings, severity, recommendations, and a risk score, with terminal, JSON, Markdown, and SARIF output for review or development workflows. It fits teams that install third-party skills or maintain their own skill collections. Treat the report as evidence for a review decision: scan results do not guarantee a skill is safe.