NVIDIA has released SkillSpector, an open-source security scanner that vets “skills” for AI agents for risks before you install them. The tool inspects skills for Claude Code, Codex and MCP – and warns of hidden dangers.
What SkillSpector checks
You point the scanner at a directory, a ZIP file, a single SKILL.md or a Git URL. Back come a list of findings, a risk score and concrete recommendations. It detects vulnerabilities, malicious patterns, prompt injection, data exfiltration and supply-chain risks – 68 vulnerability patterns across 17 categories in total.
Static analysis plus optional AI
By default it runs fast static checks: an AST walk flags exec, eval, subprocess and dynamic imports; a taint tracker follows environment variables and file contents to network sinks; YARA rules match known malware, webshells and cryptominers. Optionally, an LLM-based semantic analysis can be added when comparing actual intent is needed.
Free on GitHub
SkillSpector is available for free on GitHub (NVIDIA/SkillSpector), currently at version 2.0.0 and in active development. Given the fast-growing number of freely shared agent skills, a vetting step before installation is sensible protection – especially in team and production environments.
Sources: Help Net Security, GitHub: NVIDIA/SkillSpector.
