NVIDIA SkillSpector
Open source (Apache 2.0) from NVIDIA and part of its Verified Skills pipeline. Point it at a skill folder, a single SKILL.md, a zip, a URL or a Git repo and it returns a 0 to 100 risk score with findings across 71 patterns in 17 categories, including prompt injection, data exfiltration, privilege escalation, excessive agency, MCP tool poisoning and known CVEs (looked up on OSV.dev). The --no-llm flag runs the fast static checks only; without it, a second pass asks an LLM provider you configure to judge intent. Reports come out as terminal text, JSON, Markdown or SARIF for CI. The README cites research finding vulnerabilities in 26.1% of 31,132 analyzed skills. Version 2.13.1 shipped Oct 23, 2026. Needs Python 3.12+, or build the included Dockerfile.
uv tool install git+https://github.com/NVIDIA/skillspector.git && skillspector scan ./my-skill/ --no-llmUse the NVIDIA SkillSpector skill for this task: [describe the task]. Before you start, tell me which steps of the skill apply and anything you need from me. When done, list what you changed and how you checked it.Paste into your agent after installing the skill. Replace the [brackets].