SkillSpector is an open-source security scanner from NVIDIA that answers one question before you add a skill to a coding agent: is this skill safe to install? NVIDIA's skill documentation also uses it as the security tier (Tier 1) of SkillEvaluator, its pre-publication check for skills. It accepts Git repositories, URLs, zip files, directories and single files, and it targets skills for agents such as Claude Code and Codex as well as MCP skills.
Scanning runs in two stages. Fast, deterministic static analysis looks for suspicious strings, dangerous APIs, dependency risk and mismatches between declared and actual permissions, and it can look up known vulnerable dependencies on OSV.dev with an offline fallback. An optional LLM stage compares what a skill claims to do with what its code appears to do. Findings cover 17 categories, including prompt injection, data exfiltration, privilege escalation, supply chain, excessive agency, memory poisoning and tool misuse, and each scan returns a 0-100 risk score with severity labels and a recommendation.
Reports can be printed to the terminal or written as JSON, Markdown or SARIF, which makes the scanner usable as a CI gate. The README also documents baselines for suppressing accepted findings, comparison of MCP registry snapshots, an MCP server mode and extensions for Pi and OpenCode. NVIDIA's guidance says critical or high findings should block a release until fixed or formally accepted, and that hidden instructions and under-declared capabilities should be removed or declared first.
SkillSpector is free under the Apache License 2.0. Static scans need no model; the semantic stage uses whichever model provider you configure, and that usage is billed by the provider. A clean report is not a guarantee: NVIDIA frames the score as one release gate alongside reviewing the permissions a skill actually requests.
