SkillSpector

SkillSpector is a command-line security scanner from NVIDIA that analyzes AI agent skills (as used by Claude Code, Codex CLI, Gemini CLI and similar tools) before installation. It accepts Git repositories, URLs, zip files, directories or single files, and reports vulnerabilities, malicious patterns and security risks. The README cites a research dataset in which 26.1% of 31,132 analyzed skills contained vulnerabilities and 5.2% showed likely malicious intent.

Analysis runs in two stages: fast static analysis, plus optional LLM-based semantic evaluation. The README lists 71 vulnerability patterns across 17 categories, including prompt injection, data exfiltration, privilege escalation, supply chain, MCP tool poisoning and YARA signatures. Supply-chain checks query OSV.dev for live CVE data with an offline fallback. Output is available as terminal, JSON, Markdown and SARIF reports, with a 0-100 risk score and baseline-based suppression of known findings. It can also scan MCP Registry entries.

It is installed with uv or pip (Python), or run through a locally built Docker image, and can be used as an MCP server or as a Pi or OpenCode extension. LLM analysis needs a configured provider (the README mentions DeepSeek as the default, plus options such as AWS Bedrock and Google Vertex), and can be disabled with --no-llm. SkillSpector is part of NVIDIA's Verified Skills pipeline. Some file-output operations are unavailable on Windows.