SkillSpector is NVIDIA’s open-source security scanner for AI agent skills. It inspects skills for Claude Code, Codex, and MCP-based agents before you install them, and it has quickly passed 20,700 stars on GitHub following its release, with active updates in October 2026 under the Apache-2.0 license.
Agent skills are powerful: a single skill file can instruct your coding agent to run commands, read files, and call external services. That makes malicious or sloppy skills a real supply-chain risk. SkillSpector addresses it by statically analyzing skill packages for dangerous patterns: hidden prompt-injection instructions, attempts to exfiltrate data or credentials, suspicious network or shell behavior, and other vulnerabilities that are easy to miss when eyeballing a markdown file. Think of it as an antivirus and linter combined, purpose-built for the agent-skill ecosystem.
The need for a tool like SkillSpector grows with the skill ecosystem itself. Marketplaces and GitHub are filling with community skills of wildly varying quality, and each one effectively asks for broad permissions over your development environment. SkillSpector gives that install decision an evidence base: it produces a structured report of what a skill contains, which instructions look manipulative, and which behaviors could move data off the machine, so reviewers can approve, reject, or sandbox with confidence. Run in CI, it can also gate internal skill registries, ensuring that only vetted skills reach a team’s agents.
SkillSpector is free and open source under the Apache-2.0 license. There is no paid edition; it is a community security tool from NVIDIA.
SkillSpector is for developers and security teams who install third-party agent skills, publish skills for others, or manage fleets of coding agents and need assurance that a skill will not quietly compromise their machines or data.