NVIDIA SkillSpector Scans AI Agent Skills for Security Flaws

NVIDIA released SkillSpector, a security scanner for AI agent skills targeting Claude Code, Codex, and MCP. It detects prompt injection, data exfiltration risks, and known vulnerabilities before a skill is installed. The project gained 166 GitHub stars on day one.

The scanner addresses the agent-native supply chain problem where third-party skills execute with elevated context and tool access. As teams adopt reusable skills, the trust boundary shifts from code review to capability audit. SkillSpector formalizes that audit step, making static analysis a precondition for installation rather than a post-incident forensic tool.

For operators, this changes deployment workflow: skill acquisition now includes a security gate analogous to container image scanning. Builders who previously relied on manual review or sandbox testing can automate pre-install checks. The second-order effect is pressure on skill marketplaces to integrate such scanning natively, making unsigned or unscanned skills operationally unacceptable in production environments. Expect internal skill registries to adopt similar tooling as a standard CI step.