NVIDIA SkillSpector: Security Scanner for AI Agent Skills
NVIDIA SkillSpector is an open-source security scanner designed to detect prompt injection, data exfiltration, and supply-chain risks in AI agent skills before
SkillSpector is an open-source security scanner designed to inspect AI agent skills—such as those for Claude Code, Codex, and Model Context Protocol (MCP)—for prompt injection, data exfiltration, supply-chain risks, and malicious patterns before they are installed. As the ecosystem expands, SkillSpector offers checks that help developers review third-party skills before use.
Image source: NVIDIA official documentation
Core Features and Detection Architecture
SkillSpector systematically analyzes agent skills for hidden vulnerabilities using a robust set of capabilities:
- 71 Vulnerability Patterns Across 17 Categories: Comprehensive checks covering prompt injection, sensitive data exfiltration, unauthorized network requests, and supply-chain weaknesses.
- Two-Stage Analysis Pipeline: Combines rapid static analysis to filter out known signatures with selective LLM semantic analysis for deep contextual threat diagnosis.
- Real-Time CVE Integration: Integrates with OSV.dev for real-time vulnerability lookups, with an offline fallback when live lookup is unavailable.
- Multiple Reporting Formats: Generates output in terminal, JSON, Markdown, and SARIF formats, allowing seamless integration into CI/CD pipelines.
Target Audience and Value
SkillSpector is ideal for developers and security teams who frequently integrate AI agents and custom skills. By incorporating automated security checks before pulling code from untrusted sources, it helps identify potential privilege-escalation and data-exfiltration risks before installation.
Caveats
As open-source software, users should review official license terms and scanning policies before deploying it in strict production environments.