Why Your “Skill Scanner” Is Just False Security (and Maybe Malware)

This report evaluates community AI 'Skill' scanners and finds critical shortcomings: one scanner (SkillGuard) was itself malicious, and others (Skill Defender, Ferret Scan) produced false negatives against a crafted 'vercel' exfiltration skill. It argues that regex/denylist approaches are insufficient for detecting intent-based attacks and recommends AI-native, LLM-driven behavioral analysis (mcp-scan/Evo by Snyk), continuous inventory, and continuous scanning to mitigate supply-chain and agentic-exfiltration risks.