SkillCloak Lets Malicious AI Agent Skills Evade Static Scanners with Self-Extracting Packing

Researchers from the Hong Kong University of Science and Technology have demonstrated SKILLCLOAK, a technique that enables malicious AI agent skills to evade static scanners by using self-extracting packing and obfuscation methods. These cloaked skills hide payloads in ignored directories like .git/ or use character substitution and command splitting to bypass pattern-based detection, remaining undetected in over 90% of cases across eight tested scanners. The study also introduces SKILLDETONATE, a runtime behavioral checker that detects 97% of attacks by monitoring system-level activity, highlighting the insufficiency of static analysis alone. Real-world marketplaces already host malicious skills using similar evasion tactics, indicating active exploitation.

AI-extracted · verify before operational use