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

Researchers at the Hong Kong University of Science and Technology have developed a method to bypass AI coding agent scanners by using self-extracting packing and character substitution to disguise malicious skills. Their research, detailed in a preprint titled "Cloak and Detonate," highlights a significant gap in current security measures, as malicious skills are routinely slipping past scanners. They also created a runtime checker, SKILLDETONATE, that monitors skill behavior at the operating system level to identify attacks, achieving a lower false-positive rate than traditional scanners. The research demonstrates that malicious skills are already prevalent in public marketplaces, and attackers are utilizing the techniques described to evade detection.