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

2026-07-07

Researchers show that simple byte changes and a self-extracting packing method called SKILLCLOAK let malicious AI agent skills bypass eight popular static scanners in over 90% of tests, while a sandboxed runtime monitor, SKILLDETONATE, catches most cloaked skills instead.