SkillCloak is a technique developed by researchers at the Hong Kong University of Science and Technology that rewrites malicious AI agent skills to evade static scanners. It uses self-extracting packing and character manipulation to bypass detection, achieving over 90% evasion across eight tested scanners.