Tag
Introduces skill cascading attacks on skill-based agent systems, where multiple benign-looking skill modifications collectively cause harm while evading detection, and presents an automated red-teaming framework and benchmark to study this threat.