SkillTrustBench is the premier benchmark for evaluating AI security scanners that detect malicious agent skills. Jointly released by Tencent Zhuque Lab and The Chinese University of Hong Kong (Shenzhen). The dataset contains 5,520 samples across 9 attack categories (T01–T09) and 5 agent-dependency layers (A–E).
Please enable JavaScript to view the interactive leaderboard and dataset overview.