ProtectSkills/MaliciousAgentSkillsBench
MaliciousAgentSkillsBench A security benchmark dataset of Claude Code Agent Skills, from the USENIX Security 2026 paper "Do Not Mention This to the User": Detecting and Understanding Malicious Agent Skills in the Wild โ the first systematic study of malicious skills in the Claude Code ecosystem. ๐ Paper: https://arxiv.org/abs/2602.06547 ๐ป Code & full evaluation framework: https://github.com/protectskills/MaliciousAgentSkillsBench ๐ฆ Permanent archive:โฆ See the full description on the dataset page: https://huggingface.co/datasets/ProtectSkills/MaliciousAgentSkillsBench.
Update dataset to paper final release: 98,380-skill snapshot; 157 confirmed malicious (add Severity); refresh dataset card
Update README.md
Upload 3 files
initial commit
