Researchers identify 627 malicious ClawHub skills using instruction and code analysis

Researchers report identifying 627 malicious skills among 66,157 ClawHub skills after screening with a new security analysis framework and confirming findings through auxiliary dynamic validation and manual review. Their preprint introduces the Supply-Chain Scanner for Agent Skills, which examines natural-language instructions alongside executable code. The scanner translates instructions into a constrained Python-style representation, then links their dependencies with those of bundled code to trace security risks across both. On a manually verified benchmark containing 117 malicious and 117 benign skills, it achieved average recall of 95.2% and accuracy of 85.6% across five language models. Precision was 80.2%, and average token consumption was 13.7 million.