Evidence · DiggingBeagle record
From Instructions to Exploits: Applying Program Analysis to Malicious Agent Skills
Cross-language/code dependency-graph scanner evaluated on 66,157 ClawHub skills; reports 2,473 candidates and 627 malicious samples after follow-up. Corpus outcomes do not count compromised users.
- Source role
- primary disclosure
Evidence record
Cross-language/code dependency-graph scanner evaluated on 66,157 ClawHub skills; reports 2,473 candidates and 627 malicious samples after follow-up. Corpus outcomes do not count compromised users.
Claim-level citations (2)
- supportsSkill scanner correlated natural-language instructions with executable malicious flows: The authors collected 66,157 ClawHub skills and, with additional dynamic and manual checking, report 627 confirmed malicious specimens in their sample.
Abstract; corpus scan and validation description
- supportsSkill scanner correlated natural-language instructions with executable malicious flows: The study distinguishes 2,473 initially suspicious candidates from the smaller confirmed group; neither figure measures infected end users.
Results; large-scale skill scanning and manual confirmation