Malicious Skill Detection on ClawHub Unsafe File Ops 1.0 (n=10)
100Catch RateSkillVetBench
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| SkillVetBench2026.05 | 100 | 91 | 95 | 0 | |
| LLMEvaluation Protocol=few-shot2026.05 | 80 | 89 | 84 | 20 | |
| SkillProbe2026.05 | 80 | 89 | 84 | 20 | |
| SkillSieve2026.05 | 80 | 89 | 84 | 20 | |
| LLMEvaluation Protocol=0-shot2026.05 | 70 | 88 | 78 | 30 | |
| CodeBERT2026.05 | 60 | 100 | 75 | 40 | |
| ClawScan2026.05 | 50 | 100 | 67 | 50 | |
| ClawVet2026.05 | 40 | 100 | 57 | 60 | |
| VirusTotal2026.05 | 20 | 100 | 33 | 80 |