Malicious Skill Detection on SKILLVETBENCH labeled set 1.0 (78 malicious + 22 benign skills)
0.95BalanceSKILLVETBENCH
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| SKILLVETBENCH2026.06 | 0.95 | 100 | 96 | 0 | |
| SkillSieve2026.06 | 0.84 | 85 | 90 | 15 | |
| SkillProbe2026.06 | 0.82 | 81 | 88 | 19 | |
| LLMPrompting strategy=few-shot2026.06 | 0.8 | 78 | 88 | 22 | |
| LLMPrompting strategy=0-shot2026.06 | 0.76 | 74 | 87 | 26 | |
| CodeBERT2026.06 | 0.68 | 70 | 95 | 30 | |
| ClawScan2026.06 | 0.56 | 48 | 100 | 52 | |
| ClawVet2026.06 | 0.53 | 41 | 100 | 59 | |
| VirusTotal2026.06 | 0.46 | 33 | 100 | 67 |