Malicious Skill Detection on ClawHub Prompt Injection 1.0 (n=19)
100Catch RateSkillVetBench
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| SkillVetBench2026.05 | 100 | 95 | 97 | 0 | |
| SkillSieve2026.05 | 79 | 88 | 83 | 21 | |
| SkillProbe2026.05 | 74 | 88 | 80 | 26 | |
| LLMEvaluation Protocol=few-shot2026.05 | 68 | 87 | 76 | 32 | |
| LLMEvaluation Protocol=0-shot2026.05 | 63 | 80 | 71 | 37 | |
| ClawScan2026.05 | 16 | 100 | 27 | 84 | |
| ClawVet2026.05 | 10 | 100 | 18 | 90 | |
| VirusTotal2026.05 | 0 | — | — | 100 | |
| CodeBERT2026.05 | 0 | — | — | 100 |