Skill Poisoning Detection on Skill-Inject
99PrecisionLLM-only
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| LLM-only2026.05 | 99 | 90 | 94 | |
| BIVModel=Claude Sonnet 4.52026.05 | 98 | 98 | 98 | |
| Rule-based2026.05 | 83 | 21 | 33 | |
| RouteGuard2026.04 | 80.19 | 42.18 | 75.28 | |
| Inj2026.04 | 69.93 | 15.15 | 23.07 | |
| Malicious Or Not2026.04 | 67.9 | 15.27 | 24.51 | |
| MSkills2026.04 | 67.59 | 23.88 | 32.89 | |
| Probe2026.04 | 63.18 | 17.33 | 26.98 | |
| Wild2026.04 | 62.51 | 18.42 | 28.14 | |
| MWild2026.04 | 60.98 | 15.39 | 24.11 | |
| Scan2026.04 | 0 | 0 | 0 |