Pre-load security auditing on SkillGuardBench (Boundary)
99.17Overall Exact Match RateSKILLGUARD-ROBUST
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| SKILLGUARD-ROBUST2026.04 | 99.17 | 100 | 98.67 | 100 | 98.94 | 100 | 100 | |
| SKILLGUARD-CALIBRATE2026.04 | 97.13 | 100 | 95.75 | 100 | 95.94 | 100 | 99.28 | |
| SKILLGUARD-VERIFY2026.04 | 95.13 | 100 | 91.75 | 100 | 91.94 | 100 | 98.48 | |
| BUNDLEJUDGE2026.04 | 90.11 | 98.78 | 83.5 | 81.12 | 87.94 | 86.34 | 97.71 | |
| Qwen2.5-14BPackage Mode=full-package2026.04 | 85.55 | 96.65 | 77.33 | 75.82 | 81.88 | 81.1 | 95.48 |