Skill Retrieval on SkillResolve-Bench 1.0 (test)
76.6Recall@3SkillResolve
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| SkillResolveEvaluation Protocol=Same-protocol held-out training2026.06 | 76.6 | 69.9 | 81.8 | 72.1 | 0 | 0 | |
| Attribution-listwiseEvaluation Protocol=Same-protocol held-out training2026.06 | 67.6 | 59.6 | 74.7 | 62.6 | 0 | 0 | |
| SkillRouterEvaluation Protocol=Zero-shot fixed-pool references2026.06 | 65.4 | 53.4 | 69.9 | 55.3 | 69.3 | 71.9 | |
| RRFEvaluation Protocol=Zero-shot fixed-pool references2026.06 | 63.1 | 50.4 | 72.2 | 54.2 | 64.6 | 72.8 | |
| BGEEvaluation Protocol=Zero-shot fixed-pool references2026.06 | 46.1 | 35.2 | 55.2 | 38.9 | 46.1 | 53.9 |