Agent Task Completion Average on tau2-bench, SkillsBench, and ALFWorld
58.7Average Pass RateSkillsInjector
Evaluation Results
| Method | Links | |
|---|---|---|
| SkillsInjector2026.05 | 58.7 | |
| Graph of Skills2026.05 | 53.6 | |
| SkillRouter2026.05 | 53.5 | |
| LLM-as-selector2026.05 | 49.7 | |
| Dense Cosine2026.05 | 48.5 | |
| BM252026.05 | 46.7 | |
| Random-skill2026.05 | 42.6 | |
| No-skill2026.05 | 40.2 | |
| Full-library2026.05 | 24.8 |