Agent Task Completion on tau2-bench telecom
67Pass RateSkillsInjector
Evaluation Results
| Method | Links | |
|---|---|---|
| SkillsInjector2026.05 | 67 | |
| SkillRouter2026.05 | 62.8 | |
| Graph of Skills2026.05 | 60.4 | |
| LLM-as-selector2026.05 | 55.8 | |
| Dense Cosine2026.05 | 54.7 | |
| BM252026.05 | 51.2 | |
| Random-skill2026.05 | 41.9 | |
| No-skill2026.05 | 40 | |
| Full-library2026.05 | 24.6 |