Agent Task Completion on tau2-bench airline
60Pass RateSkillsInjector
Evaluation Results
| Method | Links | |
|---|---|---|
| SkillsInjector2026.05 | 60 | |
| Graph of Skills2026.05 | 56.1 | |
| SkillRouter2026.05 | 54 | |
| LLM-as-selector2026.05 | 49.6 | |
| Dense Cosine2026.05 | 45.2 | |
| BM252026.05 | 43.6 | |
| Random-skill2026.05 | 42.4 | |
| No-skill2026.05 | 37.6 | |
| Full-library2026.05 | 24.4 |