| name: Claw-Eval | |
| description: > | |
| Claw-Eval is a transparent benchmark for AI agents acting in the real world. | |
| 300 human-verified tasks across 24 categories, graded on completion, safety, | |
| and robustness with Pass³ methodology (N=3 independent trials). | |
| evaluation_framework: claw-eval | |
| tasks: | |
| - id: general | |
| config: default | |
| split: general | |
| - id: multimodal | |
| config: default | |
| split: multimodal | |
| - id: multi_turn | |
| config: default | |
| split: multi_turn | |