| name: "SKBench-Env" | |
| description: > | |
| SKBench repackaged as a Harbor agent environment: the agent fetches each | |
| cited source and judges whether the quoted text survives SKBench's | |
| mechanical normalization check. Eval-only. | |
| evaluation_framework: harbor | |
| tasks: | |
| - id: skbench_env | |
| config: default | |
| split: test | |