Dataset

Datasets are registered in seta_env/dataset/datasets.yaml.

python -m seta_env.dataset.download --list                          # see available
python -m seta_env.dataset.download terminal-bench-core_migrated    # download to dataset/

To add a new dataset, add an entry to seta_env/dataset/datasets.yaml with repo and optional subfolder.

Filtering tasks based on prior eval results

load_harbor_dataset(<dataset_dir>) auto-detects <dataset_dir>/task_filter.txt (one task_id per line, blank lines and # comments ignored) and only loads the listed tasks. Generate it with seta_env.dataset.filter_tasks from an evaluated_tasks.csv. See evaluation.md for the end-to-end workflow.