@daniel can add some color here as well, but the gist of it is not clear how to handle scaffolds/ datasets, e.g. for SWE environments we have
- scaffolds: mini swe agent plus, deepswe
- datasets: deepswe, swe bench verified, swe smith/ etc.
should there be one env per scaffold with configurable dataset? one env per dataset with configurable scaffold? one env per scaffold and dataset?