Is it ok to directly wrap the gym env in `create_single_env` in local_evaluation.py when evaluating the model in aicrowd?

[CH_do]

Hi, is it should be the observations (not observations_agent) here?