Batch classification is tightly linked to the existing Trained Model. The automated batch container does not allow you to add new ground truth points as this will change the current model.
The training part of the supervised classification is outside of the automated container and is a part of the supervised classification process. A possible solution would be to train again by adding new ground truth data to create a new model file. You can then use the new model file in the batch classification process.
Hope this helps!