Agent-fix should not discard generated code, even if it fails. This will allow the user to observe what the model tried to do to fix the issue. This code should be saved in a folder titled something like <issue>-<issue-number> each file is a self contained script with test cases that can be saved as <issue-number>-<run-humber>-<time>.py