run_eval.py appears inconsistent with task image workflow: missing /utils/encrypt.py and build_oh_runtime_images.sh uses unsupported args #2

Hi, thanks for releasing OpenAgentSafety!

I am trying to run the OpenHands evaluation locally and ran into an issue that seems related to how task images / runtime images are wired up.

Environment

• OS: macOS on Apple Silicon
• Python: 3.12
• Docker Desktop
• Repository: OpenAgentSafety
• Command: running evaluation/run_eval.sh / evaluation/run_eval.py on tasks under workspaces/tasks

What happened

After getting the OpenHands runtime container to start successfully, the evaluation fails during task initialization:

Runtime is ready.
ERROR:root:  File ".../OpenAgentSafety/evaluation/run_eval.py", line 402, in <module>
    init_task_env(runtime, args.server_hostname, env_llm_config, args.task_path)
  File ".../OpenAgentSafety/evaluation/run_eval.py", line 243, in init_task_env
    assert obs.exit_code == 0
           ^^^^^^^^^^^^^^^^^^

ERROR:root:<class 'AssertionError'>:

After adding debug output, the failing command appears to be:

python /utils/encrypt.py && rm /utils/evaluator.py /utils/encrypt.py

However, for the task I tested, /utils/encrypt.py does not exist after run_eval.py copies the local task files into the runtime.

I checked the task directory and confirmed that encrypt.py is not present in the local task utils/ directory.

However, the README says each task is packaged as a Docker image containing task assets such as utils/init.sh, instruction/task.md, and utils/eval.py.

Could you advise what the intended fix should be?

Thanks!