The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/run_training.py accepts an arbitrary command via the --command argument and executes it using subprocess.run(). This allows the execution of any binary available in the system path with user-controlled arguments.
Evidence: In scripts/run_training.py, the execute_command function calls subprocess.run(split_command(command), ...) where command is derived from CLI arguments.
[DYNAMIC_EXECUTION]: The script scripts/write_outputs.py dynamically loads and executes a Python module from a path calculated at runtime that is located outside the skill's own directory structure.
Evidence: In scripts/write_outputs.py, the load_shared_module function uses importlib.util to load a script from parents[3] / 'shared' / 'scripts' / 'write_run_bundle.py', which resolves to a path three levels above the script's location, creating a dependency on external code.

run-train