The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes subprocess.run across multiple scripts, including scripts/check_pas_control_prereqs.py and scripts/run_deft_docker_action.py, to interact with system tools like Docker, Kubectl, and virtualenv. This is the primary mechanism for environment verification and job submission.
[DYNAMIC_EXECUTION]: In scripts/run_deft_cli.py, the skill uses os.execve to execute model-related commands (clip, embedding, tmm) within a controlled virtual environment. The script implements security measures, such as verifying the entrypoint file type and its SHA256 digest before execution.
[DATA_EXPOSURE_AND_EXFILTRATION]: The helper scripts/check_pas_credentials.py checks for the existence of necessary environment variables like NGC_KEY and HF_TOKEN. The logic is designed to report only the presence of these variables and explicitly avoids logging or displaying their actual values.
[EXTERNAL_DOWNLOADS]: The workflow involves pulling container images from the NVIDIA NGC registry and base model weights from Hugging Face. These downloads are standard for the vendor's software ecosystem and are conducted after the user approves the run plan during the pre-flight phase.

tao-run-deft-pas