The Agent Skills Directory

[COMMAND_EXECUTION] (SAFE): The skill uses subprocess.run to call nvidia-smi for GPU diagnostics, which is standard practice in ML environments and does not present a security risk.
[CREDENTIALS_UNSAFE] (SAFE): Properly demonstrates the use of platform-native secret management (modal.Secret) for API tokens like HF_TOKEN, following security best practices rather than hardcoding credentials.
[EXTERNAL_DOWNLOADS] (SAFE): Downloads and installs standard, well-known libraries such as modal, torch, and transformers from official registries (PyPI) and uses official base images from NVIDIA.

modal-serverless-gpu