The Agent Skills Directory

[SAFE]: The skill serves as a legitimate technical resource and template collection for the Modal serverless GPU cloud service. All provided code examples align with standard usage of the official modal library.
[COMMAND_EXECUTION]: The skill includes code snippets using subprocess.run to execute nvidia-smi for GPU diagnostics and torch.distributed.launch for training orchestration. These are standard operations for the described ML workloads.
[EXTERNAL_DOWNLOADS]: The documentation references downloading dependencies and container images from well-known and trusted sources, including the official NVIDIA container registry and Hugging Face's GitHub repository.
[INDIRECT_PROMPT_INJECTION]: The skill provides examples of web endpoints and S3 bucket mounts that ingest external data. While this creates a potential attack surface for the user's deployed applications, the skill correctly demonstrates the use of modal.Secret for secure credential management and provides standard patterns for handling input data.
[DYNAMIC_EXECUTION]: The skill mentions modal.Sandbox for running code in isolated environments, which is a core platform feature used for safe execution of untrusted workloads.

modal-serverless-gpu