The Agent Skills Directory

[SAFE]: The skill serves as a documentation guide for the Modal serverless platform. The instructions and code examples provided are consistent with standard development workflows for cloud-based machine learning.
[EXTERNAL_DOWNLOADS]: The skill references the installation of the modal CLI and library via pip, which is the primary intended dependency. It also references official repositories and packages from well-known organizations like Hugging Face and NVIDIA for container builds.
[COMMAND_EXECUTION]: Code snippets include the use of subprocess.run to execute nvidia-smi and torch.distributed.launch. These are standard practices in machine learning for verifying GPU status and initiating distributed training.
[CREDENTIALS_SAFE]: The documentation includes examples of managing secrets using Modal's built-in Secret management system. It uses placeholders (e.g., HF_TOKEN=hf_xxx) and instructs users to store credentials securely in the cloud environment rather than hardcoding them in scripts.
[INDIRECT_PROMPT_INJECTION]: The skill identifies surfaces where user-provided text (prompts) are passed to inference models (e.g., TextGenerator().generate(prompt)). This is a standard functional requirement for LLM applications and is documented within the context of model inference.

modal