The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes code examples that utilize subprocess.run to execute system utilities such as nvidia-smi and pip list. These are intended for environment verification and debugging in a GPU computing context.
[INDIRECT_PROMPT_INJECTION]: Several examples, such as the text generation endpoint in SKILL.md, ingest user-supplied strings as prompts for machine learning models. This creates a surface for indirect prompt injection which is typical for model serving applications.
Ingestion points: The prompt parameter in the TextGenerator.generate method located in SKILL.md.
Boundary markers: No specific delimiters or safety instructions for input handling are included in the basic examples.
Capability inventory: The skill demonstrates the ability to execute system commands via subprocess.run (SKILL.md) and documents the platform's ability to run code in sandboxed environments via modal.Sandbox.exec (advanced-usage.md).
Sanitization: The provided examples do not demonstrate explicit input sanitization or filtering.
[EXTERNAL_DOWNLOADS]: The documentation includes examples of installing dependencies directly from Hugging Face's official GitHub repository using pip.

infrastructure-modal