The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs users to install the modal package via pip and specifies dependencies like torch, transformers, and numpy for container images. These are well-known packages used for the platform's intended serverless compute purpose.
[COMMAND_EXECUTION]: Provides standard CLI commands for the Modal platform, including modal token new for authentication, and modal run / modal deploy for executing and deploying serverless functions.
[INDIRECT_PROMPT_INJECTION]: The skill demonstrates building web endpoints and ML serving classes that process external data, creating a potential surface for indirect prompt injection if the ingested data is used in LLM prompts.
Ingestion points: Untrusted data enters the skill via the predict(data: dict) web endpoint and the predict(self, text: str) method in the ML Model Serving section of SKILL.md.
Boundary markers: The provided code snippets do not include explicit delimiters or instructions to ignore embedded directives in the input data.
Capability inventory: The skill provides capabilities for remote code execution on serverless infrastructure, network API exposure, and GPU-accelerated computation.
Sanitization: The example code does not include input validation or sanitization routines for the data processed by the models or endpoints.

modal