The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches the kev_serve.py deployment script and the kev library from the author's GitHub repository (github.com/jaredpalmer/kev). These are official vendor resources required for the model's operation.
[COMMAND_EXECUTION]: The skill uses the modal CLI to automate the provisioning of cloud infrastructure, including deploying the server, setting up volumes for model weights, and managing the application lifecycle.
[INDIRECT_PROMPT_INJECTION]: The skill creates an API that accepts external text inputs for classification.
Ingestion points: The state field in the /v1/systemone API endpoint in scripts/kev_serve.py receives untrusted data.
Boundary markers: None explicitly defined in the deployment script; input is passed directly to the model's server logic.
Capability inventory: scripts/kev_serve.py performs network operations to serve the API via uvicorn and writes cached model weights to a Modal volume using cache.commit().
Sanitization: The underlying 'Kev' model is a specialized classifier that produces structured output (choices and scores) rather than free-form text, which restricts the potential for malicious instructions to influence downstream systems.

kev-deploy