The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The Python implementation uses joblib.load("model.pkl") to load the machine learning model. joblib internally uses the pickle module, which is unsafe for loading untrusted data. A maliciously crafted model.pkl file could execute arbitrary code on the server during the application's startup phase.
[COMMAND_EXECUTION]: The skill generates Dockerfiles and Kubernetes manifests that contain shell commands for installing dependencies and running the API server. These automated generation steps involve executing system-level operations based on user-provided configurations.
[EXTERNAL_DOWNLOADS]: The Dockerfile instructions include pip install commands that download packages from external registries, creating a dependency on external code and potential exposure to supply chain risks.
[PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection as it processes untrusted external artifacts like model files and requirements lists to generate deployment code.
Ingestion points: model.pkl and requirements.txt files provided by the user or target environment.
Boundary markers: Absent; there are no explicit delimiters or instructions to ignore embedded commands within the processed artifacts.
Capability inventory: File system access, Docker image construction, and Kubernetes deployment manifest generation.
Sanitization: Input validation using Pydantic is implemented for the API endpoints, but no validation or safety checks are applied to the model loading process itself.

model-deployment