The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill installs the pyhealth library from official repositories and utilizes pre-trained models from HuggingFace, which is a recognized trusted organization.
[COMMAND_EXECUTION]: Instructions involve standard local environment configuration and model execution commands typical for healthcare research.
[DATA_EXFILTRATION]: No evidence of credential harvesting or unauthorized network operations; the skill focuses on local processing of standardized clinical datasets.
[PROMPT_INJECTION]: The instructions are technical and focused on library usage, without attempts to bypass agent safety filters or override system behavior. It also correctly identifies and handles the surface for indirect prompt injection within clinical notes by treating them as classification inputs.
Ingestion points: Clinical text datasets referenced in references/datasets.md.
Boundary markers: Managed by fixed tokenization schemes within the underlying models.
Capability inventory: Training, evaluation, and inference on clinical tasks; no arbitrary code execution capabilities.
Sanitization: Standard NLP preprocessing as documented in references/preprocessing.md.

pyhealth