The Agent Skills Directory

[SAFE]: The skill follows security best practices for environment management and clinical data handling. No malicious patterns, obfuscation, or persistence mechanisms were identified.\n- [EXTERNAL_DOWNLOADS]: The skill fetches synthetic clinical data from Google Cloud Storage (storage.googleapis.com) and installs dependencies from official registries. These are recognized as safe, well-known services and trusted sources for clinical research data.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes clinical Electronic Health Record data, which is a potential surface for indirect injection. Evidence Chain: (1) Ingestion points: root and ehr_root parameters in dataset classes across SKILL.md and reference files; (2) Boundary markers: The skill includes explicit 'Integrity rules' and behavioral constraints for the agent; (3) Capability inventory: Performs file system reads/writes for data caching and training models using PyTorch; (4) Sanitization: The 'Agent operating procedure' mandates verifying de-identification status and human review of all clinical outputs.

pyhealth