The Agent Skills Directory

[SAFE]: The skill provides standard functionality for medical image processing. Analysis of the scripts and instructions revealed no signs of malicious patterns, obfuscation, or unauthorized data access.
[PROMPT_INJECTION]: The skill ingests external data from DICOM files (metadata and pixel data), creating a surface for potential indirect prompt injection.
Ingestion points: pydicom.dcmread in SKILL.md, scripts/anonymize_dicom.py, scripts/dicom_to_image.py, and scripts/extract_metadata.py.
Boundary markers: None specified for the data ingestion process.
Capability inventory: Local file system read and write operations via pydicom.dcmread and Dataset.save_as.
Sanitization: A dedicated script scripts/anonymize_dicom.py is provided to sanitize sensitive Patient Health Information (PHI).

pydicom