pydicom

Pass

Audited by Gen Agent Trust Hub on Aug 17, 2026

Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes medical imaging data in DICOM format, which contains numerous metadata tags (e.g., PatientName, StudyDescription, InstitutionalDepartmentName) that can store arbitrary text. If an agent extracts and displays this metadata without proper sanitization, it could be susceptible to instructions embedded within the data by a malicious actor.
  • Ingestion points: The skill utilizes pydicom.dcmread() to load files in SKILL.md and scripts/extract_metadata.py.
  • Boundary markers: No specific delimiters or "ignore instructions" warnings are implemented when processing metadata.
  • Capability inventory: The skill possesses file-writing capabilities through ds.save_as() and can execute local scripts via uv run.
  • Sanitization: The provided scripts print or save metadata values directly without filtering or escaping content.
  • [EXTERNAL_DOWNLOADS]: The skill includes instructions to install several third-party Python libraries using the uv package manager. These are well-known and standard packages within the medical imaging and data science communities.
  • Packages: pydicom, pillow, numpy, matplotlib, pylibjpeg, pylibjpeg-libjpeg, pylibjpeg-openjpeg, python-gdcm.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 17, 2026, 02:14 AM