alterlab-pydicom
Pass
Audited by Gen Agent Trust Hub on May 5, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its handling of untrusted DICOM metadata.
- Ingestion points: Metadata is ingested from external files in
scripts/extract_metadata.py,scripts/anonymize_dicom.py, andscripts/dicom_to_image.py. - Boundary markers: None. Extracted metadata (e.g., Patient Name, Study Description) is printed to the console or written to files without delimiters or instructions to the agent to treat the content as data only.
- Capability inventory: The skill has file-writing capabilities (
pydicom.Dataset.save_as,PIL.Image.save, and standard Python file I/O) across all scripts. - Sanitization: No sanitization or validation is performed on the DICOM tag values before they are presented to the agent or saved.
- [SAFE]: The skill utilizes standard, reputable libraries for its functionality, including
pydicom,numpy, andPillow. Installation commands target the official Python Package Index (PyPI). - [SAFE]: No obfuscation, data exfiltration, or persistence mechanisms were detected in the skill instructions or provided Python scripts.
Audit Metadata