The Agent Skills Directory

[SAFE]: The skill implements a legitimate utility for processing DICOM medical imaging data without hidden malicious functionality.
The implementation uses established open-source libraries (pydicom and nibabel) to handle specialized file formats.
File system operations are localized to the input and output directories specified by the user.
No network connectivity, credential access, or persistence mechanisms are requested or present in the code.
[INDIRECT_PROMPT_INJECTION]: The skill ingests external DICOM data and includes metadata strings from the headers in its JSON output summary, which constitutes a potential attack surface.
Ingestion points: Files in the user-provided dicom_dir are parsed by the pydicom library in scripts/series_to_volume.py.
Boundary markers: The summary data is exported in a structured JSON format, but the metadata values (e.g., StudyDescription, SeriesDescription) are extracted from the files without sanitization.
Capability inventory: The skill's capabilities are strictly limited to file system read and write operations; it lacks network access or shell execution privileges.
Sanitization: No specific filtering or escaping for malicious prompt patterns is performed on the extracted DICOM header strings.
[DYNAMIC_EXECUTION]: The test suite utilizes dynamic module loading to verify the core script logic.
The regression test tests/test_orientation_regression.py uses importlib.util to load and execute scripts/series_to_volume.py. This is a standard procedure for Python unit testing and uses a static, local file path within the skill's own directory structure.

dicom-series-to-volume