The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from medical imaging files, which is a necessary part of its core functionality.
Ingestion points: pydicom.dcmread is used to load data from .dcm files in scripts/anonymize_dicom.py, scripts/dicom_to_image.py, and scripts/extract_metadata.py.
Boundary markers: There are no explicit boundary markers or instructions to the agent to ignore potentially malicious content embedded within DICOM metadata tags.
Capability inventory: The skill possesses file system write capabilities via ds.save_as() and image.save().
Sanitization: DICOM tag values are extracted and displayed or saved without specific sanitization for potential injection characters.
[METADATA_POISONING]: The skill metadata contains an inconsistency regarding the author's identity.
The SKILL.md frontmatter lists the author as 'K-Dense Inc.', whereas the provided environment context identifies the author as 'marielynneblock'. This is considered a deceptive metadata pattern, although likely benign in this context.

pydicom