The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data by scanning DICOM files in a user-provided directory. While the content of these headers is untrusted, the skill uses structured JSON output for the agent, which limits the risk of instructions embedded in headers being interpreted as commands. Findings are limited to metadata reporting.
Ingestion points: scripts/preflight_series.py reads DICOM instances from dicom_dir using the pydicom library.
Boundary markers: No explicit delimiters are used in the prompt instructions, but the tool output follows a strict JSON schema defined in validators/output_schema.json.
Capability inventory: The skill uses pydicom for header extraction and nibabel/numpy for geometric calculations. It does not perform file writes (outside of the test suite) or network operations.
Sanitization: The _safe_str helper function in scripts/preflight_series.py ensures that values extracted from DICOM tags are converted to safe string representations for JSON serialization.
[COMMAND_EXECUTION]: The skill's test suite (tests/test_preflight_series.py) uses subprocess.run to execute a local fixture generation script. This is a standard development practice for setting up testing environments and does not present a risk during the agent's normal operation of the skill.
[EXTERNAL_DOWNLOADS]: The skill manifest declares dependencies on well-known scientific libraries including pydicom, nibabel, numpy, and typer. These are standard packages for medical image processing and are fetched from official package registries.

dicom-series-preflight