The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes whole-slide images and clinical metadata from external sources such as CSV and SVS files.
Ingestion points: references/image_loading.md (SlideData.from_slide) and references/data_management.md (metadata management functions).
Boundary markers: Not present; the skill assumes trusted input from the user's data directories.
Capability inventory: The skill has capabilities to write to the file system (HDF5, CSV, AnnData formats) and perform network requests to external APIs.
Sanitization: Ingested metadata is processed using standard libraries without explicit validation or sanitization steps.
[DYNAMIC_EXECUTION]: The skill includes instructions for loading machine learning models using torch.load and onnxruntime in references/machine_learning.md, which is an expected behavior for a pathology analysis toolkit.
[EXTERNAL_DOWNLOADS]: The skill facilitates remote cell segmentation by interacting with the DeepCell project API at https://deepcell.org/api/predict, as seen in references/multiparametric.md and references/preprocessing.md.

pathml