The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes external, untrusted data which creates a surface for indirect prompt injection.
Ingestion points: The toolkit loads whole-slide images (e.g., .svs, .ndpi, .mrxs) and reads clinical metadata from CSV files (clinical_data.csv, slide_manifest.csv).
Boundary markers: The documentation does not specify the use of boundary markers or instructions for the agent to ignore embedded content within the processed images or metadata.
Capability inventory: Across its scripts, the skill facilitates file system writes (to_hdf5, np.memmap), network-based remote segmentation requests, and dynamic model loading (torch.load).
Sanitization: There is no mention of sanitization or strict validation of external CSV content or image headers.
[EXTERNAL_DOWNLOADS]: The skill facilitates data transmission to external processing services.
Remote API Integration: The SegmentMIFRemote transform connects to the DeepCell API (https://deepcell.org/api/predict) to perform remote cell segmentation, which involves sending image data to an external endpoint.
[COMMAND_EXECUTION]: The documentation provides templates for automating workflows through shell scripts.
Batch Scripts: Includes examples of SLURM job scripts (submit_jobs.sh) and Dask cluster setup that involve shell command execution for orchestrating large-scale pathology data processing.

pathml