pathml

Pass

Audited by Gen Agent Trust Hub on Nov 6, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill instructs users to install the pathml package from the official Python Package Index (PyPI) and can be configured to download pre-trained model weights for architectures like HoVer-Net and HACTNet when the pretrained=True parameter is used.
  • [DATA_EXFILTRATION]: Provides functionality to send image data to the DeepCell remote API (https://deepcell.org/api/predict) for cell segmentation tasks via the SegmentMIFRemote transform. This is a documented feature for users who lack local GPU resources.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests and processes external medical imaging files and clinical metadata.
  • Ingestion points: Whole-slide images in various proprietary formats (SVS, NDPI, DICOM) and clinical metadata manifests in CSV format.
  • Boundary markers: No specific delimiters or instructions to ignore embedded content are provided for the processing of slide metadata.
  • Capability inventory: The toolkit manages subprocesses through Dask for parallel execution and performs network requests to external APIs for remote segmentation.
  • Sanitization: Explicit sanitization of external metadata content is not detailed in the provided documentation.
  • [COMMAND_EXECUTION]: Utilizes the Dask library to coordinate parallel and distributed processing of large-scale pathology data, which involves managing subprocesses and worker nodes.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 6, 2026, 12:15 PM