pathml
Pass
Audited by Gen Agent Trust Hub on Jun 13, 2026
Risk Level: SAFEREMOTE_CODE_EXECUTIONDATA_EXFILTRATIONCOMMAND_EXECUTIONPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [REMOTE_CODE_EXECUTION]: The documentation demonstrates the use of torch.load() to import machine learning models in machine_learning.md. Because this function utilizes the pickle module, it can execute arbitrary code if the model files are sourced from untrusted providers.
- [DATA_EXFILTRATION]: The skill contains functionality for remote cell segmentation (SegmentMIFRemote) in multiparametric.md and preprocessing.md which sends image data to an external API endpoint at deepcell.org. This represents an external transfer of potentially sensitive medical imaging data.
- [COMMAND_EXECUTION]: The skill documentation in data_management.md includes examples for programmatically generating and running shell commands for SLURM job submission (sbatch) and Dask cluster management, which are used for parallelizing pathology workflows.
- [PROMPT_INJECTION]: The skill's ability to ingest and parse clinical manifests and slide metadata from external CSV files in data_management.md creates a surface for indirect prompt injection, as the agent may process untrusted textual content embedded in these files.
- [EXTERNAL_DOWNLOADS]: Instructions are provided in SKILL.md to install the pathml package and various scientific dependencies from public software registries using uv pip install.
Audit Metadata