arboreto-grn-inference
Pass
Audited by Gen Agent Trust Hub on Oct 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external gene expression datasets and transcription factor lists, representing a standard attack surface for indirect prompt injection.
- Ingestion points: External data is loaded via
pd.read_csv('expression_data.tsv'),ad.read_h5ad('preprocessed.h5ad'), andload_tf_names('tf_list.txt')as documented inSKILL.md. - Boundary markers: The instructions do not define specific delimiters or validation logic to distinguish data from instructions within the loaded files.
- Capability inventory: The skill facilitates filesystem writes (
to_csv), package installation (pip install), and network-based distributed computing coordination (dask-scheduleranddask-worker). - Sanitization: No explicit sanitization or content validation is implemented before data is passed to the underlying tree-based regression algorithms.
- [COMMAND_EXECUTION]: The skill documentation provides shell commands for environment setup and the coordination of parallel computing resources.
- Evidence:
pip install arboreto distributed networkx matplotlib,dask-scheduler, anddask-worker tcp://10.118.224.134:8786. - Purpose: These commands are necessary functional components for installing the required scientific software stack and managing the Dask distributed computing backend.
Audit Metadata