tooluniverse-regulatory-genomics
Pass
Audited by Gen Agent Trust Hub on Oct 7, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill explicitly directs the agent to write and run Python code via Bash for data processing and statistical analysis. This creates a powerful execution environment within the agent's context.
- [DYNAMIC_EXECUTION]: The instructions mandate the runtime generation of Python scripts using libraries such as pandas, scipy, and statsmodels to handle data retrieved from external tools.
- [INDIRECT_PROMPT_INJECTION]: The skill possesses a significant attack surface for indirect prompt injection. It is designed to ingest data from numerous external, potentially attacker-influenced bioinformatics sources (ENCODE, JASPAR, UniBind, RegulomeDB) and use that data in downstream reasoning and code generation.
- Ingestion points: Data retrieved from
ENCODE_search_*,jaspar_get_matrix,UniBind_get_dataset, and deep learning model predictions. - Boundary markers: None present in the skill instructions to delimit external data from system prompts.
- Capability inventory: Access to the
bashtool for Python execution and file system interaction. - Sanitization: No specific sanitization or validation logic is defined for the external genomic data before it is processed by the Python interpreter.
- [EXTERNAL_DOWNLOADS]: The skill relies on several external APIs and remote Model Context Protocol (MCP) servers (e.g., AlphaGenome, Enformer, Borzoi, NVIDIA NIM) to retrieve model predictions and genomic annotations.
Audit Metadata