biomni
Pass
Audited by Gen Agent Trust Hub on Nov 6, 2026
Risk Level: SAFEDYNAMIC_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [DYNAMIC_EXECUTION]: The framework's core functionality involves generating and executing Python code at runtime to perform biomedical analysis. While the documentation provides security warnings and configuration options for sandboxing, the default behavior allows autonomous code execution.
- [COMMAND_EXECUTION]: The environment setup script (scripts/setup_environment.py) utilizes subprocess.run to interact with the system's package manager (conda) and to execute a generated test script to verify the installation.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data such as genomic datasets (VCF, H5AD) and literature abstracts. * Ingestion points: External data files and research summaries provided via the agent.go() method. * Boundary markers: None identified in the provided documentation or scripts. * Capability inventory: Full Python code execution, local file writing (reports), and network access for database retrieval. * Sanitization: No explicit sanitization or input filtering was identified in the provided skill components.
- [EXTERNAL_DOWNLOADS]: The skill documentation instructs users to install the biomni package from an external registry and mentions the automatic download of a large biomedical data lake upon first use.
Audit Metadata