bio-generative-design
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill instructs the user to clone the official REINVENT 4 source code from GitHub and install it as a dependency for the design environment.
- Evidence: 'git clone https://github.com/MolecularAI/REINVENT4.git' in usage-guide.md.
- [COMMAND_EXECUTION]: The skill utilizes the 'reinvent' command-line interface to execute staged-learning molecular design runs using local configuration files.
- Evidence: 'reinvent -l logfile.log config.toml' in SKILL.md.
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied molecular data and configuration files which represent a potential ingestion point for untrusted data.
- Ingestion points: SMILES strings and TOML configuration files provided in example prompts and files (SKILL.md, examples/reinvent_mpo.toml, usage-guide.md).
- Boundary markers: None explicitly defined for the molecular data, though the chemical notation (SMILES) and TOML schema provide inherent structural boundaries.
- Capability inventory: The skill executes the 'reinvent' CLI tool and performs molecular scoring and optimization using the RDKit and PyTorch libraries.
- Sanitization: The skill recommends the use of a separate 'molecular-standardization' skill to process generated chemical structures.
Audit Metadata