cli-anything-unimol-tools
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external data files for molecular property prediction workflows.
- Ingestion points: The train and predict commands ingest external CSV files via the --data-path parameter.
- Boundary markers: There are no explicit instructions to the agent to treat the CSV content as untrusted or to use delimiters to prevent instruction injection from the data.
- Capability inventory: The skill uses the cli_anything.unimol_tools module to perform file operations, model training, and predictions based on the input data.
- Sanitization: The instructions do not define any validation or sanitization steps for the input SMILES strings or target values before processing.
- [COMMAND_EXECUTION]: The skill defines a complex CLI interface using python3 -m cli_anything.unimol_tools with multiple subcommands for training, prediction, and storage management, which involves direct shell interaction by the agent.
Audit Metadata