chemistry-query
Pass
Audited by Gen Agent Trust Hub on Jun 23, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses local Python scripts (scripts/query_pubchem.py, scripts/rdkit_mol.py, scripts/chain_entry.py) to perform molecular lookups, visualization, and synthesis planning. These are invoked via standard CLI patterns.
- [EXTERNAL_DOWNLOADS]: The skill communicates with the PubChem REST API (a service of the National Institutes of Health/NCBI) to resolve chemical names and fetch properties. This is an interaction with a trusted well-known scientific service.
- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) as it processes untrusted data (SMILES strings and chemical names) provided via JSON inputs in 'scripts/chain_entry.py'.
- Ingestion points: Data enters via the '--input-json' argument in 'scripts/chain_entry.py'.
- Boundary markers: None identified in the provided documentation to isolate external molecule names/SMILES from instructions.
- Capability inventory: The skill can perform network operations to PubChem and execute local subprocesses for RDKit analysis.
- Sanitization: No explicit sanitization or validation of the input strings is documented, creating a surface where malicious instructions could be embedded in molecule metadata.
Audit Metadata