alterlab-molfeat
Pass
Audited by Gen Agent Trust Hub on Sep 25, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: Fetches pre-trained molecular embeddings, graph neural network weights, and transformer models from official repositories on HuggingFace, Microsoft Research, and DGL-LifeSci. These downloads target well-known and trusted infrastructure for machine learning assets.
- [INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest and process molecular data (SMILES strings) provided by the user or external datasets. This untrusted data is processed through highly specialized chemical informatics parsers such as RDKit and Mordred, which act as rigid boundary layers that prevent typical natural language injection attacks from influencing the agent's control flow.
- Ingestion points: SMILES strings and molecule lists processed by
MoleculeTransformerandPretrainedMolTransformerinSKILL.mdandreferences/examples.md. - Boundary markers: None explicitly defined in prompts.
- Capability inventory: Execution of Python code via the
Bashtool, file read/write for configuration persistence, and network access for model retrieval. - Sanitization: Validation and parsing of chemical syntax performed by RDKit before featurization occurs.
Audit Metadata