pytdc
Pass
Audited by Gen Agent Trust Hub on Sep 25, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill facilitates the installation of the
PyTDCpackage using theuv pip installcommand. This is the primary purpose of the skill and follows standard library setup procedures. - [EXTERNAL_DOWNLOADS]: The skill downloads AI-ready datasets from the Therapeutics Data Commons platform (tdcommons.ai) for pharmaceutical tasks such as ADME, toxicity, and drug-target interaction prediction. These are well-known scientific resources from established research organizations.
- [INDIRECT_PROMPT_INJECTION]: The skill involves processing complex chemical and biological data structures, such as SMILES strings and protein sequences, which are ingested from external datasets. This ingestion represents a theoretical surface for indirect prompt injection, although it is a core requirement for the therapeutic machine learning tasks the skill supports.
- Ingestion points: Dataset loading functions in
scripts/load_and_split_data.pyandscripts/molecular_generation.pywhich fetch data from TDC. - Boundary markers: None present in the template scripts.
- Capability inventory: The skill provides templates for data processing, split generation, and model evaluation; it does not contain capabilities for arbitrary file writes or unauthorized network operations.
- Sanitization: Not explicitly handled in the provided templates; sanitization is typically managed by the underlying
tdclibrary and the specific ML models used by the researcher.
Audit Metadata