diffdock
Pass
Audited by Gen Agent Trust Hub on Sep 18, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill clones the official repository from GitHub and installs dependencies from the PyTorch package registry. It also invokes a utility function to download pretrained model weights from an external source.
- [COMMAND_EXECUTION]: The skill uses Python's
subprocess.runto execute command-line tools including the DiffDock inference script, OpenBabel for format conversion, and AutoDock Vina for rescoring. The implementation uses list-based arguments, which is a secure practice to prevent shell injection. - [INDIRECT_PROMPT_INJECTION]: 1. Ingestion points: The skill processes user-supplied PDB files for proteins and SMILES strings for ligands via the
clean_pdbandsmiles_to_sdffunctions. 2. Boundary markers: No explicit prompt boundary markers or instruction-ignoring delimiters are used. 3. Capability inventory: Includes file system writes, command execution through subprocesses, and network access for initial setup. 4. Sanitization: Inputs are parsed by specialized scientific libraries (BioPython and RDKit) before being passed to downstream CLI tools.
Audit Metadata