The Agent Skills Directory

[DYNAMIC_EXECUTION]: The scripts/setup_check.py script uses the __import__ function to dynamically verify the presence of required Python packages (such as torch, rdkit, and pandas). This is a standard practice for environment diagnostic tools, and the package names are hardcoded, preventing arbitrary code execution through this mechanism.
[EXTERNAL_DOWNLOADS]: The skill documentation provides instructions for cloning the official DiffDock source code from GitHub and pulling the official Docker image. It also notes that model checkpoints are automatically downloaded upon the first run, which is standard behavior for deep learning applications.
[INDIRECT_PROMPT_INJECTION]: The skill processes scientific data files (PDB, SMILES, SDF) and CSV configuration files. While these represent an attack surface for indirect prompt injection, the data is validated by specialized scientific libraries (RDKit) and processed by the DiffDock model, resulting in a negligible risk to the agent context.
Ingestion points: Protein structures, ligand descriptions, and batch CSV files processed by inference.py and helper scripts.
Boundary markers: None explicitly present in the data formatting, but scientific data formats provide implicit structure.
Capability inventory: The skill facilitates file system operations and command-line execution of scientific Python modules.
Sanitization: The scripts/prepare_batch_csv.py script validates SMILES strings and file paths before they are used in the workflow.

diffdock