The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/setup_check.py utilizes __import__ to dynamically verify the presence of required Python libraries (such as torch, rdkit, and pandas). This check is performed against a hardcoded list of standard scientific dependencies to ensure the environment is correctly configured.
[EXTERNAL_DOWNLOADS]: The skill instructs the user to download the DiffDock codebase and models from official sources, including a public GitHub repository and a research group's Docker image. These are established sources for the molecular docking tool described.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process molecular data from external files (PDB, SDF, CSV) and SMILES strings. It includes a validation script, scripts/prepare_batch_csv.py, which checks input formatting and validates SMILES strings to reduce processing errors. The evidence chain is as follows:
Ingestion points: The skill reads ligand data from SMILES strings or file paths, and batch instructions from CSV files via scripts/prepare_batch_csv.py.
Boundary markers: No explicit prompt boundary markers are used when interpolating file paths into shell commands, relying on the agent's logic for command construction.
Capability inventory: The skill executes shell commands for docking inference, result analysis, and rescoring using external tools.
Sanitization: Input validation is provided via scripts/prepare_batch_csv.py, which validates SMILES syntax using RDKit and verifies the existence of referenced protein/ligand files.

diffdock