The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/setup_check.py utilizes the __import__ function to load Python modules dynamically at runtime.
Evidence: module = __import__(import_name) in scripts/setup_check.py.
Context: This mechanism is used to verify that necessary scientific dependencies like numpy, torch, and rdkit are installed. The module names are hardcoded within the script and do not originate from external user input.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external molecular data, which represents a potential attack surface for indirect prompt injection.
Ingestion points: User-provided SMILES strings, amino acid sequences, and PDB file paths enter the agent context via CLI arguments and batch CSV files, as described in SKILL.md and scripts/prepare_batch_csv.py.
Boundary markers: The skill does not employ explicit delimiters or system instructions to ignore potential commands embedded within the molecular data during processing.
Capability inventory: The agent is granted capabilities for file modification (Write) and Python command execution (Bash(python:*)), which are leveraged to run the docking workflows.
Sanitization: The scripts/prepare_batch_csv.py script performs structural validation of SMILES strings via RDKit and verifies the existence of protein files, providing a basic level of input sanitization.
[EXTERNAL_DOWNLOADS]: The skill guides the user to download the primary DiffDock source code from an external repository.
Evidence: git clone https://github.com/gcorso/DiffDock.git in SKILL.md and references/workflows_examples.md.
Context: The source code is hosted on a well-known research repository associated with the original authors of the DiffDock paper.

alterlab-diffdock