The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill documentation suggests using the Python pickle module for performance optimization when handling molecular datasets.
Evidence: In SKILL.md, the 'Performance Optimization' section provides a code snippet using pickle.dump() and pickle.load() to save and load molecules.
Risk: The pickle module is not secure. Maliciously constructed data can cause arbitrary code execution during the unpickling process. This is particularly risky if users are encouraged to load pre-pickled datasets from external or untrusted sources.
[COMMAND_EXECUTION]: The provided Python scripts perform file system operations based on command-line arguments.
Evidence: scripts/molecular_properties.py, scripts/similarity_search.py, and scripts/substructure_filter.py utilize the argparse and pathlib modules to accept user-provided file paths for reading and writing molecular data (SDF, SMILES, CSV).
Risk: While standard for a data processing toolkit, these scripts allow the agent to read from and write to the local file system. Users should ensure that input paths are restricted to safe directories.
[EXTERNAL_DOWNLOADS]: The documentation provides instructions for installing the necessary chemical informatics libraries from well-known package registries.
Evidence: references/api_reference.md contains installation commands: conda install -c conda-forge rdkit and pip install rdkit-pypi.
Note: These references target official repositories for the RDKit project and are standard practice for setting up the required environment.
[SAFE]: The skill processes external molecular data formats which represent a potential surface for indirect prompt injection.
Ingestion points: File reading operations in all three utility scripts (SDMolSupplier, SmilesMolSupplier).
Boundary markers: Absent; data is ingested directly into RDKit's internal C++-backed molecule objects.
Capability inventory: The scripts are limited to file I/O and terminal output; they do not contain network requests or subprocess execution capabilities.
Sanitization: The RDKit library performs strict chemical validation and sanitization (e.g., Chem.SanitizeMol) on all imported molecular structures, which mitigates the risk of the agent interpreting the data as instructions.

rdkit