The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill's documentation in 'SKILL.md' explicitly recommends using the 'pickle' module for fast loading and storing of molecular data. The example provided uses 'pickle.load(f)', which is inherently unsafe when processing files from untrusted sources as it can lead to arbitrary code execution.
[PROMPT_INJECTION]: The skill is designed to ingest and process molecular data from external files (SMILES, SDF, MOL). This creates a surface for indirect prompt injection where malicious instructions could be embedded in molecular metadata or property fields.
Ingestion points: Molecular parsing functions in 'molecular_properties.py', 'similarity_search.py', and 'substructure_filter.py'.
Boundary markers: Absent; the skill does not implement delimiters or 'ignore' instructions for data fields.
Capability inventory: Includes file-writing capabilities to CSV and molecular formats across all scripts.
Sanitization: Standard RDKit sanitization validates chemical structures but does not filter for potential natural language instructions embedded in metadata fields.
[EXTERNAL_DOWNLOADS]: The documentation provides installation instructions for the 'rdkit' and 'rdkit-pypi' packages. These are well-known, legitimate packages from official registries (Conda and PyPI) and are considered safe.
[COMMAND_EXECUTION]: The provided scripts ('molecular_properties.py', 'similarity_search.py', 'substructure_filter.py') use standard library modules like 'argparse' to handle user input. While they don't directly execute shell commands via 'subprocess' or 'os.system', the recommendation to use 'pickle' provides a secondary path to command execution.

rdkit