The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill documentation in SKILL.md recommends using the Python pickle module to save and load molecular objects for performance optimization. Deserializing data using pickle is inherently unsafe because it can allow the execution of arbitrary code if the source file is untrusted or has been tampered with.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process molecular data from external files (SDF, SMILES) and user-provided strings. This functionality establishes a surface for indirect prompt injection attacks.
Ingestion points: Untrusted data enters the context through molecular parsing functions in SKILL.md and the provided command-line scripts: molecular_properties.py, similarity_search.py, and substructure_filter.py.
Boundary markers: The skill does not employ specific delimiters or "ignore instructions" markers when reading data from molecular files or SMILES strings.
Capability inventory: The skill possesses the ability to read and write files to the local file system (via SDMolSupplier and SDWriter) and execute shell commands via the suggested uv run usage.
Sanitization: The skill relies on RDKit's standard chemical sanitization, which validates chemical structures (valence, aromaticity) but does not filter for embedded natural language instructions.

alterlab-rdkit