The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes chemical data formats such as SMILES, SMARTS, and SDF provided by the user, which represents a potential surface for indirect prompt injection.
Ingestion points: Data parsing routines in scripts/molecular_properties.py, scripts/similarity_search.py, and scripts/substructure_filter.py.
Boundary markers: No specific delimiters are employed to separate instructional context from external data.
Capability inventory: The skill is capable of local file system operations (reading and writing) and generating molecular visualizations.
Sanitization: The RDKit library includes robust built-in chemical sanitization that validates the structural validity and valence of molecules during the parsing process.
[SAFE]: The documentation includes a performance recommendation suggesting the use of Python's pickle module for molecular data serialization.
Evidence: SKILL.md provides examples for using pickle.dump and pickle.load to cache large datasets.
Context: While deserialization of untrusted data is a known risk, the skill presents this as a performance optimization for local scientific workflows rather than a mechanism for processing untrusted remote content.

rdkit