The Agent Skills Directory

[DYNAMIC_EXECUTION]: The documentation in SKILL.md under 'Best Practices' recommends using pickle.load() for performance optimization when loading molecules. The pickle module is insecure as it can execute arbitrary code during deserialization. Recommending its use on potentially untrusted data files poses a security risk.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes external molecular data files, creating a vulnerability surface for indirect prompt injection if malicious instructions are embedded within these files.
Ingestion points: Data is read via Chem.SDMolSupplier, Chem.SmilesMolSupplier, and Chem.MolFromSmiles in the provided scripts.
Boundary markers: No delimiters or instructions are used to separate chemical data from potential natural language instructions.
Capability inventory: The scripts use file system operations to read and write molecular data and property reports.
Sanitization: The skill performs chemical sanitization but does not filter or sanitize text for prompt injection patterns.

rdkit