The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The SKILL.md file suggests using the 'pickle' module for performance optimization when saving and loading molecular objects. Specifically, it demonstrates using pickle.dump() and pickle.load() for fast loading. This is an unsafe deserialization pattern; if the agent or user loads a maliciously crafted pickle file from an untrusted source, it could result in arbitrary code execution on the system.
[PROMPT_INJECTION]: The skill is designed to ingest and process molecular data formats which constitutes a surface for indirect prompt injection. 1. Ingestion points: Molecular data is parsed via Chem.MolFromSmiles, Chem.SDMolSupplier, and Chem.SmilesMolSupplier in SKILL.md and throughout the provided scripts (molecular_properties.py, similarity_search.py, substructure_filter.py). 2. Boundary markers: The skill lacks boundary markers or explicit instructions to the agent to disregard potential instructions embedded in molecular property fields or metadata. 3. Capability inventory: The skill includes capabilities to write data to the filesystem using SDWriter in SKILL.md and CSV writer functions in the utility scripts. 4. Sanitization: While RDKit performs chemical structure validation (sanitization), it does not sanitize against logical or injection-based content in property tags or metadata fields.

rdkit