The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill recommends using Python's pickle module for performance optimization when storing and loading molecular objects. The pickle.load() function is inherently insecure and can execute arbitrary code during the deserialization process. This presents a risk of arbitrary code execution if the agent is directed to process a malicious .pkl file from an untrusted source.
[PROMPT_INJECTION]: The skill facilitates the ingestion of untrusted external data through various cheminformatics file formats (SMILES, SDF, MOL) without defining boundary markers or security sanitization, creating a surface for indirect prompt injection.
Ingestion points: Chem.SDMolSupplier, Chem.SmilesMolSupplier, Chem.MolFromSmiles, and Chem.ForwardSDMolSupplier in SKILL.md are used to read external data into the agent's context.
Boundary markers: The instructions do not define or implement delimiters or warnings to ignore embedded instructions within the processed molecular data.
Capability inventory: The skill includes capabilities such as pickle.load (arbitrary code execution), and file-writing operations like Draw.MolToFile and img.save in SKILL.md.
Sanitization: While the skill mentions Chem.SanitizeMol, this function is designed for chemical validity (e.g., valence and aromaticity) and does not provide security sanitization for embedded natural language instructions or malicious data payloads.

rdkit