The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides numerous functions for ingesting external molecular data from various file formats and remote sources. This creates a vulnerability surface where malicious instructions could be embedded in chemical data files (e.g., in SDF properties or CSV fields) to influence the agent's behavior during processing.
Ingestion points: Functions such as dm.read_sdf(), dm.read_csv(), dm.read_smi(), dm.read_excel(), dm.open_df(), and dm.to_mol() in SKILL.md are designed to load data from external files or strings.
Boundary markers: The provided code snippets do not implement boundary markers or instructions to ignore natural language content within the chemical data inputs.
Capability inventory: The skill possesses capabilities for local file writing (dm.to_sdf(), dm.to_xlsx()) and network operations to cloud storage via fsspec (s3://, gs://, https://).
Sanitization: While the skill emphasizes chemical structure sanitization using dm.sanitize_mol() and dm.standardize_mol(), it does not provide mechanisms to filter or sanitize potential natural language prompt injections contained within the data.
[EXTERNAL_DOWNLOADS]: The skill documentation instructs the installation of the datamol Python package using uv pip install datamol. This is the intended core library for the skill's cheminformatics functionality.

datamol