The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill supports fetching molecular datasets from remote locations via HTTP/HTTPS and cloud storage protocols.
Evidence: Examples in SKILL.md and references/io_module.md show the use of dm.read_csv("https://example.com/data.csv") and dm.read_sdf("s3://bucket/compounds.sdf").
[DATA_EXFILTRATION]: The library includes functionality to write processed molecular data directly to remote cloud storage buckets (S3, GCS).
Evidence: Code snippets such as dm.to_sdf(mols, "s3://bucket/output.sdf") in SKILL.md and references/io_module.md demonstrate the capability to transmit data to external infrastructure.
[PROMPT_INJECTION]: The skill processes various untrusted molecular data formats (SDF, SMILES, CSV), representing an indirect prompt injection surface.
Ingestion points: Data ingestion occurs through dm.read_sdf, dm.read_csv, dm.read_excel, and dm.to_mol across all referenced documentation files.
Boundary markers: The skill documentation does not provide specific instructions for using boundary markers or delimiters to isolate parsed data from the agent's internal instruction context.
Capability inventory: The skill has access to the local file system (read/write), network operations via fsspec, and parallelized CPU execution.
Sanitization: While functions like dm.sanitize_mol are used to ensure the chemical validity of molecules, they do not perform sanitization of metadata or property fields for malicious natural language instructions.

datamol