The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill implements remote file I/O capabilities through the fsspec library, allowing users to read molecular datasets directly from Amazon S3, Google Cloud Storage, and web URLs (HTTP/HTTPS). This is a standard feature of the datamol package designed for handling large-scale chemical data.
[DATA_EXFILTRATION]: The skill's I/O module supports writing molecules and DataFrames to remote cloud storage locations. While this allows data to be moved to external infrastructure, it is a documented and expected function for saving research outputs in distributed environments.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface because it processes untrusted external data files (SDF, CSV, Excel). If a data file contains instructions hidden in molecular metadata fields, it could potentially influence the agent's behavior during analysis.
Ingestion points: Data enters the agent's context through functions such as dm.read_sdf(), dm.read_csv(), and dm.open_df() as documented in references/io_module.md.
Boundary markers: The skill does not implement explicit delimiters or instructions to ignore natural language content within data fields.
Capability inventory: The toolset includes significant capabilities such as writing to the local and remote file systems (dm.to_sdf(), dm.to_xlsx()) and performing network operations via cloud storage protocols.
Sanitization: The library provides extensive chemical sanitization (dm.sanitize_mol()) to ensure structural validity, but it does not filter or sanitize natural language instructions from data attributes.

datamol