The Agent Skills Directory

[SAFE]: The skill implements standard cheminformatics workflows using the datamol library, which is a legitimate and well-known Python wrapper for RDKit.
[EXTERNAL_DOWNLOADS]: The skill instructs the user to install 'datamol' via pip. Datamol is an established open-source library in the scientific community.
[COMMAND_EXECUTION]: The library uses parallel processing via the n_jobs parameter to spawn subprocesses. This is a standard performance feature for molecular computations and does not involve the execution of arbitrary or untrusted shell commands.
[DATA_EXFILTRATION]: The skill documents capabilities to read from and write to remote cloud storage (AWS S3, Google Cloud Storage, Azure) and HTTP endpoints via the fsspec library. These are documented functional requirements for handling distributed molecular datasets and do not show patterns of sensitive credential harvesting or unauthorized data transfer.
[PROMPT_INJECTION]: The skill processes molecular data files from untrusted external sources, which constitutes an indirect prompt injection surface. Ingestion points: dm.read_sdf, dm.read_csv, dm.read_excel, and dm.open_df in SKILL.md and references/io_module.md. Boundary markers: Not explicitly demonstrated for data files. Capability inventory: Subprocess calls for parallelization and network/file system writes for saving results. Sanitization: The skill includes chemical structure sanitization (dm.sanitize_mol) which verifies molecular validity but does not include sanitization for non-chemical metadata fields against malicious instructions.

datamol