The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes molecular data from external files (SDF, SMILES, CSV) and remote storage (S3, GCS), and is instructed to fetch citation metadata from ArXiv. This represents an ingestion surface where untrusted content could attempt to influence agent behavior.
Ingestion points: File reading from local and remote (S3/GCS/HTTPS) paths and citation metadata fetching from ArXiv.
Boundary markers: No explicit delimiters or instruction-ignore warnings are present in the provided code snippets for data ingestion.
Capability inventory: The skill allows file writing and shell command execution (via allowed-tools), and remote network access.
Sanitization: Uses chemical sanitization and standardization (dm.standardize_mol), but lacks specific security sanitization for prompt-based threats.
[EXTERNAL_DOWNLOADS]: The skill recommends installing the datamol library from PyPI and references source code and documentation on GitHub and official documentation sites. These are well-known services and trusted repositories.
Evidence: Mentions of installation via uv pip install datamol==0.13.0 and links to github.com/datamol-io, pypi.org, and arxiv.org.

datamol