The Agent Skills Directory

[PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection by ingesting untrusted molecular data (SMILES strings) that are processed within the agent's context. While these strings are chemically structured, they represent a vector where embedded instructions could potentially influence the agent if it interprets the data beyond its intended featurization purpose.
Ingestion points: Untrusted SMILES strings and molecular datasets are ingested in batch processing functions across SKILL.md and references/examples.md (e.g., smiles_list, screening_library).
Boundary markers: No explicit delimiters or instructions to ignore embedded content are documented in the provided code snippets.
Capability inventory: The skill utilizes Read, Write, Edit, and Bash tools to manage files and execute Python-based ML pipelines.
Sanitization: No specific sanitization or filtering for prompt injection markers is performed on input strings; the skill relies on standard chemical validation via datamol and rdkit.
[EXTERNAL_DOWNLOADS]: The skill installs the molfeat library and its associated scientific ecosystem (such as PyTorch, DGL, and Transformers) from the Python Package Index (PyPI). It also references external research packages like map4 from public GitHub repositories for specialized featurization tasks.
[COMMAND_EXECUTION]: The skill uses shell access to perform environment setup, library installation via uv pip, and execution of molecular processing scripts. These operations are within the expected scope of a cheminformatics development tool.

molfeat