The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches molecular benchmark datasets from MoleculeNet and pulls pretrained transformer models from well-known Hugging Face repositories (such as ChemBERTa and MolFormer). These downloads are performed through official library APIs and are necessary for the skill's primary scientific functions.
[COMMAND_EXECUTION]: The skill uses the Bash tool to run Python scripts for training and evaluating models. This execution is scoped to the provided scripts and standard machine learning workflows.
[PROMPT_INJECTION]: The skill ingests external data (SMILES strings and CSV files) which constitutes an indirect prompt injection surface. However, the skill relies on standard scientific parsing libraries (DeepChem/RDKit) and does not exhibit any patterns suggesting exploitation of the agent's logic through these inputs.
Ingestion points: Data loading methods in scripts/graph_neural_network.py and scripts/predict_solubility.py.
Boundary markers: None identified for molecular string inputs.
Capability inventory: Permission to use Bash, Write, and Edit tools.
Sanitization: Uses standard scientific data loaders from the DeepChem ecosystem.

deepchem