The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external molecular data files (CSV, SDF, FASTA) using various loaders defined in references/api_reference.md and implemented in the scripts/ directory. This creates a standard surface for indirect prompt injection if the ingested data contains unexpected content that the agent is subsequently tasked to process.
Ingestion points: CSVLoader, SDFLoader, and FASTALoader are used in scripts/graph_neural_network.py, scripts/predict_solubility.py, and scripts/transfer_learning.py to import external datasets.
Boundary markers: The instructions do not define specific delimiters or warnings for the agent when interpreting the content of these external files.
Capability inventory: The skill utilizes the Bash tool to execute Python scripts that perform complex computations and can write results to the filesystem.
Sanitization: The 'Agent operating procedure' in SKILL.md explicitly instructs the agent to 'sanitize molecules before computing', which serves as a mitigation measure.
[EXTERNAL_DOWNLOADS]: The skill facilitates the download of pre-trained machine learning models and benchmark datasets from established, well-known services.
Evidence: scripts/transfer_learning.py and references/workflows.md describe the use of dc.models.HuggingFaceModel to fetch models such as seyonec/ChemBERTa-zinc-base-v1, ibm/MoLFormer-XL-both-10pct, and Rostlab/prot_bert from Hugging Face.
Evidence: references/api_reference.md and references/core_capabilities.md describe the use of dc.molnet.load_* functions to fetch standard MoleculeNet benchmark datasets via the DeepChem project's infrastructure.
Context: These operations target well-known repositories and services essential for the skill's primary purpose in scientific research.

deepchem