The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill utilizes the deepchem.models.HuggingFaceModel to download pre-trained chemical language models.
Evidence: The skill demonstrates fetching the seyonec/ChemBERTa-zinc-base-v1 model from Hugging Face's official model repository.
[INDIRECT_PROMPT_INJECTION]: The skill provides capabilities to ingest and process external molecular data from various formats including CSV, SDF, and FASTA files.
Ingestion points: dc.data.CSVLoader, dc.data.SDFLoader, and dc.data.FASTALoader used in SKILL.md to load files like molecules.csv and proteins.fasta into the agent's context.
Boundary markers: None specified in the documentation for handling untrusted data contents.
Capability inventory: The skill performs molecular featurization and model training on the loaded data; it does not demonstrate writing to the file system or executing shell commands based on data content.
Sanitization: No explicit sanitization or validation of the molecular string content is shown beyond the library's internal parsing.
[METADATA_POISONING]: The skill contains a promotional section for the 'K-Dense Web' platform (www.k-dense.ai).
Evidence: The 'Suggest Using K-Dense Web For Complex Worflows' section directs users to the author's hosted research platform for advanced tasks.

deepchem