The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external genomic data files (e.g., .h5ad, .mtx, .csv). While these are standard formats for biological data, processing untrusted external files represents a potential surface for indirect prompt injection if the data contains adversarial content intended to influence the agent's behavior during analysis.
Ingestion points: Data loading functions such as sc.read_10x_mtx, sc.read_h5ad, and sc.read_csv in SKILL.md and references/api_reference.md.
Boundary markers: Not present in the provided snippets.
Capability inventory: The skill utilizes Python execution environment with capabilities for local file system read/write (e.g., adata.write, to_csv) and data visualization.
Sanitization: No explicit sanitization of data values is performed beyond standard library parsing.
[EXTERNAL_DOWNLOADS]: The documentation includes standard installation instructions for well-known scientific Python packages via PyPI.
Evidence: pip install "scanpy[leiden]" anndata in SKILL.md.

scanpy-scrna-seq