The Agent Skills Directory

[SAFE]: The skill utilizes well-known and reputable scientific Python packages including cellxgene-census, scanpy, anndata, and tiledbsoma for biological data retrieval and analysis.
[SAFE]: Network operations are exclusively directed toward official CZ CELLxGENE Census endpoints to fetch public transcriptomics data, which is the primary and intended function of the skill.
[SAFE]: Analysis of the code patterns for querying, machine learning integration (via tiledbsoma-ml), and spatial data processing reveals standard API usage with no evidence of obfuscation or malicious intent.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of cell metadata and gene expression data from external scientific sources. While this represents a data-processing attack surface, the risk is minimal given the structured nature of the data and the reputation of the source.
Ingestion points: cellxgene_census.get_obs(), cellxgene_census.get_anndata(), and axis_query() methods used in SKILL.md and reference files.
Boundary markers: Absent; the skill relies on library-level parsing of structured scientific data.
Capability inventory: The skill is configured with Read, Write, Edit, and Bash tools.
Sanitization: Data is processed as structured AnnData or SOMA objects by standard scientific libraries; no specific prompt-level sanitization of ingested metadata is present.

cellxgene-census