anndata
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill demonstrates reading datasets from remote URLs and cloud storage (S3) using
ad.read_h5ad,fsspec, andurllibinreferences/io_operations.md. These patterns are consistent with standard research workflows for accessing public datasets. - [INDIRECT_PROMPT_INJECTION]: The skill provides methods for ingesting data from numerous file formats (H5AD, Zarr, CSV, MTX, Excel, Loom), creating a theoretical surface for indirect prompt injection if malicious instructions are embedded in scientific metadata.
- Ingestion points: Multiple data loading methods are documented across
SKILL.md,references/io_operations.md, andreferences/manipulation.md. - Boundary markers: None specified; the skill uses standard library parsers for the respective file formats.
- Capability inventory: The skill facilitates file writing and data transformation via
anndata,pandas,numpy, andscanpy. - Sanitization: Standard data parsing is employed without additional filtering for potential instructions hidden in unstructured text fields.
Audit Metadata