The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as it is designed to ingest and process untrusted data from multiple file formats and remote URLs.
Ingestion points: The skill provides numerous methods to load data from external files, including read_h5ad, read_csv, read_zarr, read_mtx, read_10x_h5, read_loom, read_excel, and read_text (detailed in references/io_operations.md).
Boundary markers: The instructions do not define boundary markers or explicit prompts to ignore instructions embedded within the processed data files.
Capability inventory: The skill documentation includes capabilities for file system writes (write_h5ad, write_zarr, write_csvs), network access via fsspec for remote Zarr stores, and integration with analysis tools like scanpy and muon (found in SKILL.md and references/io_operations.md).
Sanitization: There is no mention of sanitization or validation of the content within the ingested data structures to prevent instruction-like strings from being interpreted by an agent.
[EXTERNAL_DOWNLOADS]: The skill includes instructions and examples for fetching data from external URLs.
Evidence: references/io_operations.md provides examples using fsspec to access Zarr stores from URLs and urllib.request.urlretrieve to download H5AD files. These are documented as standard data loading patterns for the library's primary purpose.

anndata