scvi-tools-single-cell

Pass

Audited by Gen Agent Trust Hub on Apr 28, 2026

Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PROMPT_INJECTION]: The skill presents a surface for indirect prompt injection via the ingestion of biological datasets.
  • Ingestion points: The instructions utilize scanpy.read_h5ad() to load external AnnData files in multiple workflow examples and scvi.model.SCANVI.load() to load saved models.
  • Boundary markers: The skill lacks explicit delimiters or instructions to treat data content (such as cell labels or metadata) as non-executable or to ignore embedded instructions within those fields.
  • Capability inventory: The agent is instructed to use capabilities for writing to the file system via model.save() and pandas.DataFrame.to_csv(), and can execute shell commands for environment setup.
  • Sanitization: No specific input validation or sanitization of the structured data content is implemented before the data is processed by the model or used for downstream tasks like differential expression.
  • [EXTERNAL_DOWNLOADS]: The skill performs routine software and data acquisition.
  • The Quick Start section uses scvi.data.heart_cell_atlas_subsampled() to download a reference dataset from the scvi-tools official repository.
  • The Prerequisites section provides instructions to install standard bioinformatics packages including scvi-tools and scanpy from the Python Package Index (PyPI).
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 28, 2026, 02:13 PM