scanpy

Pass

Audited by Gen Agent Trust Hub on Jun 27, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [SAFE]: The skill uses established scientific computing libraries including Scanpy, Pandas, NumPy, and Matplotlib to perform data analysis tasks.
  • [EXTERNAL_DOWNLOADS]: Provides links to official Scanpy documentation and the scverse ecosystem tutorials. These references point to well-known, trusted resources within the bioinformatics community.
  • [COMMAND_EXECUTION]: Includes Python scripts (scripts/qc_analysis.py) and templates (assets/analysis_template.py) that perform local file operations such as reading input datasets and writing results, plots, and metadata to user-specified directories.
  • [PROMPT_INJECTION]: The skill processes user-supplied data files (e.g., .h5ad, .csv, .h5), which constitutes a theoretical surface for indirect prompt injection via metadata or feature names.
  • Ingestion points: Data reading functions sc.read_h5ad, sc.read_10x_mtx, sc.read_10x_h5, and sc.read_csv found in SKILL.md, assets/analysis_template.py, and scripts/qc_analysis.py.
  • Boundary markers: None present.
  • Capability inventory: Local file writing via adata.write and to_csv; directory creation using os.makedirs.
  • Sanitization: None implemented for natural language instructions embedded in scientific data structures.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 27, 2026, 11:42 AM
Security Audit — agent-trust-hub — scanpy