tooluniverse-single-cell

Warn

Audited by Gen Agent Trust Hub on Oct 7, 2026

Risk Level: MEDIUMDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [DYNAMIC_EXECUTION]: The script scripts/scrna_qc.py utilizes the __import__ function within its preflight() routine to dynamically check for the existence of libraries such as scanpy, anndata, and numpy. While functionally intended for environment validation, dynamic module loading is a technique that can be leveraged to load arbitrary code from computed paths at runtime.
  • [INDIRECT_PROMPT_INJECTION]: The skill is architected to ingest and process external, untrusted data formats (H5AD, CSV, and TSV count matrices). This creates a vulnerability surface where malicious instructions could be embedded within metadata or data fields to influence the agent's downstream behavior.
  • Ingestion points: Data loading logic in SKILL.md and scripts/scrna_qc.py (via sc.read_h5ad and pd.read_csv).
  • Boundary markers: The skill does not implement delimiters or explicit 'ignore instructions' warnings when processing these files.
  • Capability inventory: The skill has the capability to write files, execute shell commands (via Python scripts), and invoke a wide range of biological analysis tools via ToolUniverse.
  • Sanitization: No evidence of sanitization, validation, or escaping of the content retrieved from external data files was found.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Oct 7, 2026, 01:49 PM