alterlab-scanpy
Pass
Audited by Gen Agent Trust Hub on Sep 25, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted scientific data files (.h5ad, .csv, .mtx) in scripts/qc_analysis.py and assets/analysis_template.py, which serves as a potential vector for indirect prompt injection.
- Ingestion points: Data loading functions sc.read_h5ad, sc.read_csv, sc.read_10x_h5, and sc.read_10x_mtx are used to bring external data into the analysis environment.
- Boundary markers: No explicit boundary markers or instructions to ignore embedded instructions within data fields were found in the provided code or templates.
- Capability inventory: The skill possesses file-write capabilities (adata.write, to_csv, os.makedirs) to save analysis results and figures to the local file system.
- Sanitization: The skill relies on standard library parsing from Scanpy and AnnData without additional validation or sanitization of non-numeric metadata content that could contain malicious instructions.
Audit Metadata