bio-single-cell-batch-integration
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests user-supplied data files (H5AD and RDS), which represents a vulnerability surface for indirect prompt injection if those files contain malicious instructions in metadata or unstructured data fields.
- Ingestion points:
sc.read_h5adinexamples/harmony_integration.pyandreadRDSinexamples/harmony_integration.Rare used to load external datasets into the agent's context. - Boundary markers: No specific boundary markers or instructions are employed to distinguish data from potential embedded instructions.
- Capability inventory: The skill allows the agent to write files to the local file system (
adata.writeinexamples/harmony_integration.pyandsaveRDSinexamples/harmony_integration.R) and generate visual output (plt.savefig,pdf). - Sanitization: Relies on standard scientific data parsers for format validation; however, there is no specific sanitization for natural language instructions potentially embedded within the data structures.
- [EXTERNAL_DOWNLOADS]: The skill requires the installation of standard bioinformatics packages from public registries.
- Python packages:
scanpy,harmonypy,scvi-tools,scanorama,bbknn, andscib-metricsare listed as prerequisites to be installed via pip. - R packages:
Seurat,harmony, andbatchelorare referenced for installation via CRAN or Bioconductor.
Audit Metadata