tooluniverse-multi-omics-integration
Pass
Audited by Gen Agent Trust Hub on Mar 8, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill presents a vulnerability surface for indirect prompt injection via the processing of external data files.
- Ingestion points: The workflow in
phase_details.md(Phase 1) loads data from various external sources including CSV/TSV matrices, HDF5 files, VCF genomic variants, and peak tables. - Boundary markers: The skill does not implement boundary markers or instructions to the model to ignore potential commands embedded within the data fields (e.g., malicious strings in gene identifiers or metadata) when generating the analysis report in Phase 8.
- Capability inventory: The skill is capable of performing statistical analyses, multi-omics clustering (NMF), and orchestrating other ToolUniverse skills for gene enrichment and variant analysis.
- Sanitization: There is no explicit sanitization or validation of input identifiers (like gene names, protein IDs, or sample labels) before they are interpolated into the Markdown report template provided in Phase 8.
Audit Metadata