external-model-validation
Pass
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [SAFE]: The skill follows professional engineering standards for medical data processing tools. It includes thorough input validation, Random Seed setting for reproducibility, and explicit Research Use notices requiring de-identification of data.
- [INDIRECT_PROMPT_INJECTION]: Surface identified in scripts/functions.R where the skill reads user-provided CSV data. Ingestion points: read_expression_data, read_clinical_data, and read_model_data. Boundary markers: CSV structure. Capability inventory: standard R numeric operations and system2 call in test scripts. Sanitization: Numeric type enforcement and column name checks ensure the data is processed as statistical input and not interpreted as agent instructions.
- [DYNAMIC_EXECUTION]: The project uses the source() function in scripts/main.R to load internal utility modules. This use of dynamic execution is restricted to local paths within the skill's own directory structure and follows standard R modularity patterns.
Audit Metadata