bio-causal-genomics-mendelian-randomization
Pass
Audited by Gen Agent Trust Hub on Aug 25, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [UNVERIFIABLE_DEPENDENCIES]: The skill recommends installing several R packages from GitHub repositories maintained by academic institutions (e.g., MRC Integrative Epidemiology Unit at the University of Bristol) and specialized researchers. These tools, such as TwoSampleMR, ieugwasr, and MR-PRESSO, are the standard infrastructure for Mendelian Randomization studies.
- [DYNAMIC_EXECUTION]: The skill generates and executes R code to perform statistical calculations and visualizations. It also utilizes the genetics.binaRies package to retrieve the plink binary for local linkage disequilibrium clumping, which is a standard procedure in genomic analysis.
- [DATA_EXPOSURE]: The guide provides instructions for managing OpenGWAS API tokens in the .Renviron file. This follows common security practices for secure credential management in the R environment and does not involve hardcoded or exfiltrated secrets.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests GWAS summary statistics from local files (TSVs). This represents a data ingestion surface, though the risk is assessed as minimal because the data is processed as structured numerical input for statistical functions and not interpreted as natural language instructions. 1. Ingestion points: exposure_gwas.tsv, outcome_gwas.tsv processed via read_exposure_data and read_outcome_data. 2. Boundary markers: Absent; the skill relies on the structural integrity of the input TSV files. 3. Capability inventory: File system access for reading/writing GWAS data and execution of R code blocks (examples/cis_mr_drug_target.R, examples/two_sample_mr.R). 4. Sanitization: No explicit sanitization of text fields within the TSV data for prompt injection content.
Audit Metadata