two-sample-mr-research-planner
Pass
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill includes R code frameworks that fetch genetic data from the IEU Open GWAS database, a well-known academic service hosted by the University of Bristol. This is a standard and expected operation for the skill's medical research purpose.
- [COMMAND_EXECUTION]: The instructions generate R code templates for the user to execute locally. These templates utilize industry-standard packages (TwoSampleMR and MRPRESSO) and incorporate safety measures like mandatory placeholder identifiers and automated feasibility checks for instrument strength.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests user research directions to generate study plans. Ingestion points: The agent context is updated by user-provided exposure and outcome descriptions. Boundary markers: The skill contains 'Hard Rules' and 'Input Validation' sections that restrict the agent from performing non-MR tasks or overclaiming causal results. Capability inventory: The skill produces R code templates for statistical analysis and network requests to scientific APIs. Sanitization: The skill employs placeholder values for data identifiers and includes logic to redirect out-of-scope requests.
- [SAFE]: All referenced external resources, including genomic databases and CRAN packages, are legitimate academic tools. The skill does not contain obfuscation, persistence mechanisms, or unauthorized privilege escalation attempts.
Audit Metadata