Two-Sample MR Exposure-Screening Reference-Grounded Research Planner

Problem

Two-sample Mendelian randomization planning often collapses into a generic tool list or literature summary. This skill turns an exposure/outcome request into an executable study design, with explicit scope control and reference-grounded support.

Core behavior and workflow

• Input validation: expects outcome + one exposure or exposure family + validation emphasis; redirects clinical treatment, raw individual genotype pipelines, non-genetic observational work, one-sample MR-centered plans, or off-topic requests.
• Four workload configs: always emits Lite / Standard / Advanced / Publication+, recommends one primary path, and explains why other levels are less suitable.
• Study pattern selection: supports single-exposure MR, exposure-panel screening, robustness/sensitivity, extended causal architecture, and replication/triangulation.
• Reference layer: requires design-support references with resolvable identifiers such as DOI, PMID, PMCID, PMC, Crossref, or official resource pages; unverified candidates become evidence-gap notes rather than citations.
• Dependency consistency check: downstream claims are only allowed after the required GWAS source, ancestry assumptions, instrument logic, harmonization, sensitivity, MVMR, reverse-MR, or replication modules are declared.
• Output discipline: includes core question, config table, workflow, figure plan, validation hierarchy, minimal executable version, publication upgrade path, and self-critical risk review.

Boundaries

Useful for researchers designing two-sample MR studies, public-summary-statistics screening, IVW primary analysis, heterogeneity/pleiotropy checks, and MVMR or reverse-MR extensions. It is not for clinical diagnosis, prescribing, individual risk prediction, mechanism confirmation, or therapeutic certainty; datasets are references and must be reselected based on access, quality, and methodological fit.