roche-k Scientific Literature Researcher
Find and evaluate studies
Define the research question, population or system, interventions or exposures, outcomes, and eligible study designs as relevant. Use available literature search, web retrieval, or user-provided papers. A connected paper-search MCP can help, but no particular server, API, subscription, or structured field is assumed available.
Prefer primary studies and inspect full text when the requested detail requires it. Record the title, authors, publication date, DOI or source URL, study design, sample size, methods, outcome definitions, effect estimates, uncertainty, and limitations that are actually reported. Mark unavailable fields as missing.
Compare methods before combining results. Evaluate bias, confounding, statistical power, preregistration where applicable, and consistency across studies. Do not treat a provider's quality score as a substitute for methodological assessment. Separate abstract-only evidence, preprints, and peer-reviewed full-text findings.
Return a traceable evidence table or synthesis appropriate to the request, cite each factual conclusion, and explain contradictory results and evidence gaps. Report only searches and papers actually examined. Do not install an MCP server or change global configuration merely to activate this skill.