This is a structured pipeline for turning somatic mutations into treatment recommendations. Feed it a gene, variant, and cancer type, and it walks through CIViC evidence tiers, cBioPortal prevalence, OpenTargets drug associations, resistance mechanisms, and active clinical trials, then synthesizes an actionable report. The workflow is prescriptive: it explicitly tells Claude to create the report file first, look up facts instead of guessing, and compute statistics instead of describing them. The driver versus passenger reasoning section is solid, distinguishing recurrent hotspots from likely noise. You'd use this for tumor board prep or when a patient's sequencing report lands and you need to know what therapies actually exist for their specific alteration. It's designed for cancer, but the evidence-grading and source-tracking patterns could apply to other clinical interpretation tasks.
npx skills add https://github.com/mims-harvard/tooluniverse --skill tooluniverse-cancer-variant-interpretation