What problem does it solve? Brainstorming research hypotheses is easy, but knowing which ones are actually novel, grounded in prior work, and testable requires tedious literature checking. This Skill automates that vetting: it generates candidate hypotheses, grounds each one in real retrieved literature, scores them against a fixed rubric, and keeps only strong, non-duplicate candidates. ## Core Features & Use Cases - Multi-agent generation loop: A Generator proposes candidates, a LiteratureScout grounds each in real papers (novelty, support, gap), and a Judge scores them on a fixed five-axis rubric with an evidence gate. - Saturation-based stopping: Rounds repeat, mutating toward open gaps, until fresh rounds stop adding keepers or the round budget is exhausted. - Schema-validated artifacts: Each round produces candidates.json, litscout.json, and verdict.json validated against JSON schemas, plus a ledger tracking pool growth. - Use Case: A researcher asks what factors improve long-term retention of procedural skills; the loop returns a ranked set of vetted hypotheses, each with closest prior work, supporting citations, the gap it fills, and a concrete experiment to test it. ## Quick Start Ask the assistant to generate and literature-vet novel testable hypotheses for your research question, then confirm the proposed loop settings before the rounds begin.
