What problem does it solve? Turning a broad research direction into a set of vetted, experiment-ready hypotheses requires literature surveying, novelty checking, impact assessment, critical review, and experiment planning — a slow, error-prone manual process. This Skill automates that entire claim stage, producing 10 ranked, reviewer-ready research proposals from a single research direction. ## Core Features & Use Cases - Batch Hypothesis Generation: Brainstorms roughly 30 candidate ideas across three angle-partitioned rounds, then cuts them to 10 through novelty gates, impact scoring, and external critical review. - Multi-Signal Selection: Combines novelty verification, impact scoring, and senior-reviewer feedback to veto fatal design flaws and rank survivors by impact first. - Per-Claim Deliverables: Each surviving idea gets its own directory with a refined proposal, an experiment plan with a phenomenon-validation gate, and a claim.json paper skeleton written for a human expert reviewer. - Use Case: A mechanistic interpretability researcher provides a direction like "attention head behavior in in-context learning" and receives 10 ranked, falsifiable hypotheses, each with a concrete experiment ladder, controls, and risk analysis. ## Quick Start Run the hypothesis-batch skill with a research direction such as "mechanisms of in-context learning in transformer models" to generate ten reviewed and experiment-planned research proposals.
