What problem does it solve? Starting from a vague research direction often leads to unfocused brainstorming, untested ideas, and wasted compute. This Skill anchors the problem first, establishes a measurable baseline, and narrows the direction to a single falsifiable, literature-grounded idea before any experimentation begins. ## Core Features & Use Cases - Problem Anchoring: Freezes one problem, one primary metric, one evaluation setting, and a compute bound in a PROBLEM_ANCHOR.md file before generating ideas. - Baseline Establishment: Identifies the strongest runnable reference implementation and measures a reproducible baseline before proposing any new method. - Guided Idea Selection: Ranks mapped candidate ideas by failure match, evidence, falsifiability, and screening cost, then tests one candidate at a time. - Use Case: A researcher says "explore improving retrieval quality for long-context QA." The Skill anchors the metric and budget, measures the existing baseline, deep-reads the nearest literature, and carries one selected idea through experiments and optionally a paper. ## Quick Start Ask the agent to use discover-research-idea to turn your broad research direction into one literature-grounded, experimentally tested idea.
