What problem does it solve? Turning a rough research idea into a concrete, focused, venue-ready proposal is hard: ideas drift from the original problem, accumulate unnecessary modules, and lack method-level specificity. This Skill runs a structured multi-round refinement loop that anchors the problem, extracts a logical skeleton, and uses an external LLM reviewer to stress-test each revision. ## Core Features & Use Cases - Problem Anchor + Skeleton Extraction: Freezes the immutable bottom-line problem and maps the logical path a reviewer must follow, preventing drift across revision rounds. - Iterative External Review Loop: Sends proposals to GPT-5.4 via Codex MCP for 7-dimension scored reviews (problem fidelity, method specificity, contribution quality, frontier leverage, feasibility, validation focus, venue readiness), then revises until score >= 9 or max 3 rounds. - Full Audit Trail: Saves skeleton, per-round proposals and reviews, score history, review summary, final proposal, and refinement report to refine-logs/. - Use Case: A researcher has a vague idea like "use LLMs to improve retrieval in robotics." Run this Skill to produce a focused proposal with one dominant mechanism-level contribution, a concrete training recipe, and an evaluation sketch ready for a top venue. ## Quick Start Ask the AI to refine your research idea by saying: refine this idea into a venue-ready proposal, followed by your problem description and rough approach.
