drafting-research-plan-from-idea-folder

What problem does it solve? Turning a folder of raw ML paper ideas into a rigorous, publication-ready research plan is hard: ideas must be audited honestly, theory must be developed without overstating proofs, and experiments must fit real compute budgets. This Skill automates that entire pipeline and produces a single academic-Chinese work.md. ## Core Features & Use Cases - Three-Stage Pipeline: Runs idea audit, theory construction, and experiment program design in strict order, with each stage's output feeding the next. - Honest Gating: Forces a proceed status (PROCEED, PROCEED WITH MAJOR RISK, NO-GO) and forbids upgrading proof sketches into theorems. - Resource-Bounded Experiments: Maps every experiment to a claim and enforces hard limits on prompts, GPU count, wall-clock, and annotation budgets with an MVP fallback plan. - Use Case: Given a folder containing outputs/IDEA_DISCOVERY_REPORT.md and refine-logs/FINAL_PROPOSAL.md, produce a complete work.md with thesis, theory derivations, and a claim-driven experiment plan ready for an ML conference submission. ## Quick Start Use this Skill on my idea folder at ./my-paper-idea to audit the strongest idea and write a publication-grade work.md with theory and experiment plans.