research-refine

To turn a vague research idea into an implementable proposal, you provide a general problem and approach. The skill locks in a problem anchor and applies frontier-aware refinement to generate a concrete, paper-ready method plan.

Problem anchoring in research planning preserves an immutable problem definition across iterative refinement rounds. It is needed to prevent problem drift and ensure your final proposal stays focused on the original bottleneck.

To design minimal validation experiments for a research proposal, define 1–3 decisive experiments per core claim. This claim-driven approach prevents oversized benchmark plans and ensures targeted testing of your bottleneck mechanism.

Yes, you can use LLM review to refine your approach and decompose a research problem. The skill uses reviewer-based scoring across problem fidelity and method specificity to iteratively anchor and improve your proposal generation.

Research planning supports checkpoint recovery for iterative proposal refinement via refine-logs state persistence. This allows you to recover progress and resume refining your approach without losing previous method design decisions.

Use frontier-aware method design over standard benchmarking when you need the smallest adequate mechanism. It proposes foundation-model-era primitives only when they naturally fit your bottleneck, avoiding oversized validation plans.