research-refine
Turn a vague research direction into a problem-anchored, elegant, frontier-aware, implementation-oriented method plan via iterative GPT-5.4 review. Use when the user says "refine my approach", "帮我细化方案", "decompose this problem", "打磨idea", "refine research plan", "细化研究方案", or wants a concrete research method that stays simple, focused, and top-venue ready instead of a vague or overbuilt idea.
Profile →npx skills add brycewang-stanford/Auto-Empirical-Research-Skills --skill research-refine --agent claude-code
Same command for any agent — swap --agent for codex, cursor, copilot.
Weekly change comes from our own snapshots, not the repository page — it measures attention, not adoption.
What it does
The skill guides a user from a vague research direction to a concrete plan by preserving an immutable Problem Anchor and producing a stepwise, implementable proposal focused on a single dominant contribution. It structures the work into phases (anchor, proposal, review, refine) with defined checkpoints, state persistence to refine-logs/, and a fixed review round flow up to MAX_ROUNDS. It enforces a minimal intervention approach, prefers a single dominant thesis, and encourages using modern primitives only when they fit the bottleneck. It requires saving a full anchored proposal and subsequent refinement rounds to refine-logs/ with a formal OUTPUT_DIR and REFINE_STATE.json state file after phase boundaries.
How it works
- Phase 0: Freeze the Problem Anchor by extracting the Bottom-line problem, Must-solve bottleneck, Non-goals, Constraints, and Success condition; write refine-logs/REFINE_STATE.json with phase "anchor" and round 0.
- Phase 1: Build the Initial Proposal through Step 1.1–1.6, including scanning grounding material, identifying the technical gap, choosing the sharpest route, and concretizing the method with detailed sections (Problem Anchor, Technical Gap, Method Thesis, Contribution Focus, Proposed Method, Training Plan, Failure Modes, Novelty and Elegance, Claim-Driven Validation, Experiment Handoff, Compute & Timeline). Save to refine-logs/round-0-initial-proposal.md and update refine-logs/REFINE_STATE.json with phase "proposal" and round 0.
- Phase 2: External Method Review (Round 1): Send the full proposal to a reviewer (GPT-5.4) via mcp__codex__codex with a prompt that emphasizes a method-first critique focusing on Problem Fidelity, Method Specificity, and other dimensions; score 7 dimensions from 1-10 (including Problem Fidelity, etc.). The process references the provided proposal content and adheres to the specified scoring schema.
- Phase 3–5: Revisions, re-evaluations, and logging until SCORE_THRESHOLD or MAX_ROUNDS are reached, with appropriate updates to refine-logs and checkpointSTATE. The workflow requires persistence to refine-logs/REFINE_STATE.json after each phase boundary and maintains a full history in refine-logs/.
When to use it
Use when the research problem is visible but the technical route is fuzzy, and you want a concrete, focused, implementable plan with a minimal intervention and a clear single dominant contribution.
What it can touch
- Tools: Bash, Read, Write, Edit, Grep, Glob, WebSearch, WebFetch, Agent, mcp__codex__codex, mcp__codex__codex-reply. These are used to scan literature, construct the proposal, and perform external review via Codex/MCP prompts.
Caveats
- License: NOASSERTION.
- The workflow hinges on Phase transitions and checkpoint recovery to avoid loss of progress.
- The method emphasizes a single dominant contribution and warns against contribution sprawl and bloated experiment menus.
- All content references the anchored problem; drift is flagged if reviewer feedback would change the problem.
# Research Refine: Problem-Anchored, Elegant, Frontier-Aware Plan Refinement Refine and concretize: **$ARGUMENTS** ## Overview Use this skill when the research problem is already visible but the technical route is still fuzzy. The goal is not to produce a bloated proposal or a benchmark shopping list. The goal is to turn a vague direction into a **problem -> focused method -> minimal validation**
What does the research-refine skill do?
Turn a vague research direction into a problem-anchored, elegant, frontier-aware, implementation-oriented method plan via iterative GPT-5.4 review. Use when the user says "refine my approach", "帮我细化方案", "decompose this problem", "打磨idea", "refine research plan", "细化研究方案", or wants a concrete research method that stays simple, focused, and top-venue ready instead of a vague or overbuilt idea.
How do I install it?
Run `npx skills add brycewang-stanford/Auto-Empirical-Research-Skills --skill research-refine --agent claude-code` — it drops the skill into your project so the agent can pick it up. Swap the --agent value for codex, cursor or copilot if you use one of those.
Where does this skill come from?
From brycewang-stanford/Auto-Empirical-Research-Skills, a repository with 3,244 stars. We read it straight from the repository tree rather than a submitted listing, so what you see here is what is actually published.
Is a popular skill a good skill?
Not necessarily. Stars measure attention, not adoption — a repository can trend for a week and be abandoned. That is why we show the weekly change from our own snapshots next to the total, instead of a single flattering number.