research-refine skill

Override for Codex users who want Gemini, not a second Codex agent, to act as the reviewer. Install this package after skills/skills-codex/*.

Research Refine: Problem-Anchored, Elegant, Frontier-Aware Plan Refinement

Gemini overlay assurance: reviewindependence: cross-family and acceptancestatus: accepted.

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 document that is concrete enough to implement, elegant enough to feel paper-worthy, and current enough to resonate in the foundation-model era.

Four principles dominate this skill:

1. Do not lose the original problem. Freeze an immutable Problem Anchor and reuse it in every round.
2. The smallest adequate mechanism wins. Prefer the minimal intervention that directly fixes the bottleneck.
3. One paper, one dominant contribution. Prefer one sharp thesis plus at most one supporting contribution.
4. Modern leverage is a prior, not a decoration. When LLM / VLM / Diffusion / RL / distillation / inference-time scaling naturally fit the bottleneck, use them concretely. Do not bolt them on as buzzwords.

User input (PROBLEM + vague APPROACH)
  -> Phase 0 (Gemini): Freeze Problem Anchor
  -> Phase 1 (Gemini): Scan grounding papers -> identify technical gap -> choose the sharpest route -> write focused proposal
  -> Phase 2 (Codex/Gemini): Review for fidelity, specificity, contribution quality, and frontier leverage
  -> Phase 3 (Gemini): Anchor check + simplicity check -> revise method -> rewrite full proposal
  -> Phase 4 (Codex, same Gemini thread): Re-evaluate revised proposal
  -> Repeat Phase 3-4 until OVERALL SCORE >= 9 or MAX_ROUNDS reached
  -> Phase 5: Save full history to refine-logs/
  -> Optional handoff: /experiment-plan for a detailed execution-ready experiment roadmap

Constants

• REVIEWERMODEL = gemini-review — Gemini reviewer invoked through the local gemini-review MCP bridge. Set GEMINIREVIEW_MODEL if you need a specific Gemini model override.
• MAXROUNDS = 5** — Maximum review-revise rounds.
• SCORETHRESHOLD = 9** — Minimum overall score to stop.
• OUTPUTDIR = refine-logs/** — Directory for round files and final report.
• MAXLOCALPAPERS = 15 — Maximum local papers/notes to scan for grounding.
• MAXCOREEXPERIMENTS = 3 — Default cap for core validation blocks inside this skill.
• MAXPRIMARYCLAIMS = 2 — Soft cap for paper-level claims. Prefer one dominant claim plus one supporting claim.
• MAXNEWTRAINABLECOMPONENTS = 2** — Soft cap for genuinely new trainable pieces. Exceed only if the paper breaks otherwise.

Override via argument if needed, e.g. /research-refine "problem | approach" -- max rounds: 3, threshold: 9.

Output Structure

refine-logs/
├── round-0-initial-proposal.md
├── round-1-review.md
├── round-1-refinement.md
├── round-2-review.md
├── round-2-refinement.md
├── ...
├── REVIEW_SUMMARY.md
├── FINAL_PROPOSAL.md
├── REFINEMENT_REPORT.md
└── score-history.md

Every round-N-refinement.md must contain a full anchored proposal, not just incremental fixes.

Workflow

Phase 0: Freeze the Problem Anchor

Before proposing anything, extract the user's immutable bottom-line problem. This anchor must be copied verbatim into every proposal and every refinement round.

Write:

• Bottom-line problem: What technical problem must be solved?
• Must-solve bottleneck: What specific weakness in current methods is unacceptable?
• Non-goals: What is explicitly not the goal of this project?
• Constraints: Compute, data, time, tooling, venue, deployment limits.
• Success condition: What evidence would make the user say "yes, this method addresses the actual problem"?

If later reviewer feedback would change the problem being solved, mark that as drift and push back or adapt carefully.

Phase 1: Build the Initial Proposal

Step 1.1: Scan Grounding Material

Check papers/ and literature/ first. Read only the relevant parts needed to answer:

• What mechanism do current methods use?
• Where exactly do they fail for this problem?
• Which recent LLM / VLM / Diffusion / RL era techniques are actually relevant here?
• What training objectives, representations, or interfaces are reusable?
• What details distinguish a real method from a renamed high-level idea?

If local material is insufficient, search recent top-venue/arXiv work online. Focus on method sections, training setup, and failure modes, not just abstracts.

Step 1.2: Identify the Technical Gap

Do not stop at generic research questions. Make the gap operational:

1. Current pipeline failure point: where does the baseline break?
2. Why naive fixes are insufficient: larger context, more data, prompting, memory bank, or stacking more modules.
3. Smallest adequate intervention: what is the least additional mechanism that could plausibly fix the bottleneck?
4. Frontier-native alternative: is there a more current route using foundation-model-era primitives that better matches the bottleneck?
5. Core technical claim: what exact mechanism claim could survive top-venue scrutiny?
6. Required evidence: what minimum proof is needed to defend that claim?

Step 1.3: Choose the Sharpest Route

Before locking the method, compare two candidate routes if both are plausible:

• Route A: Elegant minimal route — the smallest mechanism that directly targets the bottleneck.
• Route B: Frontier-native route — a more modern route that uses LLM / VLM / Diffusion / RL / distillation / inference-time scaling only if it gives a cleaner or stronger story.

Then decide:

• Which route is more likely to become a strong paper under the stated constraints?
• Which route has the cleaner novelty story relative to the closest work?
• Which route avoids contribution sprawl?

If both routes are weak, rethink the framing instead of combining them into a larger system by default.

Step 1.4: Concretize the Method First

The proposal must answer "how would we actually build this?" Prefer method detail over broad experimentation and prefer reuse over invention.

Cover:

1. One-sentence method thesis: the single strongest mechanism claim.
2. Contribution focus: one dominant contribution and at most one supporting contribution.
3. Complexity budget: what is frozen or reused, what is new, and what tempting additions are intentionally excluded.
4. System graph: modules, data flow, inputs, outputs.
5. Representation design: what latent, embedding, plan token, reward signal, memory state, or alignment space is used?
6. Training recipe: data source, supervision, pseudo-labeling, negatives, curriculum, losses, weighting, stagewise vs joint training.
7. Inference path: how the trained components are used at test time and what signals flow where.
8. Why the mechanism stays small: why a larger stack is unnecessary.
9. Exact role of any frontier primitive: if you use an LLM / VLM / Diffusion / RL component, specify whether it acts as planner, teacher, critic, reward model, generator prior, search controller, or distillation source.
10. Failure handling: what could go wrong and what fallback or diagnostic exists?
11. Novelty and elegance argument: why this is more than naming a module and why the paper still looks focused.

If the method is still only described as "add a module" or "use a planner," it is not concrete enough.

Step 1.5: Design Minimal Claim-Driven Validation

Experiments exist to validate the method, not to dominate the document.

For each core claim, define the smallest strong experiment that can validate it:

• the claim being tested
• the necessary baseline or ablation
• the decisive metric
• the expected directional outcome

Additional rules:

• Ensure one experiment block directly supports the Problem Anchor.
• If complexity risk exists, include one simplification or deletion check.
• If a frontier primitive is central, include one necessity check showing why that choice matters.
• Default to 1-3 core experiment blocks and leave the full execution roadmap to /experiment-plan.

Step 1.6: Write the Initial Proposal

Save to refine-logs/round-0-initial-proposal.md.

Use this structure:

# Research Proposal: [Title]

## Problem Anchor
- Bottom-line problem:
- Must-solve bottleneck:
- Non-goals:
- Constraints:
- Success condition:

## Technical Gap
[Why current methods fail, why naive bigger systems are not enough, and what mechanism is missing]

## Method Thesis
- One-sentence thesis:
- Why this is the smallest adequate intervention:
- Why this route is timely in the foundation-model era:

## Contribution Focus
- Dominant contribution:
- Optional supporting contribution:
- Explicit non-contributions:

## Proposed Method
### Complexity Budget
- Frozen / reused backbone:
- New trainable components:
- Tempting additions intentionally not used:

### System Overview
[Step-by-step pipeline or ASCII graph]

### Core Mechanism
- Input / output:
- Architecture or policy:
- Training signal / loss:
- Why this is the main novelty:

### Optional Supporting Component
- Only include if truly necessary:
- Input / output:
- Training signal / loss:
- Why it does not create contribution sprawl:

### Modern Primitive Usage
- Which LLM / VLM / Diffusion / RL-era primitive is used:
- Exact role in the pipeline:
- Why it is more natural than an old-school alternative:

### Integration in

Phase 2: External Method Review (Round 1)

Send the full proposal to Gemini for an elegance-first, frontier-aware, method-first review. The reviewer should spend most of the critique budget on the method itself, not on expanding the experiment menu.

mcp__gemini-review__review_start:
  prompt: |
    You are a senior ML reviewer for a top venue (NeurIPS/ICML/ICLR).
    This is an early-stage, method-first research proposal.

    Your job is NOT to reward extra modules, contribution sprawl, or a giant benchmark checklist.
    Your job IS to stress-test whether the proposed method:
    (1) still solves the original anchored problem,
    (2) is concrete enough to implement,
    (3) presents a focused, elegant contribution,
    (4) uses foundation-model-era techniques appropriately when they are the natural fit.

    Review principles:
    - Prefer the smallest adequate mechanism over a larger system.
    - Penalize parallel contributions that make the paper feel unfocused.
    - If a modern LLM / VLM / Diffusion / RL route would clearly produce a better paper, say so concretely.
    - If the proposal is already modern enough, do NOT force trendy components.
    - Do not ask for extra experiments unless they are needed to prove the core claims.

    Read the Problem Anchor first. If your suggested fix would change the problem being solved,
    call that out explicitly as drift instead of treating it as a normal revision request.

  

After this start call, immediately save the returned jobId and poll mcpgemini-reviewreview_status with a bounded waitSeconds until done=true. Treat the completed status payload's response as the reviewer output, and save the completed threadId for any follow-up round.

CRITICAL: Save the returned jobId, poll mcpgemini-reviewreview_status until done=true, then save the completed threadId from the status result for all later rounds.

CRITICAL: Save the FULL raw response verbatim.

Save review to refine-logs/round-1-review.md with the raw response in a block.

Phase 3: Parse Feedback and Revise the Method

Step 3.1: Parse the Review

Extract: