research-paper-writing

これはレビュー時に保存された抜粋です。完全で最新の内容は外部ソースを確認してください。

---
name: research-paper-writing
description: >
  Draft and revise ML, CV, NLP, and systems research papers with strong claim-evidence
  flow, reviewer-aware structure, and submission-ready section planning. Use when the
  user needs help with an abstract, introduction, related work, method, experiments,
  ablations, discussion, figures/tables, checklist coverage, or rebuttal / response-letter
  writing for a paper. Triggers on: research paper, academic paper, ML paper, NeurIPS paper,
  ICLR paper, CVPR paper, experiments section, ablation plan, related work, contribution
  framing, reviewer response, rebuttal, camera-ready revision.
allowed-tools: Bash Read Write Edit Glob Grep WebFetch
license: MIT
metadata:
  tags: research-paper-writing, academic-writing, ml, cv, nlp, experiments, rebuttal
  version: "1.1"
  source: akillness/jeo-skills
---

# research-paper-writing

Use this skill when the job is not generic prose cleanup, but turning research work into a **defensible manuscript package**.

The center of gravity is:
- contribution framing,
- claim → evidence alignment,
- section-level structure,
- figure/table support,
- reviewer-facing completeness,
- rebuttal and camera-ready revision.

Read these support notes before drafting or revising:
- [references/claim-evidence-packet.md](references/claim-evidence-packet.md)
- [references/experiment-and-figure-coverage.md](references/experiment-and-figure-coverage.md)
- [references/rebuttal-and-camera-ready.md](references/rebuttal-and-camera-ready.md)

## When to use this skill

- Drafting or rewriting an abstract around concrete claims and measurable evidence
- Structuring an introduction so motivation, gap, method, and contributions land quickly
- Turning notes, code results, or slide bullets into a method section with reproducible detail
- Designing experiment, ablation, error-analysis, or figure/table coverage for a paper
- Writing related work that positions the paper instead of listing references
- Tightening a rebuttal, reviewer response, or camera-ready revision plan under strict limits
- Checking whether the manuscript actually supports each promised contribution

## When not to use this skill

- The request is an internal design doc, ADR, architecture note, or engineering runbook → use `technical-writing`
- The request is an API portal, SDK reference, or developer-facing product documentation problem → use `api-documentation`
- The request is an end-user tutorial, onboarding guide, or FAQ → use `technical-writing`
- The request is a slide deck, investor pitch, roadmap deck, or game pitch → use `presentation-builder`
- The request is broad marketing or launch messaging rather than academic publication → use `marketing-automation`
- The request is only citation collection / literature discovery without manuscript writing or revision pressure

## Instructions

### Step 1: Lock the paper contract

Before writing, define this packet:

```yaml
paper_contract:
  target_venue: "conference / journal / workshop"
  primary_problem: "one-sentence problem statement"
  core_claim: "the single strongest contribution"
  evidence_required:
    - benchmark or empirical proof
    - ablation or mechanism check
    - failure case / limitation / scope note
  audience: "reviewers / readers / subcommunity"
  constraints:
    page_or_word_limit: "..."
    anonymity_or_style_rules: "..."
    must_keep_sections:
      - abstract
      - intro
      - method
      - experiments
      - rebuttal / response letter
```

If the core claim or required evidence is fuzzy, fix that first. Weak framing leaks into every section.

### Step 2: Map each claim to proof

Build a quick claim-evidence table before drafting:

| Claim | Evidence needed | Supporting figure/table | Risk if missing |
|-------|-----------------|-------------------------|-----------------|
| ... | ... | ... | ... |

Do not let a headline contribution survive without an experiment, ablation, analysis, or explicit scope boundary.

### Step