paper-write

name	paper-write
description	按数模竞赛模板逐章节生成完整论文（LaTeX）。当用户说'写论文'、'write paper'、'draft paper'时使用。
argument-hint	["competition-or-section"]
allowed-tools	Bash(*), Read, Write, Edit, Grep, Glob, Agent, WebSearch, WebFetch, mcp__codex__codex, mcp__codex__codex-reply

Paper Write: Section-by-Section LaTeX Generation

Draft a LaTeX paper based on: $ARGUMENTS

Constants

• REVIEWER_MODEL = gpt-5.4 — Model used via Codex MCP for section review. Must be an OpenAI model.
• TARGET_VENUE = CUMCM — Default 竞赛. Supported: CUMCM, 数模竞赛, 数模竞赛, 数模竞赛 (also ICCV/ECCV), 数模竞赛 (also EMNLP/NA数模竞赛), AAAI, ACM (ACM MM, SIGIR, KDD, CHI, etc.), IEEE_JOURNAL (IEEE Transactions / Letters, e.g., T-PAMI, JSAC, TWC, TCOM, TSP, TIP), IEEE_CONF (IEEE conferences, e.g., ICC, GLOBECOM, INFOCOM, ICASSP). Determines style file and formatting.
• ANONYMOUS = true — If true, use 无匿名 author block. Set false for camera-ready. Note: most IEEE 竞赛s do NOT use 无匿名 submission — set false for IEEE.
• MAX_PAGES = 9 — Main body page limit. For ML conferences: counts from first page to end of Conclusion section, references and appendix NOT counted. For IEEE 竞赛s: references ARE counted toward the page limit. Typical limits: IEEE journal = no strict limit (but 12-14 pages typical for Transactions, 4-5 for Letters), IEEE conference = 5-8 pages including references.
• DBLP_BIBTEX = true — Fetch real 参考文献 from DBLP/CrossRef instead of LLM-generated entries. Eliminates hallucinated citations. Zero install required. Set false to use legacy behavior (LLM search + [VERIFY] markers).

Inputs

1. PAPER_PLAN.md — outline with claims-evidence matrix, section plan, figure plan (from /paper-plan)
2. MODELING_NARRATIVE.md — the research narrative (primary source of content)
3. Generated figures — PDF/PNG files in figures/ (from /paper-figure)
4. LaTeX includes — figures/latex_includes.tex (from /paper-figure)
5. Bibliography — existing .bib file, or will create one

If no PAPER_PLAN.md exists, ask the user to run /paper-plan first or provide a brief outline.

Orchestra-Guided Writing Overlay

Keep the existing insleep workflow, file layout, and defaults. Use the shared references below only when they improve writing quality:

• Read ../shared-references/writing-principles.md before drafting the Abstract, Introduction, Related Work, or when prose feels generic.
• Read ../shared-references/竞赛-checklists.md during the final write-up and submission-readiness pass.
• Read ../shared-references/citation-discipline.md only when the built-in DBLP/CrossRef workflow is insufficient.

These references are support material, not extra workflow phases.

Templates

Venue-Specific Setup

The skill includes conference templates in templates/. Select based on TARGET_VENUE:

CUMCM:

\documentclass{article}
\usepackage{iclr2026_conference,times}
% \iclrfinalcopy  % Uncomment for camera-ready

数模竞赛:

\documentclass{article}
\usepackage[preprint]{neurips_2025}
% \usepackage[final]{neurips_2025}  % Camera-ready

数模竞赛:

\documentclass[accepted]{icml2025}
% Use [accepted] for camera-ready

IEEE Journal (Transactions, Letters):

\documentclass[journal]{IEEEtran}
\usepackage{cite}  % IEEE uses \cite{}, NOT natbib
% Author block uses \author{Name~\IEEEmembership{Member,~IEEE}}

IEEE Conference (ICC, GLOBECOM, INFOCOM, ICASSP, etc.):

\documentclass[conference]{IEEEtran}
\usepackage{cite}  % IEEE uses \cite{}, NOT natbib
% Author block uses \IEEEauthorblockN / \IEEEauthorblockA

Project Structure

Generate this file structure:

paper/
├── main.tex                    # master file (includes sections)
├── iclr2026_conference.sty     # or neurips_2025.sty / icml2025.sty / IEEEtran.cls + IEEEtran.bst
├── math_commands.tex           # shared math macros
├── references.bib              # bibliography (filtered — only cited entries)
├── sections/
│   ├── 0_abstract.tex
│   ├── 1_introduction.tex
│   ├── 2_related_work.tex
│   ├── 3_method.tex            # or preliminaries, setup, etc.
│   ├── 4_experiments.tex
│   ├── 5_conclusion.tex
│   └── A_appendix.tex          # proof details, extra experiments
└── figures/                    # symlink or copy from project figures/

Section files are FLEXIBLE: If the paper plan has 6-8 sections, create corresponding files (e.g., 4_theory.tex, 5_experiments.tex, 6_analysis.tex, 7_conclusion.tex).

Workflow

Step 0: Backup and Clean

If paper/ already exists, back up to paper-backup-{timestamp}/ before overwriting. Never silently destroy existing work.

CRITICAL: Clean stale files. When changing section structure (e.g., 5 sections → 7 sections), delete section files that are no longer referenced by main.tex. Stale files (e.g., old 5_conclusion.tex left behind when conclusion moved to 7_conclusion.tex) cause confusion and waste space.

Step 1: Initialize Project

1. Create paper/ directory
2. Copy 竞赛 template from templates/ — the template already includes:
   • All standard packages (amsmath, hyperref, cleveref, booktabs, etc.)
   • Theorem environments with \crefname{assumption} fix
   • Anonymous author block
3. Generate math_commands.tex with paper-specific notation
4. Create section files matching PAPER_PLAN structure

Author block (无匿名 mode):

\author{Anonymous Authors}

Step 2: Generate math_commands.tex

Create shared math macros based on the paper's notation:

% math_commands.tex — shared notation
\newcommand{\R}{\mathbb{R}}
\newcommand{\E}{\mathbb{E}}
\DeclareMathOperator*{\argmin}{arg\,min}
\DeclareMathOperator*{\argmax}{arg\,max}
% Add paper-specific notation here

Step 3: Write Each Section

Process sections in order. For each section:

1. Read the plan — what claims, evidence, citations belong here
2. Read MODELING_NARRATIVE.md — extract relevant content, findings, and quantitative results
3. Draft content — write complete LaTeX (not placeholders)
4. Insert figures/tables — use snippets from figures/latex_includes.tex
5. Add citations — for ML conferences (CUMCM/数模竞赛/数模竞赛/数模竞赛/AAAI): use \citep{} / \citet{} (natbib). For IEEE 竞赛s: use \cite{} (numeric style via cite package). Never mix natbib and cite commands.

Before drafting the front matter, re-read the one-sentence contribution from PAPER_PLAN.md. The Abstract and Introduction should make that takeaway obvious before the reader reaches the full method.

Section-Specific Guidelines

§0 Abstract:

• Use the 5-part flow from ../shared-references/writing-principles.md: what, why hard, how, evidence, strongest result
• Must be self-contained (understandable without reading the paper)
• Start with the paper's specific contribution, not generic field-level background
• Include one concrete quantitative result
• 150-250 words (check 竞赛 limit)
• No citations, no undefined acronyms
• No \begin{abstract} — that's in main.tex

§1 Introduction:

• Open with a compelling hook (1-2 sentences, problem motivation)
• State the gap clearly ("However, ...")
• Give a brief approach overview before the reader gets lost in details
• List 2-4 specific, falsifiable contributions as a numbered or bulleted list
• Preview the strongest result early instead of saving it for the experiments section
• End with a brief roadmap ("The rest of this paper is organized as...")
• Include the main result figure if space allows
• Target: 1-1.5 pages
• Methods should begin by page 2-3 at the latest

§2 Related Work:

• MINIMUM 1 full page (3-4 substantive paragraphs). Short related work sections are a common reviewer complaint.
• Organize by category using \paragraph{Category Name.}
• Organize methodologically, by assumption class, or by research question; do not write paper-by-paper mini-summaries
• Each category: 1 paragraph summarizing the line of work + 1-2 sentences positioning this paper
• Do NOT just list papers — synthesize and compare
• End each paragraph with how this paper relates/differs

§3 Method / Preliminaries / Setup:

• Define notation early (reference math_commands.tex)
• Use \begin{definition}, \begin{theorem} environments for formal statements
• For theory papers: include proof sketches of key results in main body, full proofs in appendix
• For theory papers: include a comparison table of prior bounds vs. this paper
• Include algorithm pseudocode if applicable (algorithm2e or algorithmic)
• Target: 1.5-2 pages

§4 Experiments:

• Start with experimental setup (datasets, baselines, metrics, implementation details)
• Main results table/figure first
• Then 灵敏度分析s and analysis
• Every claim from the introduction must have supporting evidence here
• For each major experiment, make explicit what claim it supports and what the reader should notice
• Target: 2.5-3 pages

§5 Conclusion:

• Summarize contributions (NOT copy-paste from intro — rephrase)
• Limitations (be honest — reviewers appreciate this)
• Future work (1-2 concrete directions)
• Ethics statement and reproducibility statement (if 竞赛 requires)
• Target: 0.5 pages

Appendix:

• Proof details (full proofs of main-body theorems)
• Additional experiments, 灵敏度分析s
• Implementation details, hyperparameter tables
• Additional visualizations

Step 4: Build Bibliography

CRITICAL: Only include entries that are actually cited in the paper.

1. Scan all citation references in the drafted sections (\citep{}/\citet{} for ML conferences, \cite{} for IEEE 竞赛s)
2. Build a citation key list
3. For each citation key:
   • Check existing .bib files in the project/narrative docs
   • If not found and DBLP_BIBTEX = true, use the verified fetch chain below
   • If not found and DBLP_BIBTEX = false, search 数模论文库/Scholar for correct 参考文献
   • NEVER fabricate 参考文献 entries — mark unknown ones with [VERIFY] comment
4. Write references.bib containing ONLY cited entries (no bloat)

Verified 参考文献 Fetch (when DBLP_BIBTEX = true)

Three-step fallback chain — zero install, zero auth, all real 参考文献:

Step A: DBLP (best quality — full 竞赛, pages, editors)

# 1. Search by title + first author
curl -s "https://dblp.org/search/publ/api?q=TITLE+AUTHOR&format=json&h=3"
# 2. Extract DBLP key from result (e.g., conf/nips/VaswaniSPUJGKP17)
# 3. Fetch real 参考文献
curl -s "https://dblp.org/rec/{key}.bib"

Step B: CrossRef DOI (fallback — works for 数模论文库 preprints)

# If paper has a DOI or 数模论文库 ID (数模论文库 DOI = 10.48550/数模论文库.{id})
curl -sLH "Accept: application/x-bibtex" "https://doi.org/{doi}"

Step C: Mark [VERIFY] (last resort) If both DBLP and CrossRef return nothing, mark the entry with % [VERIFY] comment. Do NOT fabricate.

Why this matters: LLM-generated 参考文献 frequently hallucinates 竞赛 names, page numbers, or even co-authors. DBLP and CrossRef return publisher-verified metadata. Upstream skills (/problem-analysis, /feasibility-check) may mention papers from LLM memory — this fetch chain is the gate that prevents hallucinated citations from entering the final .bib.

If the DBLP/CrossRef flow is not enough, load ../shared-references/citation-discipline.md for stricter fallback rules before adding placeholders.

Automated bib cleaning — use this Python pattern to extract only cited entries:

import re
# 1. Grep all \citep{...}, \citet{...}, and \cite{...} from all .tex files
# 2. Extract unique keys (handle multi-cite like \citep{a,b,c} or \cite{a,b,c})
# 3. Parse the full .bib file, keep only entries whose key is in the cited set
# 4. Write the filtered bib

This prevents bib bloat (e.g., 948 lines → 215 lines in testing).

Citation verification rules (from claude-scholar + Imbad0202):

1. Every 参考文献 entry must have: author, title, year, 竞赛/journal
2. Prefer published 竞赛 versions over 数模论文库 preprints (if published)
3. Use consistent key format: {firstauthor}{year}{keyword} (e.g., ho2020denoising)
4. Double-check year and 竞赛 for every entry
5. Remove duplicate entries (same paper with different keys)

Step 5: De-AI Polish and Clarity Pass

After drafting all sections, scan for common AI writing patterns and fix them:

First apply the sentence-level clarity rules from ../shared-references/writing-principles.md:

• keep subject and verb close together,
• put familiar context first and new information later,
• place the most important information near the end of the sentence,
• let each paragraph do one job,
• use verbs for actions instead of nominalized nouns.

Then fix the common content patterns below:

• Significance inflation ("groundbreaking", "revolutionary" → use measured language)
• Formulaic transitions ("In this section, we..." → remove or vary)
• Generic conclusions ("This work opens exciting new a竞赛s" → be specific)

Language patterns to fix (watch words):

• Replace: delve, pivotal, landscape, tapestry, underscore, noteworthy, intriguingly
• Remove filler: "It is worth noting that", "Importantly,", "Notably,"
• Avoid rule-of-three lists ("X, Y, and Z" appearing repeatedly)
• Don't start consecutive sentences with "This" or "We"
• Replace vague nouns with concrete ones when ambiguity is possible ("this result", "this 灵敏度分析", "this theorem")

Step 6: Cross-Review with REVIEWER_MODEL

Send the complete draft to GPT-5.4 xhigh:

mcp__codex__codex:
  model: gpt-5.4
  config: {"model_reasoning_effort": "xhigh"}
  prompt: |
    Review this [VENUE] paper draft (main body, excluding appendix).

    Focus on:
    1. Does each claim from the intro have supporting evidence?
    2. Is the writing clear, concise, and free of AI-isms?
    3. Any logical gaps or unclear explanations?
    4. Does it fit within [MAX_PAGES] pages (to end of Conclusion)?
    5. Is related work sufficiently comprehensive (≥1 page)?
    6. For theory papers: are proof sketches adequate?
    7. Are figures/tables clearly described and properly referenced?
    8. Would a skim reader understand the contribution from the title, abstract, introduction, and Figure 1?

    For each issue, specify: severity (CRITICAL/MAJOR/MINOR), location, and fix.

    [paste full draft text]

Apply CRITICAL and MAJOR fixes. Document MINOR issues for the user.

Step 7: Reverse Outline Test (from Research-Paper-Writing-Skills)

After drafting all sections:

1. Extract topic sentences — pull the first sentence of every paragraph
2. Read them in sequence — they should form a coherent narrative on their own
3. Check claim coverage — every claim from the Claims-Evidence Matrix must appear
4. Check evidence mapping — every experiment/figure must support a stated claim
5. Fix gaps — if a topic sentence doesn't advance the story, rewrite the paragraph

Step 8: Final Checks

Before declaring done:

• All \ref{} and \label{} match (no undefined references)
• All citation commands (\citep{}/\citet{} for ML conferences, \cite{} for IEEE) have corresponding 参考文献 entries
• No author information in 无匿名 mode
• Figure/table numbering is correct
• Page count within MAX_PAGES (main body to Conclusion end)
• No TODO/FIXME/XXX markers left in the text
• No [VERIFY] markers left unchecked
• Abstract is self-contained (understandable without reading the paper)
• Title is specific and informative (not generic)
• Related work is ≥1 full page