auto-paper-improvement-loop

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Auto Paper Improvement Loop: Review → Fix → Recompile

自动论文优化循环：评审→修改→重新编译

Autonomously improve the paper at: $ARGUMENTS

自主优化以下路径中的论文：$ARGUMENTS

Context

背景

This skill is designed to run after Workflow 3 (

/paper-plan

→

/paper-figure

→

/paper-write

→

/paper-compile

). It takes a compiled paper and iteratively improves it through external LLM review.

Unlike

/auto-review-loop

(which iterates on research — running experiments, collecting data, rewriting narrative), this skill iterates on paper writing quality — fixing theoretical inconsistencies, softening overclaims, adding missing content, and improving presentation.

本技能设计用于在工作流3（

/paper-plan

→

/paper-figure

→

/paper-write

→

/paper-compile

）之后运行。它接收已编译的论文，通过外部LLM评审来迭代优化论文。

与

/auto-review-loop

（针对研究迭代——开展实验、收集数据、重写叙事）不同，本技能针对论文写作质量迭代——修正理论不一致性、弱化过度表述、补充缺失内容、优化呈现形式。

Constants

常量配置

MAX_ROUNDS = 2 — Two rounds of review→fix→recompile. Empirically, Round 1 catches structural issues (4→6/10), Round 2 catches remaining presentation issues (6→7/10). Diminishing returns beyond 2 rounds for writing-only improvements.
REVIEWER_MODEL =
gpt-5.4
— Model used via Codex MCP for paper review.
REVIEW_LOG =
PAPER_IMPROVEMENT_LOG.md
— Cumulative log of all rounds, stored in paper directory.
HUMAN_CHECKPOINT = false — When
```
true
```
, pause after each round's review and present score + weaknesses to the user. The user can approve fixes, provide custom modification instructions, skip specific fixes, or stop early. When
```
false
```
(default), runs fully autonomously.

💡 Override:

/auto-paper-improvement-loop "paper/" — human checkpoint: true

MAX_ROUNDS = 2 — 执行2轮评审→修改→重新编译。根据经验，第1轮可解决结构性问题（评分从4/10提升至6/10），第2轮可解决剩余的呈现问题（评分从6/10提升至7/10）。仅针对写作优化的话，2轮之后收益会递减。
REVIEWER_MODEL =
gpt-5.4
— 用于论文评审的模型，通过Codex MCP调用。
REVIEW_LOG =
PAPER_IMPROVEMENT_LOG.md
— 所有轮次的累积日志，存储在论文目录中。
HUMAN_CHECKPOINT = false — 设为
```
true
```
时，每轮评审结束后暂停，向用户展示评分和问题点。用户可批准修改、提供自定义修改指令、跳过特定修改，或提前终止。默认值为
```
false
```
，全程自主运行。

💡 覆盖配置：

/auto-paper-improvement-loop "paper/" — human checkpoint: true

Inputs

输入

Compiled paper —
```
paper/main.pdf
```
+ LaTeX source files
All section
.tex
files — concatenated for review prompt

已编译的论文 —
```
paper/main.pdf
```
+ LaTeX源文件
所有章节的
.tex
文件 — 合并后用于评审提示词

State Persistence (Compact Recovery)

状态持久化（快速恢复）

If the context window fills up mid-loop, Claude Code auto-compacts. To recover, this skill writes

PAPER_IMPROVEMENT_STATE.json

after each round:

json

{
  "current_round": 1,
  "threadId": "019ce736-...",
  "last_score": 6,
  "status": "in_progress",
  "timestamp": "2026-03-13T21:00:00"
}

On startup: if

PAPER_IMPROVEMENT_STATE.json

exists with

"status": "in_progress"

AND

timestamp

is within 24 hours, read it +

PAPER_IMPROVEMENT_LOG.md

to recover context, then resume from the next round. Otherwise (file absent,

"status": "completed"

, or older than 24 hours), start fresh.

After each round: overwrite the state file. On completion: set

"status": "completed"

如果在循环过程中上下文窗口耗尽，Claude Code会自动压缩上下文。为了恢复，本技能会在每轮结束后写入

PAPER_IMPROVEMENT_STATE.json

文件：

json

{
  "current_round": 1,
  "threadId": "019ce736-...",
  "last_score": 6,
  "status": "in_progress",
  "timestamp": "2026-03-13T21:00:00"
}

启动时：如果

PAPER_IMPROVEMENT_STATE.json

存在且

"status": "in_progress"

，同时

timestamp

在24小时内，则读取该文件和

PAPER_IMPROVEMENT_LOG.md

恢复上下文，然后从下一轮继续。否则（文件不存在、

"status": "completed"

，或时间超过24小时），从头开始。

每轮结束后：覆盖状态文件。完成后：将

"status"

设为

"completed"

。

Workflow

工作流

Step 0: Preserve Original

步骤0：保存原始版本

bash

cp paper/main.pdf paper/main_round0_original.pdf

bash

cp paper/main.pdf paper/main_round0_original.pdf

Step 1: Collect Paper Text

步骤1：收集论文文本

Concatenate all section files into a single text block for the review prompt:

bash

undefined

将所有章节文件合并为单个文本块，用于评审提示词：

bash

undefined

Collect all sections in order

按顺序收集所有章节

for f in paper/sections/*.tex; do echo "% === $(basename $f) ===" cat "$f" done > /tmp/paper_full_text.txt

undefined

for f in paper/sections/*.tex; do echo "% === $(basename $f) ===" cat "$f" done > /tmp/paper_full_text.txt

undefined

Step 2: Round 1 Review

步骤2：第1轮评审

Send the full paper text to GPT-5.4 xhigh:

mcp__codex__codex:
  model: gpt-5.4
  config: {"model_reasoning_effort": "xhigh"}
  prompt: |
    You are reviewing a [VENUE] paper. Please provide a detailed, structured review.

    ## Full Paper Text:
    [paste concatenated sections]

    ## Review Instructions
    Please act as a senior ML reviewer ([VENUE] level). Provide:
    1. **Overall Score** (1-10, where 6 = weak accept, 7 = accept)
    2. **Summary** (2-3 sentences)
    3. **Strengths** (bullet list, ranked)
    4. **Weaknesses** (bullet list, ranked: CRITICAL > MAJOR > MINOR)
    5. **For each CRITICAL/MAJOR weakness**: A specific, actionable fix
    6. **Missing References** (if any)
    7. **Verdict**: Ready for submission? Yes / Almost / No

    Focus on: theoretical rigor, claims vs evidence alignment, writing clarity,
    self-containedness, notation consistency.

Save the threadId for Round 2.

将完整论文文本发送至GPT-5.4 xhigh：

mcp__codex__codex:
  model: gpt-5.4
  config: {"model_reasoning_effort": "xhigh"}
  prompt: |
    你正在评审一篇[VENUE]会议的论文。请提供详细、结构化的评审意见。

    ## 完整论文文本：
    [粘贴合并后的章节内容]

    ## 评审说明
    请以资深机器学习评审专家（[VENUE]会议级别）的身份进行评审。请提供：
    1. **总体评分**（1-10分，6分=弱接收，7分=接收）
    2. **摘要**（2-3句话）
    3. **优势**（分点列出，按重要性排序）
    4. **问题点**（分点列出，按严重程度排序：CRITICAL（严重）> MAJOR（主要）> MINOR（次要））
    5. **针对每个严重/主要问题点**：具体、可执行的修改建议
    6. **缺失的参考文献**（如果有）
    7. **结论**：是否可提交？是/接近/否

    重点关注：理论严谨性、表述与证据的一致性、写作清晰度、内容自洽性、符号一致性。

保存threadId用于第2轮评审。

Step 2b: Human Checkpoint (if enabled)

步骤2b：人工检查点（若启用）

Skip if
HUMAN_CHECKPOINT = false
.

Present the review results and wait for user input:

📋 Round 1 review complete.

Score: X/10 — [verdict]
Key weaknesses (by severity):
1. [CRITICAL] ...
2. [MAJOR] ...
3. [MINOR] ...

Reply "go" to implement all fixes, give custom instructions, "skip 2" to skip specific fixes, or "stop" to end.

Parse user response same as

/auto-review-loop

: approve / custom instructions / skip / stop.

如果
HUMAN_CHECKPOINT = false
则跳过。

展示评审结果并等待用户输入：

📋 第1轮评审完成。

评分：X/10 — [结论]
主要问题点（按严重程度）：
1. [严重] ...
2. [主要] ...
3. [次要] ...

回复“go”执行所有修改，提供自定义指令，“skip 2”跳过特定修改，或“stop”终止流程。

解析用户输入的逻辑与

/auto-review-loop

相同：批准/自定义指令/跳过/终止。

Step 3: Implement Round 1 Fixes

步骤3：实施第1轮修改

Parse the review and implement fixes by severity:

Priority order:

CRITICAL fixes (assumption mismatches, internal contradictions)
MAJOR fixes (overclaims, missing content, notation issues)
MINOR fixes (if time permits)

Common fix patterns:

Issue	Fix Pattern
Assumption-model mismatch	Rewrite assumption to match the model, add formal proposition bridging the gap
Overclaims	Soften language: "validate" → "demonstrate practical relevance", "comparable" → "qualitatively competitive"
Missing metrics	Add quantitative table with honest parameter counts and caveats
Theorem not self-contained	Add "Interpretation" paragraph listing all dependencies
Notation confusion	Rename conflicting symbols globally, add Notation paragraph
Missing references	Add to `references.bib` , cite in appropriate locations
Theory-practice gap	Explicitly frame theory as idealized; add synthetic validation subsection

解析评审意见，按严重程度实施修改：

优先级顺序：

严重问题的修改（假设与模型不匹配、内部矛盾）
主要问题的修改（过度表述、内容缺失、符号问题）
次要问题的修改（时间允许的情况下）

常见修改模式：

问题	修改模式
假设与模型不匹配	重写假设使其与模型匹配，添加正式命题填补差距
过度表述	弱化表述：将“validate”改为“demonstrate practical relevance”，“comparable”改为“qualitatively competitive”
缺失指标	添加包含真实参数计数和说明的量化表格
定理不自洽	添加“解释”段落，列出所有依赖项
符号混淆	全局重命名冲突符号，添加符号说明段落
缺失参考文献	添加至 `references.bib` ，在合适位置引用
理论与实践脱节	明确将理论表述为理想化情况；添加合成验证小节

Step 4: Recompile Round 1

步骤4：重新编译第1轮修改后的论文

bash

cd paper && latexmk -C && latexmk -pdf -interaction=nonstopmode -halt-on-error main.tex
cp main.pdf main_round1.pdf

Verify: 0 undefined references, 0 undefined citations.

bash

cd paper && latexmk -C && latexmk -pdf -interaction=nonstopmode -halt-on-error main.tex
cp main.pdf main_round1.pdf

验证：0个未定义引用，0个未定义文献。

Step 5: Round 2 Review

步骤5：第2轮评审

Use

mcp__codex__codex-reply

with the saved threadId:

mcp__codex__codex-reply:
  threadId: [saved from Round 1]
  model: gpt-5.4
  config: {"model_reasoning_effort": "xhigh"}
  prompt: |
    [Round 2 update]

    Since your last review, we have implemented:
    1. [Fix 1]: [description]
    2. [Fix 2]: [description]
    ...

    Please re-score and re-assess. Same format:
    Score, Summary, Strengths, Weaknesses, Actionable fixes, Verdict.

使用保存的threadId调用

mcp__codex__codex-reply

：

mcp__codex__codex-reply:
  threadId: [第1轮保存的ID]
  model: gpt-5.4
  config: {"model_reasoning_effort": "xhigh"}
  prompt: |
    [第2轮更新]

    自上次评审后，我们已完成以下修改：
    1. [修改1]：[描述]
    2. [修改2]：[描述]
    ...

    请重新评分和评估。格式要求与之前相同：评分、摘要、优势、问题点、可执行修改建议、结论。

Step 5b: Human Checkpoint (if enabled)

步骤5b：人工检查点（若启用）

Skip if
HUMAN_CHECKPOINT = false
. Same as Step 2b — present Round 2 review, wait for user input.

如果
HUMAN_CHECKPOINT = false
则跳过。与步骤2b相同——展示第2轮评审结果，等待用户输入。

Step 6: Implement Round 2 Fixes

步骤6：实施第2轮修改

Same process as Step 3. Typical Round 2 fixes:

Add controlled synthetic experiments validating theory
Further soften any remaining overclaims
Formalize informal arguments (e.g., truncation → formal proposition)
Strengthen limitations section

与步骤3流程相同。第2轮的典型修改包括：

添加控制合成实验以验证理论
进一步弱化剩余的过度表述
将非正式论证形式化（例如，将截断改为正式命题）
强化局限性章节

Step 7: Recompile Round 2

步骤7：重新编译第2轮修改后的论文

bash

cd paper && latexmk -C && latexmk -pdf -interaction=nonstopmode -halt-on-error main.tex
cp main.pdf main_round2.pdf

bash

cd paper && latexmk -C && latexmk -pdf -interaction=nonstopmode -halt-on-error main.tex
cp main.pdf main_round2.pdf

Step 8: Format Check

步骤8：格式检查

After the final recompilation, run a format compliance check:

bash

undefined

最终重新编译后，运行格式合规性检查：

bash

undefined

1. Page count vs venue limit

1. 页数与会议限制对比

PAGES=$(pdfinfo paper/main.pdf | grep Pages | awk '{print $2}') echo "Pages: $PAGES (limit: 9 main body for ICLR/NeurIPS)"

PAGES=$(pdfinfo paper/main.pdf | grep Pages | awk '{print $2}') echo "页数：$PAGES（ICLR/NeurIPS会议主文限制为9页）"

2. Overfull hbox warnings (content exceeding margins)

2. 内容超出边距警告（Overfull hbox）

OVERFULL=$(grep -c "Overfull" paper/main.log 2>/dev/null || echo 0) echo "Overfull hbox warnings: $OVERFULL" grep "Overfull" paper/main.log 2>/dev/null | head -10

OVERFULL=$(grep -c "Overfull" paper/main.log 2>/dev/null || echo 0) echo "内容超出边距警告数：$OVERFULL" grep "Overfull" paper/main.log 2>/dev/null | head -10

3. Underfull hbox warnings (loose spacing)

3. 间距过松警告（Underfull hbox）

UNDERFULL=$(grep -c "Underfull" paper/main.log 2>/dev/null || echo 0) echo "Underfull hbox warnings: $UNDERFULL"

UNDERFULL=$(grep -c "Underfull" paper/main.log 2>/dev/null || echo 0) echo "间距过松警告数：$UNDERFULL"

4. Bad boxes summary

4. 排版问题汇总

grep -c "badness" paper/main.log 2>/dev/null || echo "0 badness warnings"


**Auto-fix patterns:**

| Issue | Fix |
|-------|-----|
| Overfull hbox in equation | Wrap in `\resizebox` or split with `\split`/`aligned` |
| Overfull hbox in table | Reduce font (`\small`/`\footnotesize`) or use `\resizebox{\linewidth}{!}{...}` |
| Overfull hbox in text | Rephrase sentence or add `\allowbreak` / `\-` hints |
| Over page limit | Move content to appendix, compress tables, reduce figure sizes |
| Underfull hbox (loose) | Rephrase for better line filling or add `\looseness=-1` |

If any overfull hbox > 10pt is found, fix it and recompile before documenting.

grep -c "badness" paper/main.log 2>/dev/null || echo "0个badness警告"


**自动修正模式：**

| 问题 | 修正方法 |
|-------|-----|
| 公式中内容超出边距 | 使用`\resizebox`包裹，或用`\split`/`aligned`拆分 |
| 表格中内容超出边距 | 缩小字体（`\small`/`\footnotesize`）或使用`\resizebox{\linewidth}{!}{...}` |
| 正文中内容超出边距 | 改写句子或添加`\allowbreak`/`\-`换行提示 |
| 超过页数限制 | 将内容移至附录、压缩表格、缩小图片尺寸 |
| 间距过松 | 改写内容以优化行填充，或添加`\looseness=-1` |

如果发现任何超过10pt的内容超出边距问题，在记录前先修正并重新编译。

Step 9: Document Results

步骤9：记录结果

Create

PAPER_IMPROVEMENT_LOG.md

in the paper directory:

markdown

undefined

在论文目录中创建

PAPER_IMPROVEMENT_LOG.md

：

markdown

undefined

Paper Improvement Log

论文优化日志

Score Progression

评分变化

Round	Score	Verdict	Key Changes
Round 0 (original)	X/10	No/Almost/Yes	Baseline
Round 1	Y/10	No/Almost/Yes	[summary of fixes]
Round 2	Z/10	No/Almost/Yes	[summary of fixes]

轮次	评分	结论	主要修改
第0轮（原始）	X/10	否/接近/是	基线版本
第1轮	Y/10	否/接近/是	[修改摘要]
第2轮	Z/10	否/接近/是	[修改摘要]

Round 1 Review & Fixes

第1轮评审与修改

<details> <summary>GPT-5.4 xhigh Review (Round 1)</summary>

[Full raw review text, verbatim]

</details>

<details> <summary>GPT-5.4 xhigh评审意见（第1轮）</summary>

[完整原始评审文本，原文保留]

</details>

Fixes Implemented

已实施的修改

[Fix description]
[Fix description] ...

[修改描述]
[修改描述] ...

Round 2 Review & Fixes

第2轮评审与修改

<details> <summary>GPT-5.4 xhigh Review (Round 2)</summary>

[Full raw review text, verbatim]

</details>

<details> <summary>GPT-5.4 xhigh评审意见（第2轮）</summary>

[完整原始评审文本，原文保留]

</details>

Fixes Implemented

已实施的修改

[Fix description]
[Fix description] ...

[修改描述]
[修改描述] ...

PDFs

PDF版本

```
main_round0_original.pdf
```
— Original generated paper
```
main_round1.pdf
```
— After Round 1 fixes
```
main_round2.pdf
```
— Final version after Round 2 fixes

undefined

```
main_round0_original.pdf
```
— 原始生成论文
```
main_round1.pdf
```
— 第1轮修改后版本
```
main_round2.pdf
```
— 第2轮修改后版本（最终版）

undefined

Step 9: Summary

步骤9：总结

Report to user:

Score progression table
Number of CRITICAL/MAJOR/MINOR issues fixed per round
Final page count
Remaining issues (if any)

向用户汇报：

评分变化表
每轮解决的严重/主要/次要问题数量
最终页数
剩余问题（如果有）

Feishu Notification (if configured)

飞书通知（若已配置）

After each round's review AND at final completion, check

~/.claude/feishu.json

After each round: Send
```
review_scored
```
— "Round N: X/10 — [key changes]"
After final round: Send
```
pipeline_done
```
— score progression table + final page count
If config absent or mode
```
"off"
```
: skip entirely (no-op)

在每轮评审完成后以及最终完成时，检查

~/.claude/feishu.json

：

每轮评审后：发送
```
review_scored
```
消息——“第N轮：X/10分 — [主要修改]”
最终轮次后：发送
```
pipeline_done
```
消息——评分变化表 + 最终页数
如果配置文件不存在或模式为
```
"off"
```
：完全跳过（不执行任何操作）

Output

输出

paper/
├── main_round0_original.pdf    # Original
├── main_round1.pdf             # After Round 1
├── main_round2.pdf             # After Round 2 (final)
├── main.pdf                    # = main_round2.pdf
└── PAPER_IMPROVEMENT_LOG.md    # Full review log with scores

paper/
├── main_round0_original.pdf    # 原始版本
├── main_round1.pdf             # 第1轮修改后
├── main_round2.pdf             # 第2轮修改后（最终版）
├── main.pdf                    # = main_round2.pdf
└── PAPER_IMPROVEMENT_LOG.md    # 包含评分的完整评审日志

Key Rules

核心规则

Large file handling: If the Write tool fails due to file size, immediately retry using Bash (
```
cat << 'EOF' > file
```
) to write in chunks. Do NOT ask the user for permission — just do it silently.
Preserve all PDF versions — user needs to compare progression
Save FULL raw review text — do not summarize or truncate GPT-5.4 responses
Use
mcp__codex__codex-reply
for Round 2 to maintain conversation context
Always recompile after fixes — verify 0 errors before proceeding
Do not fabricate experimental results — synthetic validation must describe methodology, not invent numbers
Respect the paper's claims — soften overclaims rather than adding unsupported new claims
Global consistency — when renaming notation or softening claims, check ALL files (abstract, intro, method, experiments, theory sections, conclusion, tables, figure captions)

大文件处理：如果Write工具因文件大小失败，立即使用Bash命令（
```
cat << 'EOF' > file
```
）分块写入重试。无需向用户请求权限——直接静默执行。
保留所有PDF版本——用户需要对比优化过程
保存完整原始评审文本——不要总结或截断GPT-5.4的回复
第2轮使用
mcp__codex__codex-reply
——保持对话上下文
修改后始终重新编译——在继续前验证无错误
不要编造实验结果——合成验证必须描述方法，而非虚构数据
尊重论文原有表述——弱化过度表述而非添加无依据的新表述
全局一致性——重命名符号或弱化表述时，检查所有文件（摘要、引言、方法、实验、理论章节、结论、表格、图片说明）

Typical Score Progression

典型评分变化

Based on end-to-end testing on a 9-page ICLR 2026 theory paper:

Round	Score	Key Improvements
Round 0	4/10 (content)	Baseline: assumption-model mismatch, overclaims, notation issues
Round 1	6/10 (content)	Fixed assumptions, softened claims, added interpretation, renamed notation
Round 2	7/10 (content)	Added synthetic validation, formal truncation proposition, stronger limitations
Round 3	5→8.5/10 (format)	Removed hero fig, appendix, compressed conclusion, fixed overfull hbox

+4.5 points across 3 rounds (2 content + 1 format) is typical for a well-structured but rough first draft. Final: 8 pages main body, 0 overfull hbox, ICLR-compliant.

基于对一篇9页ICLR 2026理论论文的端到端测试：

轮次	评分	主要改进
第0轮	4/10（内容）	基线版本：假设与模型不匹配、过度表述、符号问题
第1轮	6/10（内容）	修正假设、弱化表述、添加解释、重命名符号
第2轮	7/10（内容）	添加合成验证、形式化截断命题、强化局限性章节
第3轮	5→8.5/10（格式）	移除主图、移至附录、压缩结论、修正内容超出边距问题

3轮共提升4.5分（2轮内容优化+1轮格式优化）是结构完整但初稿粗糙的论文的典型情况。最终版本：主文8页，0个内容超出边距问题，符合ICLR会议要求。