多智能体学术稿件审阅 | Multi-Agent Academic Manuscript Review

一个 Claude Code 技能，模拟由6位专家组成的审稿团队，在投稿前对学术稿件进行全面评审。生成中英双语彩色批注 .docx 文件，含逐段评审、量化评分矩阵和改进建议优先级排序。

A Claude Code skill that simulates a panel of 6 expert reviewers to critique academic manuscripts before journal submission. Generates a bilingual (English-Chinese) annotated .docx file with color-coded, section-by-section reviews and a quantitative scoring matrix.

---

功能概述 | What It Does

输入稿件（docx、PDF 或粘贴文本），即可生成一份专业审稿文档：6位专业AI审稿人逐段点评，每位审稿人按1–10分制打分，评分标准根据目标期刊校准。

Give it a manuscript (docx, PDF, or pasted text), and it produces a professional review document where 6 specialized AI agents critique every section — from Title to Supplementary Materials — each scoring on a 1–10 scale calibrated to your target journal’s standards.

6位审稿专家 | The 6-Agent Review Panel

可选：Agent ML 针对定量密集型论文（高级统计/机器学习专家）。

Optional: Agent ML for quantitative-heavy papers (advanced statistics / machine learning specialist).

输出内容 | Output

每个稿件章节包含：

• 中英双语文本展示（英文原文 + 中文翻译）
• 各审稿人彩色标注评审面板
• 逐章评分及评分理由
• 最终评分矩阵表（审稿人 × 章节）
• 改进建议优先级列表（关键 / 重要 / 次要）

Each manuscript section gets:

• Bilingual text display (English original + Chinese translation)
• Color-coded review panels from each relevant agent
• Per-section scores with justification
• A final scoring matrix table (Agents × Sections)
• Prioritized improvement list (Critical / Major / Minor)

---

安装 | Installation

作为 Claude Code 技能安装 | As a Claude Code Skill

将本目录复制到 Claude Code 技能文件夹：

Copy this directory to your Claude Code skills folder:

cp -r . ~/.claude/skills/manuscript-review

或直接克隆 | Or clone directly:

git clone https://github.com/Shaowen-Ye/manuscript-review-skill.git ~/.claude/skills/manuscript-review

依赖 | Dependencies

# Node.js docx 包（用于生成 .docx 输出）
# Node.js docx package (for generating .docx output)
npm install docx

# Pandoc（可选，用于将 .docx 输入转为可读文本）
# Pandoc (optional, for converting .docx input to readable text)
brew install pandoc   # macOS
# or: apt install pandoc  # Linux

---

使用方法 | Usage

在 Claude Code 中直接说：

In Claude Code, simply say:

审稿 my_paper.docx

或 | or:

Please review my manuscript for PNAS submission

触发短语 | Trigger phrases（中英文均可）：

• “审稿”、“多Agent审稿”、“文稿审阅”、“投稿前审阅”、“逐段审稿”
• “manuscript review”, “agent team review”, “paper review panel”

配置选项 | Configuration Options

触发后，技能将请您确认：

When triggered, the skill will ask you to confirm:

1. 目标期刊 Target journal — PNAS, Nature, Science, Ecology Letters 等
2. 审稿深度 Review depth — 快速（每位审稿人2-3句）或深度（每位审稿人4-6句，默认）Quick (2-3 sentences per agent) or Deep (4-6 sentences, default)
3. 附加审稿人 Additional agents — 是否启用 Agent ML
4. 语言对 Language pair — 默认英中，可配置 Default English-Chinese, configurable
5. 输出文件路径 Output file path

---

评分校准 | Scoring Calibration

评分基于目标期刊**录用论文前10%**的标准校准：

Scores are calibrated to the top 10% of accepted papers at the target journal:

初稿通常应在5-7分之间。 首轮审稿几乎不会给出9分以上的评分。

Most first-draft manuscripts should score 5-7. A score of 9+ is virtually never given on first review.

---

审阅章节 | Sections Reviewed

审阅覆盖标准学术稿件的全部组成部分：

The review covers every component of a standard manuscript:

• 标题、重要性声明、摘要 Title, Significance Statement, Abstract
• 引言 Introduction
• 材料与方法 Materials and Methods
• 结果 Results
• 讨论 Discussion
• 图、表 Figures, Tables
• 附录方法、附图与附表 Supplementary Methods, Supplementary Figures & Tables

---

工作原理 | How It Works

1. 读取 Read — 提取稿件文本（通过 pandoc 转换 docx、PDF 或粘贴文本）Extracts manuscript text (from docx via pandoc, PDF, or pasted text)
2. 配置 Configure — 设置目标期刊、审稿深度、输出偏好 Sets target journal, review depth, output preferences
3. 生成 Generate — 使用脚手架模板创建 Node.js 脚本来构建批注 docx Creates a Node.js script using the scaffold template that builds the annotated docx
4. 输出 Output — 运行脚本生成最终双语审稿文档 Runs the script to produce the final bilingual review document

脚手架模板（template/scaffold.js）提供所有 docx 格式化基础设施——彩色审稿人面板、双语文本助手、图片嵌入、评分表构建器等。

The scaffold template (template/scaffold.js) provides all the docx formatting infrastructure — color-coded agent panels, bilingual text helpers, figure embedding, scoring table builder, etc.

---

文件结构 | File Structure

manuscript-review/
├── SKILL.md                    # 主技能定义 Main skill definition (Claude reads this)
├── references/
│   ├── agent-profiles.md       # 审稿人详细人设 Detailed agent personas
│   └── docx-js-api.md          # docx API 精简参考 Condensed docx API reference
├── template/
│   └── scaffold.js             # docx 生成模板 docx generation template
├── 中文说明.md                  # 中文参考文档 Chinese reference documentation
├── CHANGELOG.md                # 版本变更记录 Version history
├── LICENSE                     # MIT 许可证 MIT license
└── README.md                   # 本文件 This file

---

自定义 | Customization

添加自定义审稿人 | Adding Custom Agents

除默认6位外，您可以请求添加额外审稿人。Agent ML 已为定量型论文预设。自定义审稿人遵循相同模式：人设、审稿维度、必问问题和配色方案。

You can request additional agents beyond the default 6. Agent ML is pre-defined for quantitative papers. Custom agents follow the same pattern: a persona, review dimensions, must-ask questions, and a color scheme.

适应不同学科 | Adapting to Different Disciplines

本技能根据您的学科调整 Agent D（应用方向）和 Agent E（期刊标准）：

The skill adapts Agent D (application focus) and Agent E (journal standards) based on your discipline:

---

贡献 | Contributing

欢迎贡献！可改进的方向：

Contributions are welcome! Areas for improvement:

• 更多学科的审稿人配置 Additional agent profiles for specialized domains
• 支持更多输出语言（不限于英中） Support for more output languages beyond English-Chinese
• 增强图表分析能力 Enhanced figure/table analysis capabilities
• 集成参考文献管理工具 Integration with reference management tools
• 期刊特定格式模板 Journal-specific formatting templates

---

许可证 | License

MIT 许可证——详见 LICENSE。

MIT License — see LICENSE for details.