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

代号 Agent	角色 Role	审稿重点 Focus
A	架构军师 Architecture Strategist	结构、逻辑链、内部一致性、叙事节奏 Structure, logic chain, internal consistency, narrative flow
B	理论导师 Theory Mentor	创新性、文献空白、概念精确度、显著性 Novelty, literature gaps, conceptual precision, significance
C	方法统计审稿人 Methods & Stats Reviewer	统计严谨性、可重复性、效应量、ML可行性 Statistical rigor, reproducibility, effect sizes, ML feasibility
D	应用保护顾问 Application Advisor	实际相关性、管理建议、政策衔接 Practical relevance, management implications, policy connections
E	期刊主编 Journal Editor	期刊适配度、格式合规、影响力、竞争定位 Journal fit, formatting compliance, impact, competitive positioning
F	资深合作者 Senior Collaborator	战略定位、脆弱点评估、未来方向 Strategic positioning, vulnerability assessment, future directions

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: