lit-review

文献综述

v1.0 — 结构化文献综述工作流，支持多会话连续推进

管理研究项目的文献综述：搭建新综述、继续现有综述，或查看所有综述的状态。综述采用标准化模板，并支持多种输出格式（Markdown、Quarto、Beamer）。

参数： $ARGUMENTS

• setup = 首次运行配置
• 不带参数或使用 status = 显示所有综述的状态面板
• new [name] = 搭建新综述（例如，new remote-ed）
• [name] = 继续现有综述（例如，soft-skills）
• [name] referee = 对现有综述进行审稿

---

说明

步骤 0：读取配置

1. 检查 ~/.config/lit-review/config.yml 是否存在。

2. 如果配置不存在，且 $ARGUMENTS 不是 setup： 告知用户：“未找到配置。运行 /lit-review setup 即可开始。”然后停止。

3. 如果配置不存在，且 $ARGUMENTS 是 setup： 进入模式 S。

4. **如果配置存在：**读取配置并提取：

   • base_dir — 综述的存储位置
   • author — 幻灯片／文档标题中显示的姓名
   • voice_file — 可选的写作风格 .md 文件路径（null = 使用默认风格）
   • output_format — markdown、quarto 或 beamer

5. 如果已设置 voice_file 且文件存在，则读取该文件。否则使用默认风格：清晰、直接的学术文风。

6. 如果 [base_dir]/BEST_PRACTICES.md 存在，则读取其中的约定。

7. 根据 $ARGUMENTS 路由到相应模式。

---

模式 S：设置（setup）

S1. 询问配置问题

向用户询问以下 4 个问题（可以采用对话形式，也可以逐一询问）：

1. 综述应存储在哪里？（绝对路径，例如 ~/research/literature-reviews）
2. 幻灯片／文档标题中应显示什么姓名？（例如，Jane Smith）
3. 你有写作风格文件吗？（描述你偏好的学术写作风格的 .md 文件路径，或跳过）
4. 你偏好哪种输出格式？（markdown、quarto 或 beamer — 默认值：markdown）

S2. 创建配置文件

mkdir -p ~/.config/lit-review

写入 ~/.config/lit-review/config.yml：

# lit-review configuration
# Created [DATE]

base_dir: [user's answer]
author: [user's answer]
voice_file: [path or null]
output_format: [markdown|quarto|beamer]

S3. 创建基础目录和初始文件

mkdir -p [base_dir]

使用以下内容创建 [base_dir]/BEST_PRACTICES.md：

# Literature Reviews — Best Practices

*Lessons learned from building structured, multi-session literature reviews.*

---

## What works well

### 1. Standardized folder structure

Every review uses the same template. This means anyone (or any agent) can pick up any review cold and know exactly where to look:

```
[review-name]/
├── AGENT_INSTRUCTIONS.md    # What this review is about
├── RESEARCH_QUESTIONS.md    # What we're trying to learn
├── MASTER_LIST.md           # What we've found (single source of truth)
├── SEARCH_PROCESS.md        # How we searched (databases, terms, criteria)
├── [name]_lit.bib           # Machine-readable citations
├── literature_review.*      # The deliverable (.md, .qmd, or .tex)
├── CONTINUATION_LOG.md      # What's been done / what's next
└── pdfs/                    # Full-text papers
```

This consistency is the single biggest accelerator. A new review can be built from scratch in one session because the structure is predictable.

### 2. The "Roadmap to Paper" framing

Every review starts with a table mapping **paper sections -> literature streams -> key papers**. This keeps the entire review anchored to its purpose: supporting a specific paper. The "-> Paper" callouts throughout make it easy to see how each piece of literature connects to what you're writing.

### 3. Referee pass as a final step

After the initial build, doing a "referee pass" — asking "what would a reviewer at [target journal] expect to see?" — consistently catches 5-15 foundational papers that weren't in the initial search. These tend to be canonical theory papers or seminal empirical work that reviewers treat as table stakes.

### 4. MASTER_LIST as the single source of truth

Tracking every paper in one markdown table with columns for Author, Year, Title, Key finding, and In bib makes it easy to see gaps and avoid duplicates. The "Papers to obtain" section at the bottom is a useful running to-do list.

### 5. Continuation logs enable multi-session work

The dated log entries with "What was done / What's next / Blockers" format means context doesn't get lost between sessions. This is critical when reviews span multiple days or when different agents pick up the work.

### 6. Key Takeaways sections

Adding a "Key Takeaways" section at the end of each Part forces synthesis rather than just summarization. These are the most useful for actual paper-writing because they distill what the literature means for your paper.

---

## What could be improved

### 1. PDF collection is the weakest link

The `pdfs/` folders tend to be sparse. Many papers are behind paywalls. Without full-text access, deep dives are based on abstracts and prior knowledge rather than careful reading. This is the biggest quality gap.

### 2. Bib files need a clear convention

If a paper already has a `.bib` file, create a supplementary bib for new references only. If no paper `.bib` exists, create a standalone bib. Decide this at review creation time and document it in AGENT_INSTRUCTIONS.md.

### 3. Deep dives could go deeper

Reviews tend to summarize papers well but rarely go into methodology details, effect sizes with confidence intervals, or careful comparison of estimates across studies. Full-text PDFs help enormously here.

---

## Suggestions for future reviews

### Process improvements

1. **Start with source documents when possible.** Reviews with paper drafts are much more targeted than reviews built from verbal descriptions. Even a rough 2-page outline with key citations dramatically improves quality.

2. **Do the referee pass early, not just at the end.** Foundational papers should be in the first search round. Consider starting with: "What 10 papers would a reviewer at [target journal] definitely expect to see?"

3. **Consider a two-pass approach.** First pass: broad coverage. Second pass: deep dives on the 5-10 most important papers with methodology, effect sizes, and direct comparison to your study.

4. **Add a "Positioning" section.** Explicitly state: "Here is where our paper fits in this literature, and here is what is new." This is different from "contributions" — it's about the intellectual neighborhood.

5. **Standardize diagrams.** Mermaid diagrams for experimental design or conceptual frameworks are useful — make them a standard component.

6. **Add a "Comparison Table."** A table comparing your study to the 3-5 closest papers on key dimensions (sample size, setting, design, outcomes, findings) is almost always useful for positioning.

---

## What you (the human) can do to make this more effective

### Before a session

1. **Provide the paper draft and/or PAP.** This is the single highest-leverage input. Even an incomplete draft with a bibliography gives the agent a massive head start.
2. **Point to the `.bib` file explicitly.** And specify whether it's read-only or editable.
3. **Name the target journal.** This shapes the referee pass.
4. **Flag any papers you already know are missing.** This focuses the search rather than making the agent guess.
5. **Specify read/write permissions clearly.** This avoids the risk of accidentally modifying shared files.

### During a session

6. **Drop PDFs into the `pdfs/` folder.** If you have Zotero or institutional access, exporting key PDFs between sessions would dramatically improve deep-dive quality.
7. **React to the deliverable, not just the master list.** The output file is the deliverable; the master list is bookkeeping. "Part III is too thin" is more actionable than "find more papers."

### After a session

8. **Render the output locally.** Check formatting — tables and diagrams can break in ways that aren't visible in raw source. Report any rendering issues in the next session.
9. **Update the continuation log if you do manual work.** If you add papers, edit the review, or download PDFs between sessions, note it in the log so the next session doesn't duplicate effort.

### Structural suggestions

10. **Consider a shared Zotero export.** If you're running multiple reviews, a Zotero collection exported to BibTeX + PDF folder per review is the gold standard.
11. **A "seed papers" convention.** Before starting a new review, list the 5-10 papers you consider most important. This anchors search and ensures the review reflects your intellectual priorities.

---

## Quick-start checklist for a new review

```
[ ] Create folder: [base_dir]/[name]/
[ ] Create subfolder: [base_dir]/[name]/pdfs/
[ ] Provide source documents (paper draft, PAP, existing bib)
[ ] Specify: target journal, read/write permissions, key gaps to fill
[ ] Agent creates: AGENT_INSTRUCTIONS.md, RESEARCH_QUESTIONS.md, SEARCH_PROCESS.md
[ ] Agent searches and builds: MASTER_LIST.md, [name]_lit.bib
[ ] Agent builds: literature_review.* (output file in configured format)
[ ] Referee pass: add foundational papers
[ ] Agent updates: CONTINUATION_LOG.md, SEARCH_PROCESS.md
[ ] Human: render output, drop PDFs, review and iterate
```

---

*This is a living document. Update it as patterns emerge.*

创建 [base_dir]/README.md：

# Literature Reviews

| Review | Topic | Sources | Last Session | Status |
|--------|-------|---------|--------------|--------|

S4. 汇报

### Setup Complete

**Config**: ~/.config/lit-review/config.yml
**Reviews directory**: [base_dir]
**Output format**: [format]
**Files created**: BEST_PRACTICES.md, README.md

Ready to start. Run `/lit-review new [name]` to scaffold your first review.

---

模式 A：状态面板（不带参数或使用 status）

A1. 扫描所有综述文件夹

列出 [base_dir] 中的所有子目录。对于每个综述文件夹，读取：

• CONTINUATION_LOG.md — 最近一条记录（日期 + 下一步事项）
• MASTER_LIST.md — 主要表格中的来源数量
• AGENT_INSTRUCTIONS.md — 提取主题

A2. 显示状态面板

## LITERATURE REVIEW STATUS — [DATE]

| Review | Topic | Sources | Last Session | Status |
|--------|-------|---------|--------------|--------|
| [name] | [topic from AGENT_INSTRUCTIONS] | [count] | [date] | [inferred] |
| ... | | | | |

### What's Next (by review)
- **[name]**: [items from continuation log]
- ...

A3. 提示

Options:
  [name]       — continue a review (e.g., "soft-skills")
  new [name]   — scaffold a new review
  done         — exit

---

模式 B：继续现有综述（[name]）

B1. 读取综述上下文

并行读取 [name] 的以下文件：

1. [base_dir]/[name]/AGENT_INSTRUCTIONS.md — 范围、主题、来源文档
2. [base_dir]/[name]/CONTINUATION_LOG.md — 上次进展到的位置
3. [base_dir]/[name]/RESEARCH_QUESTIONS.md — 研究问题结构及论文映射
4. [base_dir]/[name]/MASTER_LIST.md — 当前来源清单

如果缺少任何文件，请指出并提出根据标准模板创建该文件。

B2. 展示续写摘要

显示：

## [REVIEW NAME] — Continuing

**Topic**: [from AGENT_INSTRUCTIONS]
**Last session**: [date from continuation log]
**Sources tracked**: [count from MASTER_LIST]

### Where we left off
[What's next items from most recent continuation log entry]

### Blockers from last session
[Any blockers noted]

然后询问：

What would you like to work on?
  1. Search for new papers (web search + forward/backward citation)
  2. Build or extend the output file (slides/document)
  3. Deep dive on specific papers (requires PDFs)
  4. Update MASTER_LIST and .bib
  5. Referee pass (check for foundational/canonical papers)
  6. Generate PDF download list (DOI links for papers missing PDFs)
  7. Something else — describe what you need

B3. 执行工作

遵循用户的指示。针对每种工作类型：

搜索新论文：

• 使用网络搜索查找符合综述研究问题的论文
• 对于每篇找到的论文：提取作者、年份、标题、主要发现、RQ 对应关系
• 对照 MASTER_LIST 检查，避免重复
• 跨综述去重：同时扫描 [base_dir] 下所有其他综述文件夹中的 MASTER_LIST.md。如果某篇论文已出现在其他综述中，请注明：“也见于：[other-review]（RQ X.Y）”。这有助于呈现共同的学术研究领域，并避免重复搜索。
• 将新条目添加到 MASTER_LIST.md
• 将 BibTeX 条目添加到 [name]_lit.bib
• 使用搜索词、所用数据库和日期更新 SEARCH_PROCESS.md
• 报告：“找到 N 篇新论文。已添加到 MASTER_LIST 和 .bib。[M 篇也出现在其他综述中。]”

构建／扩展输出文件：

• 遵循 BEST_PRACTICES.md 中的标准结构：
  • 路线图（文献脉络 -> 论文各节、关键论文表格）
  • 第 I–VI 部分：主题模块
  • 关键论文深度分析（方法、效应量、与论文的联系）
  • 每部分末尾提供关键要点
  • 对比表（标准组成部分——见下方模板）
  • 定位部分：“我们的论文定位于何处；创新之处是什么”
  • 第 VII 部分：研究空白、贡献、定位
• 使用 -> Paper 标注与论文写作相关的提示
• 使用 Mermaid 图呈现概念框架
• 如果配置了语气风格文件，则遵循该文件；否则使用清晰、直接的学术文风

对比表模板（每篇综述都要包含，通常放在第 VII 部分或深度分析之后）：

## How Our Study Compares

| Dimension | [Our study] | [Closest paper 1] | [Closest paper 2] | [Closest paper 3] |
|-----------|------------|-------------------|-------------------|-------------------|
| **Setting** | | | | |
| **Sample** | | | | |
| **Design** | | | | |
| **Treatment/Signal** | | | | |
| **Key outcomes** | | | | |
| **Main finding** | | | | |
| **What's different** | *Our contribution* | | | |

> **-> Paper**: This table maps directly to the "Contribution" section. Each row in "What's different" is a talking point.

论文深度分析：

• 检查 pdfs/ 文件夹中是否有可用的全文 PDF
• 对每篇论文说明：方法、样本、含效应量的主要结果、局限性
• 将研究发现对应到研究问题
• 说明与母论文设计／框架的联系

更新 MASTER_LIST 和 .bib：

• 核对输出文件中是否有未列入 MASTER_LIST 的论文
• 确保 MASTER_LIST 中的所有条目都标注了 RQ 对应关系
• 检查 .bib 是否完整（DOI、期刊名称、页码）

审稿人检查：

• 参见下方的模式 D。

生成 PDF 下载清单：

• 扫描 MASTER_LIST.md，找出 PDF? 列为空、“No”或缺失的所有论文
• 对于每篇缺少 PDF 的论文：
  • 如果 .bib 文件中有 DOI，则查找该 DOI
  • 构造 Google Scholar 搜索 URL：https://scholar.google.com/scholar?q=[author]+[year]+[title keywords]
  • 构造 DOI 链接：https://doi.org/[DOI]（如有 DOI）
• 在 [base_dir]/[name]/PDF_DOWNLOAD_LIST.md 创建清单文件：

# PDF Download List — [name]
*Generated [DATE]. [N] papers missing PDFs out of [M] total.*

Download PDFs and save to `pdfs/` using naming convention: `Author-Year-ShortTitle.pdf`

## Priority (key papers for deep dives)
- [ ] **Author Year** — Title
  - DOI: [https://doi.org/10.xxxx](https://doi.org/10.xxxx)
  - Scholar: [search](https://scholar.google.com/scholar?q=...)
  - Save as: `Author-Year-ShortTitle.pdf`

## Other papers
- [ ] **Author Year** — Title
  - Scholar: [search](...)
  - Save as: `Author-Year-ShortTitle.pdf`

• 报告：“已生成下载清单：N 篇论文需要 PDF。其中 K 篇有 DOI 链接，J 篇仅有 Scholar 链接。文件已保存到 PDF_DOWNLOAD_LIST.md。”

B4. 更新续接日志

每次会话结束时必须执行。 在 CONTINUATION_LOG.md 中添加一条带日期的记录：

### [DATE] — [Brief description of session]

**What was done**
- [Specific accomplishments — be precise]

**What's next**
- [ ] [Actionable items]

**Blockers**
- [Any issues, or "None."]

B5. 会话完成报告

### Session Complete

**Review**: [name]
**Papers added this session**: N
**Output sections added/modified**: N

Next steps: [top 2-3 items from continuation log]

如果输出格式为 quarto，请添加：“运行 quarto render literature_review.qmd 检查格式。” 如果输出格式为 beamer，请添加：“运行 pdflatex literature_review.tex 和 bibtex literature_review 检查格式。”

---

模式 C：新建综述（new [name]）

C1. 收集信息

询问用户：

1. 这项工作支持什么研究项目？（论文标题、项目名称）
2. 是否有源文件？（论文草稿、PAP、现有 .bib 文件——请提供路径）
3. 目标期刊是什么？（这会影响审稿人检查）
4. 主要文献脉络有哪些？（3–8 个主题）
5. 是否已知具体的研究空白？（文献综述应重点关注的主题）

C2. 创建文件夹结构

mkdir -p [base_dir]/[name]/pdfs

C3. 搭建文件框架

根据标准模板创建所有文件。前 6 个文件适用于所有输出格式：

AGENT_INSTRUCTIONS.md — 填写以下内容：

• 主题、范围、受众
• 上级项目和论文参考
• 源文档路径（论文草稿、PAP、.bib）
• 目标期刊
• 内容主题（根据用户输入）
• PDF 命名约定：Author-Year-ShortTitle.pdf
• 会话开始前／结束后的检查清单

RESEARCH_QUESTIONS.md — 填写以下内容：

• 5-8 个主题组，每组 3-5 个问题
• 映射表：论文要素 -> 文献重点
• 依据：论文草稿（如有）、用户指定的主题、目标期刊的要求
• 如果没有论文草稿：根据主题创建占位研究问题，并标记为“草稿 — 请根据源文档完善”

MASTER_LIST.md — 填写以下内容：

• 种子论文部分（如有，从论文草稿/.bib 中提取）
• 留空以下部分：网络搜索结果、待获取的论文
• 列标题：作者、年份、标题、主要发现、研究问题、是否在 bib 中、是否有 PDF？

[name]_lit.bib — 填写以下内容：

• 种子论文的 BibTeX 条目（如果提供了源 .bib）
• 否则：创建带有标题注释的空文件

SEARCH_PROCESS.md — 搭建框架：

# Search Process — [name]

*Documentation of search methodology for reproducibility.*

## Databases Searched
- [ ] Google Scholar
- [ ] NBER Working Papers
- [ ] EconLit / JSTOR
- [ ] SSRN
- [ ] Web of Science
- [ ] Other: ___

## Search Terms

| Search round | Date | Terms | Database | Results reviewed | Papers added |
|-------------|------|-------|----------|-----------------|--------------|
| 1 | [DATE] | [terms] | [database] | [N] | [M] |

## Inclusion Criteria
- [Define based on scope from AGENT_INSTRUCTIONS.md]

## Exclusion Criteria
- [Define based on scope]

## Forward/Backward Citation Searches
| Seed paper | Direction | Papers found |
|-----------|-----------|-------------|

## Notes
- [Any deviations from search plan, surprises, or methodological notes]

CONTINUATION_LOG.md — 首条记录：

### [DATE] — Initial scaffold

**What was done**
- Created review folder and all template files (including SEARCH_PROCESS.md)
- [If source docs available: Extracted N seed papers from paper draft/.bib]
- [If source docs available: Drafted research questions based on paper structure]

**What's next**
- [ ] Review and approve research questions (RESEARCH_QUESTIONS.md)
- [ ] Web search for papers across all RQ themes
- [ ] Build MASTER_LIST with initial sources
- [ ] Begin output file (Part I)

**Blockers**
- [Any missing source documents or unclear scope]

特定格式的输出文件 — 取决于配置中的 output_format 设置：

如果是 markdown：

# Literature Review: [Title]

*Author: [author from config]*
*Last updated: [DATE]*

## Roadmap

| Paper Section | Literature Stream | Key Papers |
|--------------|-------------------|------------|
| [placeholder] | [placeholder] | [placeholder] |

## Part I: [First literature stream]

[Placeholder slides/sections with Key Takeaways]

## Part II: [Second literature stream]

[Placeholder]

...

## How Our Study Compares

[Comparison table placeholder]

## Positioning

*Where our paper fits in this literature and what is new.*

## Gaps and Contributions

[Placeholder]

## References

如果是 quarto：

---
title: "Literature Review: [Title]"
subtitle: "[Subtitle based on themes]"
author: "[author from config]"
date: today
format:
  revealjs:
    theme: default
    slide-number: true
    preview-links: auto
bibliography: [name]_lit.bib
---

然后搭建框架，包含：路线图幻灯片、每个文献流各自的部分（带有“关键要点”占位幻灯片）、对比表幻灯片、定位幻灯片、第 VII 部分：研究缺口与贡献、参考文献。

如果是 beamer：

\documentclass[aspectratio=169]{beamer}
\usetheme{metropolis}
\usepackage{natbib}
\usepackage{booktabs}
\usepackage{graphicx}

\title{Literature Review: [Title]}
\subtitle{[Subtitle based on themes]}
\author{[author from config]}
\date{\today}

\begin{document}

\maketitle

\section{Roadmap}
% Roadmap frame: lit streams -> paper sections

\section{[First literature stream]}
% Placeholder frames with Key Takeaways

\section{[Second literature stream]}
% Placeholder frames

\section{Comparison}
% How Our Study Compares table

\section{Positioning}
% Where our paper fits

\section{Gaps and Contributions}
% Placeholder

\begin{frame}[allowframebreaks]{References}
\bibliographystyle{apalike}
\bibliography{[name]_lit}
\end{frame}

\end{document}

C4. 阅读源文档（如果提供）

如果用户提供了论文草稿：

• 阅读草稿（遵守文件权限——可以读取 .tex、.md、.pdf；不要打开 .dta、.csv、.sav）
• 提取：研究问题、引用的论文、主要主题、方法论、贡献主张
• 使用提取的信息填写 RESEARCH_QUESTIONS.md，并为 MASTER_LIST 填入种子论文

如果用户提供了现有的 .bib：

• 阅读该文件
• 提取条目，填入 MASTER_LIST 的种子论文部分
• 如果该文件为只读：在 AGENT_INSTRUCTIONS.md 中注明需要补充的 bib 文件

C5. 更新 README

将新的综述添加到 [base_dir]/README.md 的活动综述表中。

C6. 汇报

### New Review Scaffolded: [name]

**Files created**: [N] (standard template + [format] output)
**Seed papers**: N (from source documents)
**Research questions**: N themes, M questions

**Review before proceeding:**
- [ ] RESEARCH_QUESTIONS.md — approve or refine RQs
- [ ] AGENT_INSTRUCTIONS.md — check scope and themes

Ready to start searching? Run `/lit-review [name]` to begin.

---

模式 D：审稿人检查（[name] referee）

D1. 阅读综述背景

与 B1 相同——阅读所有综述文件。

D2. 确认目标期刊

从 AGENT_INSTRUCTIONS.md 中获取。如果未指定，请询问用户。

D3. 检查基础性论文

询问：“[目标期刊] 的审稿人肯定会希望看到引用哪些论文？”

在每个文献流中搜索经典论文。常见类别：

• 基础理论：该领域的开创性理论模型
• 开创性实证研究：该领域具有里程碑意义的研究
• 关键方法论研究：与研究设计相关的参考文献（例如，论文中使用的方法创新）
• 近期高影响力研究：过去 3 年顶级期刊上有关同一主题的论文

D4. 与 MASTER_LIST 交叉核对（并跨综述去重）

对于确定的每篇基础性论文：

• 检查它是否已在当前综述的 MASTER_LIST 中
• 跨综述检查：同时扫描 [base_dir] 下所有其他综述文件夹中的 MASTER_LIST.md。如果其他综述中已有该论文，请注明：“也在：[other-review]”——这有助于识别共通的基础性文献，避免重复深入研究。
• 如果当前综述中缺少该论文：将其添加到 MASTER_LIST，并映射到相应的研究问题
• 将 BibTeX 条目添加到 .bib
• 在“审稿人检查新增内容”部分中标记

D5. 汇报

### Referee Pass Complete: [name]

**Target journal**: [journal]
**Papers checked**: N foundational/canonical papers
**Already in review**: M
**Added**: K new papers

**New additions:**
- [Author Year] — [why a reviewer would expect this]
- ...

**Potential gaps remaining:**
- [Any areas where coverage is thin]

使用审稿人检查结果更新 CONTINUATION_LOG.md。

---

错误处理

• 未找到配置：“在 ~/.config/lit-review/config.yml 中未找到配置。运行 /lit-review setup 以创建配置。”
• 未找到评审文件夹：“在 [base_dir]/[name]/ 中未找到评审。你指的是以下评审之一吗：[list existing reviews]？或者运行 /lit-review new [name] 来创建评审。”
• 现有评审缺少文件：“[name] 评审缺少 [file]。要我根据标准模板创建该文件吗？”
• 新建评审时没有源文档：使用占位研究问题（RQ）和空的 MASTER_LIST 继续。在 CONTINUATION_LOG 中注明需要源文档以便更好地确定目标。
• 无法获取用于深入分析的 PDF：“pdfs/ 中没有全文。将基于摘要和已有知识进行分析。如需更深入的分析，请将 PDF 放入 [base_dir]/[name]/pdfs/[Author-Year-ShortTitle].pdf。”

---

约定

• 文件命名：使用小写字母和连字符，不使用空格
• PDF 命名：Author-Year-ShortTitle.pdf
• 写作风格：如果已配置风格文件，则遵循该文件；否则使用清晰、直接的学术文风
• 引用：在 .bib 文件中使用 BibTeX；正文中的作者和年份格式保持一致；有 DOI 时添加链接
• Bib 约定：如果论文自带只读 .bib 文件，则为新增引用单独创建补充 bib 文件
• 续接日志：每次会话后都必须更新——日期、已完成事项、后续安排、阻碍
• MASTER_LIST 作为 SSOT：在此跟踪所有论文；不得重复；为所有条目映射研究问题（RQ）
• 搜索记录：每次搜索会话后，都要在 SEARCH_PROCESS.md 中更新搜索词、数据库和日期
• 跨评审去重：添加论文时，检查其他评审中的 MASTER_LIST 文件，并标记重复文献