paper-check

Academic Paper Review Skill

A professional academic paper quality inspection tool that provides a systematic paper review workflow to help improve paper quality and publication success rate.

Use Cases

Comprehensive quality inspection before paper submission
Guidance for revising in response to reviewer comments
Staged checks during the paper writing process
Academic writing standards training

Core Features

Global Review: Four major dimensions: logical flow, terminology consistency, tense, and person
Section-by-Section Review: Seven sections: title, abstract, introduction, methods, results, discussion, and conclusion
Actionable Suggestions: Each issue is provided with a specific revision plan and example rewrite
Evidence Location: All judgments cite the original sentences and can be traced and verified

How to Use

Basic Usage

# 全局检查
codex e -m gemini-2.0-flash-exp "使用 paper-check skill 对我的论文进行全局检查" --attach paper.pdf

# 特定部分检查
codex e -m gemini-2.0-flash-exp "使用 paper-check skill 检查我的摘要部分" --attach paper.pdf

Recommended Workflow

After completing the first draft: First conduct a global review to locate structural problems
After revision: Refine section by section (by priority: Abstract → Introduction → Methods → Results → Discussion → Conclusion → Title)
Before submission: Conduct another round of global review to ensure consistency

Review Modules

0. Global Review

Before proceeding to section-by-section refinement, first conduct a global review, focusing on the four major dimensions of logic, terminology, tense, and person.

Core Checkpoints:

Logic: Transitions between paragraphs, coherence of the argument chain
Terminology: Consistency of key terms and abbreviations throughout
Tense: Use past tense for methods/results, present tense for introduction/discussion
Person: Whether the use of person complies with journal standards

Output Format:

Global diagnostic summary (Top 5 issues + revision roadmap)
Four-dimension issue list (logical flow/terminology/tense/person)
Terminology Glossary
Each issue includes: location/type/description/evidence/revision suggestion/example rewrite

1. Title Optimization

The title should satisfy accuracy, conciseness, and attractiveness at the same time.

Check Dimensions:

Accuracy: Whether it covers the core content (Task + Method or Task + Setting)
Conciseness: Whether there are redundant words, repeated structures, or uninformative adjectives
Attractiveness: Whether it contains 2-4 high-value search keywords

Output Format:

Fact Sheet (key facts extracted from the paper)
Title Scorecard (scores on the three dimensions of the current title + diagnosis)
Actionable Edits (list of actionable revision suggestions)
3 Candidate Titles (including strategy labels/keywords/risk notes)
Recommended Title (preferred title + rationale + journal fit statement)

2. Abstract QA

The abstract should be independent, complete, condensed text containing the four elements of background, methods, results, and conclusion.

Check dimensions:

Structural completeness: All four elements—Background/Methods/Results/Conclusion—are present
Data support: Key results include specific values/metrics/comparisons
Consistency: The abstract is consistent with the main text in research purpose, methods, results, and conclusions
Standalone clarity: Whether it can be understood without reading the main text (term definitions, background, baselines, etc.)

Output format:

Sentence-level annotation table (annotate each sentence as B/M/R/C)
Data support checklist (extract values/metrics, mark vague statements)
Consistency check (item-by-item comparison of abstract vs. main text)
Standalone clarity check (undefined terms/missing background, etc.)
Actionable revision suggestions + example of an improved abstract

3. Introduction Structure Audit

The introduction should present a clear "funnel-shaped" logical structure: gradually narrowing from the broader context to the specific research question.

Check dimensions:

Funnel structure: Broad context → subfield → specific problem → this paper's problem
Research gap: Clearly identify prior shortcomings/unresolved issues
Research question/hypothesis: Clear RQ or testable hypothesis
Roadmap: Explain the paper's structural organization

Output format:

Sentence numbering and segmentation (S1, S2, S3...)
Four check results (each includes: conclusion/evidence sentence/problem diagnosis/fix suggestions)
Top 5 issue list (ranked by severity)
Minimal-change fix plan
Structural rewrite plan (suggested paragraph order)

4. Methods Auditing

The core criterion for the methods section is "reproducibility."

Check dimensions:

Reproducibility: Data sources, sample size, preprocessing, model structure, training details, evaluation metrics, etc.
Rationale: Whether each key choice explains "why it was chosen"
Ethics statement: Ethics approval information for human subjects/animal experiments
Tense consistency: Use past tense in English, and keep Chinese narration consistent

Output format:

Conclusion overview (the 3 most serious issues + consequences)
Issue list table (category/severity/location/issue description/revision suggestion/recommended rewrite)
Four specialized checks (reproducibility/rationale/ethics statement/tense consistency)

5. Results QA

The results section should objectively present facts and avoid interpretation or inference of meaning.

Check dimensions:

Purity: Does not include explanation/causality/inference of meaning/value judgments
Figure and table citation coverage: All figures and tables are fully cited in the main text
Data description quality: Distill trends rather than recount numbers one by one

Output format:

Purity issue list (label/original sentence/issue type/suggested handling)
Figure and table citation check (cited list + missing items + formatting issues)
Data description improvements (original sentence/issue/suggested rewrite)

6. Discussion Depth Review

The discussion section should provide an in-depth interpretation of the meaning of the results: explain the reasons, compare with previous work, acknowledge limitations, and clarify the value.

Check Dimensions:

Depth of Explanation and Comparison: Whether each result explains the reasons and is compared with previous work
Limitations: Whether limitations in methodology, data, measurement, generalizability, etc. are adequately discussed
Value and Contribution: Whether the theoretical value and practical value are clear

Output Format:

Extraction of main results (3-8 items, with source text locations)
Explanation and comparison depth review table (mechanism explanation / comparison relationships / cited evidence for each result)
Limitations review (paragraph locations + list of limitations to be added + sentence templates)
Value and contribution review (evidence locations for theoretical value / practical value + concretized rewriting)
Quantitative scoring (explanation depth / comparison quality / limitations / value clarity, out of 10)
Minimal revision checklist Top 5

7. Conclusion Review

The conclusion should directly respond to the research questions raised in the introduction and summarize the main findings in one's own words.

Check Dimensions:

Response to Questions: Whether it answers RQ1/RQ2/... in the introduction
Summary Method: Whether it avoids copying the abstract
New Information Check: Whether it introduces new information that did not appear earlier
Future Outlook: Whether it proposes specific and actionable future directions based on the limitations

Output Format:

Overall conclusion (A/B/C rating + 3 most critical issues)
Item-by-item check (response to questions / summary method / new information / future outlook)
Example of an optimized concluding paragraph (100-180 words)

Output Specifications

Issue Location Format

All issues must include the following fields:

**位置**: 章节-段落编号 或 段落首句
**问题类型**: 逻辑/术语/时态/人称/可复现性/...
**问题描述**: 一句话说明问题
**证据**: 引用原句（可截取关键片段）
**修改建议**: 具体怎么改
**示例改写**: 给出1个更优表达（可选但优先提供）

Severity Grading

Critical: Affects the acceptability of the paper (ethical deficiencies, non-reproducibility, logical breaks)
Major: Affects the quality of the paper (confusing terminology, tense errors, missing key information)
Minor: Affects the reading experience (unclear expression, inconsistent formatting)

Table Output Format

Use Markdown tables to present structured information:

| # | 位置 | 问题类型 | 严重程度 | 问题描述 | 修改建议 |
|---|------|----------|----------|----------|----------|
| 1 | 引言-S3 | 逻辑跳跃 | Major | 从宏观背景直接跳到方法 | 补充子领域现状与具体问题 |

Constraints and Principles

Hard Constraints

Do not fabricate facts: All judgments must be based on the paper content provided by the user
Evidence must be traceable: Every issue must cite the original sentence
Suggestions must be actionable: Revision suggestions must be specific to "which fields/parameters/steps to add"
Do not change research facts: Only optimize expression and fill in missing information; do not alter research conclusions

Default Assumptions

When the user does not provide the following information, use the default assumptions:

Paper Language: Automatically determine based on the input (Chinese/English)
Target Journal: Follow common conventions for English scientific writing, with "consistency first"
Research Type: Infer from the text; if it cannot be determined, mark "Needs Verification"
Writing Preference: By default, allow first person "we" and prefer active voice

Handling Insufficient Information

When the information is insufficient to make a judgment:

Clearly mark "Needs Verification"
List the minimal set of information that the user needs to provide
Continue completing the check based on default assumptions (and state the assumptions at the beginning)

Best Practices

Suggested Check Order

Draft stage: Global check → Introduction → Methods → Results
Polishing stage: Abstract → Discussion → Conclusion → Title
Before submission: Global check (ensure consistency)

Tips for Collaborating with AI

1. Provide sufficient context (Context is King)

Clear context can significantly improve the quality of the check. Include the following information:

Target journal/conference name
Research type (experimental/theoretical/review)
Special requirements (word limit, formatting specifications)

# ✅ 好的做法
codex e -m gemini-2.0-flash-exp \
  "使用 paper-check skill 检查我的方法部分。
   目标期刊: Nature Communications
   研究类型: 人类受试实验
   已获伦理批准: IRB-2024-001
   字数限制: 方法部分不超过1500字" \
  --attach paper.pdf

# ❌ 不够好的做法
codex e -m gemini-2.0-flash-exp "检查方法" --attach paper.pdf

2. Process long papers in batches (Chunking Strategy)

For long papers (>10 pages), it is recommended to check them in batches by chapter to avoid context window overflow:

# 按章节逐个精修
codex e -m gemini-2.0-flash-exp \
  "使用 paper-check skill 检查引言部分（第1-3页）" \
  --attach paper.pdf

# 或使用 rg 定位特定章节后再检查
codex e -m gemini-2.0-flash-exp \
  "检查 Methods 章节，重点关注可复现性" \
  --attach paper.pdf

3. Iterative Refinement Workflow

Adopt a closed-loop process of "identify problems → revise → verify":

# 第一轮：全局诊断
codex e -m gemini-2.0-flash-exp \
  "使用 paper-check skill 进行全局检查，输出 Top 5 问题清单" \
  --attach paper_v1.pdf

# 修改后第二轮：针对性验证
codex e -m gemini-2.0-flash-exp \
  "重点检查术语一致性和时态问题，对比 v1 版本" \
  --attach paper_v2.pdf

# 投稿前最终检查
codex e -m gemini-2.0-flash-exp \
  "最终全局检查，确认所有 Critical 问题已解决" \
  --attach paper_final.pdf

4. Use Web Search to look up journal specifications

For unfamiliar journals, enable web search to obtain the latest submission guidelines:

codex e -m gemini-2.0-flash-exp \
  --enable web_search_request \
  "查询 IEEE Transactions on Pattern Analysis 的摘要字数要求，
   然后检查我的摘要是否符合规范" \
  --attach paper.pdf

5. Use Plan-First mode for complex revisions

For papers that require extensive revisions, first have the AI generate a revision plan:

# 第一步：生成修改计划
codex e -m gemini-2.0-flash-exp \
  "使用 paper-check skill 检查全文，生成分步修改计划（5-7步）" \
  --attach paper.pdf

# 第二步：逐步执行修改
codex e -m gemini-2.0-flash-exp \
  "执行修改计划的第1步：修复引言部分的逻辑跳跃问题" \
  --attach paper.pdf

6. Save check results for tracking

Save the check report as a Markdown file for easy version comparison:

codex e -m gemini-2.0-flash-exp \
  "使用 paper-check skill 全局检查，将报告保存为 review_report_v1.md" \
  --attach paper.pdf

7. Team collaboration: unify check standards

Define the team's agreed-upon check standards in the project CODEX.md:

## 论文检查标准
- 摘要必须包含具体数值（不接受"显著提升"等空泛表述）
- 方法部分必须包含超参数表格
- 讨论部分必须至少包含3条局限性
- 所有图表必须在正文中被引用

8. Avoid common pitfalls

❌ Do not check all dimensions at once (details are easily missed)
❌ Do not ignore "Needs Verification" flagged by the AI (these are usually key missing pieces of information)
❌ Do not blindly accept all suggestions (the AI may misjudge domain-specific terminology)
✅ Prioritize fixing Critical-level issues
✅ Manually review AI-suggested rewrites
✅ Keep the original version for rollback

Frequently Asked Questions

Q: There are too many check results. How should I prioritize them?

A: Follow this priority order:

Critical issues (ethics, reproducibility, logical breaks)
Structural issues (introduction funnel, four elements of the abstract, depth of discussion)
Consistency issues (terminology, tense, person)
Expression optimization (title, data descriptions)

Q: How should I handle cases where "Paper did not specify"?

A: The check tool will mark them as "Needs Verification" and suggest:

The minimal set of supplementary information
Common default values (e.g., Dropout=0.1)
Referring to practices in papers in the same field

Q: What are the differences between checking Chinese and English papers?

A: The main differences:

Tense checks: English strictly distinguishes past/present tense; Chinese checks consistency of time reference
Person checks: English checks we/our; Chinese checks mixed use of "author/this paper/this study"
Terminology checks: Checking Chinese-English terminology correspondence during additional checks

Q: How do I verify the effect of the modifications?

A: Suggested workflow:

Revise the paper according to the inspection report
Use the same module to check again
Compare the two inspection results to confirm the issues have been resolved
Finally, do a round of global checking to ensure the revisions did not introduce new issues

Technical details

Supported input formats

PDF (recommended, supports chart recognition)
Word/DOCX
Markdown
Plain text

Recommended models

Gemini 2.0 Flash Exp: Fast, suitable for first-draft checking
Gemini 2.0 Pro: In-depth analysis, suitable for the polishing stage
GPT-4 Turbo: Alternative option (requires Codex configuration)

Performance optimization

For long papers (>10 pages), it is recommended to check in batches
Use --enable web_search_request to look up journal submission guidelines
Use --dangerously-bypass-approvals-and-sandbox to speed up execution (local only)

Reference resources

License

MIT License - free to use, modify, and distribute

Contributing

Issues and Pull Requests to improve this Skill are welcome!

Note: This Skill provides writing suggestions and does not replace human peer review. Before submitting, be sure to carefully read the submission guidelines of the target journal.

About this skill

Use Cases

Core Features

How to Use

Basic Usage

Recommended Workflow

Review Modules

0. Global Review

1. Title Optimization

2. Abstract QA

3. Introduction Structure Audit

4. Methods Auditing

5. Results QA

6. Discussion Depth Review

7. Conclusion Review

Output Specifications

Issue Location Format

Severity Grading

Table Output Format

Constraints and Principles

Hard Constraints

Default Assumptions

Handling Insufficient Information

Best Practices

Suggested Check Order

Tips for Collaborating with AI

Frequently Asked Questions

Q: There are too many check results. How should I prioritize them?

Q: How should I handle cases where "Paper did not specify"?

Q: What are the differences between checking Chinese and English papers?

Q: How do I verify the effect of the modifications?

Technical details

Supported input formats

Recommended models

Performance optimization

Reference resources

License

Contributing