LaTeX Assistant for Chinese Academic Theses

Core Principles

Never modify content within
```
\cite{}
```
,
```
\ref{}
```
,
```
\label{}
```
, or math environments
Never fabricate bibliography entries out of thin air
Never modify professional terms without permission
Always output modification suggestions in the form of comments first
Chinese documents must be compiled with XeLaTeX or LuaLaTeX

Parameter Convention ($ARGUMENTS)

```
$ARGUMENTS
```
is used to receive critical information such as main file path, target chapter, module selection, etc.
If
```
$ARGUMENTS
```
is missing or ambiguous, first ask for: path to main
```
.tex
```
file, target scope, required module.
Process paths literally; do not infer or complete unprovided paths.

Execution Constraints

Only execute scripts/compile commands when explicitly requested by the user.
Confirm with the user before performing destructive operations such as cleaning (
```
--clean
```
/
```
--clean-all
```
).

Unified Output Protocol (All Modules)

Each suggestion must include fixed fields:

Severity: Critical / Major / Minor
Priority: P0 (Blocking) / P1 (Important) / P2 (Improvable)

Default Comment Template (diff-comment style):

latex

% <Module> (Line <N>) [Severity: <Critical|Major|Minor>] [Priority: <P0|P1|P2>]: <Issue Summary>
% Original: ...
% Modified: ...
% Reason: ...
% ⚠️  [To Be Verified]: <Mark when evidence/data is needed>

Failure Handling (Global)

When tools/scripts cannot be executed, output a comment block containing the cause and suggestions:

latex

% ERROR [Severity: Critical] [Priority: P0]: <Brief Error Description>
% Cause: <Missing script/tool or invalid path>
% Suggestion: <Install tool/verify path/retry command>

Common scenarios:

Script not found: Verify
```
scripts/
```
path and working directory
Compiler missing: Suggest installing TeX Live/MiKTeX and adding it to PATH
File not found: Ask the user to provide the correct
```
.tex
```
path
Compilation failed: Prioritize locating the first error and request a log snippet

Modules (Independent Call)

Except for "Structure Mapping" which requires execution first in full review or multi-file scenarios, all other modules can be called independently.

Module: Compilation

Trigger Words: compile, 编译, build, xelatex, lualatex

Default Behavior: Use

latexmk + XeLaTeX

to automatically handle all dependencies (bibtex/biber, cross-references, indexes, glossaries) and automatically determine the optimal number of compilations. This is the recommended solution for Chinese theses.

Tools (Aligned with VS Code LaTeX Workshop):

Tool	Command	Parameters
xelatex	`xelatex`	`-synctex=1 -interaction=nonstopmode -file-line-error`
lualatex	`lualatex`	`-synctex=1 -interaction=nonstopmode -file-line-error`
latexmk	`latexmk`	`-synctex=1 -interaction=nonstopmode -file-line-error -xelatex -outdir=%OUTDIR%`
bibtex	`bibtex`	`%DOCFILE%`
biber	`biber`	`%DOCFILE%`

Compilation Configurations:

Configuration	Steps	Applicable Scenario
latexmk	latexmk -xelatex (auto)	Default - Automatically handles all dependencies (recommended)
XeLaTeX	xelatex	Quick single compilation
LuaLaTeX	lualatex	Complex font requirements
xelatex -> bibtex -> xelatex×2	xelatex → bibtex → xelatex → xelatex	Traditional BibTeX workflow
xelatex -> biber -> xelatex×2	xelatex → biber → xelatex → xelatex	Modern biblatex (recommended for new theses)

Usage:

bash

# Default: latexmk + XeLaTeX automatically handles all dependencies (recommended)
python scripts/compile.py main.tex                          # Auto-detection + latexmk

# Single compilation (quick build)
python scripts/compile.py main.tex --recipe xelatex         # Single XeLaTeX compilation
python scripts/compile.py main.tex --recipe lualatex        # Single LuaLaTeX compilation

# Explicit bibliography workflow (for precise control)
python scripts/compile.py main.tex --recipe xelatex-bibtex  # Traditional BibTeX
python scripts/compile.py main.tex --recipe xelatex-biber   # Modern biblatex (recommended)

# Specify output directory
python scripts/compile.py main.tex --outdir build

# Auxiliary functions
python scripts/compile.py main.tex --watch                  # Watch mode
python scripts/compile.py main.tex --clean                  # Clean auxiliary files
python scripts/compile.py main.tex --clean-all              # Clean all (including PDF)

Auto-detection: The script automatically selects XeLaTeX when detecting ctex, xeCJK, or Chinese characters.

Module: Structure Mapping

Trigger Words: structure, 结构, 映射, map

Execute first in full review/multi-file scenarios: Analyze multi-file thesis structure

bash

python scripts/map_structure.py main.tex

Output Content:

File tree structure
Template type detection
Chapter processing order

Thesis Structure Requirements:

Section	Required Content
Preamble	Cover, Declaration, Abstracts (Chinese/English), Table of Contents, List of Symbols
Main Body	Introduction, Related Work, Core Chapters, Conclusion
Postamble	Bibliography, Acknowledgments, List of Published Papers

See STRUCTURE_GUIDE.md

Module: National Standard Format Checking

Trigger Words: format, 格式, 国标, GB/T, 7714

Check compliance with GB/T 7714-2015 standard:

bash

python scripts/check_format.py main.tex
python scripts/check_format.py main.tex --strict

Check Items:

Category	Specification
Bibliography	biblatex-gb7714-2015 format
Figure/Table Captions	Songti 5th size, captions below figures/above tables
Equation Numbering	Chapter-based numbering such as (3.1)
Title Styles	Font and size for each level of titles

See GB_STANDARD.md

Module: Academic Expression

Trigger Words: expression, 表达, 润色, 学术表达, 口语化

Colloquial → Academic Conversion:

❌ Colloquial	✅ Academic
很多研究表明	A large body of research indicates
效果很好	Demonstrates significant advantages
我们使用	This paper adopts
可以看出	It can be concluded that
比较好	Is relatively superior

Forbidden Subjective Terms:

❌ 显然, 毫无疑问, 众所周知, 不言而喻 (Obviously, Undoubtedly, As we all know, Self-evidently)
✅ 研究表明, 实验结果显示, 可以认为, 据此推断 (Research indicates, Experimental results show, It can be argued that, It can be inferred that)

Usage: The user provides paragraph source code, and the Agent analyzes it and returns the polished version with a comparison table.

Output Format (Markdown Comparison Table):

markdown

| Original | Improved Version | Issue Type | Optimization Reason |
|------|----------|----------|----------|
| 我们使用了ResNet模型。 | This paper adopts the ResNet model as a feature extractor. | Colloquial Expression | "我们使用" → "This paper adopts" (academic standard); added description of model purpose |
| 效果很好，可以看出性能提升明显。 | Experimental results show that this method has significant performance advantages. | Colloquial + Subjective Expression | Avoid colloquial terms like "很好" (very good) and "可以看出" (it can be seen); use "Experimental results show" to enhance objectivity |
| 显然，这种方法更优越。 | Experimental results show that this method outperforms the baseline method on multiple metrics. | Overly Subjective | Removed "显然" (obviously); supported conclusion with experimental results; clarified comparison object |

Alternative Format (In-source Comments):

latex

% ═══════════════════════════════════════════
% Modification Suggestion (Line 23) [Severity: Major] [Priority: P1]
% ═══════════════════════════════════════════
% Original: 我们使用了ResNet模型。
% Modified: This paper adopts the ResNet model as a feature extractor.
% Improvements:
% 1. "我们使用" → "This paper adopts" (academic standard)
% 2. Added description of model purpose
% Reason: Colloquial expression does not comply with academic standards
% ═══════════════════════════════════════════

See ACADEMIC_STYLE_ZH.md

Module: Logical Cohesion & Methodology Depth

Trigger Words: logic, coherence, 逻辑, 衔接, methodology, 方法论, 论证, argument

Objective: Ensure logical fluency between paragraphs and enhance the rigor of methodology.

Key Check Areas:

1. Paragraph-level Logical Cohesion (AXES Model):

Component	Description	Example
Assertion	Clear topic sentence stating the core viewpoint	"Attention mechanisms can improve sequence modeling performance."
Xample	Specific evidence or data supporting the assertion	"In experiments, the attention mechanism achieved 95% accuracy."
Explanation	Analyze why the evidence supports the assertion	"This improvement stems from its ability to capture long-range dependencies."
Significance	Connection to broader arguments or the next paragraph	"This finding provides a basis for the architecture design in this paper."

2. Transition Signals:

Relationship Type	Chinese Signals	English Equivalents
Progressive	此外, 进一步, 更重要的是	furthermore, moreover
Contrast	然而, 但是, 相反	however, nevertheless
Causal	因此, 由此可见, 故而	therefore, consequently
Sequential	首先, 随后, 最后	first, subsequently, finally
Illustrative	例如, 具体而言, 特别是	for instance, specifically

3. Methodology Depth Check List:

Each assertion is supported by evidence (data, citations, or logical reasoning)
Method selection is fully justified (why this method instead of others?)
Research limitations are explicitly acknowledged
Premise assumptions are clearly stated
Reproducibility details are sufficient (parameters, datasets, evaluation metrics)

4. Common Issues:

Issue Type	Manifestation	Correction Method
Logical Gap	Lack of cohesion between paragraphs	Add transition sentences to clarify paragraph relationships
Unsupported Assertion	Claims without evidence	Add citations, data, or reasoning
Shallow Methodology	"This paper adopts X" without justification	Explain why X is suitable for the problem
Implicit Assumptions	Premise conditions not explicitly stated	Explicitly state assumption conditions

Output Format:

latex

% Logical Cohesion (Line 45) [Severity: Major] [Priority: P1]: Logical gap between paragraphs
% Issue: Jumps directly from problem description to solution without transition
% Original: 数据存在噪声。本文提出一种滤波方法。
% Modified: The data contains noise, which interferes with subsequent analysis. Therefore, this paper proposes a filtering method to address this issue.
% Reason: Added causal transition to connect problem and solution

% Methodology Depth (Line 78) [Severity: Major] [Priority: P1]: Lack of justification for method selection
% Issue: No explanation for method selection
% Original: 本文采用ResNet作为骨干网络。
% Modified: This paper adopts ResNet as the backbone network, whose residual connection structure can effectively alleviate the gradient vanishing problem and performs excellently in feature extraction tasks.
% Reason: Justified architecture selection with technical principles

Chapter-specific Guidelines:

Chapter	Key Logical Cohesion Focus	Key Methodology Depth Focus
Introduction	Smooth cohesion of Problem → Gap → Contribution	Justify research significance
Related Work	Group by topic, explicit comparison	Position relationship with previous work
Methodology	Logical progression between steps	Justify each design choice
Experiments	Flow of Setup → Results → Analysis	Explain evaluation metric selection
Discussion	Cohesion of Findings → Implications → Limitations	Acknowledge research boundaries

Best Practices (Refer to Elsevier、Proof-Reading-Service):

One topic per paragraph: Each paragraph focuses on a single core viewpoint
Topic sentence first: State the paragraph's assertion at the beginning
Complete evidence chain: Each assertion requires support (data, citations, or logic)
Explicit transitions: Use signal words to indicate paragraph relationships
Argue, don't describe: Explain "why", not just "what"

Module: Long & Complex Sentence Analysis

Trigger Words: long sentence, 长句, 拆解, simplify

Trigger Condition: Sentence >60 characters or >3 clauses

Output Format:

latex

% Long & Complex Sentence Detected (Line 45, 87 characters total) [Severity: Minor] [Priority: P2]
% Main Clause: Our method achieves excellent performance on multiple datasets.
% Modifiers:
%   - [Attributive] Deep learning-based
%   - [Method] By introducing attention mechanisms
%   - [Condition] On the premise of ensuring real-time performance
% Suggested Rewrite:
%   This paper proposes a deep learning-based method. By introducing attention mechanisms,
%   this method achieves excellent performance on multiple datasets while ensuring real-time performance.

Module: Bibliography

Trigger Words: bib, bibliography, 参考文献, citation, 引用

bash

python scripts/verify_bib.py references.bib
python scripts/verify_bib.py references.bib --tex main.tex    # Check citations
python scripts/verify_bib.py references.bib --standard gb7714 # National standard check

Check Items:

Completeness of required fields
Duplicate entry detection
Unused entries
Missing citations
Compliance with GB/T 7714 format

Module: Template Detection

Trigger Words: template, 模板, thuthesis, pkuthss, ustcthesis, fduthesis

bash

python scripts/detect_template.py main.tex

Output includes template recognition results and key requirement summaries (from

references/UNIVERSITIES/

Supported Templates:

Template	University	Special Requirements
thuthesis	Tsinghua University	Figure/Table Numbering: Figure 3-1
pkuthss	Peking University	Requires symbol explanation chapter
ustcthesis	University of Science and Technology of China	-
fduthesis	Fudan University	-
ctexbook	General	Complies with GB/T 7713.1-2006

See UNIVERSITIES/

Module: De-AI Editing

Trigger Words: deai, 去AI化, 人性化, 降低AI痕迹, 自然化

Objective: Reduce AI writing traces while maintaining LaTeX syntax and technical accuracy.

Input Requirements:

Source Code Type (Required): LaTeX
Chapter (Required): Abstract / Introduction / Related Work / Methodology / Experiments / Results / Discussion / Conclusion / Other
Source Code Snippet (Required): Paste directly (retain original indentation and line breaks)

Usage Examples:

Interactive Editing (Recommended for single chapters):

python

python scripts/deai_check.py main.tex --section introduction
# Output: Interactive questions + AI trace analysis + rewritten source code

Batch Processing (For entire chapters or documents):

bash

python scripts/deai_batch.py main.tex --chapter chapter3/introduction.tex
python scripts/deai_batch.py main.tex --all-sections  # Process entire document

Workflow:

Syntax Structure Recognition: Detect LaTeX commands and fully retain:
- Commands:
```
\command{...}
```
  ,
```
\command[...]{}
```
- Citations:
```
\cite{}
```
  ,
```
\ref{}
```
  ,
```
\label{}
```
  ,
```
\eqref{}
```
  ,
```
\autoref{}
```
- Environments:
```
\begin{...\end{...}
```
- Math:
```
$...$
```
  ,
```
\[...\]
```
  , equation/align environments
- Custom macros (not modified by default)
AI Trace Detection:
- Empty slogans: "重要意义" (great significance), "显著提升" (significant improvement), "全面系统" (comprehensive and systematic), "有效解决" (effectively solve)
- Overly definitive: "显而易见" (obviously), "必然" (inevitably), "完全" (completely), "毫无疑问" (undoubtedly)
- Mechanical parallelism: Tripartite parallelism with no substantive content
- Template expressions: "近年来" (in recent years), "越来越多的" (more and more), "发挥重要作用" (play an important role)
Text Rewriting (Only modify visible text):
- Split long sentences (>50 characters)
- Adjust word order to conform to natural expression
- Replace vague statements with specific claims
- Delete redundant phrases
- Add necessary subjects (without introducing new facts)
Output Generation:
- A. Rewritten Source Code: Complete source code with minimal invasive modifications
- B. Change Summary: 3-10 key points explaining modification types
- C. To Be Verified Mark: Mark assertions that require evidence support

Hard Constraints:

Never modify:
```
\cite{}
```
,
```
\ref{}
```
,
```
\label{}
```
, math environments
Never add: Facts, data, conclusions, metrics, experimental settings, citation numbers, literature keys
Only modify: Plain paragraph text, Chinese expressions in chapter titles

Output Format:

latex

% ============================================================
% De-AI Editing (Line 23 - Introduction)
% ============================================================
% Original: 本文提出的方法取得了显著的性能提升。
% Modified: The method proposed in this paper shows performance improvement in experiments.
%
% Modification Notes:
% 1. Removed empty slogan: "显著" (significant) → deleted
% 2. Retained original claim, avoided adding specific metrics or comparison benchmarks
%
% ⚠️  [To Be Verified: Requires experimental data support, add specific metrics]
% ============================================================

\section{Introduction}
The method proposed in this paper shows performance improvement in experiments...

Chapter-specific Guidelines:

Chapter	Focus	Constraints
Abstract	Objective/Method/Key Results (with numbers)/Conclusion	Forbid vague contributions
Introduction	Importance → Gap → Contribution (verifiable)	Restrained wording
Related Work	Group by research line, specific differences	Specific comparisons
Methodology	Reproducibility first (process, parameters, metric definitions)	Implementation details
Results	Only report facts and numerical values	Do not explain causes
Discussion	Mechanisms, boundaries, failures, limitations	Critical analysis
Conclusion	Answer research questions, no new experiments	Executable future work

AI Trace Density Detection:

bash

python scripts/deai_check.py main.tex --analyze
# Output: AI trace density score for each chapter + suggestions for improvement

Refer to DEAI_GUIDE.md

Module: Title Optimization

Trigger Words: title, 标题, 标题优化, 生成标题, 改进标题

Objective: Generate and optimize thesis titles based on academic thesis standards and best practices.

Usage Examples:

Generate Title from Content:

bash

python scripts/optimize_title.py main.tex --generate
# Analyze abstract/introduction and propose 3-5 title candidates

Optimize Existing Title:

bash

python scripts/optimize_title.py main.tex --optimize
# Analyze current title and provide improvement suggestions

Check Title Quality:

bash

python scripts/optimize_title.py main.tex --check
# Evaluate title based on best practices (score 0-100)

Title Quality Standards (Based on GB/T 7713.1-2006 and international best practices):

Standard	Weight	Description
Conciseness	25%	Remove redundant phrases like "关于...的研究" (A study on...), "...的探索" (An exploration of...), "新型" (new type), "改进的" (improved)
Searchability	30%	Core terms (method + problem) appear in the first 20 characters
Length	15%	Optimal: 15-25 characters; Acceptable: 10-30 characters
Specificity	20%	Specific method/problem names, avoid vague statements
Standardization	10%	Comply with academic thesis title standards, avoid obscure abbreviations

Title Generation Workflow:

Step 1: Content Analysis Extract from abstract/introduction:

Research Problem: What challenge is being addressed?
Research Method: What method is proposed?
Application Field: What application scenario?
Core Contribution: What are the main outcomes? (Optional)

Step 2: Keyword Extraction Identify 3-5 core keywords:

Method Keywords: "Transformer", "Graph Neural Network", "Reinforcement Learning"
Problem Keywords: "Time Series Prediction", "Fault Detection", "Image Segmentation"
Field Keywords: "Industrial Control", "Medical Imaging", "Autonomous Driving"

Step 3: Title Template Selection Common patterns for academic theses:

Pattern	Example	Applicable Scenario
Research on Problem Based on Method	"Research on Time Series Prediction Method Based on Transformer"	Method innovation
Problem and Method in Field	"Graph Neural Network Method for Fault Detection in Industrial Systems"	Application-oriented
Method for Problem and Its Application	"Attention Mechanism for Time Series Prediction and Its Application in Industrial Control"	Theory + Application
Method Research for Field	"Deep Learning-based Predictive Maintenance Method for Smart Manufacturing"	Field-specific

Step 4: Generate Title Candidates Generate 3-5 title candidates with different focuses:

Method-focused
Problem-focused
Application-focused
Balanced (Recommended)
Concise variant

Step 5: Quality Scoring Each candidate title receives:

Overall score (0-100)
Sub-scores for each standard
Specific improvement suggestions

Title Optimization Rules:

❌ Remove Invalid Vocabulary:

Avoid Using	Reason
关于...的研究 (A study on...)	Redundant (all theses are research)
...的探索 (An exploration of...)	Redundant and non-specific
新型 / 新颖的 (New / Novel)	Publication implies novelty
改进的 / 优化的 (Improved / Optimized)	Non-specific; need to explain how it's improved
基于...的 (Based on...)	Can be simplified to direct expression

✅ Recommended Structure:

Good: Transformer Method for Time Series Prediction in Industrial Control Systems
Bad: A Study on Time Series Prediction in Industrial Control Systems Based on Transformer

Good: Graph Neural Network Fault Detection Method and Its Industrial Application
Bad: A Study on the New and Improved Graph Neural Network-based Fault Detection Method

Good: Attention Mechanism for Multivariate Time Series Prediction
Bad: A Study on the Improved Multivariate Time Series Prediction Model Based on Attention Mechanism

Keyword Layout Strategy:

First 20 characters: Most important keywords (method + problem)
Avoid starting with: "关于" (about), "对于" (for), "针对" (for) (can be placed in the middle)
Prefer using: Nouns and technical terms, rather than verbs and adjectives

Abbreviation Usage Guidelines:

✅ Acceptable	❌ Avoid in Titles
AI, Machine Learning, Deep Learning	Lab-specific abbreviations
LSTM, GRU, CNN	Chemical formulas (unless extremely common)
IoT, 5G, GPS	Non-standard method name abbreviations
DNA, RNA, MRI	Obscure field-specific abbreviations

University Template Special Requirements:

Tsinghua University (thuthesis):

Chinese title: No more than 36 Chinese characters
English title: Corresponding translation of Chinese title
Avoid abbreviations and formulas
Example: "Application of Deep Learning in Predictive Maintenance for Smart Manufacturing"

Peking University (pkuthss):

Chinese title: Concise, generally no more than 25 characters
Subtitles allowed (separated by em dash)
Example: "Graph Neural Network Fault Detection Method — Research for Industrial Control Systems"

General Requirements (ctexbook):

Comply with GB/T 7713.1-2006 standard
Chinese title: 15-25 characters is optimal
English title: Corresponding translation, pay attention to articles and prepositions
Example: "Transformer-Based Time Series Prediction Method and Application"

Output Format:

latex

% ============================================================
% Title Optimization Report
% ============================================================
% Current Title: "A Study on Time Series Prediction Based on Deep Learning"
% Quality Score: 48/100
%
% Detected Issues:
% 1. [Critical] Contains "关于...的研究" (A study on...) (remove redundant phrase)
% 2. [Important] Method description is too broad ("Deep Learning" is too general)
% 3. [Minor] Length is acceptable (18 characters) but can be more specific
%
% Recommended Titles (Sorted by Score):
%
% 1. "Transformer Method for Time Series Prediction in Industrial Control Systems" [Score: 94/100]
%    - Conciseness: ✅ (19 characters)
%    - Searchability: ✅ (Method + problem in first 15 characters)
%    - Specificity: ✅ (Transformer, instead of "Deep Learning")
%    - Field Relevance: ✅ (Industrial Control Systems)
%    - Standardization: ✅ (Complies with academic thesis standards)
%
% 2. "Attention Mechanism for Multivariate Time Series Prediction" [Score: 89/100]
%    - Conciseness: ✅ (17 characters)
%    - Searchability: ✅ (Core terms at the beginning)
%    - Specificity: ✅ (Attention Mechanism, Multivariate)
%    - Suggestion: Consider adding application field
%
% 3. "Deep Learning Time Series Prediction Method and Its Application in Smart Manufacturing" [Score: 81/100]
%    - Conciseness: ⚠️ (24 characters, acceptable)
%    - Searchability: ✅
%    - Specificity: ⚠️ ("Deep Learning" is still too general)
%    - Field Relevance: ✅ (Smart Manufacturing)
%
% Keyword Analysis:
% - Primary: Transformer, Time Series, Prediction
% - Secondary: Industrial Control, Attention, LSTM
% - Searchability: "Transformer Time Series" appears in 456 papers on CNKI (good balance)
%
% Suggested LaTeX Update:
% \title{Transformer Method for Time Series Prediction in Industrial Control Systems}
% \englishtitle{Transformer-Based Time Series Forecasting for Industrial Control Systems}
% ============================================================

Chinese-English Title Alignment:

When translating titles, note:

Chinese "基于X的Y" is usually translated as "X-Based Y" or "Y via X"
Avoid word-for-word translation; maintain English expression habits
Use Title Case for English titles (capitalize first letter of major words)

Chinese Title	English Title
工业系统故障检测的图神经网络方法	Graph Neural Networks for Fault Detection in Industrial Systems
基于注意力机制的时间序列预测研究	Attention-Based Time Series Forecasting
深度学习在智能制造中的应用	Deep Learning Applications in Smart Manufacturing

Interactive Mode (Recommended):

bash

python scripts/optimize_title.py main.tex --interactive
# Step-by-step guided title creation with user input

Batch Mode (Multiple Theses):

bash

python scripts/optimize_title.py chapters/*.tex --batch --output title_report.txt

Title Comparison Test (Optional):

bash

python scripts/optimize_title.py main.tex --compare "Title A" "Title B" "Title C"
# Compare multiple title candidates and provide detailed scoring

Best Practices Summary:

Keywords First: Place method + problem in the first 20 characters
Be Specific: "Transformer" > "Deep Learning" > "Machine Learning"
Remove Redundancy: Delete "关于" (about), "研究" (study), "新型" (new), "基于" (based on)
Control Length: Target 15-25 characters (Chinese)
Test Searchability: Can your thesis be found using these keywords?
Avoid Obscurity: Only use widely recognized terms (AI, LSTM, CNN)
Comply with Standards: Follow university templates and GB/T 7713.1-2006 standard

Refer to GB_STANDARD.md, UNIVERSITIES/

Complete Workflow (Optional)

For full review, execute in the following order:

Structure Mapping → Analyze thesis structure
National Standard Format Checking → Fix format issues
De-AI Editing → Reduce AI writing traces
Academic Expression → Improve expression
Long & Complex Sentence Analysis → Simplify complex sentences
Bibliography → Verify citations

Output Report Template

markdown

# LaTeX Academic Thesis Review Report

## Overview
- Overall Status: ✅ Compliant / ⚠️ Needs Revision / ❌ Major Issues
- Compilation Status: [status]
- Template Type: [detected template]

## Structural Completeness (X/10 Pass)
### ✅ Completed Items
### ⚠️ Items to Be Improved

## National Standard Format Review
### ✅ Compliant Items
### ❌ Non-Compliant Items

## Academic Expression (N Suggestions)
[Grouped by Priority]

## Long & Complex Sentence Analysis (M Items)
[Detailed Analysis]

Best Practices

This skill follows Claude Code Skills best practices:

Skill Design Principles

Single Responsibility: Each module handles one specific task (KISS principle)
Least Privilege: Only request necessary tool access permissions
Explicit Trigger: Use specific keywords to call modules
Structured Output: All suggestions use a unified diff-comment format

Usage Guide

Check Compilation First: Ensure the document can be compiled normally before other checks
Iterative Optimization: Apply only one module at a time to control modification scope
Protect Critical Elements: Never modify
```
\cite{}
```
,
```
\ref{}
```
,
```
\label{}
```
, or math environments
Verify Before Submission: Carefully review all suggestions before accepting modifications

Integration with Other Tools

Track modification history with version control (git)
Achieve real-time preview with LaTeX Workshop
Export suggestions for joint review with supervisors or collaborators

Reference Documents

STRUCTURE_GUIDE.md: Thesis structure requirements
GB_STANDARD.md: GB/T 7714 format specifications
ACADEMIC_STYLE_ZH.md: Chinese academic writing standards
FORBIDDEN_TERMS.md: Protected terms
COMPILATION.md: XeLaTeX/LuaLaTeX compilation guide
UNIVERSITIES/: University template guides
DEAI_GUIDE.md: De-AI writing guide and common patterns

latex-thesis-zh

NPX Install

Tags

SKILL.md Content (Chinese)

LaTeX Assistant for Chinese Academic Theses

Core Principles

Parameter Convention ($ARGUMENTS)

Execution Constraints

Unified Output Protocol (All Modules)

Failure Handling (Global)

Modules (Independent Call)

Module: Compilation

Module: Structure Mapping

Module: National Standard Format Checking

Module: Academic Expression

Module: Logical Cohesion & Methodology Depth

Module: Long & Complex Sentence Analysis

Module: Bibliography

Module: Template Detection

Module: De-AI Editing

Module: Title Optimization

Complete Workflow (Optional)

Output Report Template

Best Practices

Skill Design Principles

Usage Guide

Integration with Other Tools

Reference Documents