Skip to content
Projects

stephenlzc/humanize-mba-text-skill

View on GitHub

stephenlzc/humanize-mba-text-skill has 62 GitHub stars and 8 forks. GitHot measured 0 stars in 24 hours. Last checked 2026-11-15 18:38 UTC. Last push 2026-07-08.

How these counts are measured

Measured star growth

in 24h
0
in 7 days
Update delayed
in 15 days
Update delayed
in 1 month
Update delayed

Net star changes over overlapping periods, not daily totals. Missing measurements are left blank. Stars measure attention, not software quality.

About

去除中文MBA论文AI写作痕迹的检测与改写工具 | AI writing pattern detection and removal tool for Chinese MBA thesis

AI

README_EN.md

Original

Humanize MBA Text - Remove AI Writing Traces

Python 3.7+ License: MIT Claude Skill Version: 1.5 Kimi CLI 中文 English 한국어 日本語

AI Writing Trace Detection and Removal Tool Specifically for Chinese MBA Theses

#AI-Detection #Academic-Writing #MBA-Thesis #Claude-Skill #Text-Humanization #ChatGPT-Alternative #LLM-Writing #Research-Tools #Academic-Integrity #Chinese-NLP


🎯 Project Introduction

This is an AI writing trace detection and removal tool specifically designed for Chinese MBA theses. Based on the academic norms and practical requirements of MBA theses, it uses multi-dimensional detection methods to identify AI-generated features in text and provides specific modification suggestions to help you rewrite AI-generated text into natural, humanized academic writing style.

✨ Version 1.5 New Features (Paragraph-Level Sentence CV + Report Readability)

  • 📐 Paragraph-level sentence-length CV: Adapts the AI_artifact_detection method, using the coefficient of variation (CV) of sentence length within a paragraph as the core AI signal. Chinese is measured in CJK characters, English in words
  • 📊 Chinese CV report: Detection Markdown reports now automatically append a five-section CV report — Overall Statistics / CV Distribution / Uniform Paragraph Details / Section Summary / Sentence-Length Histogram
  • 🔁 Before/After comparison: New scripts/detect_compare.py CLI takes before/after texts and outputs a Chinese Before/After comparison Markdown
  • 🌐 Optional English detection mode: sentence_length.analyze(text, language="en") supports sentence-length CV detection for English prose
  • 🎯 Content snippets with context: The detailed issue list now shows ~20 characters before and after each match for quick location
  • 📝 Full modification suggestions: Suggestions are no longer truncated; all guidance is displayed in full
  • 📋 Uniform paragraphs as full table: All paragraphs with CV < 0.30 are listed in a table, no longer capped at 20
  • ⏰ Real generation timestamp: Report footer now shows the actual generation time (YYYY-MM-DD HH:MM:SS)

Older versions (1.4 / 1.3 / 1.2) are archived in FEATURE.md.

Core Features

  • ✅ Layered AI Detection: Regex rules → prose-structure statistics (5 dimensions) → semantic-chain patterns (10 dimensions)
  • ✅ Chapter-specific Rules: Optimization strategies for 5 chapters: Introduction, Theory, Analysis, Solutions, and Conclusion
  • ✅ MBA Thesis Standards: Complies with Chinese university MBA thesis word count, structure, and format requirements
  • ✅ Structured Rewrite Plan: Every issue carries location + skeleton + recommended replacements + target word count
  • ✅ Auto-fix: Automatically handles simple issues like Chinese-English mixed spacing
  • ✅ Smart Feedback: Generates detailed modification suggestions and before/after comparison examples
  • ✅ Claude Skill Integration: Can be used directly as a Claude Code Skill

Technical Reference

This project's optimization strategy design references the three-dimensional collaborative optimization concept from the thesis-optimizer project:

  • 🔍 Dimension 1: AI Detection Rate Reduction

    • Sentence Structure Diversification
    • Academic Tone Naturalization
    • Logical Chain Humanization
  • 📉 Dimension 2: Plagiarism Rate Reduction

    • Deep Semantic Rewriting
    • Citation Standardization
    • Professional Terminology Processing
  • ✨ Dimension 3: Academic Polishing Enhancement

    • Expression Precision
    • Academic Standardization
    • Readability Optimization

🎓 MBA Thesis Core Principles

This tool is designed based on the academic norms of Chinese MBA theses and follows the following core principles:

1. Practice Orientation

  • Must originate from actual enterprise management and solve specific management problems
  • Avoid pure theoretical discussions

2. Focused Deep Research ("Small Topic, Deep Analysis")

  • Clear and focused topic selection, "small topic with deep analysis"
  • Core concepts should not exceed 2-3
  • Avoid overly broad or vague topics

3. Data Traceability

  • All data must indicate sources
  • Ensure accuracy and credibility
  • Remove vague expressions like "relevant data shows"

4. Theoretical Support

  • Apply 1-2 relevant theories as analytical frameworks
  • Avoid discussing issues in isolation without theoretical grounding

5. Structural Standards

  • Main text word count ≥ 30,000 words
  • Each chapter must have at least 4 sections (including chapter summary)
  • Each section must be substantial, avoid sections shorter than one page

6. Academic Integrity

  • Plagiarism rate < 15%
  • Do not fabricate data; all citations must be verifiable

🚀 Quick Start

Installation

# Clone the repository
git clone https://github.com/stephenlzc/humanize-mba-text-skill.git
cd humanize-mba-text-skill

# Install dependencies (if using transformers models)
pip install transformers torch

Basic Usage

1. Basic Detection

# Use basic rule detection
python scripts/detect_ai_patterns.py your_text.txt --format markdown --output report.md

2. Multi-dimensional Fusion Detection (Recommended)

# Use multi-dimensional detection
python scripts/multi_detector.py your_text.txt --format markdown --output report.md --plan plan.json

3. Generate Modification Suggestions

# Generate feedback based on detection results
python scripts/feedback_generator.py detection_result.json --text your_text.txt --output feedback.md

4. Apply Auto-fixes

# Auto-fix simple issues like spacing
python scripts/feedback_generator.py detection_result.json --text your_text.txt --apply

Using as Claude Skill

  1. Clone this repository to the Claude Code skills directory:
cd ~/.config/opencode/skills
git clone https://github.com/stephenlzc/humanize-mba-text-skill.git
  1. Trigger the Skill in Claude Code:
Remove AI traces: [paste your text]

Or

Help me remove AI writing traces from this text

📊 Detection Dimensions

1. Rule Matching Detection

Identifies the following AI writing characteristics:

  • AI Vocabulary: 赋能 (empower), 抓手 (handle), 闭环 (closed loop), 痛点 (pain point), 赛道 (track), etc.
  • Vague Attribution: "Studies have shown", "Experts believe", etc.
  • Overemphasis: "Key", "Core", "Crucial", etc.
  • Macro Narratives: "Era", "Trend", "Wave", etc.
  • Surface Analysis: "Highlights", "Reflects", etc. -ing ending analyses
  • Cliché Conclusions: "In summary", "From this we can see", etc.
  • Excessive Connectives: "First... Second... Finally", etc.
  • Chinese-English Mixed Spacing: "MBA 论文" → "MBA论文"

2. Statistical Analysis Detection

  • Sentence Length Uniformity: AI text typically has more uniform sentence lengths
  • Vocabulary Diversity: AI text has lower vocabulary diversity
  • Punctuation Distribution: Analyzes punctuation usage patterns
  • Paragraph Structure: Detects paragraph length distribution

3. Linguistic Feature Detection

  • Connective Density: Statistics on logical connective usage frequency
  • Formal Expression Patterns: Identifies overly formal academic expressions
  • Sentence Complexity: Analyzes complex sentence structure usage

4. Prose-Structure Analyzers (v1.3)

Each analyzer returns an AnalyzerIssue (analyzer_id / severity / confidence / location / evidence / suggestion) plus statistical metrics:

Dim analyzer_id Trigger
6 uniform_sentence_length Sentence-length variance/CV exceeds threshold (paragraph + global)
9 uniform_paragraph_length Inter-paragraph CJK length CV < 0.25
10 paragraph_edge_template_repeat 3+ consecutive paragraphs share identical leading/trailing fingerprint
8a paragraph_structure_uniformity 3+ consecutive paragraphs share the same 4-tuple structure fingerprint
8b chapter_template_repeat 3+ sections inside one chapter reuse the same chapter-template

5. Semantic-Chain Analyzers (v1.3)

Cross-paragraph and cross-chapter AI patterns that no per-paragraph rule can catch:

Dim analyzer_id Trigger
3 chain_three_part_rule 3+ consecutive paragraphs all use 「一是…二是…三是」 or 「首先…其次…最后」
3 chain_author_listing 4+ 「Author(year) 指出/认为」 references stacked inside one chapter
3 chain_method_name One paragraph piles 2+ method/model/theory names without per-method elaboration
3 chain_abstract_template One paragraph hits 3+ abstract-template phrases
3 chain_conclusion_echo Conclusion chapter's opening duplicates the abstract's character Jaccard ≥ 0.30
4 chain_vague_problem_statement 2+ vague problem statements with no metric within 30 chars
4 chain_unsupported_quantification 2+ percentage/ranking claims with no 「根据/来源/问卷/N=」 anchor within 80 chars
4 chain_macro_narrative 3+ macro-narrative phrases inside a 1000-char window
5 evidence_chain_completeness 2+ quantified/research claims lacking method/source/N= anchors (cross-rule)
5 cross_section_problem_trace Chapter-3 problems ↔ chapter-5 recommendations keyword overlap < 30%

6. Structured Rewrite Plan (v1.3)

detect_ai_patterns.AIPatternDetector.generate_report() and multi_detector.FusionEngine.detect() now expose a modify_plan key — one ModifyEntry per issue:

{
  "analyzer_id": "chain_unsupported_quantification",
  "severity": "high",
  "location": "global",
  "evidence": "7 percentage/ranking assertions, 5 lack 「根据/来源/问卷/N=」 anchors within 80 chars",
  "suggestion": "Anchor every quantitative claim with method/source/N marker",
  "rewrite_template": "For each quantified claim, supply sample, time window, and statistical scope.",
  "recommended_replacements": [
    "Per 'per the Dec 2023 customer survey (N=120), …'",
    "Sample description: '5-point Likert scale'",
    "When unverifiable: 'this metric needs further verification'",
  ],
  "target_word_count_range": [60, 140],
  "before_after_example": {
    "before": "客户满意度提升20%。",
    "after": "根据2023年12月客户问卷(N=120),客户满意度从3.8分升至4.5分。"
  }
}

7. Per-sentence High-Risk Annotations (v1.4)

report["high_risk_annotations"] is a new third array — one entry per sentence — that bridges the gap between matches[] (hundreds of regex hits, each with only a generic suggestion string) and modify_plan[] (4 aggregated rows with no line numbers):

{
  "sentence_index": 17,
  "char_offset_start": 1287,
  "char_offset_end": 1335,
  "sentence_text": "数字化转型赋能业务创新,形成从生产到服务的完整闭环。",
  "line_number": 24,
  "severity": "high",
  "triggered_rules": [
    {
      "rule_id": "ai_buzzwords",
      "pattern_name": "AI词汇/互联网黑话",
      "pattern_type": "regex",
      "evidence": "赋能",
      "confidence": 1.0,
      "severity": "high",
      "phrase_replacements": [
        "赋能 → 促进/支持",
        "闭环 → 闭合回路/完整流程",
        "抓手 → 切入点/措施"
      ],
      "before_after_example": {
        "before": "数字化转型赋能业务创新,形成从生产到服务的完整闭环。",
        "after": "数字化改造使生产到售后的流程数据打通,订单处理时长从5天缩短至2天。"
      }
    }
  ],
  "rewrite_template": "把互联网黑话替换为传统管理学术语。",
  "recommended_replacements": [
    "赋能 → 促进/支持",
    "闭环 → 闭合回路/完整流程"
  ]
}

Field reference:

Field Meaning
sentence_text Original sentence as-is
char_offset_start/end Char offsets in the original text (inclusive of terminator)
line_number 1-based line in the original text
severity Worst severity across all rules that fired in this sentence
triggered_rules[].rule_id e.g. ai_buzzwords, empty_solution_verbs, chain_three_part_rule
triggered_rules[].phrase_replacements From TOML phrase_replacements — drop-in copy-paste swaps
triggered_rules[].before_after_example From TOML [[categories.examples]] — real rewrite pairs
rewrite_template From rewrite_planner._SKELETONS — the highest-priority skeleton for this sentence
recommended_replacements Deduped union of every triggered rule's phrase_replacements

Guarantees:

  • Sorted high → low severity, then char_offset_start ascending — ready to feed an LLM humanizer agent sentence-by-sentence
  • Multiple rules hitting the same sentence → one annotation; triggered_rules[] aggregates them
  • Existing matches / modify_plan / summary / metrics fields completely unchanged

🎓 Chapter-specific Writing Guidelines

This tool provides detailed writing guidelines for 5 core chapters of MBA theses:

Chapter 1: Introduction

Common Issues:

  • ❌ Grand narrative opening: "With the development of the economy..."
  • ❌ Vague research significance: "Has important theoretical and practical value"
  • ❌ Literature review listing: A says, B says, C says
  • ❌ Simple listing of research methods

Improvement Strategies:

  • ✅ Get straight to the point: Directly state the research enterprise and specific problem
  • ✅ Specific significance: Explain what problem is solved and what value it brings
  • ✅ Thematic review: Organize by themes with critical analysis
  • ✅ Specific methods: Explain data sources, samples, and analysis tools

📄 Detailed Guide: chapter-1-introduction.md

Chapter 2: Theoretical Foundation

Common Issues:

  • ❌ Textbook-style definition stacking
  • ❌ Theory listing: Introducing 5-6 theories
  • ❌ Disconnect between theory and practice

Improvement Strategies:

  • ✅ Concise definition: 2-3 core concepts
  • ✅ Selective theories: 1-2 core theories, with rationale for selection
  • ✅ Application-oriented: Theory provides framework for subsequent analysis

📄 Detailed Guide: chapter-2-theory.md

Chapter 3: Current Situation and Problem Analysis

Common Issues:

  • ❌ Enterprise overview data stacking with low research relevance
  • ❌ Qualitative status description lacking data
  • ❌ Vague problem identification: "Poor management", "Low efficiency"
  • ❌ Surface-level cause analysis: Remaining at phenomenon description

Improvement Strategies:

  • ✅ Focus on relevance: Only provide background relevant to the research
  • ✅ Data-driven: Use specific indicators and time series data
  • ✅ Specific problems: Each problem has measurement indicators and research data
  • ✅ Deep exploration: Use theoretical framework to analyze root causes

📄 Detailed Guide: chapter-3-analysis.md

Chapter 4: Solutions and Recommendations

Common Issues:

  • ❌ Vague recommendations: "Strengthen management", "Optimize processes"
  • ❌ Lack of operability
  • ❌ Ignoring constraints
  • ❌ Template application: SWOT becomes formalistic

Improvement Strategies:

  • ✅ Specific and actionable: Clearly define what to do, how to do it, and who will do it
  • ✅ Phased implementation: Distinguish short-term, medium-term, and long-term
  • ✅ Consider constraints: Analyze resource, capability, and cultural limitations
  • ✅ Quantified effects: Expected effects should be quantified where possible

📄 Detailed Guide: chapter-4-solutions.md

Chapter 5: Conclusion

Common Issues:

  • ❌ Repetition of abstract content
  • ❌ Simple listing of each chapter's content
  • ❌ Vague innovation points: "First study", "Fills a gap"
  • ❌ Avoiding research limitations

Improvement Strategies:

  • ✅ Research findings: Highlight core findings without repeating the abstract
  • ✅ Specific innovation: Objectively explain theoretical, methodological, and application innovations
  • ✅ Honest limitations: Specifically analyze limitations and their impact
  • ✅ Future outlook: Propose specific and feasible research directions

📄 Detailed Guide: chapter-5-conclusion.md


📝 Format Standards

Independent format standard documentation covers:

  • Chinese-English/Number mixed typesetting standards
  • Figure and table numbering and layout standards
  • Citation and reference standards
  • Number and unit standards
  • Punctuation standards
  • Paragraph and hierarchy standards

📄 Detailed Standards: format-standards.md


📈 Detection Results Interpretation

The detection report contains the following dimensions:

  • AI Generation Probability: 0-100%, higher score indicates more obvious AI traces
  • Risk Level: 🔴 High Risk / 🟡 Medium Risk / 🟢 Low Risk
  • Detector Consistency: Results consistency among three detection methods
  • Priority Fix Items: Modification suggestions sorted by severity
  • Chapter-specific Suggestions: Targeted suggestions based on detected chapter type

Modification Strategies

Different strategies based on AI probability score:

🔴 High Risk (>70%): Deep Rewriting

  • Comprehensive reconstruction of paragraph structure
  • Remove all AI characteristic vocabulary
  • Supplement specific data and cases
  • Estimated time: 2-3 hours

🟡 Medium Risk (40-70%): Targeted Optimization

  • Fix high-priority AI characteristics
  • Adjust clichés and template expressions
  • Supplement key data support
  • Estimated time: 1-2 hours

🟢 Low Risk (<40%): Detail Polishing

  • Fix small amounts of AI traces
  • Optimize language expression
  • Final proofreading
  • Estimated time: 30 minutes-1 hour

💡 Usage Examples

Example 1: Removing AI Vocabulary

Original:

Digital transformation has become a key handle for promoting high-quality 
enterprise development, empowering business innovation and creating 
significant value for enterprises.

Rewritten:

This study explores the impact of digital transformation on enterprise 
performance. By analyzing XX Company's financial data from 2018-2023, 
it was found that digital investment is positively correlated with 
revenue growth.

Example 2: Removing Vague Attribution

Original:

Research has shown that corporate culture has an important impact on 
organizational performance.

Rewritten:

Schein's (2010) research indicates that strong corporate culture is 
positively correlated with organizational performance (r=0.42, p<0.05).

Example 3: MBA Thesis Focused Deep Research

Original:

Research on Enterprise Digital Transformation

Rewritten:

Research on Process Optimization in Digital Transformation of XX 
Company's Production Department

Example 4: Data Traceability

Original:

Enterprise revenue increased by 20%, employee satisfaction is 85%.

Rewritten:

According to XX Company's 2023 annual report, enterprise revenue 
increased by 20% year-over-year. Based on a questionnaire survey 
conducted in December 2023 (N=120), employee satisfaction is 85%.

Example 5: Fixing Chinese-English Mixed Spacing

Original:

MBA thesis writing needs to pay attention to AI trace issues.
Research in 2023 shows that 15% of enterprises have such problems.

Rewritten:

MBA thesis writing needs to pay attention to AI trace issues.
Research in 2023 shows that 15% of enterprises have such problems.

📁 Project Structure

humanize-mba-text-skill/
├── SKILL.md                          # Claude Skill main file
├── README.md                         # Chinese README
├── README_EN.md                      # English README (this file)
├── README_JP.md                      # 日本語 README
├── README_KR.md                      # 한국어 README
├── LICENSE                           # MIT License
│
├── references/                       # Reference documents + AI detection rules
│   ├── ai-writing-patterns.md        # AI writing characteristics guide
│   ├── chapter-1-introduction.md     # Chapter 1: Introduction guide
│   ├── chapter-2-theory.md           # Chapter 2: Theory foundation guide
│   ├── chapter-3-analysis.md         # Chapter 3: Analysis guide
│   ├── chapter-4-solutions.md        # Chapter 4: Solutions guide
│   ├── chapter-5-conclusion.md       # Chapter 5: Conclusion guide
│   ├── format-standards.md           # Format standards
│   ├── strategy_ai_reduction.md      # AI Detection Rate Reduction Strategy
│   ├── strategy_plagiarism.md        # Plagiarism Rate Reduction Strategy
│   ├── strategy_polishing.md         # Academic Polishing Strategy
│   ├── chinese-paper-humanization-rules.toml   # monolithic rule entry (fallback)
│   └── rules/                        # ⭐v1.3: progress-loadable AI rules
│       ├── index.toml                #   lightweight manifest
│       ├── categories/
│       │   ├── structure.toml        #   structure category
│       │   ├── rhythm_quality.toml  #   rhythm category
│       │   ├── formatting.toml       #   formatting category
│       │   ├── content.toml          #   content (incl. macro-narrative, vague problem, no-source quantification)
│       │   ├── evidence.toml         #   evidence (incl. data-without-method, causal leap)
│       │   └── language.toml         #   language category
│       ├── chapter-categories.toml  #   chapter-type classifier
│       └── metrics.toml              #   shared metrics wordlists
│
└── scripts/                          # Detection scripts
    ├── rule_loader.py                # ⭐v1.3: progressive TOML loader
    ├── detect_ai_patterns.py         # AIPatternDetector entry (with modify_plan)
    ├── multi_detector.py             # FusionEngine entry
    ├── feedback_generator.py         # Feedback generator
    └── analyzers/                    # ⭐v1.3: analyzer package
        ├── __init__.py               #   exposes run_prose_analyzers / run_semantic_chain_analyzers / build_modify_plan
        ├── _types.py                 #   AnalyzerIssue / AnalyzerReport data contracts
        ├── _segments.py              #   shared segmentation helpers
        ├── _regex_categories.py      #   hit-level wrapper (used by chain layer)
        ├── sentence_length.py        #   dimension 6
        ├── paragraph_length.py       #   dimension 9
        ├── paragraph_edges.py        #   dimension 10
        ├── paragraph_structure.py    #   dimension 8a
        ├── chapter_template.py       #   dimension 8b
        ├── semantic_chain.py         #   dimensions 3/4/5: 10 chain analyzers
        └── rewrite_planner.py        #   AnalyzerIssue → structured ModifyEntry

🔧 Advanced Usage

Interactive Modification

python scripts/multi_detector.py your_text.txt --interactive

Export Modification Plan

python scripts/multi_detector.py your_text.txt --plan modification_plan.json

Batch Processing

# Process multiple files
for file in *.txt; do
    python scripts/multi_detector.py "$file" --output "reports/${file%.txt}_report.md"
done

⚠️ Important Notes

  1. Academic Integrity: Maintain academic integrity when rewriting; do not fabricate data; all data must have authentic sources
  2. Data Verification: For uncertain citations and data, users are advised to verify original sources
  3. Theory Adaptation: Users are advised to select appropriate theoretical frameworks based on specific research questions
  4. Enterprise Authorization: If involving internal enterprise data, written authorization from the enterprise or data anonymization is recommended
  5. Plagiarism Control: Plagiarism check is recommended after rewriting to ensure rate < 15%
  6. Personalized Adjustment: Rewritten text requires users to adjust according to specific school and advisor requirements
  7. Scope Limitation: This tool focuses on removing AI traces and format standards, not on academic content depth review
  8. References: Rewriting does not automatically generate references; users need to supplement complete references based on actual citations

🤝 Contributing Guidelines

Issues and Pull Requests are welcome!

Submitting Issues

  • Describe the problem clearly
  • Provide example text (can be anonymized)
  • Explain expected behavior

Submitting Pull Requests

  1. Fork this repository
  2. Create your feature branch (git checkout -b feature/AmazingFeature)
  3. Commit your changes (git commit -m 'Add some AmazingFeature')
  4. Push to the branch (git push origin feature/AmazingFeature)
  5. Open a Pull Request

📝 Changelog

v1.4.0 (2026-07-05)

  • 🧭 high_risk_annotations[] per-sentence aggregation: New top-level key on detect_ai_patterns.generate_report JSON output; merges regex hits and structural / chain issues by sentence; each row carries sentence-level char offsets, original sentence text, all triggered rules, per-rule phrase replacements, and real rewrite pairs
  • 🔁 TOML phrase_replacements field: Phrase-level replacement mappings for 4 high-impact rules (ai_buzzwords / empty_solution_verbs / vague_attribution / unsupported_quantification); hook reserved for a v2 LLM fallback
  • 📐 ModifyEntry.before_after_example: Additive field on every modify_plan row — None or {before, after} — sourced from [[categories.examples]]
  • 🆕 scripts/analyzers/high_risk_annotator.py: New module: sentence segmentation (with char offsets) + dual-track bucketing (regex by line_number / issues by location + evidence substring) + severity sorting; decoupled from detect_ai_patterns.py via a lazy import
  • ✅ Backward compatible: Existing matches / modify_plan / summary / metrics field order and types are untouched; target_word_count_range remains a 2-element list; all 117 tests pass

v1.3.0 (2026-07-05)

Main release — feat: rule-driven AI-trace detection with prose + semantic-chain analyzers

  • 🔬 Externalized rules: All AI-detection rules moved to external TOML, split across 7 categories; scripts/rule_loader.py provides progressive loading
  • 📊 Prose-structure analyzers: 5 new statistical / fingerprint analyzers (dimensions 6 / 8a / 8b / 9 / 10)
  • 🔗 Semantic-chain analyzers: 10 new cross-paragraph / cross-chapter chain analyzers (dimensions 3 / 4 / 5)
  • 📋 Structured rewrite plan: detect_ai_patterns.generate_report now returns a modify_plan key
  • 🎯 Unified rule source: AIPatternDetector / StatisticalDetector / FeedbackGenerator share the same TOML rule document
  • 🛠 New scripts/analyzers/ package: single home for the prose + chain analyzer code

v1.3.0 increments

  • 📐 SKILL.md Harris-style split: 576 → 197 lines (-66%); identification / three-dimensional optimization / MBA checklist / chapter-specific rewrite data moved into references/strategy_optimization.toml / mba_rewrite_checklist.toml / chapter_rewrite_rules.toml; SKILL.md becomes a thin navigation index
  • 🔁 Cross-reference with humanizer-academic-zh: detect (us) ↔ rewrite (them) paths complement each other; patterns 21–25 wire into references/rules/categories/*.toml through the progressive manifest
  • 📊 Rule data upgrade:
    • filler_phrases (Pattern 21): 5 → 8 regex patterns
    • chatbot_conversation_residue (Pattern 23): 4 → 7 regex patterns
    • New sycophantic_praise (Pattern 25): 9 regex patterns + 3 before/after examples, severity=medium / weight=0.5
  • 🌐 Multi-language README sync: v1.3.0 content brought into Chinese / English / Japanese / Korean READMEs; both 致谢 / Acknowledgments and 🔗 Related Projects cross-link to humanizer-academic-zh

v1.2.0 (2024-03-11)

  • 📚 Three-Dimensional Optimization Strategy: Added 3 new strategy documents
    • strategy_ai_reduction.md: AI Detection Rate Reduction Strategy
    • strategy_plagiarism.md: Plagiarism Rate Reduction Strategy
    • strategy_polishing.md: Academic Polishing Strategy
  • 🔍 Enhanced Detection Capability: Optimized rule matching, statistical analysis, and language feature detection
  • 🎓 Refined MBA Standards: Further refined MBA thesis core principles
  • 📝 Improved Format Standards: Expanded format standard documentation
  • 🏗️ Optimized Project Structure: Added English and Korean README files

v1.1.0 (2024-02-05)

  • 📚 Refactored Chapter Documents: Split chapter rules into 5 independent files
    • chapter-1-introduction.md: Detailed introduction guide
    • chapter-2-theory.md: Detailed theory foundation guide
    • chapter-3-analysis.md: Detailed analysis guide
    • chapter-4-solutions.md: Detailed solutions guide
    • chapter-5-conclusion.md: Detailed conclusion guide
    • format-standards.md: Independent format standards document
  • 🎓 Refined MBA Standards: Added MBA thesis core principles section
  • 🎯 Strengthened Practice Orientation: Greater emphasis on data support, theoretical application, and specific cases
  • 📝 Improved Format Standards: Independent format standards covering Chinese-English mixed typesetting, figures/tables, citations, etc.
  • 🏗️ Optimized Project Structure: Clearer file organization and indexing

v1.0.0 (2024-02-04)

  • ✨ Initial release
  • 🎯 Support for multi-dimensional AI feature detection
  • 📚 Added chapter-specific rules
  • 🔧 Implemented auto-fix functionality
  • 🎓 Integrated Claude Skill

📄 License

This project uses MIT License - see LICENSE file for details


🙏 Acknowledgments

  • Thanks to Claude Code for providing the Skill framework
  • Thanks to Kimi CLI for providing the Agent parallel execution tools
  • Thanks to all users who provided feedback and suggestions
  • Special thanks to the Haimbeau1o/thesis-optimizer project, whose three-dimensional collaborative optimization strategy (AI detection rate reduction, plagiarism rate reduction, academic polishing enhancement) inspired this project's design
  • Rewrite strategy reference: cangtianhuang/humanizer-academic-zh — this project takes the detect + quantify path, while that project takes the rewrite + polish path; its 25-pattern taxonomy (organised into 5 categories: expression / language / style / fillers / communication residue) is cross-referenced from our SKILL.md's "supplementary reference" section, especially patterns 21-25 that we do not currently model

📮 Contact


🔗 Related Projects

  • thesis-optimizer - Academic Paper Intelligent Optimization System (Computer Deep Learning Direction)
  • cangtianhuang/humanizer-academic-zh - Humanizer Chinese Academic Edition: a complementary rewrite + polish Skill to our detect + quantify path. Its 25-pattern taxonomy is grouped into 5 categories (expression / language / style / fillers / communication residue). v1.3.0 of this project merged patterns 21–25 into the progressive-load TOML (references/rules/categories/*.toml) as filler_phrases (boosted), chatbot_conversation_residue (boosted), sycophantic_praise (new), etc.

If this project helps you, please give it a ⭐ Star!