Thesis Generator: Outline → Complete Thesis

Output Format Selection

The user can specify the desired output format. If not specified, default to LaTeX.

NYCU Thesis Format Specification

The thesis format is based on the NYCU master's thesis template, derived from Chris Huang's thesis: "Leveraging Reinforcement Learning for LLM-based Generic Honeypots with MITRE Engage Framework" (June 2025).

Required Thesis Structure (in order)

1.  Title Pages (Chinese + English)
2.  Chinese Abstract (摘要)
3.  English Abstract
4.  Acknowledgments (誌謝)
5.  Table of Contents (目錄)
6.  List of Figures (圖目錄)
7.  List of Tables (表目錄)
8.  Chapter 1: Introduction
9.  Chapter 2: Background and Related Work
10. Chapter 3: Problem Formulation
11. Chapter 4: Solution Approach / Methodology
12. Chapter 5: Implementation
13. Chapter 6: Experiment Results / Evaluation
14. Chapter 7: Conclusion and Future Work
15. References
16. Appendix (optional)

Title Page Elements

Chinese Title Page:

國立陽明交通大學
[Institute Name]
碩士論文

[Chinese Thesis Title]

研究生：[Student Chinese Name]（[Student English Name]）
指導教授：[Advisor Chinese Name]（[Advisor English Name]）

中華民國[Year]年[Month]月

English Title Page:

[Institute Name]
National Yang Ming Chiao Tung University
Master Thesis

[English Thesis Title]

Student: [Student English Name]
Advisor: Dr. [Advisor English Name]

[Month] [Year]
Taiwan, Republic of China

Core Transformation: Outline → Thesis

The Expansion Principle

The outline uses arrow notation (→) to chain key concepts concisely. The thesis expands each arrow chain into full academic paragraphs with:

1. Topic sentences — State the main point clearly
2. Evidence and explanation — Provide details, data, citations
3. Transitions — Connect to the next concept logically
4. Citations — Reference relevant literature
5. Figures/Tables — Formal captions and cross-references

Expansion Rules

Expansion Examples

Outline (arrow notation):

Honeypot → security tool → simulate real system → decoy hacker → analysis
attack method → prevent hacker attack → past research → simulate-type
honeypot → e.g., Cowrie, Kippo → partial command set → easy expose flaw

Thesis (full academic prose):

Honeypots are security tools designed to simulate real systems, acting as decoys
to attract and monitor malicious activities. By analyzing the behaviors and
techniques employed by attackers, defenders can gain valuable insights into
potential threats and develop more effective countermeasures. Over the years,
researchers have proposed a variety of honeypot implementations. One major
category is simulation-type honeypots, such as Cowrie [3] and Kippo [4], which
attempt to imitate system behavior using limited, predefined command sets.
While lightweight and easier to deploy, these systems are constrained in realism
and can be easily identified by experienced attackers.

Key patterns observed:

• Each → in the outline becomes a clause or sentence
• Parenthetical examples (e.g., Cowrie, Kippo) become formal citations
• Informal language → formal academic tone
• Implicit connections → explicit topic sentences and transitions
• Single concept chains → multi-paragraph sections with structure

Chapter-by-Chapter Expansion Guide

Chapter 1: Introduction

Goal: Motivate the research problem, present contributions, and outline the thesis.

IMPORTANT: No section headings in Introduction. Chapter 1 should NOT have any numbered subsections (1.1, 1.2, etc.). Instead, the content should flow as continuous paragraphs under the chapter heading only.

Outline structure (typically 1.1–1.7 in the outline):

• The outline uses subsections with parenthetical purposes, but these are for organization only
• In the final thesis, all subsection content is merged into flowing paragraphs
• No add_section_heading() calls in Chapter 1

Expansion rules:

1. Do NOT create section headings (1.1, 1.2, etc.) — only the chapter heading "Chapter 1: Introduction"
2. Merge outline subsections into flowing paragraphs with smooth transitions
3. Each outline subsection → roughly 1 paragraph in the final thesis
4. Keep the logical flow: problem → challenge → why existing methods fail → our approach → contributions → organization
5. Contributions → formal bullet list introduced by a sentence ending with a colon
6. Thesis organization → final paragraph describing each chapter's content
7. Length: approximately 3 pages total (roughly 6-8 paragraphs at 12pt Times New Roman, 1.5 line spacing)

Typical Introduction structure in expanded thesis:

[Chapter heading only — no section numbers]
Paragraph 1: Background, motivation, attack lifecycle importance
Paragraph 2: Current detection limitations (alert-based paradigm)
Paragraph 3: Log analysis challenges (volume, complexity)
Paragraph 4: Proposed approach (two-stage: extraction + LLM)
Paragraph 5: Problem statement + contributions intro
  - Bullet: Contribution 1
  - Bullet: Contribution 2
Paragraph 6: Thesis organization

Chapter 2: Background and Related Work

Goal: Provide technical foundation and position the work in literature.

Outline structure:

• 2.1 Background with subsections (2.1.1, 2.1.2, ...)
• 2.2 Related Works with bullet comparisons

Expansion rules:

1. Each background subsection → 1-3 pages explaining the concept
2. Follow "What is X? → Why X matters → How X relates to our work" pattern
3. Include formal definitions, equations where appropriate
4. Add figures (architecture diagrams, concept illustrations)
5. Related Works → organized by methodology/approach, not paper-by-paper
6. Include comparison table: Table 2.X: Comparison of Related Work
7. End with clear gap statement: "Unlike prior work, our approach..."
8. Length: 8-15 pages total

Background subsection template:

[Concept Name]

[Definition paragraph: What is X?]

[Importance paragraph: Why does X matter?]

[Technical details: How does X work? Include equations/figures if needed]

[Connection paragraph: How X relates to our approach]

Chapter 3: Problem Formulation

Goal: Formally define the research problem with mathematical notation.

Expansion rules:

1. Notation Table → formal Table 3.1: Summary of Notations
2. Each notation entry → add description and context
3. Problem Overview → 1-2 paragraphs explaining the two sub-problems
4. Problem Statements → formal mathematical formulation:
   • Given: (inputs)
   • Find: (outputs)
   • Objective: (optimization goal)
   • Subject to: (constraints)
5. Use equation environments for formal definitions
6. Add explanatory paragraphs after each formal statement
7. Length: 3-5 pages

Chapter 4: Solution Approach / Methodology

Goal: Present the proposed method in detail.

Expansion rules:

1. Solution Overview → 1-2 pages with architecture figure (Figure 4.1)
2. Each module → dedicated subsection with:
   • Purpose and motivation
   • Algorithm/procedure description
   • Formal definitions (input/output)
   • Pseudocode or algorithm box if applicable
   • Figure showing the module's workflow
3. Use formal algorithm environments for key procedures
4. Include equations for formulas, scoring functions, etc.
5. Cross-reference between modules to show data flow
6. Length: 8-15 pages

Chapter 5: Implementation

Goal: Describe how the system was built and configured.

Expansion rules:

1. Tools table → formal Table 5.1: Open-Source Tools and Libraries
2. Implementation details → describe system architecture with figure
3. Include:
   • Hardware/software environment
   • Dataset description and statistics
   • Configuration parameters
   • Step-by-step pipeline description
4. Each step → 1-2 paragraphs with technical specifics
5. Length: 4-8 pages

Chapter 6: Experiment Results / Evaluation

Goal: Present experimental methodology, results, and analysis.

Expansion rules:

1. Each experiment → dedicated subsection with:
   • Research question or hypothesis
   • Experimental setup (metrics, baselines, parameters)
   • Results presentation (tables, figures)
   • Analysis and discussion
2. Every table/figure needs:
   • Formal numbering (Table 6.X, Figure 6.X)
   • Descriptive caption
   • In-text cross-reference
   • Discussion of key observations
3. Include statistical measures (averages, standard deviations)
4. Compare with baselines explicitly
5. Discuss limitations of results
6. Length: 8-15 pages

Chapter 7: Conclusion and Future Work

Goal: Summarize contributions and propose future directions.

Expansion rules:

1. Summary paragraph → restate problem and approach (1 paragraph)
2. Key findings → summarize main experimental results (1-2 paragraphs)
3. Contributions → restate in context of results (1 paragraph)
4. Limitations → honest assessment (1 paragraph)
5. Future Work → specific, actionable directions (2-3 paragraphs)
6. Length: 2-3 pages

Workflow

Step 1: Read the Outline

# Extract text content from the outline document
python -m markitdown paper_outline_v1.docx > outline_content.md

Or using python-docx:

from docx import Document
doc = Document("paper_outline_v1.docx")
for para in doc.paragraphs:
    print(f"[{para.style.name}] {para.text}")

Step 2: Identify Structure and Metadata

From the outline, extract:

1. Thesis title (Chinese and English)
2. Student name and advisor names
3. Chapter structure (chapter titles, section numbers)
4. Key content per section (arrow chains, tables, figures)
5. References list

Step 3: Determine Output Format

Check user's specification:

• If user says "docx" or "Word" → generate .docx using python-docx
• If user says "latex" or "tex" → generate .tex files
• If not specified → default to LaTeX

Step 4: Expand Each Chapter

For each chapter in the outline:

1. Parse arrow chains → identify concept flow
2. Expand to paragraphs → apply expansion rules above
3. Formalize elements:
   • Tables → numbered with captions
   • Figures → numbered with captions and cross-references
   • Equations → numbered in equation environments
   • Citations → formal [X] or \cite{X} references
4. Add transitions between sections and paragraphs
5. Maintain academic tone throughout

Step 5: Generate Front Matter

Create:

• Title pages (Chinese + English)
• Abstract (Chinese + English) — derive from Introduction + Conclusion
• Table of Contents, List of Figures, List of Tables

Step 6: Generate the Thesis Document

Option A: LaTeX Output (Default)

Generate a LaTeX project structure:

thesis_v1/
├── main.tex              # Master document with \input{} for each chapter
├── chapters/
│   ├── abstract_zh.tex   # Chinese abstract
│   ├── abstract_en.tex   # English abstract
│   ├── chapter1.tex      # Introduction
│   ├── chapter2.tex      # Background and Related Work
│   ├── chapter3.tex      # Problem Formulation
│   ├── chapter4.tex      # Solution Approach
│   ├── chapter5.tex      # Implementation
│   ├── chapter6.tex      # Evaluation
│   └── chapter7.tex      # Conclusion
├── figures/              # Figure files
├── tables/               # Standalone table files (optional)
├── references.bib        # Bibliography
└── nycu-thesis.cls       # NYCU thesis class (if available)

main.tex template:

\documentclass[12pt, a4paper]{report}
\usepackage[utf8]{inputenc}
\usepackage{amsmath, amssymb, amsthm}
\usepackage{graphicx}
\usepackage{booktabs}
\usepackage{hyperref}
\usepackage{algorithm}
\usepackage{algorithmic}
\usepackage{xeCJK}  % For Chinese characters

\title{[Thesis Title]}
\author{[Student Name]}
\date{[Month Year]}

\begin{document}

% Title pages
\input{chapters/titlepage}

% Abstracts
\input{chapters/abstract_zh}
\input{chapters/abstract_en}

% Front matter
\tableofcontents
\listoffigures
\listoftables

% Main chapters
\input{chapters/chapter1}
\input{chapters/chapter2}
\input{chapters/chapter3}
\input{chapters/chapter4}
\input{chapters/chapter5}
\input{chapters/chapter6}
\input{chapters/chapter7}

% References
\bibliographystyle{IEEEtran}
\bibliography{references}

\end{document}

Option B: DOCX Output

Use python-docx to generate the thesis document:

from docx import Document
from docx.shared import Pt, Inches, RGBColor
from docx.enum.text import WD_ALIGN_PARAGRAPH

doc = Document()

# Set default font
style = doc.styles['Normal']
font = style.font
font.name = 'Times New Roman'
font.size = Pt(12)

# Generate title pages, abstracts, TOC placeholder, chapters...
# (See detailed generation code below)

doc.save("thesis_v1.docx")

DOCX Generation Rules:

• Font: Times New Roman, 12pt for body text
• Headings: Chapter titles in 16pt bold, Section titles in 14pt bold
• Line spacing: 1.5 or double spacing
• Margins: 1 inch all sides (or per NYCU requirements)
• Page numbers: bottom center
• Tables: use borders with proper formatting
• Figures: centered with caption below
• Equations: centered with right-aligned numbering

Step 7: Review and Refine

After generating:

• Verify all outline sections are expanded
• Check transitions between sections
• Ensure all figures/tables have captions and cross-references
• Verify citation format consistency
• Check for [TODO] or [PLACEHOLDER] markers
• Ensure consistent terminology throughout
• Verify formal mathematical notation

DOCX Generation Details (python-docx)

Document Setup

from docx import Document
from docx.shared import Pt, Inches, RGBColor, Cm
from docx.enum.text import WD_ALIGN_PARAGRAPH
from docx.enum.section import WD_ORIENT
from docx.oxml.ns import qn

def create_thesis_doc():
    doc = Document()
    
    # Set default style
    style = doc.styles['Normal']
    font = style.font
    font.name = 'Times New Roman'
    font.size = Pt(12)
    paragraph_format = style.paragraph_format
    paragraph_format.line_spacing = 1.5
    
    # Set margins
    for section in doc.sections:
        section.top_margin = Cm(2.54)
        section.bottom_margin = Cm(2.54)
        section.left_margin = Cm(3.17)
        section.right_margin = Cm(2.54)
    
    return doc

Chapter Heading

def add_chapter_heading(doc, chapter_num, title):
    """Add a chapter heading with page break."""
    doc.add_page_break()
    heading = doc.add_heading(f'Chapter {chapter_num}', level=0)
    for run in heading.runs:
        run.font.size = Pt(24)
        run.font.color.rgb = RGBColor(0, 0, 0)
    
    subtitle = doc.add_heading(title, level=1)
    for run in subtitle.runs:
        run.font.size = Pt(18)
        run.font.color.rgb = RGBColor(0, 0, 0)

Section with Academic Prose

def add_section(doc, section_num, title, content_paragraphs):
    """Add a section with multiple paragraphs."""
    heading = doc.add_heading(f'{section_num} {title}', level=2)
    for run in heading.runs:
        run.font.color.rgb = RGBColor(0, 0, 0)
    
    for para_text in content_paragraphs:
        p = doc.add_paragraph(para_text)
        p.paragraph_format.first_line_indent = Inches(0.5)
        p.paragraph_format.space_after = Pt(6)

Figure with Caption

def add_figure(doc, image_path, caption, fig_num):
    """Add a centered figure with caption."""
    p = doc.add_paragraph()
    p.alignment = WD_ALIGN_PARAGRAPH.CENTER
    run = p.add_run()
    run.add_picture(image_path, width=Inches(5))
    
    cap = doc.add_paragraph(f'Figure {fig_num}: {caption}')
    cap.alignment = WD_ALIGN_PARAGRAPH.CENTER
    cap.runs[0].font.size = Pt(10)

Table with Caption

def add_table_with_caption(doc, headers, rows, caption, table_num):
    """Add a table with caption above."""
    cap = doc.add_paragraph(f'Table {table_num}: {caption}')
    cap.alignment = WD_ALIGN_PARAGRAPH.CENTER
    cap.runs[0].font.size = Pt(10)
    cap.paragraph_format.space_after = Pt(4)
    
    table = doc.add_table(rows=1 + len(rows), cols=len(headers))
    table.style = 'Table Grid'
    
    # Header row
    for i, h in enumerate(headers):
        cell = table.rows[0].cells[i]
        cell.text = h
        for p in cell.paragraphs:
            for r in p.runs:
                r.font.bold = True
                r.font.size = Pt(10)
    
    # Data rows
    for row_idx, row_data in enumerate(rows):
        for col_idx, cell_text in enumerate(row_data):
            cell = table.rows[row_idx + 1].cells[col_idx]
            cell.text = str(cell_text)
            for p in cell.paragraphs:
                for r in p.runs:
                    r.font.size = Pt(10)

LaTeX Generation Details

Academic Prose Style

When generating LaTeX, follow these conventions:

% Chapter heading
\chapter{Introduction}

% Section with prose
\section{Background and Motivation}

Honeypots are security tools designed to simulate real systems, acting as decoys
to attract and monitor malicious activities~\cite{spitzner2003honeypots}.
By analyzing the behaviors and techniques employed by attackers, defenders can
gain valuable insights into potential threats and develop more effective
countermeasures.

% Figure
\begin{figure}[htbp]
    \centering
    \includegraphics[width=0.8\textwidth]{figures/architecture.pdf}
    \caption{System architecture overview showing the two-stage pipeline.}
    \label{fig:architecture}
\end{figure}

% Table
\begin{table}[htbp]
    \centering
    \caption{Summary of notations used in this thesis.}
    \label{tab:notations}
    \begin{tabular}{lll}
        \toprule
        \textbf{Category} & \textbf{Notation} & \textbf{Description} \\
        \midrule
        Alerts \& Logs & $A = \{a_1, a_2, \ldots\}$ & Alert set from EDR/NDR \\
        \bottomrule
    \end{tabular}
\end{table}

% Equation
\begin{equation}
    \text{Recall Rate (RR)} = \frac{|\hat{L} \cap L^*|}{|L^*|}
    \label{eq:recall}
\end{equation}

% Problem statement
\textbf{Problem 1 (Malicious Log Extraction).}
\textit{Given} a set of raw logs $L$ and a set of alerts $A$,
\textit{find} the subset of malicious logs $\hat{L} \subseteq L$ such that
the recall rate is maximized while minimizing the false positive rate.

Writing Quality Guidelines

Academic Tone Rules

Paragraph Structure

Every paragraph should follow this pattern:

1. Topic sentence — States the main point
2. Supporting evidence — Data, citations, examples
3. Analysis — Interpret the evidence
4. Transition — Connect to next paragraph/section

Citation Handling

⚠️ CRITICAL: Never hallucinate citations.

• Outline references like [1], [6], [7] → preserve exactly as-is
• If the outline provides a reference list → use those citations
• Mark any missing citations as [CITATION NEEDED]
• In LaTeX: use \cite{key} with entries in references.bib
• In DOCX: use [X] numbering matching the reference list

Figure and Table Handling

Figures from the outline (embedded as base64 images):

1. Extract and save as separate image files
2. Reference with proper numbering and captions
3. If figure quality is insufficient → add [TODO: Replace with high-resolution figure]

Tables from the outline:

1. Reformat with proper academic table style
2. Add table number and caption
3. Use booktabs style in LaTeX (no vertical lines)
4. Use "Table Grid" style in DOCX

⚠️ Important Notes

What This Skill Does

• Expands concise arrow-notation outlines into full academic prose
• Generates properly formatted thesis documents (LaTeX or DOCX)
• Follows NYCU thesis format conventions
• Preserves all technical content from the outline
• Maintains citation references from the outline

What This Skill Does NOT Do

• Generate new research content not present in the outline
• Fabricate experimental results or data
• Invent citations or references
• Translate between languages (Chinese abstract should be provided separately)
• Create figures from scratch (uses figures from outline or marks as TODO)

Quality Checklist

Before delivering the thesis:

- [ ] All outline sections expanded to full paragraphs
- [ ] Consistent terminology throughout
- [ ] All figures numbered and captioned
- [ ] All tables numbered and captioned  
- [ ] All equations numbered
- [ ] Cross-references correct (Figure X.Y, Table X.Y, Equation X)
- [ ] Citations preserved from outline
- [ ] Missing citations marked as [CITATION NEEDED]
- [ ] No informal language remaining
- [ ] Transitions between sections are smooth
- [ ] Front matter complete (title pages, abstracts, TOC)
- [ ] [TODO] markers for incomplete sections clearly marked
- [ ] Output format matches user's request (LaTeX or DOCX)

Dependencies

This skill relies on the following:

Required Python Packages

• markitdown — Extract text from .docx
• python-docx — Read/write .docx files
• pypdf / pdfplumber — PDF text extraction

For LaTeX Output

• TeX Live or MiKTeX installation (for compilation)
• xelatex recommended (for CJK character support)

Example Usage

Example 1: Generate thesis in LaTeX (default)

User: "Convert paper_outline_v1.docx into a complete thesis"

Action:

1. Read paper_outline_v1.docx using markitdown
2. Parse chapter structure and content
3. Generate LaTeX files in thesis_v1/ directory
4. Provide compilation instructions

Example 2: Generate thesis in DOCX

User: "Convert paper_outline_v1.docx into thesis_v1.docx"

Action:

1. Read paper_outline_v1.docx using markitdown or python-docx
2. Parse chapter structure and content
3. Generate thesis_v1.docx using python-docx
4. Apply NYCU thesis formatting

Example 3: Generate specific chapters

User: "Expand Chapter 4 of the outline into full thesis prose"

Action:

1. Read the outline's Chapter 4 content
2. Expand arrow notation into full paragraphs
3. Add formal figure/table references
4. Output in the previously specified format