exam-paper

name	exam-paper
description	Generate exam papers, study guides, review books, and teaching materials from any subject. Extracts and synthesizes user-provided materials, generates transfer-oriented questions, and outputs compact Markdown + PDF via Pandoc + ctexart + xelatex.
keywords	["exam-paper","试卷","出题","讲义","复习全书","举一反三","试卷排版","pandoc-exam","ctexart-exam","知识提炼","xelatex","跨学科"]

Exam Paper & Teaching Material Generation Skill

Generate exam papers, practice tests, study guides, review books, and teaching materials from any subject. Works end-to-end: extract knowledge from user materials → synthesize → generate questions that encourage "举一反三" (transfer learning) → output compact MD + PDF.

This skill is content-type centric, not subject-list centric. Whether the input is chemistry formulas, historical essays, legal cases, medical symptoms, or engineering diagrams, follow the same workflow and apply the corresponding content-type rules below.

When to use

• User provides reference materials and wants exam papers / practice questions / study guides / review books
• User asks to "提炼知识点出题" or "根据这份资料出几道题" or "生成复习全书"
• User wants teaching materials with detailed explanations + worked examples
• User needs compact, printable PDFs without page break lines
• User says "出一套xx试卷" or "根据xx内容整理讲义" or "写一本xx复习全书"

When NOT to use

• User only wants a simple text list (no PDF needed)
• User provides a LaTeX .tex source file directly and only wants minor edits
• User explicitly asks for an interactive quiz / online test platform

Workflow Overview

1. Explore materials — read user-provided files (PDF, DOCX, Markdown, images via OCR if needed). Identify the subject and scope.
2. Extract & synthesize — identify core knowledge points (考点), key formulas, common patterns, and exam traps.
3. Design document structure — choose document type (exam paper / study guide / review book / practice questions) and outline chapters/sections.
4. Apply content-type rules — decide which rules from this skill apply: formulas, tables, diagrams, code, multilingual text, citations/proofs, etc.
5. Write markdown — use proper YAML frontmatter, correct formula rendering, Chinese typography rules, and no page-break horizontal rules.
6. Generate questions — for each topic, create 基础题、变式题、综合题、开放题 with detailed answers.
7. Compile PDF — use pandoc + xelatex. Verify no rendering warnings/errors.
8. Verify — check PDF visually for formula display, layout, page breaks, orphan lines, and missing characters.

Document Types

Type A: Exam Paper (试卷)

• Purpose: simulate a real exam; compact and printable
• Layout: minimal title, no headers/footers/page numbers, 10.5–11pt, 2.0–2.5cm margins
• Content: instructions, multiple question types, answer key (optional, at end)
• Include reference data (constants, tables, terminology) if needed

Type B: Study Guide / Teaching Material (讲义)

• Purpose: systematic learning material
• Layout: title page, chapter headings, 10.5pt, 2.2cm margins
• Content: 知识梳理 + 1–2 worked examples per topic + "举一反三引导" questions
• Use comparison tables and formula boxes

Type C: Review Book (复习全书)

• Purpose: comprehensive end-of-term / exam-prep reference
• Layout: title page, multi-chapter, 10.5pt, 2.2cm margins
• Content: theory summary + examples (choice / T-F / calculation / proof / short answer) + detailed solutions
• Each chapter should be self-contained but coherent

Type D: Practice Questions Only (纯题)

• Purpose: maximize question density for drilling
• Layout: 8.5–9pt, 1.2–1.5cm margins, minimal explanations
• Content: questions grouped by topic; answers in separate file or at end

Markdown Frontmatter Templates

Type A: Exam Paper

---
documentclass: ctexart
geometry: margin=2.5cm,top=2cm,bottom=2cm
fontsize: 11pt
header-includes: |
  \setlength{\emergencystretch}{3em}
  \pagestyle{empty}
  \widowpenalty=10000
  \clubpenalty=10000
  \brokenpenalty=10000
  \usepackage{enumitem}
  \setlist[enumerate]{nosep}
  \setlength{\parskip}{0.3em}
  \setlength{\parindent}{0pt}
  \usepackage{booktabs}
  \usepackage{amsmath}
  \usepackage{amssymb}
  \usepackage{graphicx}
mainfont: "Noto Serif CJK SC"
sansfont: "Noto Serif CJK SC"
---

Type B/C: Study Guide / Review Book

---
title: "学科名 期末复习全书（详细版）"
subtitle: "Subtitle"
fontsize: 10.5pt
documentclass: ctexart
geometry: margin=2.2cm
colorlinks: true
header-includes: |
  \setlength{\emergencystretch}{3em}
  \linespread{1.25}
  \setlength{\tabcolsep}{3pt}
  \renewcommand{\arraystretch}{1.05}
  \widowpenalty=10000
  \clubpenalty=10000
  \brokenpenalty=10000
  \usepackage{amsmath}
  \usepackage{amssymb}
  \usepackage{booktabs}
  \usepackage{tikz}
  \usetikzlibrary{calc,positioning}
  \usepackage{tcolorbox}
  \tcbuselibrary{skins,breakable}
  \renewenvironment{quote}{\begin{tcolorbox}[colback=blue!3!white,colframe=blue!40!black,boxrule=0.4pt,arc=1.5mm,breakable,left=2mm,right=2mm,top=1mm,bottom=1mm]}{\end{tcolorbox}}
mainfont: "Noto Serif CJK SC"
sansfont: "Noto Serif CJK SC"
---

Type D: Practice Questions Only (Ultra-Compact)

---
fontsize: 8.5pt
documentclass: ctexart
geometry: margin=1.2cm
header-includes: |
  \setlength{\emergencystretch}{3em}
  \linespread{1.05}
  \setlength{\parindent}{0pt}
  \setlength{\parskip}{0.05em}
  \setlength{\tabcolsep}{2pt}
  \renewcommand{\arraystretch}{1.0}
  \widowpenalty=10000
  \clubpenalty=10000
  \brokenpenalty=10000
mainfont: "Noto Serif CJK SC"
sansfont: "Noto Serif CJK SC"
---

Critical: ALL documents must include \widowpenalty, \clubpenalty, \brokenpenalty set to 10000 to prevent orphan/widow lines and unwanted page breaks. For Chinese documents, always set mainfont and sansfont to a CJK font such as Noto Serif CJK SC.

Type E: English Document

For non-CJK documents, use a Latin font and the standard article class:

---
title: "Calculus Final Review"
subtitle: "Example English Review Book"
fontsize: 11pt
documentclass: article
geometry: margin=2.2cm
colorlinks: true
header-includes: |
  \setlength{\emergencystretch}{3em}
  \linespread{1.2}
  \widowpenalty=10000
  \clubpenalty=10000
  \brokenpenalty=10000
  \usepackage{amsmath}
  \usepackage{amssymb}
  \usepackage{booktabs}
---

CRITICAL: NO Horizontal Rules

NEVER use --- (three dashes) in markdown content. Pandoc converts --- to a horizontal rule which becomes a visible line / page break in the PDF. Use blank lines or section headers to separate content instead.

Exception: the YAML frontmatter at the very top of the file, which is delimited by ---.

Chinese Typography Rules

1. Chinese Quotes

NEVER use straight ASCII " or ' for Chinese quotes. They render incorrectly and may be interpreted as math mode delimiters.

Use full-width Chinese quotes:

Wrong	Correct
"特征反应"	“特征反应”
"三看"	“三看”
"远距离先反应"	“远距离先反应”

2. Dashes and Ellipsis

Wrong	Correct	Notes
--	——	Chinese em dash (U+2014 U+2014)
...	……	Chinese ellipsis

3. Punctuation in Math Mode

Chinese punctuation inside math mode may cause font warnings. Keep punctuation such as ，、。、； outside $...$. For example:

• Wrong: $x = 1。$
• Correct: $x = 1$。

Content-Type Rules (Universal)

Instead of memorizing per-subject rules, identify which content types appear in the material and apply the matching rules below.

1. Formulas and Symbols

• Always use LaTeX math mode for formulas, Greek letters, operators, subscripts, and superscripts.
• Common commands: \alpha, \beta, \gamma, \theta, \sigma, \mu, \pi, \sum, \int, \prod, \frac, \sqrt, \lim, \infty, \pm, \times, \div, \leq, \geq, \neq, \approx.
• Wrap Chinese text inside math mode with \text{...}.
• Convert Unicode subscripts/superscripts to LaTeX (e.g., H₂O → H$_2$O, x² → $x^2$).
• Avoid Unicode logic symbols (∧, ∨, ¬, →); use \land, \lor, \neg, \rightarrow.

Example matrix:

$$
\mathbf{A} = \begin{bmatrix} a_{11} & a_{12} \\ a_{21} & a_{22} \end{bmatrix}
$$

Example piecewise function:

$$
f(x) = \begin{cases} x, & x \geq 0 \\ 0, & x < 0 \end{cases}
$$

2. Tables and Comparisons

• Use markdown tables for comparisons, terminology mappings, and summary sheets.
• Ensure every cell containing math is fully wrapped in $...$.
• For very long tables, consider splitting or reducing font size in the frontmatter.
• Avoid multi-line content inside a single table cell; use separate rows instead.

Example:

| 概念 | 定义 | 公式 |
|-----|------|------|
| 熵 | 系统不确定性的度量 | $H(X) = -\sum p_i \log p_i$ |
| 条件熵 | 已知 $Y$ 后 $X$ 的不确定性 | $H(X|Y) = -\sum p(x,y) \log p(x|y)$ |

3. Diagrams and Figures

You have three ways to include diagrams, depending on complexity:

1. TikZ inside markdown — best for simple geometry, arrows, trees, and anything that should stay vector in PDF.
2. matplotlib helper script — best for coordinate systems, force diagrams, bar/line charts, Venn diagrams, flowcharts, and triangles.
3. External image — for hand-drawn or complex figures, include with ![caption](path.png).

3.1 TikZ in markdown

Use raw LaTeX blocks (works with --from markdown+raw_tex):

\begin{tikzpicture}
  \node (A) at (0,0) {A};
  \node (B) at (3,0) {B};
  \draw[->] (A) -- (B) node[midway,above] {$f$};
\end{tikzpicture}

More examples:

% Right triangle
\begin{tikzpicture}
  \draw (0,0) -- (4,0) -- (4,3) -- cycle;
  \node at (-0.2,-0.2) {A};
  \node at (4.2,-0.2) {B};
  \node at (4.2,3.2) {C};
\end{tikzpicture}

% Tree / hierarchy
\begin{tikzpicture}
  \node {Root}
    child {node {Left}}
    child {node {Right}};
\end{tikzpicture}

3.2 Using the bundled matplotlib helper

The skill includes tools/diagram_tools.py with ready-to-use functions:

• coordinate_system()
• function_plot()
• number_line()
• force_diagram()
• geometry_triangle()
• bar_chart()
• line_chart()
• scatter_plot()
• box_plot()
• probability_tree()
• venn_diagram()
• flowchart() — supports standard flowchart symbols

Call them from your generation script:

from tools.diagram_tools import (
    coordinate_system, function_plot, number_line,
    force_diagram, geometry_triangle,
    bar_chart, line_chart, scatter_plot, box_plot,
    probability_tree, venn_diagram, flowchart
)

# Example: coordinate system with points and segments
coordinate_system(
    points=[(1, 2), (-2, -1)],
    segments=[((0, 0), (1, 2))],
    labels={0: "P", 1: "Q"},
    output_path="output/diagram_coord.png"
)

# Example: function plot
function_plot(
    lambda x: x ** 2,
    x_range=(-3, 3),
    output_path="output/diagram_function.png",
    title=r"$y = x^2$"
)

# Example: number line
number_line(
    points=[-2, -1, 0, 1, 2],
    output_path="output/diagram_numberline.png"
)

# Example: free-body force diagram
force_diagram(
    forces=[
        {"x": 2, "y": 1, "label": "F1", "color": "red"},
        {"x": -1, "y": 2, "label": "F2", "color": "blue"},
    ],
    output_path="output/diagram_forces.png"
)

# Example: bar chart
bar_chart(
    categories=["A", "B", "C"],
    values=[10, 20, 15],
    output_path="output/diagram_bar.png",
    title="Sample Comparison"
)

# Example: standard flowchart with decision diamond and edge labels
flowchart(
    nodes={
        "开始": {"pos": (0, 0), "type": "start"},
        "输入n": {"pos": (2, 0), "type": "io"},
        "n>0?": {"pos": (4, 0), "type": "decision"},
        "计算": {"pos": (6, 1.5), "type": "process"},
        "报错": {"pos": (6, -1.5), "type": "process"},
        "输出结果": {"pos": (8, 0), "type": "io"},
        "结束": {"pos": (10, 0), "type": "end"},
    },
    edges=[
        ("开始", "输入n"),
        ("输入n", "n>0?"),
        ("n>0?", "计算", "是"),
        ("n>0?", "报错", "否"),
        ("计算", "输出结果"),
        ("报错", "输出结果"),
        ("输出结果", "结束"),
    ],
    output_path="output/diagram_flowchart.png"
)

Then embed the generated image in markdown:

![坐标系示例](output/diagram_coord.png)

3.3 External images

If the user provides images or you draw them elsewhere, embed normally:

![受力分析图](images/force_diagram.png)

Keep image resolution at least 150 DPI for print quality.

4. Code and Pseudocode

• Use fenced code blocks with language identifier for actual code.
• For pseudocode, use numbered lists or plain code blocks without language tag.
• Ensure monospace fonts contain glyphs for any special Unicode used.

```python
def foo(x):
    return x + 1

### 5. Multilingual and Specialized Text

- Classical Chinese, foreign language passages, legal articles, medical terminology: keep as normal text, but mark special terms with quotes or emphasis.
- For terminology lists, use tables with original term / definition / symbol columns.
- Romanize or transliterate where helpful, but preserve original characters for accuracy.

### 6. Theorems, Proofs, and Citations

- Use section headings or bold labels for theorems, lemmas, corollaries, and proofs.
- End proofs with a clear marker such as **证毕** or `\square` if in math mode.
- For citation-style references, use numbered lists or parenthetical notes.

## Python Raw-String Traps When Generating Markdown

If you generate markdown content using Python raw strings (`r"..."`), remember that a literal `\n` in the raw string is two characters (backslash + n), NOT a newline. You must convert it to a real newline before writing the file. **However**, naive replacement will destroy LaTeX commands that start with `\n`.

### Safe conversion pattern

```python
# Protect LaTeX commands that start with \n before converting \n to newline
content = content.replace('\\newpage', '__NEWPAGE__')
content = content.replace('\\neq', '__NEQ__')
content = content.replace('\\neg', '__NEG__')
content = content.replace('\\ni', '__NI__')
content = content.replace('\\n', '\n')
content = content.replace('__NEWPAGE__', '\\newpage')
content = content.replace('__NEQ__', '\\neq')
content = content.replace('__NEG__', '\\neg')
content = content.replace('__NI__', '\\ni')

Also protect any other \ command you use that begins with n.

Even bigger trap: Python escape sequences in LaTeX commands

In normal Python strings (not raw), backslash + one of a b f n r t v is interpreted as a control character. This corrupts many common LaTeX commands:

LaTeX command	Corrupted to	Result in PDF
\forall	form feed (\f) + orall	Missing $ inserted
\rightarrow	carriage return (\r) + ightarrow	math error
\theta	tab (\t) + heta	weird spacing / error
\times	tab (\t) + imes	weird spacing / error
\text{...}	tab (\t) + ext{...}	error
\alpha	bell (\a) + lpha	invisible / error
\beta	backspace (\b) + eta	invisible / error
\frac	form feed (\f) + rac	error

Rule of thumb: any Python string that will contain LaTeX commands should be a raw string (r"..." or r"""..."""). If you must use a normal string, double every backslash: \\forall, \\rightarrow, etc.

Recommended pattern: build chapter content with r"""...""" and make helper-function arguments raw strings as well:

def ex_choice(q, opts, ans, expl):
    return f"{box('例题·选择题', q)}\n\n{opts}\n\n{box('答案与解析', f'**答案：{ans}**\\n\\n{expl}')}\n"

ex_choice(
    r"将公式 $(\forall x)(P(x) \rightarrow Q(x))$ 化为子句集。",
    r"A. ...\n\nB. ...",
    r"B",
    r"解析中使用 $\forall$、$\rightarrow$ 等命令。"
)

Compilation Commands

Basic command

pandoc input.md -o output.pdf --pdf-engine=xelatex

Recommended command for Chinese documents with math

pandoc input.md -o output.pdf \
  --from markdown+raw_tex \
  --pdf-engine=xelatex \
  -V CJKmainfont="Noto Serif CJK SC" \
  -V mainfont="Noto Serif CJK SC" \
  -V sansfont="Noto Serif CJK SC" \
  --variable geometry:margin=2.2cm \
  --variable fontsize=10.5pt

Verbose debugging command

pandoc input.md -o output.pdf \
  --from markdown+raw_tex \
  --pdf-engine=xelatex \
  --verbose \
  2>&1 | tee pandoc.log

Always inspect pandoc.log for Missing character or Undefined control sequence warnings.

Quality Check

Before compiling, run tools/quality_check.py on your generated markdown to catch common mistakes early:

python tools/quality_check.py output/review.md

It checks for:

• Horizontal rules --- in content
• Straight quotes used as Chinese quotes
• Chinese characters inside math mode without \text{}
• Unicode subscripts/superscripts not converted to LaTeX
• Unicode logic symbols (∧, ∨, ¬, etc.) not converted
• Control characters from corrupted LaTeX commands (e.g., \forall becoming form feed)
• Unclosed inline math mode
• Table cells with unclosed math mode
• Adjacent math modes that should be merged

A non-zero exit code is returned if any ERROR-level issue is found. Integrate this step into your generation pipeline:

import subprocess
subprocess.run(["python", "tools/quality_check.py", "output/review.md"], check=True)

Question Type Templates

Use the following markdown structures to keep questions and answers visually distinct. The review-book frontmatter turns quote blocks into light-blue example boxes.

Multiple Choice

> **【例题·选择题】**
>
> 题干内容……

A. 选项一

B. 选项二

C. 选项三

D. 选项四

> **【答案与解析】**
>
> **答案：C**
>
> 解析内容……

True / False

> **【例题·判断题】**
>
> 题干内容……

> **【答案与解析】**
>
> **答案：错误**
>
> 解析内容……

Calculation / Proof

> **【例题·计算/证明题】**
>
> 题干内容……

> **【答案与解析】**
>
> 详细解答步骤……

Short Answer

> **【例题·简答题】**
>
> 题干内容……

> **【答案与解析】**
>
> 简明扼要的答案……

Question Generation Principles

Extract & Synthesize Knowledge

When user provides reference materials:

1. Read ALL materials first
2. Identify core knowledge points (考点)
3. Summarize key patterns and rules
4. Note common mistakes and exam traps

Design "举一反三" Questions

Questions should NOT just test memorization. They should guide students to:

• Transfer knowledge: Apply learned patterns to new situations
• Compare and contrast: Identify similarities/differences between similar concepts
• Analyze conditions: What changes if a variable changes?
• Multiple approaches: Solve the same problem using different methods

Question Variety

For each knowledge point, generate questions of varying difficulty:

• 基础题: Direct application of learned rules
• 变式题: Same concept, different context/presentation
• 综合题: Combining multiple knowledge points
• 开放题: "如果…会怎样？" encouraging deeper thinking

Example: "举一反三" Question Sets

For a chemistry topic on metal displacement:

1. Direct: Fe + CuSO₄ → ? (basic substitution)
2. Variant: Zn + AgNO₃ → ? (different metals, same pattern)
3. Multi-step: Zn added to Cu(NO₃)₂ + AgNO₃ mixture → filter residue?
4. Reverse reasoning: Given residue composition, what was added?
5. Real-world: Why can't iron pots store copper sulfate solution?

Guiding questions for students:

• "如果换成铁粉而不是锌粉，结论有何不同？"
• "滤液滤渣问题中，'一定'和'可能'的区别是什么？"

Subject Adaptation Framework

This skill is designed to work with any subject. Use the following framework instead of looking up a fixed subject list.

Step 1: Identify the dominant content types

Ask: what does this subject primarily use?

• Formulas and symbols
• Tables and comparisons
• Diagrams and figures
• Code and algorithms
• Long text passages
• Legal/medical/engineering cases
• Multilingual terminology

Step 2: Apply the matching Content-Type Rules

Map each dominant content type to the rules in the "Content-Type Rules" section above.

Step 3: Capture subject-specific conventions

Note down:

• Common symbols and notation
• Standard abbreviations
• Typical question formats
• Common traps or grading points
• Required reference data (constants, tables, terminology)

Step 4: Generate and compile

Use the appropriate frontmatter, question templates, and compilation command.

New Subject Quick-Start Template

When extending this skill to a new subject, fill in:

## 学科名称

- 核心内容类型：公式 / 图表 / 代码 / 文本 / 病例 / 法条 / 工程图 / …
- 常用 LaTeX 宏包/命令：
- 典型题型：选择题 / 判断题 / 计算题 / 证明题 / 简答题 / 案例分析 / …
- 常见符号与约定：
- 易错排版点：
- 推荐 frontmatter 调整（如有）：

Example Subject Snapshots

The following are illustrative applications of the framework, not an exhaustive subject list.

Example A: Natural Sciences (Chemistry / Physics / Biology)

• Convert all Unicode subscripts/superscripts to LaTeX math mode
• Chemical equations: H$_2$SO$_4$, ion charges Cu$^{2+}$
• Reaction conditions: $\xrightarrow{\text{点燃}}$
• Units: m/s$^2$, kg·m/s$^2$
• Diagrams: molecular structures, force diagrams, cell diagrams via TikZ or matplotlib

Example B: Mathematics

• Equations: $x^2 + 2x - 3 = 0$
• Functions: $f(x) = \sin x$
• Geometry: coordinates, angles, areas, TikZ diagrams
• Statistics: probability, distributions, data tables

Example C: Humanities and Languages

• Long reading passages as plain text
• Terminology tables (term / definition / context)
• Classical text annotations
• Essay prompts with grading rubrics

Example D: Social Sciences (Economics / Political Science / Sociology)

• Regression models and matrix notation in math mode
• Causal diagrams (DAGs) via TikZ
• Terminology and theory comparison tables
• Case-study questions with structured analysis

Example E: Engineering and Computer Science

• Pseudocode and actual code blocks
• Algorithms with step-by-step examples
• Circuit diagrams, flowcharts, system architecture diagrams
• Mathematical modeling and unit analysis

Example F: Medicine and Law

• Case-based questions with structured facts → analysis → conclusion
• Terminology tables with Latin/legal terms
• Diagnostic / legal reasoning flowcharts
• Citation of articles, clauses, or guidelines

Post-Compilation Checklist

1. Read PDF visually — check formula rendering, layout, page breaks
2. Check for missing Unicode — any squares/blocks mean unconverted subscripts or logic symbols
3. Check math mode — Cu$^{2+}$ not Cu$^2$$^+$, CO$_3^{2-}$ not CO$_3$$^{2-}$
4. Check Chinese in math — wrap with \text{...} when necessary
5. Check tables — all cells properly formatted
6. Check quotes — no straight ", only “”
7. Check page breaks — no --- horizontal rules in content
8. Recompile if needed

Common Pitfalls

Issue	Cause	Fix
Chemical formula shows as square	Unicode sub/sup not converted	Convert to $_n$ and $^n$
Cu$^2$$^+$ appears as two separate	Adjacent math modes	Merge: Cu$^{2+}$
CO$_3$^{2-}$ broken	Split sub/sup	Merge: CO$_3^{2-}$
Missing character in font latinmodern-math	Chinese in math mode	Wrap with \text{...}
Unicode ∧ / ∨ shows as square	Monospace font lacks glyph	Use \land / \lor
Table cells misaligned	Missing math mode delimiters	Ensure each cell's formula is complete
Chinese quotes look wrong	Using "	Use “”
Page break lines appear	--- in markdown	Remove all content ---
Widow/orphan lines	No penalty settings	Add \widowpenalty=10000 etc.
Text overflow on right	Long Chinese sentences	\emergencystretch=3em
\neg rendered as eg	Raw string \n replacement	Protect \neg before converting \n
\forall becomes form feed	Normal-string \f escape	Use raw strings for LaTeX

Output Directory Convention

• Place generated markdown and PDF under an output/ subdirectory of the current project folder.
• Use descriptive filenames in Chinese, e.g. 计量经济学_期末复习全书.pdf, 博弈论_模拟试卷_一.pdf.
• Keep intermediate markdown files for debugging: output/review.md, output/exam.md.

Complete Workflow Example

User request: "根据课件和复习资料生成人工智能期末复习全书，包含例题与解答，并输出 PDF。"

Agent steps:

1. List files in 课件/, 课件提取文本/, 复习资料/ to understand scope.
2. Determine chapters based on course syllabus (e.g., 绪论、知识表示、确定性推理、不确定性推理、搜索、机器学习、神经网络、附录).
3. Generate markdown with review-book frontmatter.
4. For each chapter:
   • Write theory summary in Chinese.
   • Add 1–2 examples per key concept using the question templates.
   • Ensure all math-mode Chinese uses \text{...}.
5. Convert literal \n to newlines safely (protect \neg, \neq, \newpage).
6. Compile with pandoc + xelatex.
7. Inspect log for Missing character warnings; fix any issues.
8. Report the generated PDF path to the user.