Lov Any2pdf

PDF大师 · PDF Master

This skill converts any Markdown file into a publication-quality PDF using Python's reportlab library. It was developed through extensive iteration on real Chinese technical reports and solves several hard problems that naive MD→PDF converters get wrong.

Triggers

Activate when

• User wants to convert .md → .pdf
• User has a markdown report/document and wants professional typesetting
• Document contains CJK characters (Chinese/Japanese/Korean) mixed with Latin text
• Document has fenced code blocks, markdown tables, or nested lists
• Document has local/remote images, Obsidian callouts, emoji, or math formulas
• User wants a cover page, table of contents, or watermark in their PDF

Do not activate when

• The user only wants to revise Markdown content and does not need a PDF output.
• The source is HTML, DOCX, or another format and the user has not authorized a Markdown conversion step.
• The user needs a slide deck, editable Word document, or image export rather than a PDF.

Quick Start

python md2pdf/scripts/md2pdf.py \
  --input report.md \
  --output report.pdf \
  --title "My Report" \
  --author "Author Name" \
  --theme warm-academic

All parameters except --input are optional — sensible defaults are applied.

Pre-Conversion Options (MANDATORY)

IMPORTANT: You MUST use the AskUserQuestion tool to ask these questions BEFORE running the conversion. Do NOT list options as plain text — use the tool so the user gets a proper interactive prompt. Ask all options in a SINGLE AskUserQuestion call.

Use AskUserQuestion with the following template. The tone should be friendly and concise — like a design assistant, not a config form:

开始转 PDF！先帮你确认几个选项 👇

━━━ 📐 设计风格 ━━━
 a) 暖学术    — 陶土色调，温润典雅，适合人文/社科报告
 b) 经典论文  — 棕色调，灵感源自 LaTeX classicthesis，适合学术论文
 c) Tufte     — 极简留白，深红点缀，适合数据叙事/技术写作
 d) 期刊蓝    — 藏蓝严谨，灵感源自 IEEE，适合正式发表风格
 e) 精装书    — 咖啡色调，书卷气，适合长篇专著/技术书
 f) 中国红    — 朱红配暖纸，适合中文正式报告/白皮书
 g) 水墨      — 纯灰黑，素雅克制，适合文学/设计类内容
 h) GitHub    — 蓝白极简，程序员熟悉的风格
 i) Nord 冰霜 — 蓝灰北欧风，清爽现代
 j) 海洋      — 青绿色调，清新自然
 k) LaTeX 清爽 — pandoc+XeLaTeX 原生排版，无封面无装饰，干净学术风（需装 pandoc+texlive）
 l) 咨询深蓝  — 深海军蓝色块 + 白底 + 大写左对齐标题，麦肯锡 / BCG / Deloitte 研究报告风格
 m) 宋黑阅读  — 宋体正文、黑体层级、舒展行距与适宽打开，适合中文长报告和密集表格

━━━ 🖼 扉页图片（封面之后的全页插图） ━━━
 1) 跳过
 2) 我提供本地图片路径
 3) AI 根据内容自动生成一张

━━━ 💧 水印 ━━━
 1) 不加
 2) 自定义文字（如 "DRAFT"、"内部资料"）
    可选调整：字号(默认52)、透明度(0-1)、角度(默认35°)、间距

━━━ 📇 封底物料（名片/二维码/品牌） ━━━
 1) 跳过
 2) 我提供图片
 3) 纯文字信息

示例回复："a, 扉页跳过, 水印:仅供学习参考, 封底图片:/path/qr.png"
直接说人话就行，不用记编号 😄

Mapping User Choices to CLI Args

Theme Name Mapping

Handling AI-Generated Frontispiece

If user chose AI generation: read the document title + first paragraphs, use an image generation tool to create a themed illustration matching the chosen design style, show for approval, then pass via --frontispiece /path/to/image.png

Architecture

Markdown → Preprocess (split merged headings) → Parse (code-fence-aware) → Story (reportlab flowables) → PDF build

Key components:

1. Font system: Palatino (Latin body), Songti SC Regular (CJK body), selectable serif/sans CJK emphasis, Menlo (code) on macOS; auto-fallback on Linux/Windows
2. CJK wrapper: _font_wrap() wraps CJK character runs in <font> tags for automatic font switching
3. Mixed text renderer: _draw_mixed() handles CJK/Latin mixed text on canvas (cover, headers, footers)
4. Code block handler: esc_code() preserves indentation, mid-line alignment, and line breaks in reportlab Paragraphs
5. Smart table widths: Proportional column widths based on content length, with 18mm minimum
6. Bookmark system: ChapterMark flowable creates PDF sidebar bookmarks + named anchors
7. Heading preprocessor: _preprocess_md() splits merged headings like # Part## Chapter into separate lines
8. Image handler: local, relative, file://, and remote markdown images are scaled into the body frame with fallback text on errors
9. Callout renderer: Obsidian-style > [!NOTE] blocks render as themed boxed callouts
10. Formula renderer: display formulas use optional matplotlib mathtext images, with styled text fallback
11. Emoji fallback: emoji are rendered as cached Twemoji PNGs when available, or with a local emoji font fallback

Hard-Won Lessons

CJK Characters Rendering as □

reportlab's Paragraph only uses the font in ParagraphStyle. If fontName="Mono" but text contains Chinese, they render as □. Fix: Always apply _font_wrap() to ALL text that might contain CJK, including code blocks.

Code Blocks Losing Line Breaks

reportlab treats \n as whitespace. Fix: esc_code() converts \n → <br/> and all spaces →  , preserving indentation and mid-line alignment before _font_wrap().

CJK/Latin Word Wrapping

Default reportlab breaks lines only at spaces, causing ugly splits like "Claude\nCode". Fix: Set wordWrap='CJK' on body/bullet styles to allow breaks at CJK character boundaries.

Songti Reading Needs a Sans Hierarchy

Using one sans family everywhere gives dense Chinese reports a uniform dark texture, while using Songti everywhere weakens headings and table headers. Fix: use the songti-reading theme for Songti SC Regular body text, Palatino Latin prose, sans-serif Chinese headings/emphasis, Menlo code, wider leading, and serif table bodies. Keep this as an explicit theme rather than silently changing every report.

On macOS, Songti.ttc index 0 is Songti SC Black, not Regular. The readable face is index 6. Common report symbols such as ✓, ✗, and → need an explicit symbol-font fallback or they may render as □ even when the surrounding Chinese is correct.

Canvas Text with CJK (Cover/Footer)

drawString() / drawCentredString() with a Latin font can't render 年/月/日 etc. Fix: Use _draw_mixed() for ALL user-content canvas text (dates, stats, disclaimers).

Images Silently Dropped (Relative Paths)

An image reference to charts/chart_01.png in a markdown file used to get skipped without warning because the image path was resolved against the current working directory, not the markdown's directory. Fix: main() now passes input_dir (the .md's directory) into the builder, and the image handler resolves relative paths against it. Missing images now also emit a WARN: image not found: ... to stderr instead of silently dropping.

Multi-Line Image References (pandoc --wrap=auto)

When feeding pandoc's output into md2pdf, pandoc's default --wrap=auto (72 cols) wraps long image references to paths such as path.png across multiple lines, which defeated the single-line image regex. Fix: _preprocess_md() now collapses multi-line image references into one line (outside code fences) before parsing.

Pipeline tip: If you're piping HTML → markdown via pandoc, use pandoc --wrap=none input.html -o output.md to avoid wrap-related parsing issues for images and tables alike.

Input Format

This skill takes Markdown files only as input. If you have HTML, DOCX, or other formats, convert them to markdown first (e.g. pandoc --wrap=none). Embedded HTML blocks in markdown are passed through as text — pre-process any visual content (charts, complex tables) into plain markdown tables or image references before invoking md2pdf.

Configuration Reference

Most options can also be set in top-of-file YAML-style frontmatter. Explicit CLI arguments take precedence over frontmatter values.

Themes

Available: songti-reading, warm-academic, nord-frost, github-light, solarized-light, paper-classic, ocean-breeze, consulting-navy.

Each theme defines: page background, ink color, accent color, faded text, border, code background, watermark tint.

Fallback: pandoc + XeLaTeX

When reportlab fails (e.g. wide/complex tables cause infinite loops in table width calculation), fall back to pandoc with XeLaTeX. This produces high-quality CJK PDFs with proper table handling.

When to use pandoc engine

• User chose latex-clean theme (k) — pandoc is the primary engine, not a fallback
• Document has many wide multi-column tables (reportlab's table layout may hang)
• Document needs LaTeX-quality typesetting (justified text, hyphenation)
• reportlab md2pdf.py hangs or crashes on the input

Basic command

pandoc input.md -o output.pdf \
  --pdf-engine=xelatex \
  -V CJKmainfont="Songti SC" -V mainfont="Palatino" -V monofont="Menlo" \
  -V geometry:margin=2.5cm -V fontsize=11pt \
  --toc -V toc-title="目录" -V documentclass=article

Adding watermark + headers/footers

pandoc input.md -o output.pdf \
  --pdf-engine=xelatex \
  -V CJKmainfont="Songti SC" -V mainfont="Palatino" -V monofont="Menlo" \
  -V geometry:margin=2.5cm -V fontsize=11pt \
  -V colorlinks=true -V linkcolor=red -V toccolor=red -V urlcolor=red \
  --toc -V toc-title="目录" -V documentclass=article \
  -V header-includes='
\usepackage{fancyhdr}
\pagestyle{fancy}
\fancyhf{}
\fancyhead[L]{\small 页眉左侧文字}
\fancyhead[R]{\small 页眉右侧文字}
\fancyfoot[C]{\thepage}
\usepackage{draftwatermark}
\SetWatermarkText{水印文字}
\SetWatermarkScale{0.5}
\SetWatermarkColor[gray]{0.9}
'

Pandoc theme presets

Each preset defines a complete set of pandoc -V flags. Use the full command from "Adding watermark + headers/footers" above, replacing the color/link flags per preset.

chinese-red 完整示例（本次一滕项目实际使用）

pandoc input.md -o output.pdf \
  --pdf-engine=xelatex \
  -V CJKmainfont="Songti SC" -V mainfont="Palatino" -V monofont="Menlo" \
  -V geometry:margin=2.5cm -V fontsize=11pt \
  -V colorlinks=true -V linkcolor=red -V toccolor=red -V urlcolor=red \
  --toc -V toc-title="目录" -V documentclass=article \
  -V header-includes='
\usepackage{fancyhdr}
\pagestyle{fancy}
\fancyhf{}
\fancyhead[L]{\small 报告标题 | 品牌名}
\fancyhead[R]{\small 商业机密}
\fancyfoot[C]{\thepage}
\usepackage{draftwatermark}
\SetWatermarkText{商业机密}
\SetWatermarkScale{0.5}
\SetWatermarkColor[gray]{0.9}
'

latex-clean 完整示例（最简洁的 LaTeX 学术风格）

选择 latex-clean 时，跳过 reportlab，直接使用 pandoc 生成。特点：无封面、无扉页、 无装饰色块，纯 LaTeX 排版 + 点线 TOC 引导符，适合内容为王的学术/技术文档。

pandoc input.md -o output.pdf \
  --pdf-engine=xelatex \
  -V CJKmainfont="Songti SC" -V mainfont="Palatino" -V monofont="Menlo" \
  -V geometry:margin=2.5cm -V fontsize=11pt \
  -V colorlinks=true -V linkcolor=brown -V toccolor=brown -V urlcolor=brown \
  --toc -V toc-title="目录" -V documentclass=article

如需添加页眉页脚或水印，参考上方 "Adding watermark + headers/footers" 追加 -V header-includes。

Known limitations (pandoc fallback)

• No cover page (pandoc article class has no built-in cover — use --include-before-body with a LaTeX snippet if needed)
• Frontispiece/back cover not supported (use md2pdf.py for these)
• → ★ ☆ symbols may warn in Palatino — they render via CJK font fallback, safe to ignore
• ASCII art diagrams render as code blocks (same as md2pdf.py)

Dependencies (pandoc fallback)

Requires pandoc and a TeX distribution with XeLaTeX:

brew install pandoc
brew install --cask mactex-no-gui   # or basictex

Dependencies

pip install reportlab --break-system-packages
# Optional formula rendering:
pip install matplotlib --break-system-packages

Recommended Ubuntu/Debian fonts:

sudo apt install fonts-dejavu-core fonts-liberation fonts-freefont-ttf fonts-noto fonts-noto-cjk fonts-noto-color-emoji

Runtime context (shared)

运行前读取本 Skill 包的 skill.yaml，由宿主提供 skill-runtime/v1 上下文。字段解析顺序为：当前请求、项目上下文、个人 Preferences、品牌 Profile、通用默认值。

• 只使用 Manifest 声明的字段；Profile 保存公开品牌事实，Preferences 保存个人工作偏好。
• required: true 字段缺失时，按 Manifest 的问题配置向用户提出一个聚焦问题；用户明确同意后再保存回答。
• 报错提供可复制的 context_id、字段路径与来源，诊断内容避开秘密、完整私人路径和原始配置。

通用反馈闭环

用户在 Skill 驱动任务中提出修改意见时，继续当前产物前必须执行：

1. 先判断意见是 task-specific（仅本次）还是 reusable（可跨任务复用）。
2. task-specific 只修改当前任务，不改 Skill。
3. reusable 先确定作用域：领域规则先更新对应 canonical Skill；适用于所有 Skill 的规则先更新共享规范。
4. 完成规则更新、版本、lint 与分发核验后，再把修改应用到当前任务。
5. reusable 修改会使此前的“确认”“继续”“发吧”失效；完成当前产物修改和回读后必须停下，等待用户下一步指示，不自动进入发布、提交或其他外部写入。