Markdown to polished PDF: 3 Claude skill paths (with proper Chinese fonts)

Generating a .pdf from Markdown sounds trivial until you discover three problems: CJK fonts render as squares, page breaks land mid-paragraph, and embedded code blocks lose their highlighting. Most pandoc one-liners fail on at least one of these. The Claude skill ecosystem has three paths that each solve a subset.

This post compares them and shows when to pick which.

Path 1: pdf-creator (WeasyPrint, CJK-aware)

pdf-creator by @daymade is the most “front-end” of the three — it uses WeasyPrint, which converts HTML/CSS to PDF. The CSS engine handles Chinese fonts, layout, and pagination better than pandoc’s LaTeX path.

npx degit daymade/claude-code-skills/pdf-creator ~/.claude/skills/pdf-creator

Use it when:

• Source is Markdown with mixed Chinese / English / code.
• You want proper Chinese font rendering without manually configuring font fallback chains.
• You need modern CSS print features (page-break-after, named pages, running headers).

Example pipeline:

Use pdf-creator to render this research note (Chinese body + English code blocks) into a print-ready A4 PDF. Use a serif body font and ensure CJK glyphs render correctly. Save to ./output/notes.pdf.

Path 2: anthropic-pdf (manipulation, not generation)

anthropic-pdf is the inverse of path 1 — it’s for working with PDFs that already exist, not generating new ones from Markdown.

npx degit anthropics/skills/skills/pdf ~/.claude/skills/pdf

Use it when:

• You already have a PDF and need to merge / split / watermark / OCR / encrypt.
• You’re building a pipeline that consumes PDFs (extract tables, extract text, fill form fields).
• You need to round-trip a PDF (read → modify → write).

It’s not a Markdown → PDF tool. But chained after pdf-creator, it’s the post-processor: generate with pdf-creator, watermark/merge with anthropic-pdf.

Example pipeline:

Step 1: Use pdf-creator to convert these three Markdown chapters to PDFs. Step 2: Use anthropic-pdf to merge them, add a “DRAFT” watermark to every page, and add page numbers in the footer.

Path 3: anthropic-docx + Word’s PDF export

anthropic-docx generates a .docx first, which you can then export to PDF either programmatically (LibreOffice headless, Microsoft Word automation) or by hand.

npx degit anthropics/skills/skills/docx ~/.claude/skills/docx

Use it when:

• The deliverable is a Word document first (legal letter, consulting report) and PDF is just a distribution format.
• You need track changes / comments in the source document, not just the PDF.
• The recipient might want to edit the source and they live in Microsoft Word.
• You need a letterhead (Word handles this more cleanly than HTML/CSS).

Example pipeline:

Use anthropic-docx to draft a 6-page consulting report from these notes. Cover page, ToC, three sections with subheadings, embedded chart, custom letterhead. Then I’ll export to PDF in Word.

Side-by-side comparison

Need	Path
Chinese / CJK content rendering correctly	pdf-creator
Print-quality long-form document	pdf-creator
Modify an existing PDF	anthropic-pdf
Letterhead, comments, track changes	anthropic-docx
Output is consumed in Word	anthropic-docx
Need to fill PDF forms	anthropic-pdf
OCR a scanned PDF	anthropic-pdf
Just want HTML → PDF, English content	pdf-creator (still wins on layout)

Common pitfalls

Pandoc + LaTeX is not on this list for a reason. It works for English academic papers but breaks badly on Chinese — XeLaTeX font config is notoriously fragile. If your team has a working pandoc + xeCJK pipeline, keep it. If you’re starting fresh, pdf-creator skips this entire mess.

LibreOffice headless conversion (soffice --convert-to pdf) is unreliable for complex layouts but works fine for simple .docx → PDF. Pair it with anthropic-docx for the doc-first workflow.

@page CSS selectors are pdf-creator’s secret weapon. Most Markdown → PDF pipelines treat the document as continuous flow. WeasyPrint respects named pages, running headers/footers, page-break-before, etc. — which is why the output looks like a real document instead of a long screenshot.

Composing all three

For a research report distributed bilingually:

1. anthropic-docx — draft in Word for editorial review.
2. Export .docx → PDF with LibreOffice headless.
3. anthropic-pdf — add watermark, page numbers, encrypt.

For a personal notes archive (Chinese-heavy):

1. Write Markdown.
2. pdf-creator — convert to print-ready PDF with CJK fonts.
3. anthropic-pdf — merge multiple notes into a quarterly archive.

For a public-facing report (English):

1. Markdown drafts.
2. pdf-creator — render with brand-styled CSS.
3. anthropic-pdf — encrypt, add metadata.

Browse all PDF-related skills →