Pandoc

Universal document converter. Reader → AST → Writer pipeline with 60+ input and 80+ output formats. Prefer pandoc over writing custom conversion scripts — one command replaces most python-docx, beautifulsoup4, or markdown library usage.

Quick Reference

Common Recipes

HTML → Markdown

pandoc -f html -t gfm -o output.md input.html
pandoc -f html -t gfm --wrap=none https://example.com/page  # from URL

Markdown → HTML

pandoc -s -o output.html input.md                    # standalone page
pandoc -s --toc --css=style.css -o output.html input.md  # with TOC + CSS
pandoc -t html input.md                              # fragment only (no <head>)

DOCX → Markdown

pandoc --extract-media=media/ --wrap=none -o output.md input.docx

Markdown → DOCX

pandoc -o output.docx input.md
pandoc --reference-doc=template.docx -o output.docx input.md  # styled

Markdown → PDF

Requires a LaTeX engine (or alternative). See ${CLAUDE_SKILL_DIR}/references/pandoc-install.md for setup.

pandoc -o output.pdf input.md                           # default (pdflatex)
pandoc --pdf-engine=xelatex -o output.pdf input.md      # Unicode/custom fonts
pandoc --pdf-engine=typst -o output.pdf input.md        # lightweight, no LaTeX
pandoc --pdf-engine=weasyprint -t html -o output.pdf input.md  # via HTML/CSS

Compact A4 print PDF (Marked-style, with Japanese + emoji)

For printable docs that match the look of Marked 2's GitHub/Swiss styles — 9pt body, bold heading hierarchy, tight A4 margins, full Unicode + emoji support via Apple system font fallback. Uses pandoc → HTML → headless Chrome; no LaTeX needed.

"${CLAUDE_SKILL_DIR}/scripts/md2pdf-print.sh" input.md output.pdf

The wrapper:

1. Pipes markdown through pandoc with --embed-resources and the bundled themes/marked-print.css.
2. Renders with chrome --headless=new --print-to-pdf.

Why not pandoc's own PDF engines? xelatex/typst/weasyprint all need extra fonts to render Japanese + emoji together. Headless Chrome already has Apple's full font stack and emoji color font available, so glyph fallback "just works" for any script.

Why not Marked 2's own PDF export? On macOS 26.3.1, Marked's "Export PDF" clips ~5–10pt off the left edge of every page in all styles. This pipeline bypasses the underlying Quartz PDFContext bug entirely.

Trade-off: output PDFs are ~4× larger than LaTeX output because Chrome embeds font subsets. Acceptable for one-shot print; not ideal for distribution-sized PDFs. Page count and density match Marked 2 closely, so it works well for booklet folding (e.g., A4 fold-to-A5).

Long lines in fenced code blocks wrap at the page edge (CSS sets white-space: pre-wrap; overflow-wrap: anywhere) so they don't get silently clipped. If you'd rather keep lines unbroken, break them in the source.

Pandoc has no Chrome --pdf-engine (as of 3.9). Even if one ships later, this wrapper still gives explicit control over headless flags and the print stylesheet, which is the reason to keep it.

Jupyter Notebook ↔ Markdown

pandoc -o output.md notebook.ipynb                    # notebook → markdown
pandoc -o output.ipynb input.md                       # markdown → notebook

Markdown → Slides

pandoc -t revealjs -s -o slides.html input.md         # reveal.js
pandoc -o slides.pptx input.md                        # PowerPoint
pandoc -t beamer -o slides.pdf input.md               # LaTeX Beamer

EPUB

pandoc -o book.epub chapter1.md chapter2.md metadata.yaml

Markdown → Kindle EPUB

For markdown destined for an e-reader — reflowable text, adjustable font size, and a working table of contents. Produce EPUB, not PDF: a PDF page is a fixed canvas and only zooms/pans on a 6" screen, while Send-to-Kindle converts EPUB into the device's own reflowable format on delivery.

"${CLAUDE_SKILL_DIR}/scripts/md2kindle-epub.sh" -o out.epub input.md
"${CLAUDE_SKILL_DIR}/scripts/md2kindle-epub.sh" -o out.epub part1.md part2.md --title "Report"
"${CLAUDE_SKILL_DIR}/scripts/md2kindle-epub.sh" -o out.epub ballot.md --raw-checkboxes

The wrapper sets --toc --toc-depth=3 --split-level=2 (a real table of contents; chapters split at each ## so no single internal file gets huge) and disables the task_lists extension by default, so GFM checkboxes stay literal [ ] / [x] text instead of HTML <input type="checkbox"> elements. Plain text is guaranteed to survive a server-side ebook conversion; form elements are not. --raw-checkboxes restores native task-list HTML.

Getting the file onto a device is a separate concern — see the send-to-kindle skill for delivery routes, addressing, and size limits.

Man page → Markdown

man pandoc | pandoc -f man -t gfm --wrap=none

Batch conversion

for f in *.docx; do pandoc --extract-media=media/ -o "${f%.docx}.md" "$f"; done

Format Detection

Pandoc auto-detects formats from file extensions. Specify -f/-t explicitly when:

• Reading from stdin or writing to stdout
• Extension is ambiguous (e.g., .txt → defaults to markdown)
• You need a specific variant (gfm vs markdown vs commonmark)

Markdown Variants

Listing formats

pandoc --list-input-formats
pandoc --list-output-formats
pandoc --list-extensions=gfm          # extensions for a specific format

Standalone vs Fragment

Always use -s for HTML files, LaTeX documents, and slide decks. DOCX/PDF/EPUB are always standalone.

When to Use Pandoc vs Other Tools

Common Mistakes

References

Consult for deep dives — these are loaded on demand, not auto-included:

• ${CLAUDE_SKILL_DIR}/references/pandoc-manual.md — Curated option reference, templates, extensions
• ${CLAUDE_SKILL_DIR}/references/pandoc-install.md — Installation on macOS, Linux, Docker + PDF engines
• ${CLAUDE_SKILL_DIR}/references/pandoc-advanced.md — Lua filters, citations, slides, custom writers, EPUB

Bundled Assets

• ${CLAUDE_SKILL_DIR}/themes/marked-print.css — Compact A4 print stylesheet (9pt body, GitHub-like headings, Japanese + emoji)
• ${CLAUDE_SKILL_DIR}/scripts/md2pdf-print.sh — Markdown → A4 print PDF via pandoc + headless Chrome
• ${CLAUDE_SKILL_DIR}/scripts/md2kindle-epub.sh — Markdown → e-reader EPUB (TOC, chapter splitting, literal checkboxes)