Pandoc

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

| Flag | Purpose | |------|---------| | -f FORMAT | Input format (auto-detected from file extension) | | -t FORMAT | Output format (auto-detected from extension) | | -o FILE | Output file (stdout if omitted) | | -s | Standalone — complete document with header/footer | | --wrap=none | Don't rewrap lines (preserve original line breaks) | | --extract-media=DIR | Extract images from DOCX/EPUB/ODT | | --toc | Generate table of contents | | --number-sections | Number section headings | | --pdf-engine=ENGINE | PDF backend (default: pdflatex) | | --reference-doc=FILE | Style template for DOCX/ODT/PPTX output | | --template=FILE | Custom output template | | --embed-resources | Embed images/CSS inline (HTML) | | -V KEY=VALUE | Set template variable | | -L SCRIPT | Apply Lua filter | | --shift-heading-level-by=N | Adjust heading levels | | --columns=N | Line wrap width (default 72) |

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 )

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.

1. 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

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

| Format | Use for | |--------|---------| | gfm | GitHub — tables, task lists, strikethrough, autolinks | | commonmark | Strict CommonMark spec | | commonmark_x | CommonMark + pandoc extensions | | markdown | Pandoc's Markdown — most features, default | | markdown_strict | Original Gruber Markdown — minimal |

Listing formats

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

Standalone vs Fragment

| Mode | Flag | Output | Use when | |------|------|--------|----------| | Fragment | (default) | Body content only | Embedding in another document | | Standalone | -s | Complete document with headers | Creating a valid file (HTML, LaTeX, etc.) |

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

When to Use Pandoc vs Other Tools

| Task | Tool | Why | |------|------|-----| | Document format conversion | pandoc | Built for this — one command | | Clean HTML → Markdown | pandoc | Handles structure well | | Complex web scraping | Dedicated scraper | Pandoc needs clean HTML input | | PDF text extraction | pdftotext, pdfplumber | Pandoc cannot read PDF | | Image format conversion | ImageMagick, sips | Not pandoc's domain | | CSV/JSON data processing | jq, csvkit, code | Pandoc reads CSV/JSON but as documents | | Markdown rendering in terminal | glow, pandoc -t ansi | Either works | | Office doc creation (complex) | python-docx, openpyxl | When pandoc's model is too simple |

Common Mistakes

| Mistake | Fix | |---------|-----| | Writing a Python script for DOCX → MD | Use pandoc --extract-media=media/ -o out.md in.docx | | Forgetting -s for standalone HTML | Add -s when output needs ` and | | PDF fails — no LaTeX installed | Install texlive/mactex, or use --pdf-engine=typst or weasyprint | | Losing images from DOCX | Add --extract-media=media/ | | Wrong markdown flavor in output | Specify -t gfm or -t commonmark explicitly | | Piping binary formats to stdout | Use -o file.docx — DOCX/PDF/EPUB must write to files | | Line wrapping mangles output | Add --wrap=none` to preserve original line breaks |

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

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: eins78
• Source: eins78/agent-skills
• License: MIT

Install and usage instructions live in the source repository linked above.