Markdown To Pdf

Markdown → PDF (pandoc + headless Chrome)

Problem

Converting Markdown to PDF through pandoc's default LaTeX route breaks on real-world GitHub-flavored documents: Unicode box-drawing characters in code blocks (├──) crash pdflatex, wide GFM tables overflow the page, remote images need manual download, and the output looks nothing like the rendered Markdown the author reviewed. Naive HTML routes drop images or paste them at native resolution so they spill off the page.

Context / Trigger Conditions

• User asks to save/convert/export a .md file as PDF.
• The document carries images (hosted PNGs, local figures), GFM tables,

fenced code blocks, blockquotes, or #-anchor TOC links.

• Requirements like "all images must be preserved" or "scaled appropriately".

Solution

Render the Markdown to HTML exactly as a browser would, then print it.

One command (preferred):

python ~/.claude/skills/markdown-to-pdf/scripts/md2pdf.py  [-o out.pdf]

The script: (1) runs pandoc -f gfm -t html5 --standalone --embed-resources with assets/print.css, executed from the Markdown's own directory so relative image paths resolve, and with remote images fetched and inlined as data URIs; (2) prints with headless Chrome (Edge/Chromium fallback) using --no-pdf-header-footer --virtual-time-budget=30000; (3) verifies with pypdf that the embedded raster-image count is at least the number of image references in the source, printing each image's page and dimensions. Exit code 2 flags a missing-image warning.

Manual equivalent (what the script automates):

pandoc input.md -f gfm -t html5 -s --embed-resources -c print.css \
    --metadata pagetitle="input" -o /tmp/input.html
"/c/Program Files/Google/Chrome/Application/chrome.exe" --headless \
    --disable-gpu --no-pdf-header-footer --virtual-time-budget=30000 \
    --print-to-pdf="out.pdf" "file:///tmp/input.html"

Why this works for images. The stylesheet constrains every image with max-width: 100%; max-height: 240mm; width/height: auto; break-inside: avoid, so each image scales to the printable width, a very tall diagram still fits a single page, aspect ratio is preserved, and no image is split across a page break. --virtual-time-budget makes Chrome wait for image loading before printing; --embed-resources makes the HTML self-contained so nothing depends on the network at print time.

Verification

1. The script's pypdf pass: page count, plus one line per embedded image with

its page number and pixel dimensions; compare against the source's image count.

1. Visual spot-check: the Read tool renders PDF pages as images — read the

pages the verifier listed and confirm each diagram is legible and fits the page.

Example

style-emulation/README.md (~800 lines: 4 hosted PNG diagrams up to 1410×2998 px, two request/command tables, a box-drawing module map, a linked TOC) → 19-page, 1.35 MB PDF; all 4 images embedded on pages 6/7/9/10, each scaled to one page; tables bordered with zebra striping; code blocks styled; TOC links live.

Notes

• Page-break trade-off: break-inside: avoid keeps images whole, so an

image that does not fit the space left on a page pushes to the next page, leaving white space behind. Acceptable for reports; mention it to the user.

• TOC anchors: pandoc's GFM reader auto-generates GitHub-style heading

ids, so existing (#section-name) links keep working in the PDF.

• SVG images may be drawn as vector art rather than raster XObjects; the

pypdf count then under-reports and the warning is a false positive — check visually.

• Same image referenced twice can be embedded once; the embedded for a per-job override (e.g. letter,

landscape).

• --no-embed reverts to link-only HTML with Chrome fetching images at

print time — useful if pandoc cannot reach a host that Chrome can.

• If headless Chrome ever hangs on a remote-image page, raise --budget.
• Locked targets (cloud sync / open viewers): Chrome can fail to

overwrite a PDF in a Dropbox/OneDrive-synced folder — or one open in Acrobat — while still exiting 0, silently leaving the stale file in place (the byte-identical output size is the tell). The script therefore prints to its temp dir and moves the result over the target with retries; if the target stays locked (e.g. the user has the old PDF open), it saves to .new.pdf and says so. The pypdf verification step is what catches this class of failure — keep it on.

Related skills

• pdf — probing, extracting, manipulating existing PDFs (and reportlab

generation from scratch). Markdown sources route here instead.

• tex2docx, manuscript-editing-template-latex — LaTeX/docx conversion

pipelines; use those for .tex/.docx, not this skill.

References

• pandoc manual, --embed-resources / GFM reader: https://pandoc.org/MANUAL.html
• Chrome headless print-to-pdf flags: https://developer.chrome.com/docs/chromium/headless

Source & license

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

• Author: kennethkhoocy
• Source: kennethkhoocy/legal-scholarship-skills
• License: MIT

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