marp-to-branded-pptx

name	marp-to-branded-pptx
description	Convert a Marp markdown deck into a fully editable, on-brand PowerPoint (.pptx) built on a Google Slides template's native layouts, carrying speaker notes across, ready to upload to Google Slides. Use when someone wants a Marp deck "in Google Slides", "in the company template", or "as an editable branded deck" rather than rasterized slide images.
allowed-tools	Bash Read Edit Write

Marp deck to branded, editable Google Slides pptx

Turns a Marp presentation.md into a .pptx that reuses a Google Slides template's layouts, theme, and backgrounds, with real editable text and the Marp speaker notes preserved. Import the result into Google Slides (Drive upload, then "Open with Google Slides").

This is different from marp --pptx, which bakes each slide into an image, and from marp --pptx-editable, which needs LibreOffice and drops speaker notes. Here the slides are generated natively with python-pptx, so they inherit the template's design and stay fully editable.

Reference implementation

A working generator is bundled with this skill at build_gslides_pptx.py (same directory as this SKILL.md). Reuse or adapt it:

./build_gslides_pptx.py \
  --md path/to/presentation.md \
  --template path/to/brand-template.pptx \
  --out path/to/output.pptx

It runs via a uv shebang (uv run --with python-pptx --with lxml), so dependencies install automatically. Use --drop-cover to omit the template's branded cover slide.

How it works (adapt these when the template differs)

1. Parse presentation.md. Split on \n---\n. For each slide pull: the <!-- _class: X --> directive, the speaker-notes comment (the comment block that does not start with _), the headings (#/## and > become the title), and the body (paragraphs, <ul>/<li>, <p class="fine">). Convert inline **bold**, <strong>, *italic*, <em>, <br> to styled runs.
2. Pick template layouts by rendered appearance, not by name. Google Slides export layout names are duplicated junk (MAIN_POINT_1_1_1_1_...). Render one slide per layout to PDF/PNG and look. Grab the content layout from a sample content slide in the template (prs.slides[i].slide_layout), and find the title/section layouts by the clean names that do exist (TITLE, SECTION_HEADER).
3. Map _class to layouts. In this deck: title to the dark TITLE layout, section/spine/quote to the dark SECTION_HEADER layout, everything else to the white headline+body content layout.
4. Fix placeholder collisions. Template title/subtitle boxes often assume a one-line title; reposition and enable MSO_AUTO_SIZE.TEXT_TO_FIT_SHAPE so multi-line titles do not overlap the subtitle.
5. Emphasis color. Map **bold** to the template's accent color (read it from ppt/theme/theme1.xml, e.g. accent4) so emphasis reads as brand, not just bold.
6. Delete the template's sample slides (keep the layouts/masters) via the _sldIdLst XML; optionally keep the branded cover.
7. Carry notes with slide.notes_slide.notes_text_frame.text.

Verifying the output

Render to a contact sheet and eyeball every slide before declaring done:

soffice --headless --convert-to pdf --outdir /tmp out.pptx   # needs LibreOffice
pdftoppm -png -r 42 /tmp/out.pdf /tmp/out                     # needs poppler
montage /tmp/out-*.png -tile 4x -geometry +3+3 /tmp/grid.png  # needs imagemagick

Read grid.png and check for overlaps, overflow, and empty placeholders. Confirm notes and editability by reopening the .pptx with python-pptx.