Branded pptx

Skill dzivkovi/agent-skills-on-databricks/skills/branded-pptx

Bring your Claude Code skill unchanged and run it as a governed Databricks job. Publish once to a Unity Catalog volume, reuse from any job, chain skills into a pipeline (markdown in, branded PowerPoint out). No external API key. Runs on Free Edition.

Install

npx -y skills add dzivkovi/agent-skills-on-databricks --skill branded-pptx

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

22 days oldThe repository was created 22 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Turn a markdown document into a branded PowerPoint (.pptx) deck: a title slide from the H1, one content slide per H2, bullets from list items, with a light brand skin (accent colour + footer). Pure-Python (python-pptx), runs on Databricks Free Edition serverless - no Node, no LibreOffice. Use when you need a real .pptx from markdown without a design toolchain. Triggers: 'make a deck', 'markdown to powerpoint', 'branded pptx', 'slides from this doc', 'pptx from markdown'.

SKILL.md

2.8 KB, as published. Nobody here has run it

Branded PPTX (MVP-2)

The MVP-2 rung of the branded-pptx ladder. It re-cuts the faithful branded-pptx skill in pure Python so it runs on Free Edition serverless. The faithful engine (pptxgenjs on Node + LibreOffice

Poppler, with a multimodal vision-in-the-loop QA cycle) needs Node and system packages that serverless does not have; MVP-4 restores it on paid classic compute. MVP-2 trades that fidelity for portability: a deterministic markdown-to-slides map with a light brand skin, no vision QA.

How to run this skill

Deterministic step - run scripts/build_pptx.py <input.md> [-o out.pptx] [--brand coral]. It parses the markdown structurally and writes a real .pptx:
- the first # H1 becomes the title slide (the first paragraph after it is the subtitle);
- each ## H2 becomes a content slide titled with that heading;
- - / * list items (and plain paragraphs) under a heading become the slide's bullets. The build is a pure function of the markdown - same input, same deck. No LLM.
Optional non-deterministic step - before building, an LLM may restructure or condense a long document into the H1/H2/bullets shape this skill expects (outline the deck). That step is advisory and out of scope for MVP-2; the deterministic builder is the contract.

Output contract

A valid .pptx (openable by PowerPoint / Keynote / Google Slides / python-pptx) with:

one title slide carrying the H1 (accent-coloured) and the subtitle;
one content slide per H2, each titled (accent-coloured) with its bullets;
a small brand footer when the chosen brand defines one.

Brands are a skin only (accent colour + footer label), defined in BRANDS in the build script - add one there without touching the layout code. This is intentionally low fidelity: it is a correct, serverless-safe deck, not a pixel-perfect brand system (that is MVP-4's job).

Why this rung exists

An LLM cannot emit a binary .pptx, and the faithful toolchain cannot run on serverless. So the deterministic Python builder is the only half that can produce the artifact here - and being deterministic, it is fully testable (build a deck, reopen it, assert the slides) without a vision loop. That honesty - shipping the rung that actually runs, and labelling its fidelity - is the point.