High-Quality PPT Generation Engine

Problem

The hard part of slide generation is not placing text on a canvas. It is constraining audience, narrative, data relationships, and visual hierarchy at the same time. Many automated tools produce decks that are technically valid but weak in communication: text-heavy pages, charts colliding with content, tables overlapping images, and late manual rework.

How It Works

This skill separates design from rendering:
- Requirement parsing: extracts topic, audience, page count, style, and compliance constraints, then builds a pyramid-style outline.
- Expression planning: each page declares an expression value, following chart/diagram > table > text to choose chart, diagram, table, or text, with D0-D8 used to justify downgrades.
- Page design draft: writes slides_plan.json with region coordinates, sizes, image placement, and bottom bands, so collisions between table, chart, and image are detected before code generation.
- JS rendering: uses PptxGenJS, icon assets, native charts, and generated images to output an editable .pptx deck with speaker notes.

Boundaries

It fits technology reports, consulting-style presentations, and product launches where structure and visual quality matter. For re-skinning an existing deck, use the existing-PPT optimization path. The skill avoids fake data; the host AI must supply real content. The Python pipeline favors staged confirmation and is visually weaker than the JS path.