cubeplexai/cubeplex/pptx

pptx — research, design, and build self-checked decks

Don't jump straight to code. A good deck is content first, design second, mechanics last. Follow the pipeline; the depth is in the reference docs — read the ones each step points to.

<skill> below = the absolute path load_skill returned for this skill. The helper library is <skill>/scripts/deckbuilder.py; the checker is <skill>/scripts/check_deck.py; reference docs live under <skill>/references/.

Pipeline

1. Understand the request & inputs. Read every uploaded file in full first.

Identify: language, audience, the occasion, any explicit page count, and whether the user gave you a finished document, an outline, or just a topic.

1. Choose the content mode — decides where slide content comes from.

See references/content.md.

• Summarize — user gave a complete document → distill it (don't invent).
• Outline — user gave structure/points → follow it, enrich each page.
• Research — user gave only a topic → research the web first.

1. Research (if Research mode, or to supplement). Two passes: broad (map the

space, tell the user the directions you'll dig into) → deep (read sources, extract real data). Attribute key facts. See references/content.md → Research. Tools: the browser skill or curl in the sandbox.

1. Write the outline to a file (/workspace/<deck>/outline.md): a typed

page sequence (cover / toc / section / content / closing), each with a conclusion-style title ("Q3 revenue up 23%", not "Results") and one core message. Default 8–16 pages; obey an explicit count. See references/content.md.

1. Design plan. Pick one theme for the whole deck (references/themes.md

catalog → read the chosen references/themes/<name>.md). Then map each outline page to a builder archetype and decide where images, charts, or tables earn their place. See references/design.md, references/layout.md, references/images.md.

1. Build with deckbuilder.py. Write one script that imports Deck, calls

the archetype methods in order, and saves to /workspace/<deck>/deck.pptx. Run it with the sandbox python (python-pptx / pillow / matplotlib are installed). API + composition rules: references/layout.md. Fonts: references/fonts.md.

1. Self-check — mandatory. Run the checker and fix every ERROR, then

re-run until it prints 0 error(s):

``bash python3 <skill>/scripts/check_deck.py /workspace/<deck>/deck.pptx ``

It flags overflow / off-canvas / low-contrast with real font metrics. Fix overflow by tightening copy or splitting the slide — never by shrinking below the builder's sizes. Read the full output; don't grep past it.

1. Deliver the .pptx as an artifact with a one-paragraph summary of the

deck's structure and key points. Then stop.

The archetypes (build vocabulary)

Deck(theme=...) then, per slide: cover · hero (full-bleed image) · section (divider) · agenda · statement (big idea) · cards (2-col) · steps (numbered process) · comparison (table) · metrics (big numbers) · chart (native column) · image_split (image + text) · closing. Signatures and density limits: references/layout.md.

Reference docs (read on demand)

Non-negotiables

• Titles are conclusions, not labels. One core idea per slide.
• Sparse content needs *more* design, not a near-empty slide — split it, use

a statement/section, or enrich with a chart/visual.

• One theme for the whole deck; don't hand-tune colors/fonts per slide —

extend deckbuilder.py if you need a new archetype.

• Never deliver without a clean checker run. No placeholder text, no empty

slides, no [TODO].

• Output is standard editable .pptx (PowerPoint / WPS / LibreOffice). This

skill creates decks; it does not read or convert existing .pptx.