From Text to Editable PPTX: An Agent Skill for Decks

Where this fits

A product manager drops a 30-page source doc into the agent chat: 'Turn this into a deck for Thursday.' pptwise gives the agent a deterministic path from that source to a native .pptx, with outlines and notes converted into typed slide components. The agent owns the narrative; the skill owns the geometry and theme, so the output is editable and consistent instead of a flat image export.

Why agents benefit

• Agents can produce a real PowerPoint file, not a PDF or HTML mimic, using DrawingML shapes and text that remain editable in PowerPoint.
• The one-way chain (intent -> narrative -> theme binding -> spec with kind -> fill -> render) prevents the model from redesigning layout at render time, reducing nondeterministic decks.
• Semantic JSON specs let the LLM describe page intent and components, while the engine handles visual decisions, which keeps token usage low and structure inspectable.
• The CLI is simple to invoke from a tool call and returns a deterministic exit code, so agents can retry on failure instead of guessing whether the artifact was created.

Practical scenarios

Quarterly board update from a research memo

An agent reads notes, builds an outline, then writes a JSON spec with cover, metrics, and risk pages. pptwise outputs a branded .pptx that the finance team can edit without asking engineering.

Sales deck from public source material

With a competitor's blog as source, the agent distills comparisons into typed component pages, keeping all claims editable for legal review.

Course lecture from an existing outline

An educator provides a bullet outline. The agent maps each section to content layouts, producing a theme-consistent deck that can be reordered and restyled later.

Add it to your agent workflow

Install the skill in an agent environment:

npx skills add liustack/pptwise --skill pptwise

Then create a minimal JSON spec:

{
  "theme": "corporate",
  "pages": [
    { "kind": "cover", "title": "Q3 Review" },
    { "kind": "bullets", "title": "Summary", "items": ["Growth up", "Costs down"] }
  ]
}

Ask the agent to run the bundled launcher (bash <skill-dir>/scripts/run.sh spec.json) and hand the result to the user. Because the output is native PPTX, it can be dropped into PowerPoint, Google Slides, or SharePoint without conversion. For more context on how this skill fits the broader agent ecosystem, see https://www.openagentskill.com/skills/liustack-pptwise.

Compare before adopting

Before wiring pptwise in, compare it with alternative deck paths: an HTML-to-PPTX converter, a Python library like python-pptx, or a dedicated design service. Look at quality signals such as whether text remains editable, whether theme tokens are consistently applied across layouts, and whether failure modes are clear from exit codes. Check maintenance freshness: is the CLI updated, are issues answered, and how young is the repo? Also judge workflow fit: can your agent reliably write semantic JSON, and do you need deterministic local rendering or asynchronous presentation generation?

Why it is worth tracking

The skill is early and has just ten stars, so adopt only after a small experiment validates it for your content types. Its quality signals are good: native DrawingML, theme binding, and a one-way chain that keeps LLMs from overdesigning. If it stays maintained and the repo gains community momentum, pptwise could become a default tool for any agent that needs editable deck output; for now, it is worth tracking and evaluating on a single, low-stakes deck.