PPTWork has two portable Anthropic-style Skills for AI agents to plan, author, and export PowerPoint decks. Both work in OpenCode, Claude Code, Cursor, Codex, or any other host that honors the SKILL.md frontmatter convention.
| Skill | Trigger | What it does |
|---|---|---|
ppt/ |
"build a deck", "make a presentation", "export to PowerPoint" | End-to-end deck pipeline: clarify → plan → author → screenshot → export |
ppt-html-authoring/ |
"design a slide", "draft this page", "rewrite this layout" | Single-page authoring: writes design.md + a self-contained slide.html |
The two skills compose: the deck pipeline delegates each page to the authoring skill. The authoring skill is also useful standalone when the user just wants one slide.
The deck skill ships Node scripts that the agent calls via bash. They replace what would otherwise be runtime tools, so the same skill works in hosts that don't support custom MCP tooling.
| Script | Purpose |
|---|---|
deck.mjs |
Deck CRUD (init / init-slide / move / delete / path / list) |
screenshot.mjs |
Headless render of slide.html → thumbnail.png |
export.mjs |
PPTX export — default editable (html2pptx-pro), opt-in --mode raster |
| Library | Role |
|---|---|
| playwright-core | Headless Chromium/Edge rendering of slide.html, screenshots |
| html2pptx-pro | Editable PPTX (HTML structure mapping) — default export engine |
| dom-to-pptx | DOM → slide elements — backup editable engine for diagnostics |
| pptxgenjs | PPTX file writing, raster export |
Every slide is just a self-contained HTML file. Authoring, revisions, and layout fixes happen in the HTML. The export is one-way:
edit slide.html → screenshot.mjs to verify → export.mjs to ship .pptx
That's it. Don't try to edit the .pptx and round-trip changes back into HTML — that path doesn't exist by design.
export.mjs <deck> loads each slide.html in headless Chromium, injects
html2pptx-pro, and maps the
live DOM into editable PowerPoint objects (text frames, lists, tables, shapes,
charts). This is the main path when the recipient needs to revise content
inside PowerPoint.
It is still an HTML-to-PPTX conversion — always open the result and verify
fonts, layout, and object ordering. If anything drifts, re-export with
--mode raster (pixel-faithful fallback) or, for one-slide diagnostics only,
--editable-engine dom-to-pptx.
Author editable decks with simple, explicit DOM structure; see
ppt/references/editable-html-rules.md.
export.mjs <deck> --mode raster captures each slide as a single full-page
PNG (at deviceScaleFactor=2) and places it onto a 13.333×7.5 inch slide via
pptxgenjs. Pixel-faithful to
the browser preview, immune to font substitution, robust against any
HTML / CSS quirks.
Text inside the .pptx isn't selectable — but that's the point: the
browser preview is the contract. To revise content, edit the slide.html
and re-run export.mjs.
cd ppt
npm install # or `bun install`Dependencies: playwright-core, html2pptx-pro, dom-to-pptx, and
pptxgenjs. The headless renderer detects a local Chrome / Edge
automatically; if neither is installed, run npx playwright install chromium
once.
cd ppt
npm install
# kick off a deck in any project root
cd /path/to/your/project
node /path/to/skills/ppt/scripts/deck.mjs init pitch
node /path/to/skills/ppt/scripts/deck.mjs init-slide pitch cover
# ... write .pptwork/pitch/cover/slide.html (single-file, 1280x720) ...
node /path/to/skills/ppt/scripts/screenshot.mjs pitch cover
node /path/to/skills/ppt/scripts/export.mjs pitch
# → .pptwork/pitch/pitch.pptxRun the smoke test to verify the toolchain end-to-end:
node ppt/test/smoke.mjsIt spins up a temporary project, walks the full pipeline, and reports pass/fail per step.
Two complete example decks ship in examples/ with
design.md, slide.html, and pre-rendered thumbnail.png per slide.
Pre-built PPTX files are in examples/exports/.
Why the shipped examples use raster export: README preview images and
the committed .pptx files under examples/exports/ were built with
--mode raster so GitHub / doc viewers show pixel-identical output to
the browser screenshots — useful for visual QA and README fidelity. This is
a packaging choice, not a platform limit: the default export path is
editable (html2pptx-pro), and the same example decks export to fully
editable PowerPoint when you omit --mode raster (editable copies are also
committed as *-editable.pptx).
| File | Mode | Text in PowerPoint |
|---|---|---|
examples/exports/pptwork-capabilities.pptx |
raster | not selectable (PNG slide) |
examples/exports/pptwork-capabilities-editable.pptx |
editable | selectable text frames |
examples/exports/ai-agent-landscape-2026.pptx |
raster | not selectable |
examples/exports/ai-agent-landscape-2026-editable.pptx |
editable | selectable text frames |
# Re-screenshot a sample slide (from repo root)
node ppt/scripts/screenshot.mjs examples/pptwork-capabilities/p01-cover
# Raster export — pixel-faithful (what README previews match)
node ppt/scripts/export.mjs pptwork-capabilities --mode raster \
--output examples/exports/pptwork-capabilities.pptx
# Editable export — default engine; revise text/shapes inside PowerPoint
node ppt/scripts/export.mjs pptwork-capabilities \
--output examples/exports/pptwork-capabilities-editable.pptxWorking copies for the export scripts live under .pptwork/ at the repo root.
Every deck lives at .pptwork/<deck-name>/ in the project root:
.pptwork/
└── <deck-name>/
├── deck.json # slide order — single source of truth
├── outline.md # optional, story arc for long decks
├── materials/ # optional, research notes / digests
│ ├── doc_raw.md
│ └── research.md
└── <slide-name>/
├── design.md # frontmatter + Content / Note / Design sections
├── slide.html # single-file, self-contained, 1280x720
└── thumbnail.png # produced by screenshot.mjs
deck.json is the only authoritative slide order. Don't hand-edit;
let deck.mjs init-slide / move / delete mutate it.
PPTWork ships two starter buckets (10 layouts total) under
ppt/assets/:
| Bucket | Theme | Layouts |
|---|---|---|
corporate-light/ |
Crisp white, navy accent (Inter + Source Sans 3) | cover, agenda, two-column + chart, KPI row, closing CTA |
claude-warm/ |
Warm cream, editorial pairing (Space Grotesk + Lora) | cover, section divider, timeline, quote, bento grid |
Each bucket follows this shape:
ppt/assets/<bucket>/
├── index.json # array of template descriptors
├── html/<id>.html # one self-contained reference slide per id
├── specs/<id>.json # optional structured zone description
└── materials/<id>/ # optional images / svgs
The story-planning phase reads index.json to pick layouts; the
authoring phase reads the matching HTML to borrow composition.
Add your own buckets alongside the starters, or skip buckets entirely —
story planning can describe layouts in plain language and authoring
composes from scratch using ppt-html-authoring/references/.
Layout inspirations are documented in ATTRIBUTIONS.md.
Preview PNGs live in examples/showcase/. Each slide
below is a real thumbnail.png from the sample decks (1280×720).
From examples/pptwork-capabilities/p01-cover/
From examples/pptwork-capabilities/p03-bento/
From examples/pptwork-capabilities/p05-pipeline/
From examples/ai-agent-landscape-2026/market-size/
From examples/ai-agent-landscape-2026/landscape/
From examples/pptwork-capabilities/p06-workflow/
examples/pptwork-capabilities/— 8-page meta demo of the skill pipelineexamples/ai-agent-landscape-2026/— 7-page content pitch alternating both themes
MIT — see LICENSE.
Filesystem-as-database deck contract and the export toolchain (headless render via playwright-core, editable conversion via html2pptx-pro / dom-to-pptx, raster assembly via pptxgenjs) are adapted from a closed-source OpenCode plugin. Released here in skill form so the same workflow runs on any agent host.
Built-in layout templates adapt MIT-licensed design patterns from
presentation-skill-pack
and ppt-agent-skill — see
ATTRIBUTIONS.md for details.