Presentation Skill

Presentation Skill helps AI coding agents create slide decks. It defaults to image-generated decks, where each slide is rendered as a complete visual scene. HTML mode generates the deck as a web page: one borderless, animated canvas with a camera, written in HTML, CSS and SVG, where diagrams draw themselves, cards split and numbers roll on each click, with exact copy in the DOM, textless generated plates, and copy drafted by a writing model and fact-checked by the builder.

This repo is a public, platform-agnostic skill package. It works with agents such as OpenCode, Claude Code, Cursor, Codex, or any terminal coding agent that can read Markdown instructions and write files.

Install Into An Agent Workspace

Give your agent this repository URL and ask it to install the skill:

Install this public skill repo into my workspace:
https://github.com/grapeot/presentation_skill

Start from my workspace AGENTS.md or CLAUDE.md. Follow any WORKSPACE.md or skills/INDEX.md routing rules. Clone or vendor the repo under an appropriate project directory. Expose exactly one root skill to my global skill index or agent instructions. Keep private aliases, local paths, credentials, endpoint defaults, and business context in a local overlay, not in the public repo.

The root skill is skills/skill_presentation.md.

Modes

Image-generated mode is the default. The agent creates a deck plan, a visual direction, slide prompts, local artifacts, and rendered images through the workspace’s configured image generation tool.

HTML mode is a first-class alternative for talks, lectures, pitches and explainers that should move, and for decks that must keep exact copy, code, real data or links. See skills/html_decks.md. It was designed and validated with Claude Opus 5.5 as the builder; other models are untested (GPT-6 Astra is the suggested alternative). Rendering mode and asset policy are independent.

scripts/presentation-skill "Visual keynote" --mode image --output deck
scripts/presentation-skill "Technical briefing" --mode html --assets mixed --output deck

--mode reveal remains a compatibility alias for html.

Editable PPTX mode is for decks a co-author will keep editing in PowerPoint or Google Slides. Every word is a native text box and pictures are separate image plates, so the text survives an import into Google Slides. The deck is built from a JSON slide spec. See skills/pptx_decks.md; it needs the [pptx] extra, and --render needs LibreOffice and poppler.

scripts/presentation-skill build-pptx deck/slides.json --render deck/verification/render

PDF Export

scripts/presentation-skill export-pdf deck                 # image deck: lossless PDF from the slide images, with clickable overlays
scripts/presentation-skill export-pdf deck --with-notes    # HTML (canvas) deck: one page per slide, each followed by its notes

export-pdf picks the exporter from the deck. Image decks need the [pdf] extra. HTML (canvas) decks are printed through headless Chromium, one page per slide at a chosen print state (by default the slide’s last step), with vector, selectable text. Effects that Preview renders differently from the browser (blurred shadows, repeating gradients, SVG patterns) are redrawn before printing, and large images are capped and JPEG-compressed. They need the [pdf-html] extra and python -m playwright install chromium. The HTML export fails on console errors, requests that leave localhost, unfilled copy slots or a page count that does not match the slide table, and writes a contact sheet next to the PDF. See skills/html_decks.md.

Local Development

uv venv .venv
uv pip install --python .venv/bin/python -e '.[dev]'
.venv/bin/python -m pytest -v
scripts/presentation-skill --help

The tests are offline and validate the planning contract, starter artifact generation, PDF export, and installation contract. The HTML PDF export tests drive a local headless Chromium when Playwright is installed (uv pip install --python .venv/bin/python -e '.[dev,pdf-html]' and .venv/bin/python -m playwright install chromium) and are skipped otherwise.