Presentations, written in HTML.
Give Claude a topic, an outline, documents, data or links. Get a designed slide deck: one file that opens in any browser, presents itself, and prints to PDF.
Four skills, one deck
Claude picks the right one from what you ask. Each loads only when it is needed, so the plugin adds about 734 tokens to a session.
Make
present-in-htmlWrites the deck from your material in a look chosen for your audience, with speaker notes, then runs the checker.
Turn q3-report.pdf into a 10-minute board deck
Edit
edit-deckOpens a click-and-type editor for wording and notes. Save writes the same file back.
Let me fix the wording myself
Review
review-deckChecks timing, jargon, story and sources against the audience and the time slot.
Will this fit in 15 minutes for managers?
Publish
publish-deckPuts the deck and its images on GitHub Pages and returns a link to share.
Put my deck online so I can send a link
What every deck can do
The engine, the theme and the slides live in one HTML file. Nothing to install to present it.
Install
Two commands in Claude Code, then ask for a presentation. The skills and the source are on GitHub.
As a plugin
Recommended. Updates arrive with /plugin update.
/plugin marketplace add hisham-abuella/present-in-html /plugin install present-in-html@present-in-html-marketplace
Then just ask
Or call a skill directly, like /present-in-html:review-deck.
Make a 10-minute deck for our Q3 board review. Revenue was $4.2M, up 18%.
Without the plugin system
Copies the skills into ~/.claude/skills/; the workflow needs the plugin.
git clone https://github.com/hisham-abuella/present-in-html.git cd present-in-html && ./install-skill.sh
For the checker
Playwright, installed once outside the plugin folder.
mkdir -p ~/.kp-tools && cd ~/.kp-tools npm i playwright && npx playwright install chromium
Template library
Every deck here is one HTML file. Open one, press F for fullscreen, P for the presenter window, S to try another theme.
Fonts and themesEvery font pairing and theme as a specimen. Type your own headline.
Nothing matches. Try another word.
Evals Work in progress
How well, how fast and at what cost do different models build a deck with this skill? Each model got the same one-shot prompt and the same course module, in its own empty folder, through the Claude Code CLI. The skill's checker and a blind judge (Opus 5.5) scored every deck: wrong claims, UI issues and seven criteria.
| Model | Runs on | Judge /10 | Time | Cost |
|---|---|---|---|---|
| Opus 5.5 | Anthropic API | 9 / 9 | 11m 44s / 11m 14s | $2.88 / $2.79 |
| Sonnet 5.5 | Anthropic API | 9 / 9 | 5m 27s / 5m 24s | $0.90 / $0.70 |
| Haiku 5.5 | Anthropic API | 7 / 7 | 7m 05s / 8m 44s | $0.24 / $0.32 |
| qwen3.8 27B | Mac Studio M4 Max, Ollama | unfinished | timed out at 120 min | $0 (local) |
Each cell is Wireless 101 / STATS 101, Module 1: judge score out of 10, wall-clock time, and API list-price cost.
So far, Sonnet 5.5 matches Opus 5.5 at about half the time and a quarter to a third of the cost, and Haiku 5.5 is a usable first draft. The local model was too slow to finish on its own. In a separate experiment, not ranked against these runs, a local MoE model (qwen3.6 35B-A3B) finished a 30-slide deck in 20 minutes, scored 9/10, by building from a plan Opus wrote once.
Preliminary: one run per model per task. Small differences may be noise, and more repeated tests are planned before drawing firm conclusions.