AIToday
AI Coding AssistantsZenn AI/MLPublished: Oct 9, 2026, 22:00 JST

Claude Code builds HTML decks; 15-deck test spans 22%–100%

Claude Code builds HTML decks; 15-deck test spans 22%–100%

The author built 18 decks in Claude Code between May and August, using HTML and a shared asset folder instead of PowerPoint, and asked Claude to write the HTML inside the repository.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, in your inbox every morning.

Context & Analysis

The author's setup separates the deck body from a shared asset folder. Each deck is its own directory with an index.html that just lines up <section> elements inside a <deck-stage> tag, while colors, type, slide layouts and a deck-stage.js script live in a common _deck_assets folder. The JS and base CSS came from a Claude Design output; the author added only a print layout to the JS, plus product-intro CSS patterns and three scripts.

The "asset" idea is about patterns migrating. Three layouts first written inside two distribution decks — a logo cover, a Before/After comparison, and a screen capture with callout lines and notes — were moved into the shared CSS, and a closing "Thank you" page followed the same path to become a default. The author chose not to retrofit two already-distributed decks onto the shared CSS, since their details differed slightly and unifying them would change how they look.

Layout checking was handed to a script. Because slides clip anything outside their frame, a PDF can pass a page-count check while content is silently cut off; in one deck, 7 of 15 slides overflowed. The script renders slides at 1:1, measures how far child elements exceed the 1920x1080 frame, and ignores decorative blur circles that are meant to sit outside. Claude runs it after writing the HTML and fixes overflows before showing the deck.

FAQ
How was the deck-making process measured?
The decks are managed in Git, so the first committed version and the current version could be compared. Of the 18 decks, the 15 that were edited after the first commit were split into sentences and checked for how many original sentences remained.
What is the first draft's quality in the author's view?
The author puts it at about 70 out of 100. That number is not fixed, though — it moves depending on how far the Markdown outline was refined before generation. The author notes this link was not isolated by the 15-deck measurement and writes it as a personal impression.
How are PDF and PowerPoint outputs made?
PDF uses the browser's print function, with print CSS assigning one slide per page. PowerPoint instead places each slide as a high-resolution image, and edits are made in the HTML, not on the PowerPoint file.

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Related Articles

Next articleOpenAI halts Dark Clark, Iranian fake-byline influence ops