How it works

Pick a source. We gather context server-side, merge your rules, and stream a complete markdown document.

PDF brand guide

Upload a guide. Text and page imagery become evidence for a faithful nine-section DESIGN.md.

PowerPoint deck

Upload a .pptx. We parse theme XML, slide text, and embedded media, then compose opinionated brand observations before sending it to Gemini.

Figma file

Connect with your token. We pull file metadata, styles, and variables to ground the doc in your system.

Website URL (last resort)

When no brand artifact is available, we fetch the page, strip noise, and let Gemini read structure, color, and type signals from the HTML.

Built for clarity

Every run produces a structured nine-section document designed to be read by both humans and AI agents — consistent format, deterministic tokens, zero guesswork.

Rules-backed output

Generation is steered by your local rules/ markdown pack so structure stays consistent.

Agent-ready

Every run targets the Agent Prompt Guide pattern—quick colors, quoted prompts, iteration steps, renderer token map, implementation warnings.

One canonical output

Every run produces the full nine-section design system spec. No tier toggle, no half-output runs.

Publish to GitHub

One click commits a Claude-Skill-shaped folder (DESIGN.md, tokens.css, tailwind preset, fonts, SKILL.md) to any repo you can push to.