Turn product briefs into UI directions, Tailwind-friendly HTML, and screenshots in minutes.
Prompt -> UI direction -> local HTML + screenshots.
StitchFlow packages Google Stitch into a portable local workflow for:
- Codex
- Claude Code
- OpenClaw
- GitHub Copilot
- Gemini CLI
- other clients that understand
SKILL.mdorAGENTS.md
Why people star and keep it:
- generate usable UI directions from plain-language briefs
- save local HTML and screenshots your team can review immediately
- reuse one workflow across multiple coding-agent clients
Generated locally from a natural-language prompt with the bundled Stitch workflow.
Use $stitchflow to generate a premium desktop analytics dashboard for a product team, with a left sidebar, KPI cards, trend charts, and clean Tailwind-ready HTML.
Use $stitchflow to explore three mobile-first checkout directions for a modern ecommerce app, with sticky CTA, compact order summary, and polished spacing.
Use $stitchflow to create a SaaS landing page for an enterprise design tool, with strong typography, product shots, pricing cards, and a credible B2B feel.
Compatibility note:
- brand name:
StitchFlow - current skill slug:
stitchflow - legacy alias still supported:
stitch-design-local - GitHub Copilot plugin slug:
stitchflow - Gemini CLI extension id:
stitchflow
git clone https://github.com/yshishenya/stitchflow.git
cd stitchflow
bash install.sh --target allThen add STITCH_API_KEY to:
${STITCH_STARTER_ROOT:-$HOME/.agents/stitch-starter}/.env
Then restart your client.
- product and design engineers who want faster UI exploration before frontend work
- agent builders who want one reusable design workflow across multiple clients
- teams that need local artifacts instead of another hosted black box
- one canonical skill:
stitchflow - one shared local toolkit:
stitch-starter - local HTML, screenshots, and run artifacts
- variants and edits from natural-language prompts
- installable packaging across multiple agent clients
English
The Stitch SDK is powerful, but most teams still need a repeatable workflow around it.
This repo removes the friction between:
- a rough product brief
- a useful UI direction
- local artifacts the team can review immediately
StitchFlow packages Stitch as a reusable agent skill, installs a local toolkit, and saves HTML, screenshots, and run metadata on your machine instead of behind another hosted workflow.
- product engineers who want to explore UI before writing frontend code
- design engineers who want faster prompt-to-HTML loops
- founders who need strong first-pass screens from natural-language briefs
- AI-agent builders who want a ready-to-run Stitch workflow across multiple clients
- one canonical skill:
stitchflow - one shared local toolkit:
stitch-starter - one installer for Codex, Claude Code, OpenClaw, and GitHub Copilot
- local HTML, screenshots, and run artifacts
- one canonical setup under
~/.agentswith compatibility links for native clients
Canonical install paths:
- skill:
${AGENT_SKILLS_HOME:-$HOME/.agents}/skills/stitchflow - toolkit:
${STITCH_STARTER_ROOT:-$HOME/.agents/stitch-starter}
Optional compatibility links:
${CODEX_HOME:-$HOME/.codex}/skills/stitchflow${CLAUDE_HOME:-$HOME/.claude}/skills/stitchflow${OPENCLAW_HOME:-$HOME/.openclaw}/skills/stitchflow${COPILOT_HOME:-$HOME/.copilot}/skills/stitchflow
Legacy alias links remain available after install:
${CODEX_HOME:-$HOME/.codex}/skills/stitch-design-local${CLAUDE_HOME:-$HOME/.claude}/skills/stitch-design-local${OPENCLAW_HOME:-$HOME/.openclaw}/skills/stitch-design-local${COPILOT_HOME:-$HOME/.copilot}/skills/stitch-design-local
Native extension / plugin entrypoints:
- GitHub Copilot:
.github/plugin/plugin.json - Gemini CLI:
gemini-extension.json
The raw Stitch SDK is flexible.
This repo is for when you want:
- a ready-to-run local workflow instead of wiring the SDK yourself
- portable skill packaging across multiple agent clients
- HTML, screenshots, and run artifacts saved locally by default
In Codex:
Use $stitchflow to generate a premium desktop dashboard for an internal analytics product.
In Claude Code:
/stitchflow landing page for a design tool aimed at enterprise product teams
In OpenClaw:
Use the stitchflow skill to explore three mobile-first UI directions for a checkout experience.
In GitHub Copilot CLI:
copilot plugin install yshishenya/stitchflowIn Gemini CLI:
gemini extensions install https://github.com/yshishenya/stitchflowDirect CLI usage:
cd "${STITCH_STARTER_ROOT:-$HOME/.agents/stitch-starter}"
npm run list
npm run generate -- --prompt "A modern SaaS dashboard with sidebar and stat cards"
npm run edit -- --prompt "Make it more premium and add stronger typography"
npm run variants -- --prompt "Explore three different visual directions" --variant-count 3- explore 3 landing page directions before writing code
- turn a PM brief into HTML and screenshots for review
- iterate on a dashboard without opening Figma
- generate local artifacts a team can review without adopting a new hosted service
Ready-to-use prompt ideas:
If you want to record a demo or launch video:
- docs/demo-script.md
- docs/launch-kit.md
- docs/catalog-submissions.md
- docs/launch-system.md
- docs/community-posts.md
Outputs go to:
${STITCH_STARTER_ROOT:-$HOME/.agents/stitch-starter}/runs/<timestamp>-<operation>-<slug>/
Typical files:
result.jsonorvariants.jsonscreen.htmlscreen.png,screen.jpeg, orscreen.webphtml-url.txtimage-url.txt
Latest single-screen pointer:
${STITCH_STARTER_ROOT:-$HOME/.agents/stitch-starter}/runs/latest-screen.json
- built on Google's Stitch SDK
- exports clean HTML and screenshots programmatically
- works across Codex, Claude Code, OpenClaw, GitHub Copilot, and Gemini CLI
- includes AGENTS.md, SKILL.md, and agent-specific manifests
- licensed under Apache-2.0
- official OpenClaw registry path via ClawHub
- official GitHub Copilot plugin install from the repository
- official Gemini CLI extension install from the repository
- community catalog checklist in docs/catalog-submissions.md
- contribution guide: CONTRIBUTING.md
- toolkit details: stitch-starter/README.md
Install it, generate one screen, and ship the best direction into code.
Русский
Stitch SDK мощный, но большинству команд нужен не просто SDK, а готовый workflow вокруг него.
Этот репозиторий убирает трение между:
- сырым продуктовым брифом
- первым сильным UI-направлением
- локальными артефактами, которые можно сразу показать команде
StitchFlow упаковывает Stitch в reusable skill для агента, ставит локальный toolkit и сохраняет HTML, скриншоты и run metadata на вашей машине.
- product engineers, которые хотят исследовать UI до написания фронтенда
- design engineers, которым нужен быстрый prompt-to-HTML цикл
- founders, которым нужны сильные первые экраны из текстового брифа
- builders агентных workflow, которым нужен готовый Stitch setup для нескольких клиентов
- один канонический skill:
stitchflow - один локальный toolkit:
stitch-starter - один installer для Codex, Claude Code, OpenClaw и GitHub Copilot
- локальные HTML, скриншоты и run artifacts
- каноническую установку в
~/.agentsи compatibility links для нативных клиентов
Канонические пути:
- skill:
${AGENT_SKILLS_HOME:-$HOME/.agents}/skills/stitchflow - toolkit:
${STITCH_STARTER_ROOT:-$HOME/.agents/stitch-starter}
Compatibility links:
${CODEX_HOME:-$HOME/.codex}/skills/stitchflow${CLAUDE_HOME:-$HOME/.claude}/skills/stitchflow${OPENCLAW_HOME:-$HOME/.openclaw}/skills/stitchflow${COPILOT_HOME:-$HOME/.copilot}/skills/stitchflow
Legacy alias links после установки тоже создаются:
${CODEX_HOME:-$HOME/.codex}/skills/stitch-design-local${CLAUDE_HOME:-$HOME/.claude}/skills/stitch-design-local${OPENCLAW_HOME:-$HOME/.openclaw}/skills/stitch-design-local${COPILOT_HOME:-$HOME/.copilot}/skills/stitch-design-local
Нативные entrypoints:
- GitHub Copilot:
.github/plugin/plugin.json - Gemini CLI:
gemini-extension.json
Raw Stitch SDK гибкий.
Этот репозиторий нужен, когда вы хотите:
- готовый локальный workflow, а не собирать обвязку самому
- переносимую skill-упаковку для нескольких agent clients
- локальные HTML, скриншоты и run artifacts по умолчанию
Быстрый старт:
git clone https://github.com/yshishenya/stitchflow.git
cd stitchflow
bash install.sh --target allПотом добавьте STITCH_API_KEY в:
${STITCH_STARTER_ROOT:-$HOME/.agents/stitch-starter}/.env
И перезапустите клиент.
В Codex:
Use $stitchflow to generate a premium desktop dashboard for an internal analytics product.
В Claude Code:
/stitchflow landing page for a design tool aimed at enterprise product teams
В OpenClaw:
Use the stitchflow skill to explore three mobile-first UI directions for a checkout experience.
В GitHub Copilot CLI:
copilot plugin install yshishenya/stitchflowВ Gemini CLI:
gemini extensions install https://github.com/yshishenya/stitchflowЧерез CLI:
cd "${STITCH_STARTER_ROOT:-$HOME/.agents/stitch-starter}"
npm run list
npm run generate -- --prompt "A modern SaaS dashboard with sidebar and stat cards"
npm run edit -- --prompt "Make it more premium and add stronger typography"
npm run variants -- --prompt "Explore three different visual directions" --variant-count 3- исследовать 3 направления лендинга до написания кода
- превратить PM brief в HTML и скриншоты для ревью
- итерировать dashboard без Figma
- получать локальные артефакты, которые команда может смотреть без нового hosted-сервиса
Готовые prompt ideas:
Для demo и запуска:
Все результаты попадают в:
${STITCH_STARTER_ROOT:-$HOME/.agents/stitch-starter}/runs/<timestamp>-<operation>-<slug>/
Обычно внутри:
result.jsonилиvariants.jsonscreen.htmlscreen.png,screen.jpegилиscreen.webphtml-url.txtimage-url.txt
Указатель на последний single-screen run:
${STITCH_STARTER_ROOT:-$HOME/.agents/stitch-starter}/runs/latest-screen.json
- построен на Google Stitch SDK
- программно экспортирует HTML и screenshots
- работает в Codex, Claude Code, OpenClaw, GitHub Copilot и Gemini CLI
- содержит AGENTS.md, SKILL.md и platform manifests
- лицензия: Apache-2.0
- официальный реестр OpenClaw через ClawHub
- официальный install GitHub Copilot plugin прямо из репозитория
- официальный install Gemini CLI extension прямо из репозитория
- чеклист по каталогам в docs/catalog-submissions.md
- гайд: CONTRIBUTING.md
- toolkit details: stitch-starter/README.md
Установите, сгенерируйте один экран и протащите лучший вариант в код.
