Google Labs Releases Stitch Skills Pack for AI Agents and Gets 248 Stars in a Day
Google Labs has published stitch-skills on GitHub, a set of ready-made instructions ("skills") that let Claude Code, Cursor, and Codex agents work with Stitch, Google's design service, through an MCP server. Given a prompt like "make a login screen," the agent produces a mockup and converts it into React code. The repository gained 248 stars in a day on GitHub Trending. This saves teams time, since they no longer need to write prompts for linking design and code from scratch.
AI-processed from GitHub Trending AI; edited by Hamidun NewsWhen an agent writes an interface, it usually works blind: it doesn't know your design system, can't see the mockup, and isn't used to checking against it. Developers have been working around this with long prompts that they copy from project to project. Google Labs offers a different approach: the stitch-skills repository, which contains ready-made "skills," meaning folders of instructions for agents such as Claude Code, Cursor, Codex, and others.
With them, an agent controls Stitch, Google's design service where application screens are drawn from text descriptions. The agent and the service communicate through an MCP server, the standard "adapter" between an AI agent and an external tool. In a single day, the repository gained 248 stars on GitHub Trending, meaning that many developers starred it in one day.
What's Inside the Repository
A skill is a folder of instructions for an agent. Each one follows the open Agent Skills standard and is structured the same way. The main file, SKILL.md, works as a "control panel": it tells the agent what to do. Alongside it are scripts for validation and network requests, a knowledge base with checklists and style guides, and examples of reference results.
The skills are grouped into three plugins. The first, stitch-design, handles design work. It can generate new screens from text or an image, edit existing ones, and create variants, such as a dark theme or denser layouts. It manages the design system: it uploads a DESIGN.md file and applies the theme to screens. It can extract DESIGN.md from frontend source code and capture a static HTML snapshot of a running application. There is also a reverse direction: ready React or Vue code can be transferred into a Stitch project.
The second plugin, stitch-build, goes from design to code. It turns Stitch screens into a system of React components, can assemble React Native with platform-specific code, and builds dashboards on React and Vite. Among its skills are a guide to shadcn/ui and generation of a video overview of the project via Remotion.
The third, stitch-utilities, is a helper. It turns vague ideas like "make a settings page" into more precise prompts for Stitch. It describes the project in a SITE.md file, which serves as the "constitution" for the build cycle. It helps generate multi-page websites and DESIGN.md files with strict typography and color standards.
How to Install
The repository supports Codex, Antigravity, Gemini CLI, Claude Code, Cursor, and OpenCode. For Claude Code, the team installs the plugins into the current project: npx plugins add google-labs-code/stitch-skills --scope project --target claude-code. For Cursor, the scope and target change: --scope workspace --target cursor. In Codex, you first add the marketplace with codex plugin marketplace add, optionally limiting cloning to the needed folders, and then select the plugins. Individual skills are installed with the command npx skills add google-labs-code/stitch-skills.
OpenCode is supported manually. The skill folders are copied into .opencode/skills/, and a remote MCP server with the address from the Stitch instructions is added to the configuration.
Limitations Worth Knowing
First, the skills themselves draw nothing. They require a configured and running Stitch MCP server with credentials in the agent's environment. Without it, they are just folders of text.
Second, the Stitch Design skills often depend on one another. If you install them selectively, you need to check yourself that all dependencies have been pulled in.
Third, manual installation in OpenCode requires care. Skill names must be lowercase with hyphens and match the folder name, or OpenCode won't load them. Tool names inside the instructions may differ from those built into OpenCode, so when adapting, it's better to replace them with your local ones.
Finally, the authors state directly that this is not an officially supported Google product and that it is not covered by the vulnerability reward program. For production, this means reading the scripts before giving an agent access to your project and network.
Why It Matters for Product Managers and Developers
The main value here is the format, not Stitch itself. The repository shows how to structure a repeatable workflow for an agent: instructions, validation scripts, checklists, and reference examples in a single folder. If your team already uses DESIGN.md or builds prototypes with agents, you can take these skills as they are, or copy the structure for your own internal processes. The authors invite contributions of new skills. They are looking for Stitch HTML converters to other UI frameworks with syntax checking, exports of static designs into files with test data, and generation of screens from a dataset.
- Nov 21, 2026Perplexity returned risky shops in 15% of shopping answers, ChatGPT in 8%
- Nov 21, 2026XYBench test: AI picks the useful advice in 71–84% of cases, but gives it itself only 35–41% of the time
- Nov 21, 2026Anthropic Paid $3.4 Billion for 1.7 GW of Gas Generators
- Nov 21, 2026Gemini 4 Hydrogen: Google's Junior Model Beats the Senior Argon and Runs 2.5x Faster Than Flash
Key Questions Answered
What's Inside the Repository?
A skill is a folder of instructions for an agent. Each one follows the open Agent Skills standard and is structured the same way. The main file, SKILL.md, works as a "control panel": it tells the agent what to do. Alongside it are scripts for validation and network requests, a knowledge base with checklists and style guides, and examples of reference results. The skills are grouped into three plugins. The first, stitch-design, handles design work.
How to Install?
The repository supports Codex, Antigravity, Gemini CLI, Claude Code, Cursor, and OpenCode. For Claude Code, the team installs the plugins into the current project: npx plugins add google-labs-code/stitch-skills --scope project --target claude-code. For Cursor, the scope and target change: --scope workspace --target cursor. In Codex, you first add the marketplace with codex plugin marketplace add, optionally limiting cloning to the needed folders, and then select the plugins.
What are the key details on Limitations Worth Knowing?
First, the skills themselves draw nothing. They require a configured and running Stitch MCP server with credentials in the agent's environment. Without it, they are just folders of text. Second, the Stitch Design skills often depend on one another. If you install them selectively, you need to check yourself that all dependencies have been pulled in. Third, manual installation in OpenCode requires care. Skill names must be lowercase with hyphens and match the folder name, or OpenCode won't load them.
Build & Ship Production AI Agents Without Boilerplate Code
Start your free 72-hour trial at Hamidun Academy (academy.hamidun.com) — interactive lessons on Claude Code & Cursor, live Tech Radar, and production agent templates.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.