stitch::generate-design
An agent skill by google-labs-code, from google-labs-code/stitch-skills. Tags: design, generative, stitch, ui-ux, web.
What it does
Generate new screens from text prompts or images, edit existing screens with prompts and design system tokens, and generate design variants using Stitch MCP. Includes prompt enhancement pipeline, design mappings, professional UI/UX terminology, design tokens and theme system capabilities.
Install
With the skills CLI, which installs into Claude Code, Codex, Cursor and other agents:
npx skills add google-labs-code/stitch-skills --skill 'stitch::generate-design'Or copy the skill folder into Claude Code's skills directory by hand (~/.claude/skills for every project, or .claude/skills inside one):
git clone --depth 1 https://github.com/google-labs-code/stitch-skills
cp -r stitch-skills/plugins/stitch-design/skills/generate-design ~/.claude/skills/'stitch::generate-design'Safety box score
Grade C, moderate risk: 32 of 100 risk points, rated 2026-10-16. The grade rates what the skill and its scripts can reach on the machine of whoever installs it, across eight categories from shell execution to secrets access. How the score works.
This skill drives Stitch MCP to create, edit, and variant-generate design screens from prompts or images, then downloads the results into `.stitch/designs` and updates local metadata. The main risks are remote network access to Stitch endpoints, shell-based downloads, automatic local file writes/overwrites, and prompt-shaping instructions that depend on the agent’s system prompt.
| Category | Level | Why |
|---|---|---|
| Code & shell execution | 1 (low) | The skill explicitly tells the agent to use Bash for downloads, e.g. `Use curl -o via run_command or similar`, but it does not ship arbitrary scripts or `eval`/`sh` payloads. |
| Network access | 2 (moderate) | It calls remote Stitch MCP tools (`list_projects`, `generate_screen_from_text`, `edit_screens`, `generate_variants`) and downloads HTML/screenshot URLs from those remote outputs; `web_fetch` is also allowed. |
| Filesystem reach | 1 (low) | It writes into the local workspace under `.stitch/designs` and updates `.stitch/metadata.json`, with explicit overwrite behavior for edited screens. |
| Secrets & credentials | 0 (none) | There are no instructions to read, request, or transmit tokens, passwords, `.env` files, browser sessions, or other secrets. |
| Privilege & persistence | 0 (none) | No sudo, global installs, services, scheduled tasks, or shell/OS configuration changes are requested. |
| Instruction hijack surface | 2 (moderate) | The skill tells the agent to `Refer to your system prompt` for MCP tool-prefix handling and to delegate to another skill first, which relies on hidden model instructions rather than only visible file contents. |
| Irreversible actions | 2 (moderate) | It can create projects, edit existing screens, and generate variants without an explicit human approval step, though these actions are bounded to Stitch design assets. |
| Transparency | 0 (none) | All provided files are plain markdown and readable; there are no binaries, base64 blobs, or decoded-at-runtime code. |
Before installing
- Read `SKILL.md` with special attention to the `.stitch/designs` and `.stitch/metadata.json` write steps.
- Use a scratch Stitch project first, since the skill can create and edit remote screens automatically.
- Confirm you are comfortable with the agent running shell downloads (`curl -o`) for generated HTML and screenshots.
- Make sure the workspace is disposable or backed up, because edited assets are overwritten.
Rated 2026-10-16 by an automated review of 4 files from the skill's repository.
Source
- Repository
- google-labs-code/stitch-skills (all skills from this repository)
- Path
- plugins/stitch-design/skills/generate-design/SKILL.md
- Branch
- main
- Collection
- stitch-design
- Updated
- 2026-09-19
Related skills
- enhance-prompt — Transforms vague UI ideas into polished, Stitch-optimized prompts.
- copywriting — When the user wants to write, rewrite, or improve marketing copy for any page — including homepage, landing pages, pricing pages, feature pages, about pages.
- algorithmic-art — Creating algorithmic art using p5.js with seeded randomness and interactive parameter exploration. Use this when users request creating art using code.
- animate — Build an animation from scratch, making the decisions in the order that determines whether it feels right — should it animate at all, what purpose, which tool.
- antfu-design — antfu-style design conventions, broadened.
- brainstorming — You MUST use this before any creative work - creating features, building components, adding functionality, or modifying behavior.