What does this skill do, and when should you use it?
This skill manages project-level design systems through the Stitch MCP server. It retrieves project screens and assets, derives design guidance from existing screens or a written description, and produces or updates DESIGN.md. It requires explicit user confirmation before uploading a design system to Stitch. It can then apply the system to existing screens and maintain .stitch/metadata.json.
It uses list_projects to locate a project, list_screens to select representative screens, get_screen to obtain screenshot and HTML download URLs, and read_url_content to read HTML. For existing screens, it relies on the design-md skill from stitch-utilities to extract a design system; for new projects or direct descriptions, it maps terms to concrete colors, fonts, and roundness values and generates .stitch/DESIGN.md. After user confirmation, it uploads DESIGN.md and calls create_design_system_from_design_md. It can also call get_project and list_design_systems to identify screen instances and the design-system asset, then call apply_design_system. Finally, it creates or updates .stitch/metadata.json.
- A designer maintaining a Stitch project needs to derive reusable design language from a representative home or dashboard screen.
- A frontend team needs to apply consistent colors, typography, and corner roundness across existing Stitch screens.
- A product team starting a new Stitch project has a written direction such as a dark theme, blue and purple colors, rounded shapes, and Inter typography.
- A developer already has .stitch/DESIGN.md and needs to upload it to Stitch and establish a project-level design system.
How do you install this skill?
- Core functionality depends on a configured and reachable Stitch MCP server, external credentials, and other skills; mainland-China network reachability is not established by the supplied files.
- The upload workflow requires an API key but does not explain secure injection, log-exposure risks, or least-privilege configuration.
- Applying a design system changes an external project, yet rollback, change preview, and a complete pre-application confirmation flow are not documented.
- The skill lacks dedicated automated tests and detailed abnormal-input handling for its key paths.
- Shell / CLI
- Network access
- Local filesystem
- MCP Server
Stitch MCP server
Install the Stitch Skills collection as documented in the README: npx skills add google-labs-code/stitch-skills. Codex users can add the repository as a marketplace; Claude Code can use npx plugins add google-labs-code/stitch-skills --scope project --target claude-code; Cursor can use npx plugins add google-labs-code/stitch-skills --scope workspace --target cursor. The target skill is plugins/stitch-design/skills/manage-design-system/. A configured and running Stitch MCP server is required; detailed MCP setup steps are not included in the supplied material.
tmp="$(mktemp -d)"
git clone --depth 1 https://github.com/google-labs-code/stitch-skills.git "$tmp"
mkdir -p ~/.claude/skills
cp -R "$tmp/plugins/stitch-design/skills/manage-design-system" ~/.claude/skills/
rm -rf "$tmp"Generated from the source repository and skill path; it copies only this skill's folder. If the author's install steps above differ, follow those first. To scope it to one project, replace ~/.claude/skills with that project's .claude/skills.
How do you use this skill?
Once installed, send your agent any of these to trigger it:
- Upload our design system from `.stitch/DESIGN.md` and apply it to all screens.
Configure and run the Stitch MCP server, then use a prompt such as: "Upload our design system from .stitch/DESIGN.md and apply it to all screens." You can also ask it to analyze a Stitch project, create DESIGN.md from a written design direction, or apply a design system to selected screens. Before uploading, it must pause, present a design-system summary, and wait for explicit confirmation.
What are this skill's strengths and limitations?
- Covers retrieval, design-system file generation, Stitch upload, and screen application in one workflow.
- Includes an explicit confirmation checkpoint before uploading design changes.
- Supports both extraction from existing screens and synthesis from written design descriptions.
- Documents the relevant Stitch MCP calls and the required screen-instance argument shape.
- Requires a configured Stitch MCP server plus network, shell, and local filesystem access.
- Design extraction depends on the separate stitch-utilities design-md skill, so selective installation may require additional dependencies.
- The confirmation checkpoint prevents a fully unattended creation workflow.
- The README does not document a test suite or platform verification results for this skill.
How does this skill compare with similar options?
Side by side with related skills; every score comes from the same FSRS standard.
| Skill | FS score | Stars | Last updated | License |
|---|---|---|---|---|
| Stitch Design System Manager this page | 45 · Not recommended | ★ 8.5k | 2mo ago | Apache-2.0 |
| Stitch Prompt Refiner | 49 · Use with care | ★ 8.5k | 2mo ago | Apache-2.0 |
| Stitch Frontend Design Extractor | 42 · Not recommended | ★ 8.5k | 2mo ago | Apache-2.0 |
| Stitch Design Taste | 42 · Not recommended | ★ 8.5k | 2mo ago | Apache-2.0 |
| VS Code Design Philosophy ✓ Microsoft · Official | 58 · Recommended | ★ 194k | 12d ago | MIT |
How did FollowSkills review this skill?
The skill requires explicit user confirmation before uploading and scopes tools to Stitch MCP, Bash, Read, Write, and web_fetch. However, the permissions are broad, local writes and script execution are allowed, the API key is passed as a command-line argument, and there is no disclosure of sensitive-data handling, least-privilege setup, upload data flow, rollback, or a second confirmation before applying changes, so points are deducted.
The retrieval, upload, creation, and application workflow is fairly specific and includes parameter formats and some invalid-argument avoidance. However, it depends on other skills, the Stitch MCP server, and an external script; it provides no tests for this skill's key paths, limited abnormal-input handling, inconsistent script-argument examples, and insufficient failure feedback, so points are deducted.
The audience and main scenarios are reasonably clear, covering existing screens, new projects, direct descriptions, and applying systems to screens. Boundaries, non-fit cases, trigger precision, environment requirements, and Chinese-language support are not fully specified, and mainland-China network reachability of Stitch is not evidenced, so points are deducted.
The documentation has sections, progressive workflow guidance, metadata and tool-schema examples, dependency notes, and known parameter constraints. The repository supplies an Apache-2.0 license, security policy, supported-version table, and CI, but the skill lacks a clear maintainer, update path, changelog, or stable versioning strategy, and relies on material outside its own path, so points are deducted.
The skill plausibly organizes DESIGN.md upload, project-level design-system creation, and application to screens, which may reduce repeated design prompting. Static files do not verify representative outputs; the workflow depends heavily on Stitch MCP, API credentials, and other skills, while metadata synchronization and recovery are incomplete, limiting evidence of direct usability and comparative benefit, so points are deducted.
The skill contains auditable procedures, JSON examples, and schema documentation, and the repository includes CI. However, the CI covers a different React skill, not this skill's key paths, and there are no third-party execution records, dedicated tests, or corroborating sources; static evidence is therefore limited and scored conservatively.
Open a dimension to read why it scored that way
Evidence confidence：Low — Mostly static review, author material or a limited demo; useful for discovery, not high-risk decisions.
See the full review method →