What does this skill do, and when should you use it?
Stitch Build Loop is a frontend-building skill for the Stitch MCP Server that uses a baton file to drive repeated website iterations. Each iteration reads the current task, generates a page with Stitch, integrates it into the site, and prepares the next task. It expects a Stitch project plus `.stitch/DESIGN.md` and `.stitch/SITE.md` files. Chrome DevTools MCP Server can provide optional visual verification.
Reads .stitch/next-prompt.md for the page name and task; reads .stitch/SITE.md and .stitch/DESIGN.md for site planning and visual rules; discovers Stitch MCP tools and creates or retrieves a Stitch project; calls generate_screen_from_text to generate a page for the requested device; persists project and screen metadata in .stitch/metadata.json; downloads HTML and screenshots into .stitch/designs/; moves generated HTML into site/public/, fixes asset paths, and updates navigation; updates .stitch/SITE.md; and writes the next iteration to .stitch/next-prompt.md.
- A frontend developer wants to build a multi-page website page by page from an initial Stitch prompt.
- A team needs generated pages to follow a shared visual system defined in `.stitch/DESIGN.md`.
- A developer wants `.stitch/next-prompt.md` to pass the next web task between human review, CI/CD, or agent-chain iterations.
- A team needs to integrate Stitch-generated static HTML into an existing `site/public/` structure.
- A developer wants optional Chrome DevTools MCP screenshots to visually check generated pages when that server is available.
How do you install this skill?
- Core operation depends on Stitch MCP, project credentials, and potentially overseas services; reachability and authentication should be verified for mainland-China users before adoption.
- The skill grants Bash, Write, and multiple MCP namespaces; review the actual tool inventory, restrict write locations, and keep recoverable copies before project creation, downloads, or site edits.
- The autonomous loop requires modifying next-prompt.md before completion and may select or invent the next page; production use should add human approval, budgets, and an explicit termination condition.
- No skill-specific tests or real execution evidence are provided; validate metadata persistence, download URL handling, relative-path repair, navigation updates, and failure messages before adoption.
- Shell / CLI
- Network access
- Local filesystem
- MCP Server
Stitch MCP Server
Install the repository's plugins/stitch-utilities plugin, or use the README command: npx skills add google-labs-code/stitch-skills. Configure and run the Stitch MCP Server before use. The source does not provide a separate installation command or version requirement for this individual skill.
tmp="$(mktemp -d)"
git clone --depth 1 https://github.com/google-labs-code/stitch-skills.git "$tmp"
mkdir -p ~/.claude/skills
cp -R "$tmp/plugins/stitch-utilities/skills/stitch-loop" ~/.claude/skills/
rm -rf "$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?
Prepare a Stitch project, .stitch/DESIGN.md, .stitch/SITE.md, and .stitch/next-prompt.md, then trigger the agent with a prompt such as: Build a 5-page portfolio website with Stitch. The skill follows the baton loop to generate and integrate pages, then updates the next task. If local design files already exist, it asks before refreshing them from Stitch.
What are this skill's strengths and limitations?
- Covers task intake, Stitch generation, HTML integration, navigation updates, and next-task preparation.
- Uses `.stitch/DESIGN.md` and `.stitch/SITE.md` to support visual consistency and roadmap-aware page selection.
- Persists Stitch project and screen metadata for later iterations.
- Can optionally use Chrome DevTools MCP for visual verification.
- Requires the Stitch MCP Server, local filesystem access, and shell access.
- Visual verification is optional and is skipped when Chrome DevTools MCP is unavailable.
- The source provides no test suite, performance benchmarks, or cross-platform validation results.
- Each iteration depends on maintaining the expected `.stitch/` file structure and updating the baton file.
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 Build Loop this page | 41 · Not recommended | ★ 8.5k | 2mo ago | Apache-2.0 |
| Stitch Design System Synthesizer | 45 · Not recommended | ★ 8.5k | 2mo ago | Apache-2.0 |
| Stitch Screen Generator | 45 · Not recommended | ★ 8.5k | 2mo ago | Apache-2.0 |
| Stitch Frontend Design Extractor | 42 · Not recommended | ★ 8.5k | 2mo ago | Apache-2.0 |
| Stitch Prompt Refiner | 49 · Use with care | ★ 8.5k | 2mo ago | Apache-2.0 |
How did FollowSkills review this skill?
The required tools and Stitch MCP dependency are disclosed, and the skill requires user confirmation before refreshing existing designs; however, Bash, Write, and external MCP access are broad, while project creation, downloads, site edits, and baton continuation are autonomous. Least-privilege controls, sensitive-data handling, rollback, and complete data-flow disclosure are incomplete, so points are deducted.
The execution protocol, file structure, metadata schema, troubleshooting table, and baton validation rules are fairly specific; however, key failures such as project creation, download, invalid page names, and permission errors are not handled in depth. There are no skill-specific tests or reproducible execution records, so the static ceiling applies and points are deducted.
The audience, iterative scenario, input files, and principal outputs are reasonably clear, and optional Chrome verification is described; however, non-fit boundaries, semantic trigger conditions, Chinese support, and mainland-China network reachability are not addressed. Core operation depends on Stitch MCP, so environmental fit remains uncertain.
The documentation is layered and includes prerequisites, workflow steps, examples, resource templates, troubleshooting, and limitation notes. The repository supplies an Apache-2.0 license, security policy, and CI, but lacks a clear maintainer, skill version, changelog, or explicit update path. The README's claim of automated validation is also stronger than the available evidence, so points are deducted.
The workflow covers reading the task, generating a page, integrating it, updating navigation and documentation, and preparing the next baton; it could complete the core iteration and produce usable HTML. However, generation quality, downloads, link repairs, and visual fidelity are unexecuted and unverified, and autonomous looping adds review cost, so the score remains below the static maximum.
The source files, examples, and metadata schema are auditable, and the repository has CI; however, the CI validates a different react-components skill, with no stitch-loop tests, real run logs, or independent corroboration. The conclusion is therefore mainly documentation-based and constrained by the static ceiling.
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 →