Global Rank · of 600 Skills
stitch-loop AI Agent Skill
View Source: google-labs-code/stitch-skills
SafeInstallation
npx skills add google-labs-code/stitch-skills --skill stitch-loop 32.4K
Installs
Stitch Build Loop
You are an autonomous frontend builder participating in an iterative site-building loop. Your goal is to generate a page using Stitch, integrate it into the site, and prepare instructions for the next iteration.
Overview
The Build Loop pattern enables continuous, autonomous website development through a "baton" system. Each iteration:
- Reads the current task from a baton file (
.stitch/next-prompt.md) - Generates a page using Stitch MCP tools
- Integrates the page into the site structure
- Writes the next task to the baton file for the next iteration
Prerequisites
Required:
- Access to the Stitch MCP Server
- A Stitch project (existing or will be created)
- A
.stitch/DESIGN.mdfile (generate one using thedesign-mdskill if needed) - A
.stitch/SITE.mdfile documenting the site vision and roadmap
Optional:
- Chrome DevTools MCP Server — enables visual verification of generated pages
The Baton System
The .stitch/next-prompt.md file acts as a relay baton between iterations:
---
page: about
---
A page describing how jules.top tracking works.
**DESIGN SYSTEM (REQUIRED):**
[Copy from .stitch/DESIGN.md Section 6]
**Page Structure:**
1. Header with navigation
2. Explanation of tracking methodology
3. Footer with linksCritical rules:
- The
pagefield in YAML frontmatter determines the output filename - The prompt content must include the design system block from
.stitch/DESIGN.md - You MUST update this file before completing your work to continue the loop
Execution Protocol
Step 1: Read the Baton
Parse .stitch/next-prompt.md to extract:
- Page name from the
pagefrontmatter field - Prompt content from the markdown body
Step 2: Consult Context Files
Before generating, read these files:
| File | Purpose |
|---|---|
.stitch/SITE.md |
Site vision, Stitch Project ID, existing pages (sitemap), roadmap |
.stitch/DESIGN.md |
Required visual style for Stitch prompts |
Important checks:
- Section 4 (Sitemap) — Do NOT recreate pages that already exist
- Section 5 (Roadmap) — Pick tasks from here if backlog exists
- Section 6 (Creative Freedom) — Ideas for new pages if roadmap is empty
Step 3: Generate with Stitch
Use the Stitch MCP tools to generate the page:
- Discover namespace: Run
list_toolsto find the Stitch MCP prefix - Get or create project:
- If
.stitch/metadata.jsonexists, use theprojectIdfrom it - Otherwise, call
[prefix]:create_project, then call[prefix]:get_projectto retrieve full project details, and save them to.stitch/metadata.json(see schema below) - After generating each screen, call
[prefix]:get_projectagain and update thescreensmap in.stitch/metadata.jsonwith each screen's full metadata (id, sourceScreen, dimensions, canvas position)
- If
- Generate screen: Call
[prefix]:generate_screen_from_textwith:projectId: The project IDprompt: The full prompt from the baton (including design system block)deviceType:DESKTOP(or as specified)
- Retrieve assets: Before downloading, check if
.stitch/designs/{page}.htmland.stitch/designs/{page}.pngalready exist:- If files exist: Ask the user whether to refresh the designs from the Stitch project or reuse the existing local files. Only re-download if the user confirms.
- If files do not exist: Proceed with download:
htmlCode.downloadUrl— Download and save as.stitch/designs/{page}.htmlscreenshot.downloadUrl— Append=w{width}to the URL before downloading, where{width}is thewidthvalue from the screen metadata (Google CDN serves low-res thumbnails by default). Save as.stitch/designs/{page}.png
Step 4: Integrate into Site
- Move generated HTML from
.stitch/designs/{page}.htmltosite/public/{page}.html - Fix any asset paths to be relative to the public folder
- Update navigation:
- Find existing placeholder links (e.g.,
href="#") and wire them to the new page - Add the new page to the global navigation if appropriate
- Find existing placeholder links (e.g.,
- Ensure consistent headers/footers across all pages
Step 4.5: Visual Verification (Optional)
If the Chrome DevTools MCP Server is available, verify the generated page:
- Check availability: Run
list_toolsto see ifchrome*tools are present - Start dev server: Use Bash to start a local server (e.g.,
npx serve site/public) - Navigate to page: Call
[chrome_prefix]:navigateto openhttp://localhost:3000/{page}.html - Capture screenshot: Call
[chrome_prefix]:screenshotto capture the rendered page - Visual comparison: Compare against the Stitch screenshot (
.stitch/designs/{page}.png) for fidelity - Stop server: Terminate the dev server process
Note: This step is optional. If Chrome DevTools MCP is not installed, skip to Step 5.
Step 5: Update Site Documentation
Modify .stitch/SITE.md:
- Add the new page to Section 4 (Sitemap) with
[x] - Remove any idea you consumed from Section 6 (Creative Freedom)
- Update Section 5 (Roadmap) if you completed a backlog item
Step 6: Prepare the Next Baton (Critical)
You MUST update .stitch/next-prompt.md before completing. This keeps the loop alive.
- Decide the next page:
- Check
.stitch/SITE.mdSection 5 (Roadmap) for pending items - If empty, pick from Section 6 (Creative Freedom)
- Or invent something new that fits the site vision
- Check
- Write the baton with proper YAML frontmatter:
---
page: achievements
---
A competitive achievements page showing developer badges and milestones.
**DESIGN SYSTEM (REQUIRED):**
[Copy the entire design system block from .stitch/DESIGN.md]
**Page Structure:**
1. Header with title and navigation
2. Badge grid showing unlocked/locked states
3. Progress bars for milestone trackingFile Structure Reference
project/
├── .stitch/
│ ├── metadata.json # Stitch project & screen IDs (persist this!)
│ ├── DESIGN.md # Visual design system (from design-md skill)
│ ├── SITE.md # Site vision, sitemap, roadmap
│ ├── next-prompt.md # The baton — current task
│ └── designs/ # Staging area for Stitch output
│ ├── {page}.html
│ └── {page}.png
└── site/public/ # Production pages
├── index.html
└── {page}.html.stitch/metadata.json Schema
This file persists all Stitch identifiers so future iterations can reference them for edits or variants. Populate it by calling [prefix]:get_project after creating a project or generating screens.
{
"name": "projects/6139132077804554844",
"projectId": "6139132077804554844",
"title": "My App",
"visibility": "PRIVATE",
"createTime": "2026-03-04T23:11:25.514932Z",
"updateTime": "2026-03-04T23:34:40.400007Z",
"projectType": "PROJECT_DESIGN",
"origin": "STITCH",
"deviceType": "MOBILE",
"designTheme": {
"colorMode": "DARK",
"font": "INTER",
"roundness": "ROUND_EIGHT",
"customColor": "#40baf7",
"saturation": 3
},
"screens": {
"index": {
"id": "d7237c7d78f44befa4f60afb17c818c1",
"sourceScreen": "projects/6139132077804554844/screens/d7237c7d78f44befa4f60afb17c818c1",
"x": 0,
"y": 0,
"width": 390,
"height": 1249
},
"about": {
"id": "bf6a3fe5c75348e58cf21fc7a9ddeafb",
"sourceScreen": "projects/6139132077804554844/screens/bf6a3fe5c75348e58cf21fc7a9ddeafb",
"x": 549,
"y": 0,
"width": 390,
"height": 1159
}
},
"metadata": {
"userRole": "OWNER"
}
}| Field | Description |
|---|---|
name |
Full resource name (projects/{id}) |
projectId |
Stitch project ID (from create_project or get_project) |
title |
Human-readable project title |
designTheme |
Design system tokens: color mode, font, roundness, custom color, saturation |
deviceType |
Target device: MOBILE, DESKTOP, TABLET |
screens |
Map of page name → screen object. Each screen includes id, sourceScreen (resource path for MCP calls), canvas position (x, y), and dimensions (width, height) |
metadata.userRole |
User's role on the project (OWNER, EDITOR, VIEWER) |
Orchestration Options
The loop can be driven by different orchestration layers:
| Method | How it works |
|---|---|
| CI/CD | GitHub Actions triggers on .stitch/next-prompt.md changes |
| Human-in-loop | Developer reviews each iteration before continuing |
| Agent chains | One agent dispatches to another (e.g., Jules API) |
| Manual | Developer runs the agent repeatedly with the same repo |
The skill is orchestration-agnostic — focus on the pattern, not the trigger mechanism.
Design System Integration
This skill works best with the design-md skill:
- First time setup: Generate
.stitch/DESIGN.mdusing thedesign-mdskill from an existing Stitch screen - Every iteration: Copy Section 6 ("Design System Notes for Stitch Generation") into your baton prompt
- Consistency: All generated pages will share the same visual language
Common Pitfalls
- ❌ Forgetting to update
.stitch/next-prompt.md(breaks the loop) - ❌ Recreating a page that already exists in the sitemap
- ❌ Not including the design system block from
.stitch/DESIGN.mdin the prompt - ❌ Leaving placeholder links (
href="#") instead of wiring real navigation - ❌ Forgetting to persist
.stitch/metadata.jsonafter creating a new project
Troubleshooting
| Issue | Solution |
|---|---|
| Stitch generation fails | Check that the prompt includes the design system block |
| Inconsistent styles | Ensure .stitch/DESIGN.md is up-to-date and copied correctly |
| Loop stalls | Verify .stitch/next-prompt.md was updated with valid frontmatter |
| Navigation broken | Check all internal links use correct relative paths |
Installs
Security Audit
Power your AI Agents with
the best open-source models.
Drop-in OpenAI-compatible API. No data leaves Europe.
Explore Inference APIGLM
GLM 5
$1.00 / $3.20
per M tokens
Kimi
Kimi K2.5
$0.60 / $2.80
per M tokens
MiniMax
MiniMax M2.5
$0.30 / $1.20
per M tokens
Qwen
Qwen3.5 122B
$0.40 / $3.00
per M tokens
How to use this skill
Install stitch-loop by running npx skills add google-labs-code/stitch-skills --skill stitch-loop in your project directory. Run the install command above in your project directory. The skill file will be downloaded from GitHub and placed in your project.
No configuration needed. Your AI agent (Claude Code, Cursor, Windsurf, etc.) automatically detects installed skills and uses them as context when generating code.
The skill enhances your agent's understanding of stitch-loop, helping it follow established patterns, avoid common mistakes, and produce production-ready output.
What you get
Skills are plain-text instruction files — not executable code. They encode expert knowledge about frameworks, languages, or tools that your AI agent reads to improve its output. This means zero runtime overhead, no dependency conflicts, and full transparency: you can read and review every instruction before installing.
Compatibility
This skill works with any AI coding agent that supports the skills.sh format, including Claude Code (Anthropic), Cursor, Windsurf, Cline, Aider, and other tools that read project-level context files. Skills are framework-agnostic at the transport level — the content inside determines which language or framework it applies to.
Chat with 100+ AI Models in one App.
Use Claude, ChatGPT, Gemini alongside with EU-Hosted Models like Deepseek, GLM-5, Kimi K2.5 and many more.