What problem does it solve?
The stitch-loop Skill removes the manual, error-prone work of repeatedly generating, integrating, and sequencing website pages by establishing an autonomous baton-passing workflow that reads a task, generates a screen with Stitch, integrates assets into the site, updates project metadata, and writes the next task.
Core Features & Use Cases
- Baton-driven Iterations: Uses a single baton file to coordinate successive agent iterations and ensure continuous progress.
- Stitch MCP Integration: Calls Stitch MCP create/get/generate APIs to create projects, generate screens, and persist screen metadata.
- Asset Retrieval & Integration: Downloads HTML and screenshots, moves HTML into site/public, fixes asset paths, and updates navigation and site documentation.
- Project State Management: Persists a .stitch/metadata.json map of project and screen identifiers and enforces sitemap and roadmap checks.
- Optional Visual Verification: Supports Chrome DevTools MCP for automated rendering and screenshot comparisons when available.
- Use cases include autonomous site expansion in CI/CD pipelines, iterative agent-driven prototyping, and low-touch content generation for marketing or product sites.
Quick Start
Use the stitch-loop skill to read the baton from .stitch/next-prompt.md, generate the page via Stitch MCP tools, integrate the generated page into site/public, update .stitch/SITE.md and .stitch/metadata.json, and write a new .stitch/next-prompt.md to continue the loop.