Stitch Agent Skills aims to make Stitch's design and generation capabilities consumable as small, reusable "skills" for coding agents. Rather than treating Stitch as an isolated tool, this repository packages common workflows—prompt enhancement, design-system synthesis, multi-page site generation, component extraction, and video walkthrough export—as discoverable skills that an agent can install and invoke automatically. That makes it easier to orchestrate Stitch-driven UI work from a range of coding agents and CLIs.
What Sets It Apart
- Agent-first packaging: Skills follow the Agent Skills open standard so multiple coding agents (Antigravity, Gemini CLI, Claude Code, Cursor and others) can discover and run them without bespoke adapters. This reduces integration work when adding Stitch capabilities to different agent runtimes.
- Focused, composable missions: Each skill is organized with a clear SKILL.md, resources, examples, and executable validators. That structure is intended to give agents the context and checks they need to perform higher-quality, few-shot-driven tasks (e.g., synthesize a DESIGN.md, convert Stitch screens to React components, or generate a multi-page site from one prompt).
- CLI-first discovery/install: The repository is designed to be consumed via a skills CLI (npx skills add ...), which automates detection of the active agent environment and installs a skill into the appropriate directory—so agents or developer workflows can add functionality with a single command rather than manual wiring.
Who It's For and Tradeoffs
Great fit if you are a developer or design-engineer who:
- Uses Stitch (or plans to) as part of a generative UI workflow and wants those capabilities exposed to agent-driven automation;
- Builds or extends coding agents and need a standardized skillset to call Stitch tasks (UI prompt enhancement, design documentation, component extraction, site generation, or video export);
- Values modular, example-driven skills that include validators and style guides to improve agent outputs.
Look elsewhere if:
- You do not use Stitch or an MCP-compatible agent runtime—this repo is optimized for the Stitch + MCP ecosystem and assumes that workflow;
- You expect a polished end-user application—this is a developer-facing collection of skills, not a consumer product or a hosted service.
Where It Fits
This repository is functionally a bridge between Stitch (design/screen generation) and agent orchestration layers. It sits alongside other agent tooling (agent skill frameworks, orchestration libraries, or LLM-based code assistants) and is most valuable when you need repeatable, validated Stitch-related tasks callable by different agent implementations.
Notable Practical Notes
- The project was created on 2026-01-16 and is provided by a Google Labs Code organization. The README explicitly states it is not an officially supported Google product and excludes the project from Google’s OSS vulnerability rewards program—important if you need formal support or security guarantees.
In short: Stitch Agent Skills reduces the integration friction of exposing Stitch workflows to agents by packaging common UI and developer workflows as discoverable, validated skills—best for teams embedding Stitch into automated agent-driven pipelines.
