google-labs-code / stitch-sdk

Only one skill in this repository teaches an agent to consume the Stitch SDK; the rest serve the people building the SDK itself.

stitch-sdk-usage is the consumer-facing skill: generate UI screens from a text prompt, edit a screen and produce variants of it, manage projects and design systems, and retrieve a screen's HTML and screenshot — through the @google/stitch-sdk npm package, its Vercel AI SDK tool bindings, or Google's ADK for TypeScript. Every one of those paths needs a STITCH_API_KEY; there is no free-standing mode.

The rest are Google Labs' own development tooling for this repository, not for someone using the published package. stitch-sdk-pipeline, stitch-sdk-domain-design, and stitch-sdk-development are stages of the SDK's own code-generation pipeline — mapping MCP tools to domain classes, then implementing them; stitch-sdk-bug-bash finds bugs in the SDK using a real API key; stitch-sdk-readme regenerates this repository's own README; and tdd-red-green-refactor and github-codebase-briefing are generic skills, a TDD workflow enforcer and a repository-briefing tool, that are not specific to Stitch at all and would do the same job bundled into any other repository.

READ THE FULL ANALYSIS

Only stitch-sdk-usage is for SDK users. If you came here to generate UI from prompts in your own app, you want exactly one of these skills; the rest exist because this is Google Labs' working repository for the SDK, not a curated skill pack for its users.

One inconsistency worth flagging: our own repository metadata does not mark this package as an official Google repository, even though it publishes to npm under the @google scope and lives in the google-labs-code GitHub organization. We have not resolved that discrepancy and are reporting it rather than picking a side.