Google Labs has published a small repository, google-labs-code/stitch-skills, that packages its Stitch interface-design tool as a set of “agent skills” instead of a standalone product. According to the project's own description, it is “a library of Agent Skills designed to work with the Stitch MCP server,” and each skill follows the Agent Skills open standard so it stays portable across coding agents — the README names Codex and Antigravity among those it supports.
The packaging is the story. Stitch turns text prompts into UI screens, and the companion stitch-sdk lets developers generate those screens programmatically and pull out their HTML and screenshots. By shipping the capability as Agent Skills behind an MCP server, Google Labs makes its design tool callable from agents built by other vendors rather than confining it to a Google-only workflow.
It is also one more step in the consolidation of the Agent Skills format, which has drifted over the past few months from hobbyist prompt packs toward vendor-shipped libraries. The same google-labs-code organization already publishes Agent Skills for a wider set of Google products and technologies, including Google Cloud.
The Stitch repository itself is young — a handful of skills, days old — and drew roughly 250 stars in a day on GitHub Trending. But it lands in a busy lane: control of the plumbing that connects coding agents to the tools they drive. For teams already putting Stitch in a design-to-code pipeline, the practical change is that the tool can now be invoked by the agent they already run, not only from Google's own surface.

