Google Labs Stitch-Skills: Agent Skills for Stitch MCP Server
WHY IT MATTERS
Google Labs released a library of Agent Skills designed for the Stitch MCP server, following the Agent Skills open standard for compatibility with Antigravity, Gemini CLI, Claude Code, and Cursor (~540 stars today).
What Happened
Google Labs published stitch-skills, a repository of Agent Skills built for the Stitch MCP server. The library implements the Agent Skills open standard, a format specification that defines how skills are packaged, declared, and discovered by agent harnesses. The repository surfaced on GitHub trending with roughly 540 stars in a single day, indicating fast pickup among agent tooling practitioners. The skills are positioned for compatibility across Antigravity, Gemini CLI, Claude Code, and Cursor.
Why It Matters
A first-party Google repository conforming to a vendor-neutral skill format is a portability signal. Until now, agent skill definitions have largely been harness-specific — a Claude Code skill and a Cursor rule and a Gemini CLI extension each required separate authoring and maintenance, which imposed a per-harness tax on anyone shipping reusable agent capability. By publishing against the Agent Skills standard and explicitly naming competitor harnesses as compatible targets, Google is treating skill definitions as an interchange layer rather than a lock-in surface. The practical beneficiary is any team maintaining a shared skill library across multiple agent environments: author once, distribute to whichever harness a given workflow runs on. The strategic read is that Google is competing on the skill ecosystem and MCP server surface rather than on harness exclusivity, which raises the cost of staying format-proprietary for other vendors.
Technical Details
The repository ships skills targeting the Stitch MCP server, meaning the skills are thin procedural wrappers — instructions, tool-call sequences, and context — that a host agent loads and executes against Stitch's MCP-exposed tools. Conformance to the Agent Skills open standard governs the packaging layout and metadata declaration, which is what enables the same skill directory to be consumed by Antigravity, Gemini CLI, Claude Code, and Cursor without per-harness rewrites. The MCP server dependency is the binding constraint: skills only function where the host harness supports MCP connections and the Stitch server is reachable and authenticated. Skill granularity and any harness-specific behavioral variance across the four named clients are not fully characterized in the repository surface, so operators should expect to validate execution semantics per harness rather than assume identical behavior.
Operational Impact
For teams already running multiple agent harnesses, this collapses a duplication cost: skill definitions become a single artifact in version control with harness-specific forks eliminated, which reduces review surface and drift between environments. Onboarding a new harness shifts from "re-author the skill set" to "point the harness at the existing skill directory and verify MCP connectivity," lowering the marginal cost of adding or swapping agent clients. The MCP server requirement pushes operators toward standardizing their tool connectivity layer — teams without an MCP-capable toolchain will need one to consume these skills at all. Expect skill libraries to start being evaluated as portable assets during vendor selection, which weakens harness-specific skill marketplaces as a differentiator and makes the underlying MCP tool surface the real switching cost.
SHARE
MORE FROM STUFFINSIDER