The AI agent ecosystem is fragmenting. Claude Code, Cursor, Gemini CLI, and Antigravity each build their own tool integrations, forcing developers to rewrite the same capabilities across incompatible platforms. Google Labs' Stitch Skills framework, which surged to 338 stars on GitHub today, attempts to solve this through standardization. The library provides a collection of pre-built Agent Skills designed to work with the Stitch MCP (Model Context Protocol) server, following an open Agent Skills standard that enables cross-platform compatibility. This is essentially Terraform-for-agents: instead of each tool defining infrastructure resources differently, Stitch defines a common interface that any agentic coding platform can adopt. A skill—say, 'execute shell command' or 'read file from repository'—follows a single schema, making it portable across Claude's Artifacts, Cursor's agent, and other emerging coding assistants.
The friction being addressed is real and concrete. Currently, if you build a skill for Claude Code that chains multiple API calls, file operations, and syntax validation, you cannot reuse that definition in Cursor without rewriting the entire integration. This creates redundant engineering work and prevents skill libraries from network effects. The Stitch approach defines skills using a standardized interface with clear input/output contracts, much like how Terraform providers abstract cloud APIs into a common language. Early signals suggest real adoption potential: Claude Code, Cursor, and Gemini CLI are listed as compatible platforms, though it remains unclear whether this represents built-in support or aspirational roadmap items. What matters is that Google Labs is positioning Stitch as infrastructure-grade standardization rather than another proprietary wrapper—the MCP protocol underpinning it was already adopted by Anthropic and others, lending legitimacy to the approach.
Yet skeptics rightfully ask: is this solving fragmentation or adding another abstraction layer? Every new standard risks becoming yet another thing developers must learn. The competitive context matters here—no other major standardization push has emerged at this scope level, giving Stitch first-mover advantage. However, adoption hinges on whether tool makers actually implement Stitch-compatible skill runners versus ignoring it. The comparison to Terraform is apt but cautionary: Terraform succeeded because cloud providers had strong incentives to support it. For Stitch to become the de facto standard, Claude, Cursor, and competitors must see skills-as-portable-units as more valuable than lock-in. The framework's open-source nature helps, but without mandatory adoption or killer apps, it risks becoming another underutilized abstraction.