Vercel and OpenAI launch a cross-vendor standard for agent skills

Vercel and OpenAI introduced Agent Plugins, an open, vendor-neutral packaging standard for extending AI coding agents, backed at launch by Amazon Web Services, GitHub, Microsoft and Cursor-maker Anysphere. The format bundles two things that agent tools had previously handled separately: Agent Skills, reusable instructions and resources an agent can draw on, and MCP server configurations, which connect an agent to external tools and services. A plugin is a directory with a fixed manifest file and set locations for its components, so any compatible client can read it the same way.

The standard shipped with support across Codex, ChatGPT, Cursor, GitHub Copilot, Kiro and VS Code from launch, meaning a plugin author does not need to build separate integrations for each tool a developer might use. The problem it addresses is fragmentation: as coding agents from different vendors converged on similar capabilities — reading documentation, calling external APIs, following house coding conventions — each had developed its own incompatible format for packaging that context, forcing developers who wanted their skills or tool configurations to work everywhere to maintain multiple versions.

The announcement is part of a broader pattern of agent-tooling vendors agreeing common protocols — following Anthropic’s Model Context Protocol, which Agent Plugins builds on directly — rather than each company building a closed ecosystem, reducing switching costs between competing coding-agent products.