AI agents' Skills standardized—ChatGPT, Codex backed, Claude left out

Vercel's Agent Plugins standard addresses a fragmentation problem in the emerging AI agent ecosystem. As different AI platforms and code editors each developed their own ways for agents to reference procedures and connect to external services, developers faced duplication: writing the same extension multiple times to support ChatGPT, VS Code, GitHub Copilot, and others. By proposing a shared format for Agent Skills (the instruction files agents consult at runtime) and MCP server configurations, Vercel aims to create a write-once, use-everywhere model that mirrors how web standards reduced friction across browsers.

The backing of OpenAI, Amazon, Cursor, and Microsoft suggests mainstream adoption is likely among leading platforms. However, Anthropic's absence is notable: neither Claude Code nor Claude Cowork currently supports the standard, despite Claude's prominence in the agent space. This gap may indicate a deliberate choice by Anthropic to maintain its own extension architecture, or it may be a matter of timing—adoption could still follow. The eight clients now supporting Agent Plugins represent a critical mass around which developers can build, but broader coverage (especially Anthropic) would signal a true industry consensus.