OpenAI, Vercel, AWS agree on plugin format
Agent Plugins 1.0.0 gives Agent Skills and MCP servers one packaging format that ChatGPT, Cursor, GitHub Copilot, VS Code, Codex, and Kiro can all load.
Vercel, AWS, Cursor-maker Anysphere, GitHub, Microsoft, and OpenAI have published Agent Plugins 1.0.0, a vendor-neutral standard for packaging the two things most agent extensions are already built from: Agent Skills (reusable instructions and resources) and MCP servers (live tool and data connections). A plugin is a directory with a plugin.json manifest in fixed locations, so any compatible client can discover and load it without a client-specific rebuild. The spec deliberately stays minimal, defining only discovery and validation, and leaves installation and distribution to each client.
Six clients support it at launch: ChatGPT, Codex, Cursor, GitHub Copilot, Kiro, and VS Code. A Technical Steering Committee with core maintainers from AWS, Cursor, Microsoft, OpenAI, and Vercel will govern the spec going forward; Anthropic is not among the initial partners.
What it means for you
The practical win is real: build a plugin once and it works across six major agent clients instead of six separate integration paths, similar to what Model Context Protocol already did for tool connections but at the packaging layer above it. If you maintain an internal skill or an MCP server your team uses across multiple tools, check whether repackaging it to the 1.0.0 manifest format is worth the afternoon it will likely take. The open question is namespacing and how cleanly client-specific features layer on top of a genuinely minimal shared contract; watch how Cline’s own plugin ecosystem and other non-launch clients respond before betting a distribution strategy entirely on this one format.