Anthropic defined the standards inside Agent Plugins. So why isn’t it helping govern the format?
Earlier this month, Vercel released Agent Plugins 1.0.0 with core maintainers from AWS, Cursor, Microsoft, and OpenAI. Google announced the The post Anthropic defined the standards inside Agent Plugins. So why isn’t it helping govern the format? appeared first on The New Stack .
Anthropic helped establish standards for Agent Plugins, but it does not seem to be involved in governing the format. In early February, Vercel, AWS, Cursor, Microsoft, and OpenAI released Agent Plugins 1.0.0, a format designed to address a pain point for developers: skills and MCP servers working independently still need to be forked and maintained twice when shipped to multiple clients.
The six companies agreed on the components' locations but did not standardize on their functionality. Section 7.1 of the spec assigns the skill format to the Agent Skills specification, while Section 7.2 assigns wire behavior and lifecycle to MCP. Both specifications, released openly, were created by Anthropic but are overseen by the Agentic AI Foundation, a Linux Foundation-directed fund.
The Agent Skills specification is now being developed openly at agentskills.io. The governance of these standards differs, with MCP belonging to an independent foundation and Agent Skills being openly developed. The Agent Plugins Technical Steering Committee includes representatives from Amazon, Cursor, Microsoft, OpenAI, and Vercel, who serve as Lead Core Maintainer.
However, Anthropic is not part of the Technical Steering Committee. The manifest structure of a plugin includes a plugin.json file at its root for basic information, separate directories for skills and MCP servers, and an mcp.json file defining the configuration shape. The specification does not govern the components themselves but rather where they are located within the plugin directory structure.
This approach avoids layering behavioral requirements on top of the separately governed specifications for Agent Skills and MCP. Clients can be either skills-only or MCP-capable, and a plugin can contain both types of components. However, a conformant client must support at least one component type: skills or MCP servers, and it must implement at least one transport method.
The compatible clients page lists various tools that support Agent Skills and MCP, including VS Code, Cursor, GitHub Copilot, ChatGPT, Codex, Kiro, Hermes Agent, OpenClaw, and Grok Bot. Despite the compatibility, the bare claim of Agent Plugins conformance does not guarantee interoperability. Instead, it is the per-client capability matrix that determines interoperability, which is documented but not part of the conformance claim itself.
It is unclear why Anthropic is absent from the governance of the Agent Plugins format, as it played a crucial role in establishing the standards.
Written by urgent.news from The New Stack's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
