OpenAI, Microsoft, Amazon Launch Agent Plugins Open Standard for AI Agents

Major AI Companies Standardize How AI Agents' Skills Are Packaged

OpenAI, Microsoft, Amazon, Cursor, and Vercel have joined forces to launch Agent Plugins 1.0, a vendor-neutral open standard designed to standardize how AI agents' skills are packaged and deployed1

1

2

2

. Vercel initiated the proposal with input from these tech giants, who all pledged engineers and resources to the project1

1

. The specification was quickly embraced by the Linux Foundation's Agentic AI Foundation (AAIF), though it operates as an independent entity under open governance1

1

.

The open standard addresses a fundamental problem: today, every AI agent platform expects different folder layouts and configurations, forcing developers to rebuild tools for each environment. Agent Plugins eliminates this friction by introducing a unified package format that enables write once, run everywhere deployment across competing products2

2

3

3

. ChatGPT, GitHub Copilot, Cursor, VS Code, Codex, and Kiro all support the standard at launch1

1

2

2

.

How Agent Plugins Combines Agent Skills and MCP Servers

Agent Plugins 1.0 packages software agent extensions in a single directory structure built on two existing components. Agent Skills hold modular, reusable instructions for how agents should perform specific tasks, while MCP Servers (Model Context Protocol) provide connectivity to external databases, APIs, shell scripts, and other resources1

1

3

3

. Together, these components make AI agent add-ons modular and portable across platforms.

The specification prescribes creating a top-level directory for each plugin containing a plugin.json manifest file at its root2

2

3

3

. This manifest provides the plugin's name, version, and schema in JSON format. Skills are filed in a skills sub-directory, while MCP server configuration is stored as an mcp.json file1

1

. Developers can also use a namespace directory to store additional extensions, though portability of these custom elements isn't guaranteed1

1

.

Agent Plugins follows the Next.js practice of using the host computer's directory format as the structure for defining components, keeping the shared format predictable without requiring it to absorb every experimental feature happening across agent products1

1

.

Deliberate Scope Limitations and What's Left to Platforms

The specification deliberately keeps its scope narrow for this 1.0 release. It defines only how plugins are packaged and discovered, leaving critical elements like marketplaces, installation, permissions, sandboxing, and trust to individual client platforms1

1

2

2

. This approach prioritizes ease of adoption while sidestepping the hardest problems in shared AI agent infrastructure.

Deciding whether a plugin is safe to run remains each client's responsibility, a live concern after fake Agent Skills slipped past security scanners earlier in 20252

2

. It's still up to agent platforms to provide user interfaces, establish security boundaries, and create marketplaces where plugins can be shared1

1

. The standardized folder structure ensures consistent integration, but elements like runtime environments and permission models remain platform-specific3

3

.

Developers have mixed reactions. Angie Jones, a developer advocate and AAIF Vice President, celebrated the move, saying "We neeeeded this" for carrying skills between tools2

2

. However, Dax Raad of SST developer-tools framework expressed skepticism, calling it "a thin standard" whose useful parts will likely end up in client-specific extensions anyway2

2

.

Interoperability Benefits and Competition Concerns

The vendor-neutral open standard offers significant advantages for developers building AI agents. The unified format eliminates platform-specific adjustments, reducing development time and allowing seamless integration into diverse environments3

3

. Enhanced portability means plugins can be easily shared and reused across different systems, reducing redundant work and fostering collaboration3

3

. The flexible architecture supports adding new features as projects evolve3

3

.

Yet standardizing the plumbing raises questions about market dynamics. A shared format could let small developers reach every major agent platform at once, democratizing access to the AI agent ecosystem2

2

. Conversely, it might cement the handful of clients that already have user bases, since standards reward whoever people already use2

2

. The fight over marketplaces and trust mechanisms that Agent Plugins deliberately excludes has barely started2

2

.

Agent Plugins will be managed under an open governance model with an independent technical oversight committee considering public input1

1

. The spec is licensed under Creative Commons Attribution 4.0 International License. Engineers from Amazon, Cursor, Microsoft, OpenAI, and Vercel currently form the core development team1

1

. Watch how competing approaches like AAIF's Skills Over MCP, which bundles skills into MCP servers themselves, fare against this plugin-centric model1

1

.