Article on # AI titans to tidy agent fron...

Vercel published the initial draft of the Agent Plugins spec on Thursday, with backing from Amazon, Cursor, Microsoft, and OpenAI. The proposal aims to standardize how AI agents package their skills and tools, so developers can move them between platforms without rewriting code. The Linux Foundation's Agentic AI Foundation (AAIF) has embraced the project, though it will operate as an independent entity. The spec bundles an AI agent's capabilities into a single directory structure. It builds on two existing standards: the AAIF-stewarded MCP protocol for connecting to external resources, and the open Agent Skills standard originally created by Anthropic in 2025. Agent Skills holds the instructions for how an agent performs a specific task, while MCP handles connectivity to outside databases, APIs, and shell scripts. Together, the trio makes AI capabilities modular - write once, run anywhere.

File-system-as-configuration

Agent Plugins follows the Next.js practice of using the host computer's directory format as the structure for defining components. The spec prescribes creating a top-level directory for each plugin, with a small manifest and schema. The manifest provides the name, version, and schema of the plugin in JSON format. Skills are filed in a skills sub-directory, and the MCP server configuration is stored as an mcp.json file. Plugin authors can use a namespace directory to store additional extensions. "This format keeps the shared format predictable without requiring it to absorb every client-specific experiment happening across agent products," the AAIF said in a blog post. Bundling the skills with the MCP configuration makes the package portable, allowing it to be copied and used across any supporting agent client. VS Code, Cursor, GitHub Copilot, ChatGPT, Codex, and Kiro currently support the standard. Developers looking to understand how MCP fits into this workflow can explore MCP Courses & Training.

The narrow way

Agent developers are adding experimental hooks, rules, and extensions into their skills, but the spec's keepers deliberately kept the scope narrow for this 1.0 release. The best extensions may be considered for future versions, though the team is making no promises. Until then, developers can add these extensions to the namespace and hope for the best regarding portability. Agent Plugins will be managed under an open governance model with an independent technical oversight committee. Engineers from Amazon, Cursor, Microsoft, OpenAI, and Vercel currently sit on the core development team. The spec is licensed under the Creative Commons Attribution 4.0 International License. Other efforts are underway to tame the agent development frontier. Google folded its Agent2Agent (A2A) communications protocol into AAIF and is building a specification for finding tools on the web called Agentic Resource Discovery. AAIF also runs a complementary project called Skills Over MCP, which shares skills by bundling them into the MCP server itself - essentially shipping "the manual with the product," as AAIF Vice President Angie Jones wrote.

Why this matters for developers

The spec directly affects how you package and distribute agent skills. If you build agentic tools for clients or internal use, adopting the Agent Plugins format now means your work will transfer across Cursor, VS Code, ChatGPT, and other supported platforms as they roll out support. The write-once-run-anywhere promise is real but limited to the narrow 1.0 scope - extensions you add to the namespace won't be portable until the spec evolves. Developers who want to prepare for this shift can review the AI for Software Developers Learning Path to build practical skills in AI coding automation and agent tooling.