Skip to content
BSWEN
Blog

What is OpenAI Codex Plugin System: A Package Manager for AI Agents

Cowrie
Cowrie
Dev @ Bswen

The Problem

AI programming agents need a way to share capabilities across teams.

Right now, if you create a useful prompt workflow for an AI agent, you have no good way to package it, version it, and distribute it to your team. You copy-paste prompts. You hope everyone uses the same version. It’s messy.

The Solution

OpenAI created a plugin system for Codex. I think of it as npm for AI agent capabilities.

Each plugin bundles three things:

  1. Skills - Prompts that guide the agent through workflows
  2. MCP Servers - Remote tools and shared context
  3. Apps - Application integrations and connectors
Plugin Architecture
+------------------+
|     Plugin       |
+------------------+
|                  |
|  +------------+  |
|  |   Skills   |  |  --> Workflow prompts
|  +------------+  |      (agent discovers progressively)
|                  |
|  +------------+  |
|  | MCP Servers|  |  --> Remote tools
|  +------------+  |      + shared context
|                  |
|  +------------+  |
|  |    Apps    |  |  --> Integrations
|  +------------+  |      + connectors
|                  |
+------------------+

Why This Matters

Prompt engineering becomes software engineering.

You can now:

  • Version your AI workflows (1.0.0, 1.1.0, etc.)
  • Distribute them to your team
  • Control what plugins are available or installed by default
  • Cache installations locally for speed

This is the same model npm uses. It’s proven. It works.

How It Works

The key file is plugin.json. Think of it as package.json for AI capabilities.

plugin.json
{
  "name": "my-codex-plugin",
  "version": "1.0.0",
  "interface": {
    "brandColor": "#007ACC",
    "logo": "./logo.svg",
    "screenshots": ["./screens/demo.png"],
    "defaultPrompt": "Use this plugin for..."
  },
  "skills": ["./skills/"],
  "mcpServers": ["./mcp/"],
  "apps": ["./apps/"]
}

The interface section is interesting. It has brandColor, logo, and screenshots. This tells me an official Marketplace is coming. You’ll browse and install plugins from a registry, just like npm.

Installation Policies

Teams can control what plugins are available:

Policy Levels
AVAILABLE            --> Users can install
INSTALLED_BY_DEFAULT --> Auto-installed for everyone
NOT_AVAILABLE        --> Hidden from team

This gives organizations control over what AI capabilities their teams can use.

Current Limitations

The system is new. Here’s what’s missing:

  1. No public directory yet - You can’t publish to a central registry
  2. Local source only - Plugins load from local files
  3. Incomplete security docs - The security model isn’t fully documented

What You Should Do Now

Start packaging your AI workflows as plugins.

Even without a public registry, you can:

  1. Create plugin.json for your existing prompts
  2. Structure your skills, MCP servers, and apps
  3. Version them properly
  4. Share within your team via git

When the official Marketplace opens, you’ll be ready.

Final Words + More Resources

My intention with this article was to help others share my knowledge and experience. If you want to contact me, you can contact by email: Email me

Here are also the most important links from this article along with some further resources that will help you in this scope:

Oh, and if you found these resources useful, don’t forget to support me by starring the repo on GitHub!

Comments

Tags: