# What Is the Purpose of the OpenAI Plugins Repository? A Complete Guide to Codex Plugin Development > Explore the openai/plugins repository, your guide to developing Codex plugins. Discover examples, tools, and best practices for building and publishing your own. - Repository: [OpenAI/plugins](https://github.com/openai/plugins) - Tags: tutorial - Published: 2026-07-03 --- **The openai/plugins repository is a curated collection of Codex plugin examples that provides reference implementations, scaffolding tools, and best practices for building, packaging, and publishing plugins for the Codex platform.** The openai/plugins repository serves as the official source of truth for developers building extensions for OpenAI's Codex platform. It contains production-grade examples like Figma, Notion, and Mixpanel that demonstrate how to structure plugin manifests, define skills, and prepare packages for the Codex marketplace. Each plugin in the repository follows a standardized architecture that ensures compatibility with the Codex runtime and marketplace. ## Core Purpose and Design Goals The repository fulfills five strategic objectives for the Codex ecosystem: - **Reference Implementation**. It provides concrete, production-grade examples (e.g., Figma, Notion, Mixpanel) that illustrate best-practice patterns for plugin structure, manifest fields, and skill definitions. - **Learning and Onboarding**. New developers can explore the folder layout, read embedded documentation, and run plugins locally to understand the end-to-end workflow from manifest creation to runtime execution. - **Reusable Building Blocks**. Common patterns (such as the `plugin-creator` skill and asset pipelines) are abstracted into reusable components that other plugins can copy or extend. - **Marketplace Readiness**. By adhering to the mandated manifest schema and optional [`.app.json`](https://github.com/openai/plugins/blob/main/.app.json) metadata, each plugin can be published to the Codex marketplace without additional conversion work. - **Community Collaboration**. The open-source nature encourages contributions, bug fixes, and new plugin ideas that benefit the broader ecosystem. ## Directory Structure and Architecture Every plugin resides under `plugins//` and follows a strict directory convention that the Codex runtime recognizes. ``` plugins/ ├─ / │ ├─ .codex-plugin/ │ │ └─ plugin.json ← mandatory manifest │ ├─ .app.json ← optional UI/marketplace metadata │ ├─ skills/ ← skill definitions (Markdown + YAML) │ ├─ agents/ ← agent scripts (optional) │ ├─ commands/ ← CLI commands (optional) │ └─ assets/ ← icons, images, etc. ``` The **[`plugin.json`](https://github.com/openai/plugins/blob/main/plugin.json)** file in `.codex-plugin/` is mandatory and defines the plugin's identity, capabilities, and entry points. The optional **[`.app.json`](https://github.com/openai/plugins/blob/main/.app.json)** supplies UI-specific fields such as branding, screenshots, and default prompts. **Skills** are Markdown files with YAML front-matter that the Codex runtime parses to expose MCP tools. ## The Plugin Manifest Schema The manifest (`plugins//.codex-plugin/plugin.json`) is the authoritative source for plugin metadata. The Figma plugin manifest illustrates the required structure: ```json { "name": "figma", "version": "2.0.9", "description": "Figma workflows for design implementation, Code Connect templates, and design system rule generation.", "author": { "name": "Figma", "url": "https://www.figma.com" }, "homepage": "https://www.figma.com", "repository": "https://github.com/openai/plugins", "license": "LicenseRef-Figma-Developer-Terms", "keywords": ["figma", "design-to-code", "ui-implementation", "code-connect", "design-system"], "skills": "./skills/", "apps": "./.app.json", "interface": { "displayName": "Figma", "shortDescription": "Design-to-code workflows powered by the Figma integration", "category": "Creativity", "capabilities": ["Interactive", "Read", "Write"], "websiteURL": "https://www.figma.com", "privacyPolicyURL": "https://www.figma.com/legal/privacy/", "termsOfServiceURL": "https://www.figma.com/legal/developer-terms/", "defaultPrompt": [ "Inspect a Figma design and implement it in code", "Create Code Connect templates for my components", "Build or update a screen in Figma" ], "brandColor": "#1ABCFE", "composerIcon": "./assets/logo-padded.png", "logo": "./assets/logo-padded.png" } } ``` Key fields include **`capabilities`** (which declares permissions like `Interactive`, `Read`, and `Write`), **`skills`** (the relative path to skill definitions), and **`interface`** (UI branding and marketplace listing data). ## Skill Definitions and MCP Integration Skills are the functional units of a Codex plugin. Each skill is defined in a [`SKILL.md`](https://github.com/openai/plugins/blob/main/SKILL.md) file located under `plugins//skills//`. The **`figma-use-slides`** skill in [`plugins/figma/skills/figma-use-slides/SKILL.md`](https://github.com/openai/plugins/blob/main/plugins/figma/skills/figma-use-slides/SKILL.md) demonstrates the standard format: - **YAML front-matter** containing `name`, `description`, and `disable-model-invocation` flags - **Markdown sections** detailing usage instructions, design thinking guidelines, and reference links - **MCP tool definitions** that the Codex runtime exposes to the language model When `disable-model-invocation` is set to `false`, the Codex runtime can automatically invoke the skill's MCP tools during agent execution. ## Scaffolding New Plugins with the Plugin-Creator Skill The repository includes a meta-skill for generating new plugins. The **`plugin-creator`** skill, located at [`.agents/skills/plugin-creator/SKILL.md`](https://github.com/openai/plugins/blob/main/.agents/skills/plugin-creator/SKILL.md), automates the creation of boilerplate directories and manifest files. When invoked, this skill generates: 1. A new `plugins//` directory 2. A valid [`.codex-plugin/plugin.json`](https://github.com/openai/plugins/blob/main/.codex-plugin/plugin.json) with populated required fields 3. Standard subdirectories (`skills/`, `assets/`, etc.) This makes the repository a self-referential toolkit: it contains the tools to extend itself. ## Practical Implementation Examples ### Loading a Manifest in Node.js To programmatically read a plugin's metadata: ```javascript const fs = require('fs'); const path = require('path'); const manifestPath = path.resolve( __dirname, 'plugins/figma/.codex-plugin/plugin.json' ); const manifest = JSON.parse(fs.readFileSync(manifestPath, 'utf8')); console.log('Plugin name:', manifest.name); console.log('Version:', manifest.version); console.log('Capabilities:', manifest.interface.capabilities); ``` ### Configuring an Agent with YAML Define an agent that uses the plugin-creator skill: ```yaml name: plugin-creator description: Scaffold a new Codex plugin model: gpt-4 skillNames: - plugin-creator parameters: pluginName: str version: str description: str ``` ### Invoking a Skill at Runtime ```javascript // Pseudocode for an agent invoking the figma-use-slides skill await codex.runSkill({ skillName: "figma-use-slides", input: { action: "create_new_slide", title: "Quarterly Results", layout: "title-subtitle-chart" } }); ``` ## Summary - The **openai/plugins repository** is a curated library of Codex plugin examples serving as reference implementations and learning resources. - Each plugin requires a mandatory **[`plugin.json`](https://github.com/openai/plugins/blob/main/plugin.json)** manifest in `.codex-plugin/` and optionally includes [`.app.json`](https://github.com/openai/plugins/blob/main/.app.json) for UI metadata. - **Skills** are defined in Markdown files with YAML front-matter under `skills/`, exposing MCP tools to the Codex runtime. - The **`plugin-creator`** skill at [`.agents/skills/plugin-creator/SKILL.md`](https://github.com/openai/plugins/blob/main/.agents/skills/plugin-creator/SKILL.md) provides automated scaffolding for new plugins. - Production examples like **Figma**, **Notion**, and **Mixpanel** demonstrate marketplace-ready patterns for capabilities, asset management, and branding. ## Frequently Asked Questions ### What is the primary purpose of the openai/plugins repository? The openai/plugins repository serves as a curated collection of Codex plugin examples that demonstrates how to build, package, and publish plugins for the Codex platform. It provides production-grade reference implementations (such as Figma and Notion) that developers can study to understand manifest schemas, skill definitions, and marketplace requirements. ### How do I create a new plugin using the openai/plugins repository? Use the **plugin-creator** skill located at [`.agents/skills/plugin-creator/SKILL.md`](https://github.com/openai/plugins/blob/main/.agents/skills/plugin-creator/SKILL.md) to scaffold a new plugin automatically. This skill generates the mandatory directory structure, creates a valid [`.codex-plugin/plugin.json`](https://github.com/openai/plugins/blob/main/.codex-plugin/plugin.json) manifest, and populates standard subdirectories like `skills/` and `assets/`, ensuring your plugin follows Codex platform conventions from the start. ### What files are required for a valid Codex plugin in the openai/plugins repository? Every Codex plugin must include a **[`plugin.json`](https://github.com/openai/plugins/blob/main/plugin.json)** file inside a `.codex-plugin/` directory. This manifest defines the plugin's name, version, capabilities, skill paths, and author metadata. While optional, you should also include a **[`.app.json`](https://github.com/openai/plugins/blob/main/.app.json)** file for UI/marketplace metadata, `skills/` directory containing [`SKILL.md`](https://github.com/openai/plugins/blob/main/SKILL.md) files, and an `assets/` directory for icons referenced by the manifest. ### How do skills work within the openai/plugins ecosystem? Skills are functional units defined in Markdown files with YAML front-matter (typically under `plugins//skills/`). They expose MCP tools to the Codex runtime through the `skillNames` parameter. The front-matter includes fields like `name`, `description`, and `disable-model-invocation`, which control whether the language model can automatically invoke the skill during agent execution.