Building Agent Skills for ChatGPT and Codex: Architecture and Best Practices

ArticleTutorial

OpenAI details the implementation of agent skills for ChatGPT and Codex based on the open agent skills standard. Skills package instructions, references, and optional scripts into directory structures centered around a SKILL.md specification.

Key points

  1. Directory layout: A skill requires a directory containing a SKILL.md file with name, description, and instructions, alongside optional folders for scripts, references, and assets.
  2. Progressive disclosure: Context budgets are protected by loading only names and descriptions initially (capped at 2% of the context window or 8,000 characters), reading full instructions only upon selection.
  3. Invocation methods: Skills activate explicitly via @ in ChatGPT or $ / /skills in Codex, or implicitly when prompt requirements match the skill description.
  4. Authoring workflows: Developers can build skills manually, generate them using @skill-creator / $skill-creator, or capture workflows directly with Record & Replay.
  5. Local scopes and distribution: Codex discovers skills from repository hierarchies (.agents/skills), user paths, and system directories, while plugins serve as the vehicle for wider packaging and MCP integration.

Why it matters

Modular skill design allows engineering teams to codify domain knowledge and routine workflows directly alongside source code. Efficient context loading prevents prompt saturation while enabling scalable agentic capabilities across CLI and desktop environments.

Analysis

By relying on progressive disclosure and strict initial context boundaries, OpenAI addresses token exhaustion when managing extensive toolkits. The repository-aware scoping hierarchy provides a clean mechanism for sharing project-specific automation in version control while keeping global agent profiles clean.

Mentioned links

Found in

Summary generated by AI (Gemini 3.8 Flash, 27 September 2026). It may contain errors – the original source is authoritative.

Related entries

Similar in content, found by the AI search.

  • Link:Anthropic

    Agent Skills: An Open Standard for Modular AI Agent Capabilities

    Agent Skills is an open format designed to equip AI agents with specialised domain knowledge and repeatable workflows. Centred around a directory containing a SKILL.md file, it relies on progressive disclosure to keep context overhead minimal.

    AI & LLMs· Resource

  • Article:Anthropic

    Extending Claude Code with Skills: Architecture and Best Practices

    Anthropic details the skills system for Claude Code. Developers can encapsulate recurring procedures and workflows inside SKILL.md files built on the open Agent Skills standard, supporting dynamic context injection and subagent delegation.

    AI & LLMs· Tool

  • Article:Barry Zhang, Keith Lazuka, Mahesh Murag

    Anthropic Introduces Agent Skills: Modular File-Based Capabilities for AI Agents

    Anthropic has introduced Agent Skills, an open specification allowing agents like Claude to dynamically load procedural knowledge and executable scripts from structured folders. The approach uses progressive disclosure to keep system prompts lean while providing access to deep domain expertise.

    AI & LLMs· Announcement

  • Article:learn.chatgpt.com

    ChatGPT and Codex: Unified Plugin Ecosystem with MCP and Connectors

    OpenAI outlines the universal plugin architecture for ChatGPT and Codex, enabling reusable workflows with skills, MCP servers, connectors, and lifecycle hooks across web, desktop, mobile, and Codex CLI surfaces.

    AI & LLMs· Resource

  • Video

    Video:The AI Automators

    Agent Plugins: New Open Standard for AI Agent Extensions

    Industry players including Google, Vercel, and AWS are backing the new 'Agent Plugins' open specification. The standard packages Agent Skills and MCP servers into a unified folder structure, allowing extensions to work across multiple coding agents like Codex, Cursor, VS Code, and GitHub Copilot. Anthropic is currently not participating, though conversion tools exist.

    AI & LLMs· Tutorial

Back to the knowledge base

Let's talk about your project

Get in touch

Locations

  • Mattersburg
    Johann Nepomuk Bergerstraße 7/2/14
    7210 Mattersburg, Austria
  • Vienna
    Ungargasse 64-66/3/404
    1030 Wien, Austria

Parts of this content were created with the assistance of AI.