On February 2, 2026, OpenAI made a significant announcement: the launch of the Codex desktop app for macOS, complete with a brand new Skills Catalog. If you've been following the AI agent space, this might feel like déjà vu—because OpenClaw and its Clawdbot assistant have been running a skills-based architecture for months.

Welcome to the era of AI agent plugins. And OpenAI just joined the party.

What Are OpenAI Skills?

According to OpenAI's official announcement, Skills are "folders of instructions, scripts, and resources that AI agents can discover and use to perform at specific tasks." The tagline? "Write once, use everywhere."

Sound familiar? It should. This is the exact same philosophy that powers OpenClaw's skill system, where reusable capability bundles extend what AI agents can accomplish beyond simple text generation.

OpenAI's Skills framework consists of three tiers:

Tier	Description	Installation
.system	Core skills automatically installed with Codex	Automatic
.curated	OpenAI-vetted skills for common workflows	Via $skill-installer command
.experimental	Community-contributed, bleeding-edge skills	Manual installation with full path

The Skills Catalog lives in an open-source GitHub repository at github.com/openai/skills, which already has over 4,500 stars and 249 forks just days after launch. The repository contains 61 commits from 15 contributors—a sign of rapid development and community interest.

The OpenClaw Connection: Skills Before They Were Cool

For those unfamiliar, OpenClaw is an open-source AI agent platform that has been pioneering extensible skill architectures since 2025. Its flagship assistant, Clawdbot, uses a remarkably similar approach to what OpenAI just announced:

Skill bundles that package instructions, scripts, and tools
Automatic discovery of relevant skills based on context
Cross-platform compatibility across CLI, desktop, and IDE
Open contribution model for community skills

The parallels are striking. When OpenAI says Codex can now "use code to get work done on your computer," they're describing exactly what OpenClaw has been enabling—browser control, file manipulation, API integrations, and workflow automation through composable skills.

Industry observers have noted the timing. As Simon Willison wrote in his newsletter: "OpenAI are quietly adopting skills, now available in ChatGPT and Codex CLI." The word "quietly" speaks volumes—this isn't a revolutionary new concept, it's OpenAI catching up to a paradigm that open-source projects like OpenClaw established.

What's in OpenAI's Skills Catalog?

The curated skills folder contains approximately 30 production-ready skills covering several categories:

Cloud Deployment Skills

cloudflare-deploy - Deploy apps to Cloudflare
netlify-deploy - Netlify deployment automation
render-deploy - Render platform deployments
vercel-deploy - Vercel hosting and deployment

Design & Development Skills

figma-implement-design - Convert Figma designs to production code
figma - General Figma integration
develop-web-game - Browser-based game development
playwright - Browser automation and testing

Documentation & Content Skills

pdf - PDF creation and manipulation
doc - Document file handling
spreadsheet - Spreadsheet operations
jupyter-notebook - Jupyter notebook integration

Productivity & Collaboration Skills

linear - Linear project management
notion-knowledge-capture - Notion documentation
notion-meeting-intelligence - Meeting notes and summaries
sentry - Error tracking integration

AI & Media Skills

imagegen - Image generation via GPT Image
sora - Video generation
speech - Text-to-speech
transcribe - Audio transcription

Security Skills

security-best-practices - Security recommendations
security-ownership-map - Code ownership mapping
security-threat-model - Threat modeling assistance

OpenClaw vs. Codex: A Feature Comparison

Let's break down how these two skills-based agent systems compare:

Feature	OpenClaw/Clawdbot	OpenAI Codex
Skill Definition	SKILL.md + scripts	SKILL.md + scripts
Open Source	Yes (fully open)	Skills repo only
Multi-Model Support	Any LLM (Claude, GPT, local)	GPT-5.3-Codex only
Browser Control	Built-in via OpenClaw browser	Via Playwright skill
Platform	Linux, macOS, Windows	macOS only (Windows coming)
Pricing	Free + API costs	Included with ChatGPT subscription
Multi-Agent	Subagent spawning	Parallel agent threads

The similarities are undeniable. Both systems use the same SKILL.md file format for skill definitions. Both emphasize composability and reusability. Both position skills as the bridge between raw AI capabilities and real-world task completion.

Why This Matters: The Convergence of AI Agent Architectures

The fact that OpenAI—the company with effectively unlimited resources—has adopted an architecture pioneered by open-source projects validates a crucial trend: skills-based extensibility is the future of AI agents.

Consider what this means:

Standardization is coming - OpenAI's adoption of the skills pattern will push the industry toward standardized skill formats. The agentskills.io open standard they reference could become the "npm" of AI capabilities.
Open source wins again - Just as React influenced how the entire industry builds UIs, OpenClaw's architecture is influencing how major AI labs think about agent extensibility.
The IDE is dead, long live the agent - Both OpenAI and OpenClaw are betting that the future isn't better IDEs—it's agent command centers that orchestrate AI workers.

The Codex App: More Than Just Skills

OpenAI's announcement included several features beyond skills:

Multi-Agent Orchestration: The Codex app lets developers run multiple agents in parallel, each working on different tasks. Agents operate in isolated worktrees so they don't conflict with each other or your local Git state.

Automations: Similar to cron jobs for AI, Automations let you schedule recurring agent tasks—daily bug triage, CI failure analysis, documentation updates. OpenAI claims they've been using these internally for months.

Personality Modes: Codex now offers two personalities—a terse, pragmatic mode and a more conversational, empathetic one. Use /personality to switch between them.

GPT-5.3-Codex: Announced just three days after the app launch, this is OpenAI's "most capable agentic coding model to date," optimized for the multi-step, tool-using workflows that skills enable.

What OpenClaw Does Differently

While OpenAI is playing catch-up on skills, OpenClaw maintains several advantages:

Model Agnosticism: OpenClaw works with any LLM—Claude, GPT-4, GPT-5, Gemini, or local models like Llama. You're not locked into one provider's ecosystem.

Full Platform Support: OpenClaw runs on Linux, macOS, and Windows today. Codex is macOS-only with Windows "coming soon."

Deeper Integration: OpenClaw's browser control, node management, and messaging capabilities are built into the core platform, not bolted on via skills.

Community-First Development: As a fully open-source project, OpenClaw's roadmap is shaped by community needs, not corporate strategy.

The Future of AI Agent Plugins

We're witnessing the birth of an ecosystem. Just as mobile apps transformed smartphones from phones into pocket computers, AI skills are transforming language models from chatbots into general-purpose workers.

The race is now on to build the definitive skill marketplace. OpenAI has capital and distribution. OpenClaw has openness and flexibility. Anthropic, with Claude Code, is also in the mix—notably, the Reddit community has already observed that Codex's skills repo "is basically the canonical awesome-list for Claude skills."

For developers, this convergence is good news. Skills written for one platform will increasingly work across others. The SKILL.md format is becoming a de facto standard. And the barrier to extending AI agents is dropping rapidly.

Getting Started with Skills

If you want to explore the new paradigm, here's how to get started with each platform:

For OpenAI Codex:

Download the Codex app from OpenAI (requires ChatGPT Plus/Pro subscription)
Use $skill-installer [skill-name] to add curated skills
Create custom skills with the built-in skill editor

For OpenClaw:

Install OpenClaw CLI (npm install -g openclaw)
Add skills to your .openclaw/skills directory
Reference skills in your AGENTS.md configuration

Conclusion: The Playbook Works

When OpenAI, the world's most valuable AI company, adopts an architecture pattern pioneered by open-source projects like OpenClaw, it's not just validation—it's vindication.

The skills-based agent model works. It scales. It composes. And now it's the industry standard.

Whether you choose OpenAI's Codex, OpenClaw's Clawdbot, or another skills-enabled agent, you're betting on a pattern that's proven itself. The question isn't whether AI agents will become extensible through skills—that's settled. The question is which ecosystem you want to build on.

For those who value openness, flexibility, and community governance, OpenClaw remains the leader. For those who want tight integration with ChatGPT and don't mind platform lock-in, Codex is a compelling option.

Either way, welcome to the age of AI agent plugins. OpenAI just made it official.

Want to build skills for AI agents? Check out the OpenAI Skills repository or dive into OpenClaw's documentation to start extending your AI workflows today.

OpenAI Skills: How Codex Just Copied OpenClaw's Playbook