How OpenAI Quietly Adopted Anthropic’s “Skills”

In the fast-moving world of AI agents, one of the biggest surprises of the past year was not a new model. It was a humble folder of text files. In October 2025, Anthropic introduced Agent Skills, a simple way to teach an AI assistant how to do a specific job. Within weeks, developers noticed that OpenAI was using the same idea in its own products, and by December the two rivals were working with the very same file format. This is the story of how OpenAI Adopted Anthropic’s “Skills”, why it happened so quickly, and what it means for anyone building with AI.

If you build or plan to build on OpenAI's platform, a program like the Certified OpenAI API Developer course covers the API tools behind these agents. The language below is simple enough for beginners and detailed enough for professionals. A quick disclosure: this article was written by Claude, an AI model made by Anthropic, so it checks claims carefully and tries to stay neutral.

Quick Answer: Did OpenAI Really Adopt Anthropic's Skills?

Yes. Anthropic launched Agent Skills in October 2025 and released the format as an open standard on December 18, 2025. Developers reported that OpenAI's Codex command-line tool had already merged experimental support for a skills format in early December, and that ChatGPT's code environment contained a skills folder by December 12. Within about 48 hours of the open standard release, OpenAI was reported to support the same SKILL.md format in both ChatGPT and Codex CLI. Later, OpenAI published official Codex documentation for skills. The word "quietly" fits because the support appeared in code and folders before any big announcement.

What Are Agent Skills?

The Basic Idea

A skill is a folder that teaches an AI agent how to do something well. At its core is one required file called SKILL.md. That file has two parts: a short header with the skill's name and description, and a body of plain Markdown instructions. A skill can also include scripts, templates, and reference files.

A Simple Analogy

Think of an agent as a smart new employee. A skill is like a laminated instruction card for a specific task, such as "how to format our quarterly report" or "how to review a pull request." The employee does not read every card at the start of the day. They look at the card titles, and only pull out the one they need.

Progressive Disclosure

This is the clever part. When an agent starts, it reads only each skill's name and short description. When a task matches a skill, the agent loads the full instructions. If those instructions point to extra files or scripts, it loads those only when needed. This keeps the agent fast and avoids stuffing its memory with instructions it does not need.

Why Understanding This Helps

Skills sit on top of large language models, which follow written instructions. Knowing how models read instructions, where they fail, and how to test them helps you write better skills. A program like the Certified Artificial Intelligence (AI) Expert course explains how AI systems learn and follow instructions, which makes it easier to judge what a skill can and cannot do.

The Timeline: From Anthropic Launch to OpenAI Support

The counts and exact dates come from developer blogs and industry write-ups, so treat the "around 30" figure as an estimate.

How OpenAI "Quietly" Adopted Skills

Signs in ChatGPT

Developer Simon Willison reported on December 12, 2025 that ChatGPT's code interpreter had a new skills folder. By asking ChatGPT to zip that folder, users could see skill files, including one with guidelines for creating PDFs, that looked very close to Anthropic's own design. The discovery was credited to other researchers as well, and OpenAI had not announced it at the time.

Signs in Codex CLI

In Codex CLI, developers found code that built a skills prompt and a command-line option to enable skills. Users could install a skill made for Claude by dropping it into a Codex skills folder, and it worked with little or no change.

Official Support Followed

OpenAI later documented skills for Codex in its developer docs. Reports describe skills living at several levels: inside a repository, in a user's home folder, and bundled with the system. Codex also includes helper skills, such as a skill creator and a skill installer, and OpenAI maintains a catalog of Codex skills on GitHub.

Why the Word "Quietly" Matters

Support appeared first in code and in file folders, then in documentation. That is why developers described it as quiet adoption. It also shows how simple the format is to implement.

Why Did OpenAI Use the Same Format?

It Is Easy to Build

Adding skills support mostly means finding skill folders and reading SKILL.md files. That takes far less work than building a full protocol. Industry write-ups point out that this low cost helped adoption spread so quickly.

Developers Already Wanted It

Developers had started writing instructions for coding agents in many different places. A shared format means one skill can work across tools, so no one has to rewrite it for each product.

Network Effects

Once many tools support the same format, every new skill becomes more valuable. Rivals gain more by joining than by building a competing format that would split the community.

Open Standards Have Precedent

The AI industry has already seen competitors share standards. Anthropic's Model Context Protocol was donated to the Linux Foundation in late 2025, and OpenAI's own AGENTS.md convention for coding agents also moved to shared governance. Skills fit the same pattern of common building blocks and competition on model quality.

Skills vs MCP vs AGENTS.md vs Other Tools

These tools work together. A skill might explain how to analyze a sales file, while MCP gives the agent access to the sales database.

Where Skills Live in Different Tools

Folder locations vary by product, and they can change, so always check the latest docs.

Several tools also read each other's folders. One report noted that Kimi's coding tool searches Claude, Codex, and generic agent skill directories, so a skill installed once may work in several tools.

Benefits for Developers and Teams

• Write once, use in many tools: A single skill can work across several agents.

• Capture team knowledge: Style guides, review checklists, and runbooks become reusable.

• Faster, cheaper agents: Progressive disclosure keeps unused instructions out of the agent's working memory.

• Version control: Skills are just files, so you can store them in Git, review changes, and share them.

• Less vendor lock-in: Teams can switch tools without losing their instructions.

• Easy sharing: Public repositories already host collections of skills from many companies and communities.

Risks and Limits

Security

A skill can include scripts and instructions that an agent may follow. A malicious or careless skill could tell an agent to leak data or run harmful commands. Treat skills like software you download: read them, prefer trusted sources, and limit what the agent can access.

Prompt Injection

Instructions hidden inside documents or web pages can try to trick an agent, and skills add one more place where bad instructions could hide.

Uneven Support

Not every tool supports every feature, and folder paths and settings differ. A skill that works well in one agent may behave differently in another.

Quality Varies

A vague or bloated skill can confuse an agent. Test skills on real tasks and keep them focused.

Fast-Changing Details

Documentation, flags, and file locations have changed during the past year, so older tutorials may be out of date.

What This Means for the AI Industry

The skills story shows how fast common building blocks can spread in AI. Competitors can still fight over model quality and price, yet agree on the plumbing that connects tools together. For businesses, that lowers the risk of betting on one vendor. For developers, it means that effort spent writing good instructions can travel with you. As AI agents blend with automation, data, and other advanced systems, many professionals widen their skills beyond one platform. You can explore a broad range of options through Deep Tech Certification programs, which cover advanced and emerging technologies and help you stay relevant as standards shift.

How to Write a Good Skill

• Pick one repeatable task. For example, "write a weekly status report."

• Name it clearly. Use a short, lowercase name with hyphens.

• Write a precise description. Say what the skill does and when to use it, since the agent uses this text to decide.

• Keep instructions step by step. Use short, direct sentences.

• Add examples. Show a sample input and output.

• Put heavy details in separate files. Link to templates or references instead of stuffing everything into SKILL.md.

• Test with real tasks. Run it, see where the agent goes wrong, and refine.

• Review for safety. Avoid secrets, and be careful with scripts.

A Minimal Example

markdown

---
name: weekly-report
description: Turns rough bullet notes into a short weekly status report. Use when the user asks for a status update or weekly summary.
---

# Weekly report

1. Group the notes into Done, In Progress, and Blocked.
2. Write one sentence per item in plain language.
3. End with a short list of next steps.
4. Keep the report under 200 words.

What This Means for Businesses

• Document your best practices as skills. Onboarding guides, brand rules, and compliance checklists can become reusable agent instructions.

• Avoid lock-in. Choose tools that support open formats where possible.

• Set rules for third-party skills. Require review before anyone installs a skill from the internet.

• Measure results. Track whether skills actually save time and improve quality.

• Train your team. Teach people how to write, test, and maintain skills.

As artificial intelligence becomes part of everyday business operations, employers increasingly need people who can apply AI knowledge responsibly and effectively. Evaluating candidates on relevant skills, reasoning, and practical understanding can support stronger hiring decisions. HireValid can be explored by organisations seeking structured candidate assessments.

Common Mistakes to Avoid

• Installing unknown skills blindly. Review them first.

• Writing huge, vague skills. Small and specific works better.

• Skipping tests. An untested skill can fail silently.

• Assuming all tools behave the same. Check each tool's documentation.

• Hard-coding secrets. Never put passwords or keys inside a skill.

• Forgetting updates. Revisit skills as tools and models change.

Simple Tips for Beginners

• Start by reading one existing skill from a public repository.

• Create a tiny skill for a task you do every week.

• Test it in one tool, then try it in another.

• Keep descriptions short and clear.

• Store skills in version control.

• Follow official docs from each vendor.

• Share what you learn with your team.

Conclusion

The way OpenAI Adopted Anthropic’s “Skills” shows how quickly useful ideas can become shared standards in AI. A simple folder with a SKILL.md file turned out to be easy to build, easy to share, and valuable enough for competitors to embrace within days of its release as an open standard. The result is a growing ecosystem where teams can capture their know-how once and use it across many agents. The smart path forward is to start small, review everything you install, and keep an eye on the documentation as the standard grows.

Great technology also needs great communication. If you want to learn how to explain and promote AI products and workflows to the right audience, a Marketing Certification is a smart next step.

FAQs

ng concept is similar, but the implementations are tied to each company's agent ecosystem. Anthropic describes Skills across Claude apps, Claude Code, and its API, while OpenAI documents Skills for ChatGPT, Codex, the Responses API, and agent workflows.

6. Why are Skills important for AI agents?

Skills allow an agent to reuse established procedures instead of rediscovering how to perform the same task every time. This can make complex workflows more consistent, structured, and easier to maintain.

7. What can an OpenAI Skill contain?

An OpenAI Skill can contain a required SKILL.md file along with instructions, scripts, reference materials, templates, and other supporting assets. These resources give an agent the information and procedures it needs to complete a specialized workflow.

8. How does OpenAI Skills discovery work?

OpenAI's architecture allows an agent to discover available Skills and determine which ones are relevant to the task. The model can then load the appropriate Skill instructions and resources instead of placing every possible workflow in its initial context.

9. Can Skills include executable code?

Yes. Skills can include scripts or other executable resources. OpenAI describes Skills as reusable workflow bundles that can combine instructions with scripts and assets, while its agent architecture can execute Skill logic through a sandbox environment.

10. Are OpenAI Skills available in ChatGPT?

OpenAI's Help Center currently describes Skills in ChatGPT as reusable, shareable workflows that help ChatGPT complete specific tasks more consistently. Availability is currently tied to eligible ChatGPT Business, Enterprise, Healthcare, and Edu workspaces and can depend on workspace settings.

11. Can developers use Skills through the OpenAI API?

Yes. OpenAI provides a Skills workflow for its API ecosystem. Developers can package instructions, scripts, and assets into reusable Skill bundles and use them with supported agent environments.

12. Can Skills be used with Codex?

Yes. OpenAI's documentation describes Skills as being supported by Codex and other OpenAI products. Developers can use Skills to give coding agents reusable project workflows, conventions, procedures, and supporting resources.

13. How do Skills differ from ordinary prompts?

A normal prompt provides instructions for a particular interaction, while a Skill packages reusable instructions and supporting resources into a structured workflow. Skills are therefore better suited to repeatable tasks that need consistent procedures, files, examples, or scripts.

14. Can Skills make AI agents more reliable?

They can improve consistency by giving agents predefined procedures, examples, and supporting resources. However, Skills do not guarantee correct results because the underlying AI model can still make mistakes and the Skill itself may contain incomplete or incorrect instructions.

15. Can multiple Skills be used together?

Yes. Skills can be designed as composable workflows. This allows an agent to combine specialized capabilities when a task requires multiple procedures or areas of expertise. Anthropic explicitly describes its Skills as composable, while OpenAI's documentation similarly describes Skills as reusable workflow building blocks.

16. Why did OpenAI adopt a Skills-based architecture?

Skills address an important problem in agent development: repeatable workflows often require specialized instructions, files, and procedures. Packaging those components separately can make agents easier to extend and maintain while reducing the need to repeatedly place large workflow descriptions in prompts.

17. Are Skills an open standard?

Anthropic announced Agent Skills as an open standard for cross-platform portability in December 2025. OpenAI's current documentation also uses a structured Skill format, including SKILL.md, making the broader Skills pattern increasingly relevant across agent ecosystems. However, compatibility should not be assumed between platforms without checking the specific implementation.

18. What are some practical examples of AI Skills?

A Skill could teach an AI agent how to analyze spreadsheets, create presentations, follow a company's brand guidelines, process documents, conduct a standardized research workflow, or perform a particular coding procedure. The same basic concept can be adapted to many business and technical workflows.

19. What does OpenAI adopting Skills mean for the future of AI agents?

It suggests that AI agents are moving toward modular architectures in which capabilities can be packaged, discovered, reused, and combined. Instead of relying entirely on increasingly long prompts, developers can build agents around reusable workflow components, tools, files, and execution environments.

20. Is OpenAI's Skills approach the same as Anthropic's?

The concepts are closely related, but they should not be treated as identical implementations. Anthropic pioneered the publicly announced Agent Skills approach in Claude, while OpenAI has developed its own Skills architecture across ChatGPT, Codex, and its agent and API ecosystem. The broader trend is toward reusable, modular capabilities for AI agents.