On October 22, 2026, Anthropic announced it had handed its Agent Skills standard to a neutral foundation, the Agentic AI Foundation. Behind this technical name lies a concrete shift for small and mid-sized businesses: a common format to teach any AI agent how to carry out a specific business task, without rewriting that know-how for every single tool. Here's what it actually changes.
In brief
- On October 22, 2026, Anthropic donated the Agent Skills standard (name, specification, GitHub repository, agentskills.io domain) to the Agentic AI Foundation, a directed fund hosted by the Linux Foundation.
- Agent Skills is an open format published in December 2025: each "skill" is a folder containing a SKILL.md file that teaches an AI agent a reusable business procedure.
- Over 70 agentic platforms have already integrated it, including ChatGPT, Cursor, GitHub Copilot, and VS Code, according to Anthropic's official announcement.
- According to Anthropic, more than one million skills have been published and nearly 280 million have been installed since launch.
- The foundation already stewards two other building blocks of agentic AI: the Model Context Protocol (MCP, donated by Anthropic) and AGENTS.md (donated by OpenAI).
What is an "Agent Skill"?
An Agent Skill is a folder containing a text file named SKILL.md, sometimes paired with scripts or reference documents. This file describes, in plain language, how to carry out a specific task: following up with a client on an unpaid invoice, structuring a meeting summary, checking a standard contract for compliance. The AI agent reads this folder and applies the procedure, instead of improvising each time.
Anthropic developed this format while working on Claude Code, its coding assistant. The team found that an AI agent handles general knowledge well, but often lacks the procedural know-how specific to a given business: its document templates, internal rules, approval steps. A skill fills that gap in a portable form that any compatible agent can read.
A simple analogy
An Agent Skill works like an internal procedure sheet you would hand to a new employee, except it is written to be understood and executed by an AI rather than a human.
Neutral governance, not abandonment
Handing a standard to a foundation does not mean walking away from it. Anthropic explains that anyone can keep using and contributing to Agent Skills without joining the foundation itself. The goal is to give the standard a neutral home, steered by the community rather than a single company, to reassure competing businesses that might want to adopt it widely.
October 2025
Skills launch
December 2025
Open standard
2025
AAIF founded
October 22, 2026
Agent Skills donated
The standard will be maintained by engineers from Anthropic, Google, OpenAI, and other companies, participating as individuals. Anthropic also announced a roadmap: versioned releases, a formal proposal process for changes, and security guidance for anyone writing or installing skills.
Already widespread adoption
These figures, reported by Anthropic, show rapid adoption for a standard that is barely a year old. They should be read with the usual caution reserved for statistics supplied by the party that created the standard, but the list of integrated platforms is independently verifiable: it includes direct competitors of Anthropic such as OpenAI (ChatGPT) and Microsoft (GitHub Copilot, VS Code).
Skills, MCP, A2A: three different building blocks not to confuse
Agentic AI rests on several complementary standards, and it is easy to mix them up. Here is how to tell them apart.
| Standard | Role | Example use for an SMB |
|---|---|---|
| MCP (Model Context Protocol) | Connects an agent to external tools and data (CRM, database, files) | The agent reads unpaid invoices from the accounting software |
| A2A (Agent-to-Agent) | Lets two AI agents, sometimes from different vendors, communicate with each other | A sales agent hands off a qualified lead to a support agent |
| Agent Skills | Teaches an agent a reusable, portable business procedure | The agent always applies the same invoice follow-up process, whichever tool is used |
In short: MCP connects the agent to the outside world, A2A connects agents to each other, and Agent Skills documents how the agent should work. The three standards complement rather than compete with one another.
Why this is good news for SMBs
For a business owner, the practical upside comes down to one word: portability. Without a common standard, documenting a procedure for the company's AI assistant means starting over every time you switch tools, or every time you add a second assistant for another team.
With Agent Skills, an SMB can in theory:
- Document a procedure once (invoice follow-up, new client onboarding, a standard meeting summary) in a SKILL.md file;
- Reuse it across Claude, ChatGPT, Copilot, or Cursor, as soon as these tools support the standard, without rewriting everything;
- Reduce dependency on a single AI vendor, a growing concern for business owners focused on controlling costs and digital sovereignty.
Without a common standard
Every AI tool requires its own instructions, prompts, or configuration. Switching vendors often means rebuilding everything from scratch.
With Agent Skills
A business procedure written once, in SKILL.md format, can be read by any compatible agent: Claude, ChatGPT, Copilot, Cursor, and others to come.
Limits and points of caution
Enthusiasm should stay measured. First, governance is still young: the roadmap (versions, proposal process, security) is still being built, and rules may change. Second, a skill is still text executed by an AI: installing a skill written by a third party without reviewing it amounts to running code of unknown origin. Anthropic promises security guidance to come, but it is not yet finalized.
Finally, the promised interoperability (a skill written for one tool working on another) still needs to be checked in practice, tool by tool. Adoption by 70 platforms does not guarantee perfect, immediate compatibility across all of them.
How an SMB can prepare starting now
Spot repetitive tasks
Write a first clear procedure
Test with the AI tool you already use
Follow the open governance
FAQ
What is Anthropic's Agent Skills standard?
It is an open format, published in December 2025, that lets you write a folder (containing a SKILL.md file) describing to an AI agent how to carry out a specific business task. It was donated to the Agentic AI Foundation on October 22, 2026.
What is the Agentic AI Foundation?
It is a directed fund hosted by the Linux Foundation, co-founded by Anthropic, OpenAI, and Block in 2025, to neutrally steward open agentic AI standards such as the Model Context Protocol, AGENTS.md, and now Agent Skills.
Does an SMB need to change AI tools to use Agent Skills?
No. The standard is already built into widely used tools such as ChatGPT, Claude, GitHub Copilot, or Cursor. The question is whether the tool already in place supports it, not whether to switch.
Does Agent Skills replace the Model Context Protocol (MCP)?
No, the two complement each other. MCP connects an agent to external tools and data, while Agent Skills teaches a reusable business procedure. An SMB can use both at the same time.
With measured optimism, this decision confirms an underlying trend: the major AI players are increasingly choosing open standardization over closed proprietary formats, which directly benefits SMBs by limiting their dependence on a single vendor. To dig deeper into the technical building blocks of agentic AI, see our article on the Model Context Protocol and all our resources on AI in business.
Sources: Anthropic, official announcement "Donating Agent Skills to the Agentic AI Foundation" (October 22, 2026); Linux Foundation / Agentic AI Foundation, founding announcement (2025).
