Skip to content
Back to Blog

AI

Agent Skills is now an open standard. Write a skill once, run it anywhere

Anthropic gave Agent Skills to the Linux Foundation on 22 October. What a vendor-neutral skill format means for marketing teams writing their procedures.

M
MyDigipal Team
Published on November 18, 2026

On 22 October, Anthropic handed Agent Skills to the Agentic AI Foundation, the Linux Foundation body that already looks after the Model Context Protocol. The format that teaches an AI agent how to do a specific job is now owned by a neutral foundation rather than by the company that invented it.

Anthropic says more than 70 agent platforms have integrated the standard, including ChatGPT, Cursor, GitHub Copilot and VS Code. It also counts 1 million skills published and nearly 280 million installs.

For a marketing team, this removes the objection we heard most often when we suggested writing procedures as skills. Nobody wanted to spend weeks documenting their reporting rules for a tool they might drop next year. That risk has shrunk a lot.

Key takeaways

  • Agent Skills now sits with the Agentic AI Foundation, next to MCP, under the Linux Foundation.
  • The same skill folder can be read by Claude, ChatGPT, Copilot, Cursor and dozens of other agents.
  • Instructions travel well between tools. Scripts, connectors and triggering still need testing on each one.
  • The procedures your team writes become a company asset that outlives any single vendor.

What changed on 22 October

A skill is still a folder with a SKILL.md file that explains a task, plus any scripts or reference documents the agent needs. The change is about who owns the format. Anthropic transferred the Agent Skills name and logo, the GitHub organisation and the agentskills.io domain. Engineers from Anthropic, Google, OpenAI and other companies will maintain the standard as individuals, with David Soria Parra and Mahesh Murag as lead maintainers.

The proposed roadmap starts with versioned releases, a formal proposal process and security guidance for people who write and install skills. Versioning matters, because until now each tool supported whatever it chose that month.

The move was widely expected. Simon Willison pointed out that he had predicted it last December, when Anthropic first called skills an open standard, writing at the time that he “wouldn’t be surprised to see this end up in the AAIF”. The timeline backs him up. Skills launched inside Claude in October 2025, became a published standard in December 2025, and have now followed MCP into the same foundation.

Why portability was the missing piece

We have already covered how to build skills, from the four pillars of a skill that scales to packaging scripts and templates into one workflow. The method still applies, and the business case is now much stronger.

In most marketing teams we work with, people do not share one AI tool. The content writer lives in ChatGPT. The developer who maintains the website uses Copilot inside VS Code. The analyst uses Claude. Until this year, writing a procedure as a skill meant picking one of them and leaving the others with a PDF.

With a shared standard, the brand voice rules, the campaign naming convention and the monthly reporting checklist can sit in one folder that every tool reads. When the team changes tools or hires someone new, the folder stays.

There is also a quieter benefit. A procedure written for an agent has to be explicit. It forces the team to write down the rule that lived in one person’s head, and that document has value even on the day nobody runs it.

70+
agent platforms that have integrated Agent Skills (Anthropic, Oct 2026)
1M
skills published since the launch
~280M
skill installs counted by Anthropic

What “runs anywhere” does not cover

A common file format does not make every agent equal. Four things still vary from one tool to the next.

Scripts. A skill that runs a Python script to clean a CSV export needs an agent with a code execution environment. Some tools have one, some do not, and some restrict it by plan.

Connectors. If your reporting skill expects to pull data through an MCP connector to Google Ads or your CRM, that connector has to exist on the other side too. The skill travels and the plumbing stays where you built it.

Triggering. Each tool decides when to load a skill by reading its description. A vague description that works in one agent can be ignored in another. In our experience, this is the most frequent reason a skill seems to do nothing.

Security. Installing a skill means installing instructions, and sometimes code, written by someone else. The security guidance is on the roadmap, which tells you it does not exist yet. Until it does, read third-party skills before you install them, the same way you would vet a browser extension.

AI FOR YOUR TEAM

Turn your team’s know-how into skills

We train marketing teams to write, test and maintain their own skills, so procedures work across the AI tools people already use.

How we write skills at MyDigipal

We use skills every day. Our campaign build rules, our conventions for client emails and the checks we run before sending a performance report all live as skills. A few habits have held up, and the open standard makes them more useful.

Write the description as a trigger. One sentence that says when to use the skill, with the words people actually type. “Use when creating or editing a Google Ads campaign” works better than “Advanced paid search knowledge”.

Keep tool-specific details out of the main file. If a step names a feature that only exists in one product, move it into a reference file. The core procedure then stays readable by any agent.

Give the reason behind each rule. “Read every budget back after writing it, because a wrong currency unit can multiply spend by 100” gets followed. “Always double-check budgets” gets skipped. Agents, like people, respect rules they understand.

Version skills like code. Ours sit in a Git repository, with changes reviewed before they go live. When a skill misbehaves, we can see what changed and when.

A skill used to be a bet on one vendor. Since October, it is a document your company owns and can carry to the next tool. MyDigipal

A portability test to run this month

One afternoon is enough to find out where you stand.

Pick the skill your team uses most. If you have none yet, write one for a task that repeats every week, such as turning a campaign export into a client summary. Then run it in two tools your team already pays for, on the same five real inputs.

Compare the outputs on a single criterion. Would you send this to a client after a light edit? Then note what broke. It is usually one of the four items above, and the fix is often small, such as a clearer description or a script moved into an optional step.

If both tools pass, the skill is portable and you can stop worrying about which assistant your next hire prefers. If one fails twice on the same step, rewrite that step rather than the whole skill.

Our take

The argument for waiting has gone. Procedures written as skills can now follow your team from tool to tool, under a foundation that no single vendor controls. Start with three to five skills that cover the work you repeat every week, store them where your team can review them, and test them on every tool in use.

If you have not built one yet, our guide to reusable Claude Skills for marketing still applies word for word, because the format it describes is the one that just became the standard.

Sources: Anthropic, Donating Agent Skills to the Agentic AI Foundation, 22 October 2026 - Simon Willison, 22 October 2026

#Agent Skills #Open standards #AI workflows #Agentic AI Foundation

Share this article

★★★★★

"We were looking to improve our digital marketing efforts but had a gap in the team and 12 internal tools. MyDigipal overcame significant challenges and successfully executed an ABM revamp across LinkedIn."

Brett Rieser
Brett Rieser
Digital Marketing Director EMEA, Genesys
★★★★★

"From the ground up, MyDigipal strategic acumen was invaluable. They didn’t just consult; they immersed themselves in our mission. High-value leads, significant uptick in conversions."

Surbhi Rathore
Surbhi Rathore
CEO and Co-founder, Symbl.ai
★★★★★

"The ABM programs MyDigipal put in place helped us identify and focus on the most valuable target accounts. We accelerated the deal closing process by 30%."

Kelly Wright
Kelly Wright
Global Marketing Director, Quantum Metric
★★★★★

"MyDigipal runs all our digital marketing. Google Ads, Meta, tracking, dashboards: we know exactly what every euro we spend brings in. No more black box."

Adam Aidoudy
CEO, Atelier PG
★★★★★

"Tracking and dashboards were a maze before. MyDigipal made the whole stack legible in 6 weeks. Now we ship campaigns on data, not gut feel."

Caroline Vermeersch
Marketing Manager, BeCom-Direct
★★★★★

"We were looking to improve our digital marketing efforts but had a gap in the team and 12 internal tools. MyDigipal overcame significant challenges and successfully executed an ABM revamp across LinkedIn."

Brett Rieser
Brett Rieser
Digital Marketing Director EMEA, Genesys
★★★★★

"From the ground up, MyDigipal strategic acumen was invaluable. They didn’t just consult; they immersed themselves in our mission. High-value leads, significant uptick in conversions."

Surbhi Rathore
Surbhi Rathore
CEO and Co-founder, Symbl.ai
★★★★★

"The ABM programs MyDigipal put in place helped us identify and focus on the most valuable target accounts. We accelerated the deal closing process by 30%."

Kelly Wright
Kelly Wright
Global Marketing Director, Quantum Metric
★★★★★

"MyDigipal runs all our digital marketing. Google Ads, Meta, tracking, dashboards: we know exactly what every euro we spend brings in. No more black box."

Adam Aidoudy
CEO, Atelier PG
★★★★★

"Tracking and dashboards were a maze before. MyDigipal made the whole stack legible in 6 weeks. Now we ship campaigns on data, not gut feel."

Caroline Vermeersch
Marketing Manager, BeCom-Direct

Put this into practice with your team

We train marketing teams to use AI in production and build custom MCP agents on the tools you already use.

My budget