Anthropic has donated Agent Skills to the Agentic AI Foundation, giving the format for reusable AI instructions a home under the Linux Foundation. Announced on 22 October, the move matters to teams trying to carry useful working procedures between compatible AI tools instead of rebuilding them for each product.
A skill packages instructions for a task together with supporting resources. The shared format can make those materials easier to reuse, but a portable folder does not guarantee identical results, available tools or permissions in every AI product. Check the receiving environment before relying on it.
What Is Inside a Skill
The standard’s overview describes a folder with a required SKILL.md file. It contains metadata and instructions, and can be accompanied by scripts, reference documents, templates and other resources. In practical terms, it is a place to describe how a recurring job should be carried out and what information the agent should consult.
For example, a publisher could write a procedure for checking an article’s dates, keeping company claims attributed and producing a consistent source list. That is an illustrative use case, not a tested result. The value would depend on the quality of the procedure and whether a reviewer could tell when the agent followed it.
The overview describes three stages: discovering a skill from its name and description, reading its full instructions when it becomes relevant, and using any additional resources needed for the task. This helps explain why a useful description matters: it tells the agent when the procedure should be considered.
What the Foundation Move Changes
Anthropic says the donation includes the name and logo, GitHub organisation and agentskills.io domain. Participation does not require joining the foundation. Its proposed roadmap covers versioned releases, a formal proposal process and security guidance for skill authors and users.
Our reading is that governance is the immediate development. A neutral home can provide a shared place to discuss changes, but the announcement does not establish that a particular workflow will run unchanged in every product. Teams should keep those two questions separate when deciding whether to adopt the format.
The specification includes an optional compatibility field for environment requirements, such as the intended product, required system packages or network access. Before moving an existing skill, review those requirements against the destination. A familiar file format is only the beginning of that check.
Start With One Repeatable Task
A Small Pilot for Your Team
Choose a bounded job.
Pick a recurring task with an output somebody can assess, such as checking headings and references in a draft. Avoid starting with a process nobody can clearly describe.
Write the acceptance criteria.
State what a complete result must contain, which sources are acceptable and when the agent should stop for missing information.
Inspect the whole folder.
Review instructions, scripts and references before use. Decide whether the requested tools and access are appropriate for the job.
Check the destination.
Confirm that the receiving agent supports skills and can meet the documented environment requirements.
Compare the outputs.
Use a non-sensitive sample in each intended environment. Have a person assess correctness and record differences before expanding the workflow.
Keep ownership visible. Assign someone to maintain the procedure, record why changes were made and retire instructions that no longer match your process. Those are editorial recommendations for making the pilot reviewable, rather than promises supplied by the format or evidence of a tested improvement.
For Australian teams already using several AI products, the useful next step is modest: choose one process you understand, document it clearly and assess how it behaves in the tools you actually use. The foundation announcement provides a timely reason to investigate reusable workflows without assuming that standardisation removes the need for judgement.
Sources and How This Was Checked
This article is based on the official pages below, read on 23 October 2026. Anthropic’s announcement was compared with the standard’s overview and specification. No skill was installed or tested across products for this article. The practical recommendations are editorial guidance. No hands-on test was performed.
Official Pages Used
Anthropic: Donating Agent Skills to the Agentic AI FoundationDonation, Linux Foundation home, open participation and proposed governance roadmap.www.anthropic.com
Agent Skills: OverviewSKILL.md folder format, supporting assets and progressive discovery, activation and execution.agentskills.io
Michel Elijah covers technology, streaming and online security for Tech X Telco. He writes practical how-to guides on everything from email troubleshooting to spotting the latest scams doing the rounds in Australia, with a focus on clear steps anyone can follow.
Spotted something wrong? Corrections are recorded in the open. Report a correction
