AI and business strategy

Keep your business instructions usable when AI tools change

Anthropic is donating Agent Skills to an independent foundation. Here is what that means for keeping business procedures useful across AI tools.

Context Aware7 min read
A print-shop owner and colleague comparing a job sheet with a sample booklet at a worktable

A print-shop owner has spent weeks showing an AI assistant how to turn customer emails into a useful job brief. Then a colleague tries another tool and starts again from an empty chat. The business process has not changed, but the instructions are scattered across conversations. This week's Agent Skills announcement is a reason to consider which parts of that knowledge the business should keep in a reusable form.

What changed this week?

On 22 October, Anthropic announced that it is donating Agent Skills to the Agentic AI Foundation. We reviewed the announcement and the existing format documentation on 23 October. This is a change in how the standard is maintained, rather than the first release of skills.

Agent Skills gets an independent home

Anthropic says the donation includes the Agent Skills name, logo, GitHub organisation and website domain. The foundation sits under the Linux Foundation, and engineers from Anthropic, Google, OpenAI and other companies will maintain the standard as individuals. Anthropic reports that more than 70 agent platforms have integrated it. That is a vendor-reported adoption figure, not evidence that every feature behaves identically across those products. Its proposed roadmap includes versioned releases and security guidance; those proposals should not be confused with features already delivered.

Anthropic: Donating Agent Skills to the Agentic AI Foundation22 October 2026

The format packages instructions and supporting material

The existing Agent Skills documentation describes a folder containing a SKILL.md instruction file, with optional reference documents, templates and scripts. A compatible agent can load the fuller instructions when a relevant task arises. For a business owner, the useful idea is a task-specific procedure that can be maintained outside an individual chat. The format does not train a new model or make the procedure correct by itself.

Agent Skills: OverviewExisting documentation; reviewed 23 October 2026

Compatibility still needs attention

The specification allows authors to describe requirements such as particular software or network access. It also marks the allowed-tools field as experimental and says support may vary between implementations. Our reading is that sharing an instruction format is useful, while the surrounding tools and permissions still need checking. Copying a folder does not prove that the same business task will work in a different product.

Agent Skills: Format specificationExisting documentation; reviewed 23 October 2026

What does the public discussion add?

Direct access to Anthropic's X profile returned an access error. Targeted search did not surface an accessible relevant post about the donation, so we could not assess the discussion there.

Earlier discussion questions whose instructions you are adopting

An earlier Hacker News discussion about Claude-focused coding skills includes concerns about importing someone else's working preferences, overly long instructions and different team members keeping different copies. This thread predates the donation and is background context, not a reaction to the new announcement. The comments suggest useful questions to ask, but do not demonstrate that one approach performs better.

Hacker News: Earlier discussion of coding-agent skillsEarlier discussion; reviewed 23 October 2026

What does this mean for your small business?

Consider an illustrative print shop preparing an internal brief from a customer's email. Staff need the product type, quantity, finished size, artwork status and requested collection date. They also need to see what is missing before anyone schedules production.

The first useful document could be a page of ordinary instructions. Explain which information to extract, where to find it and how to present it. Include a complete example and an incomplete one. A colleague should be able to read the procedure and understand the job without knowing how an AI assistant works.

For instance, a customer asks for 'the same brochures as last time' and mentions Friday. The procedure could require a reference to the previous job before carrying over its specifications. It could record Friday as the customer's requested date while leaving the production date unconfirmed. If the previous record is unavailable, the brief should identify that gap.

Keep the repeatable method separate from changing job information. The procedure describes how to prepare a brief. The current email and authorised job record supply the details for this customer. Avoid turning a reusable instruction package into a collection of past customers' personal information.

Once that procedure is useful, a skill can be one way to package it for a compatible assistant. Start with instructions and an example template. Add a connection to the job system only when the task needs it and the access is appropriate. There is no need to add executable scripts merely because the format supports them.

The business then has something specific to carry into a future tool trial: its procedure, template and examples. Staff still need to check how the new assistant reads the files, finds the right records and handles missing information. This example is a proposed workflow, not a client result or a guarantee about any AI product.

How could your strategy evolve over time?

Begin with one task that people already understand. Preparing a job brief is easier to assess than asking an assistant to 'manage operations'. Keep the output small enough that the person doing the next step can review it.

Give the procedure an owner and a clear revision date. Store the current copy somewhere the team can find, and decide how updates reach the assistants people use. A shared format will not prevent old copies from remaining on laptops.

Keep a small set of fictional or suitably de-identified examples with expected outcomes. Include a complete request, missing details, a correction later in the conversation and a reference to an unavailable record. These examples become a practical check whenever the instructions, model or connected tools change.

Judge the result against the work that follows. In the print-shop example, does the brief preserve the customer's correction? Can staff see which specifications are confirmed? Is the request for clarification useful? A polished paragraph is not enough if production staff still have to reconstruct the email.

Before changing products, try the same examples in both environments. Check whether supporting files are available and whether any scripts or connections need adjustment. Keep permissions configured in the actual systems: wording in an instruction file is not a substitute for controlling which records an assistant can access or change.

Review the instructions when a real exception exposes a gap. Make the smallest clear correction, then rerun the examples to see whether it affects another case. If a rule no longer applies, remove it rather than adding a contradictory paragraph at the end.

Our interpretation of the announcement is modest: an independently maintained format may make it easier to retain useful work across compatible tools. Whether it helps your team depends on the quality of the procedure and how it is used. A well-maintained checklist may be sufficient before you need a formal skill.

What can you do this week?

Choose a repeated task that currently requires someone to explain the same details each time:

  • Write a short procedure naming the input, expected output and person who will use the result. Keep the first version readable by a colleague.
  • Add two worked examples, including one with missing information. Show what should remain unknown and what needs clarification.
  • Separate reusable instructions from live customer records, passwords and changing business data. Identify the approved source for each required detail.
  • Try the procedure in your current assistant using test data. If packaging it as a skill is useful, check that the product supports the required files and tools.
  • Save the current version with an owner and revision date. Reuse the same examples when you change tools or update the procedure.

How Context Aware can help

We can help turn a recurring task into a clear procedure, choose suitable examples and work out whether a shared document, a skill or a small automation is the right fit.

If the task needs information from your website, inbox or job system, we can map those connections and the access they require. We can also test a proposed tool change against the same examples, so the decision reflects your workflow.

Bring us one task your team keeps explaining from scratch. We can help document it, trial the approach and make the instructions easier to maintain.

Prepared by Context Aware with AI assistance and linked primary sources. Examples are illustrative. How we write our insights.