One File, Every AI: Keeping a Whole Team on the Same Method

Keep each method in one file the team owns, in a format every assistant can read, and point every tool at that file rather than copying it. Open formats make this practical: SKILL.md skills work across many assistants, and AGENTS.md carries project instructions for most coding agents. Shared prompts pasted into chats drift within weeks.

How do you keep a team on the same AI method?

Write each method once, as a file with steps, checks and an example, and keep it somewhere the business owns and versions. Then have every assistant read that file, through a skill, a project or a repository, instead of each person keeping their own prompt. When the method changes, one edit updates everyone.

Most teams start the other way. Someone writes a good prompt and shares it in a chat thread. Colleagues paste it into their own assistant, tweak it, and save their version. Three months later there are eight variants and nobody knows which is right.

The fix is not a better prompt. It is a single source the team treats as the method, with a named owner, read directly by each tool wherever possible.

Why do shared prompt libraries drift?

Prompt libraries drift because every use is a copy. Each person adapts the prompt to their own taste, saves the change locally and never sends it back. Nobody owns the original, so improvements stay scattered and mistakes are never corrected everywhere. Within weeks, the team is running several methods while believing it runs one.

Different tools make it worse. A team rarely uses one assistant, and each has its own place for instructions. According to Microsoft and LinkedIn's 2024 Work Trend Index, 78% of AI users bring their own AI tools to work.

So the same method gets pasted into custom instructions in one tool, project instructions in another and a personal note in a third. Each copy then drifts on its own schedule, and none of them is the method the business agreed.

What file formats can every AI read?

Two open formats now cover most assistants. SKILL.md, from the Agent Skills standard, packages a method as a folder any compatible agent can load. AGENTS.md is a plain instructions file for a project that many coding agents read. Both are just text, so an assistant without native support can still read them as an uploaded file.

The Agent Skills site states the aim directly: "Build a skill once and use it across any skills-compatible agent." The guide to the Agent Skills open standard covers which tools support it.

AGENTS.md has spread fast. According to its own site, it is "used by over 60k open-source projects" and is now stewarded by the Agentic AI Foundation under the Linux Foundation. Claude Code can read an existing AGENTS.md as its project instructions.

Where should the shared file live?

Keep it in a versioned store the business controls, such as a shared repository or a managed drive, with one owner per method. Every tool should read from there or be refreshed from there. Never let an assistant's own settings be the only copy, because those belong to one tool and often one person.

A repository gives three things a chat setting cannot. Every change is recorded, so a bad edit can be reversed. Changes can be reviewed before they reach everyone. Finally, the file outlives any single subscription or employee.

For tools that cannot read the repository directly, publish from it. Upload the current file to the shared project or organisation skill, note the version, and replace it when the source changes. One source, many copies, all traceable.

What belongs in the shared file and what stays personal?

The shared file holds what everyone should do the same way: the steps, the checks, the approved facts it draws on and an example of good output. Personal preferences, such as tone tweaks or formatting habits, stay in each person's own settings. Mixing the two makes the shared file argue with itself.

Claude Code shows the same split in its own design. Its documentation describes project instructions as shared with "team members via source control", while a local file holds "personal project-specific preferences" and stays out of version control.

How do you roll a shared method out to a team?

Start with one method the team runs every week. Write it with the person who does it best, test it against real work, then retire the old prompts by name. Point every assistant at the new file, tell people where it lives and who owns it, and collect their corrections for the first fortnight.

1. Pick one method. Frequent, valuable and easy to check.
2. Write it once. With the person who does it best, plus a worked example.
3. Test it. On real tasks, in every assistant the team uses.
4. Retire the copies. Ask people to delete their old versions.
5. Name the owner. One person approves every change from now on.

The second method is quicker, and by the fifth the habit is set. The guide to writing a Claude skill covers the file itself.

Skein keeps method in shared skill files the business owns, read by whichever AI each person prefers, with the client's specifics in its own workspace; the agents page shows the work those methods run. That suits a team where several people use different assistants for the same jobs. One person with one assistant can keep their method in that tool's own settings.

Frequently asked questions

How do I share AI prompts with my team?

Write the prompt as a method file with steps, checks and an example, keep it in a shared, versioned place, and have each assistant read that file rather than a pasted copy. Name an owner who approves changes. Pasted prompts drift quickly, while a single owned file keeps everyone on the same version.

Can one file work in Claude, ChatGPT and other assistants?

Often, yes. Skills written to the open Agent Skills format load in compatible assistants, and AGENTS.md is read by many coding agents. Any assistant that accepts documents can read a plain Markdown method uploaded as a file. Test the file in each tool, because support and behaviour still differ.

What is AGENTS.md?

AGENTS.md is an open format for giving coding agents standing instructions about a project, described by its site as a README for agents. It is used across tens of thousands of open-source projects, read by many coding tools including Claude Code, and stewarded by the Agentic AI Foundation.

Should prompts live in a repository or a shared drive?

A repository is better when changes need review and history, because every edit is recorded and reversible. A managed shared drive works for teams that will not use a repository, provided one person owns each file. Either way, assistant settings should hold copies of the file, never the only version.

How do you stop people editing their own copies?

Make the shared file easier to use than a private copy, by connecting it to each assistant directly where possible. Retire old prompts by name when the new method launches. Then invite corrections through the owner, so improvements reach everyone instead of staying in one person's settings.

Related reading