Your SharePoint site can have a brain. One file in the Agent Assets library. Loaded every time AI in SharePoint opens. Written in natural language, written for AI not for humans.
The file is SHAREPOINT.md. It sits at the root of the Agent Assets library on the site. Anything you write in it becomes context that AI in SharePoint pulls into every conversation on the site, automatically, with no prompting.
Think of it as CLAUDE.md for SharePoint. If you have used Claude.md to brief Claude Code on a project, the pattern will feel familiar. The twist is that this file lives on a SharePoint site, governed by SharePoint permissions, and the whole team inherits whatever you write into it.
If you are new to AI in SharePoint, start with what SharePoint AI Skills are. This piece assumes you have the Agent Assets library activated and want to give the AI proper site context.
What goes in SharePoint.md
Anything AI in SharePoint needs to know about this site that it cannot infer from the content. In practice that means:
- The site purpose. One or two sentences. "This is the People and Culture team site. It holds policies, procedures, and templates for HR processes."
- The site URL. The canonical URL for the site. Useful when the AI needs to reference or link.
- A library map. Every document library, its path, what it contains, and any rules that apply. Repeat this for lists, registers, and request queues if you have them.
- Naming conventions. Formal patterns the AI should apply when creating or renaming files. "Policies follow the format
POL-NNN_PolicyName.pdfwhere NNN is a sequential three-digit number." - Organisational rules. Anything you would explain to a new starter. "All vendor proposals are stored in the Procurement library and tagged with the vendor name in the Vendor column."
Density matters more than prose flow. SharePoint.md is read by an AI on every chat. It does not need to be readable by a human at speed. It does need to be unambiguous.
Where SharePoint.md lives
SHAREPOINT.md sits at the root of the Agent Assets library on the site. The filename is uppercase. The library is created when you turn on the Agent Assets site collection feature.
To turn the feature on if it is not already enabled:
- Open Site information on the site.
- Click View all site settings.
- Open Site collection features.
- Find Agent Assets at the top of the list and activate it.
The Agent Assets library appears immediately, with Skills and Plans folders inside. SHAREPOINT.md goes in the root, alongside those folders, not inside them.
What it looks like in action
I tested this on a People and Culture site. The site has a Policies library, a Procedures library, a Templates library, a Forms library, and a few lists.
In SHAREPOINT.md I wrote out the library map and a naming convention for policies. The convention was POL-NNN_PolicyName.pdf with sequential numbering.
In the Policies library, the eight existing policies had inconsistent names. Some had spaces, some had underscores, none had the policy number prefix. Standard SharePoint mess.
I opened AI in SharePoint chat and prompted: "according to the naming policy convention, can you please rename all of these policies in this library?"
Before doing anything, the agent loaded SHAREPOINT.md automatically. It found the Policies library by reading the library map. It read the naming convention. It returned with two clarifying questions: should it use sequential numbers, and should it keep the existing file format?
I confirmed sequential numbers and asked it to keep the files as Word documents (the convention specified PDF, but in this case the source files were .docx and I did not want a format change).
The agent renamed all eight policies one by one, following the convention exactly: POL-001_RemoteWork.docx, POL-002_LeavePolicy.docx, and so on. It returned a summary table at the end mapping old names to new names.
None of that required me to re-explain the convention, point at the library, or correct the format mid-stream. SharePoint.md carried all of it.
Why this matters
Without SHAREPOINT.md, every chat on the site starts cold. The AI knows the site exists. It does not know what the site is for, where things live, or what the rules are. Every meaningful request needs setup. "This is the policies library. The naming convention is X. The required metadata fields are Y."
With SHAREPOINT.md, every chat starts warm. The AI already knows. The user can prompt with a one-line instruction and the work proceeds.
The team-level effect is bigger. A single editor writes SHAREPOINT.md once. Everyone with access to the site benefits, every time they open AI in SharePoint chat, without ever knowing the file exists.
How SharePoint.md and Skills fit together
SHAREPOINT.md is the context layer. It tells the AI what the site is, what is in it, and what the rules are. It is always loaded.
Skills are the action layer. They tell the AI how to perform a specific multi-step workflow. They are loaded only when invoked.
Most useful sites will have both. One SHAREPOINT.md at the root of Agent Assets, plus several Skills in the Skills folder. The brain plus the hands.
For where Skills fit in the broader Microsoft AI stack, see the pillar on extending SharePoint AI with Skills. For an end-to-end Skill build, the contract gaps analysis walkthrough covers the pattern.
Where to start this week
Pick the site you spend the most time in. Open Agent Assets. If the library is empty, create SHAREPOINT.md in the root.
Write three sections. The site purpose in two sentences. A library map listing every document library with its path and what it contains. Any naming or metadata rules that apply across the site.
Open AI in SharePoint chat. Ask it something specific that the file should help with. Confirm in the response that the AI used the context. Iterate.
Your site has a brain by the end of the morning. Every conversation after that is faster.