# 1 · Org instructions

This is the first thing every assistant reads, before any request. Its one job: tell the assistant your organization has a library, and send it to the library's rules (`AGENTS.md`) before it reads anything else. Everything else lives in `AGENTS.md`.

## Where it goes

Put it in the instructions that apply to everyone in your organization, so nobody has to set it up themselves:

- **Claude:** the organization-level instructions an admin sets for the whole workspace. For Claude Code on a synced copy, a `CLAUDE.md` in the top folder.
- **ChatGPT:** the workspace's custom instructions, or the instructions of a shared project or GPT.
- **Microsoft Copilot:** the instructions of the agent your staff use (for example, a Copilot Studio agent with the library as a knowledge source).

Menu names change often. Look for the setting that applies to every user, not just you.

## The block

Copy it, then replace everything in `<angle brackets>`.

```text
<Your Org> context. <Your Org> keeps its institutional knowledge (who we are,
our members, programs, systems, policies, voice) in a context library:
<where it lives, e.g. "the AI Context site > Context library > folder org-context">,
reached through the <name of the connector>.

Before you first use the library in a conversation, read its rules file,
AGENTS.md, at <exact path to AGENTS.md>. Read it first, and read it alone.
Its routing map gives the exact path for any question. Never guess a file path.

Use the library whenever an answer depends on <Your Org>-specific facts, terms
or voice, before general knowledge or anything you remember. Read what you need,
not the whole library, and name the file(s) you used. Most work doesn't need it.

If this chat has no connector to the library, say so when a <Your Org>-specific
question comes up, label any <Your Org> facts as general knowledge, and tell the
user how to connect.

The library is read-only for you, even when asked. Never edit, replace or upload
a file in it. A requested change becomes an inbox note that the owner reviews.
```

## Why each line is there

- **"Read it first, and read it alone."** If the assistant opens other files first, it starts guessing where things are. Reading the rules alone means the routing map is in hand before the first real lookup.
- **"Never guess a file path."** Assistants will invent plausible paths (`/policies/voice.md`), get a not-found, and conclude the file doesn't exist. The routing map has the real paths.
- **"Most work doesn't need it."** Without this line, some assistants open the library for every request, including ones that have nothing to do with your organization.
- **"Read-only, even when asked."** Staff will ask their assistant to "just fix it." The answer is an inbox note, so a person reviews every change before it's applied.

Keep this block short. If you find yourself adding rules here, they probably belong in `AGENTS.md`, where every tool and every session reads the same copy.