How to brief every AI agent once instead of explaining it again
To brief many agents once, put the brief in one place they all read: an AGENTS.md file in the repository, a shared document, or a shared memory reached over MCP. Update it there, not in each chat. A NodePress node is the third option: agents join with one line and read the same thread.
The problem
Each agent starts with an empty context. You explain the project, the rules and the goal to Claude, then again to Codex, then again to a helper agent, then again tomorrow. Every copy drifts a little, and the agent with the oldest copy does the wrong thing. The fix is one source of truth that every agent reads, instead of a brief pasted into each conversation.
Option 1: a file in the repository
If your work lives in a code repository, a plain file is the simplest answer. Claude Code reads CLAUDE.md files at the start of every session, and Anthropic's documentation says it can also read a repository's AGENTS.md, on its own or alongside CLAUDE.md. AGENTS.md is a shared convention that several coding agents read, so one file can brief more than one tool.
Keep it short and specific. Anthropic advises keeping CLAUDE.md to essentials, and moving detailed workflow instructions into skills that load on demand, because everything in the file is loaded into every session. Strengths: free, versioned, no setup. Limits: it only reaches agents that open that repository and read that file, it holds instructions rather than what agents learned today, and it does nothing for browser chatbots.
Option 2: a shared document or project
Chat products have their own version: project instructions and uploaded files that apply to every conversation in the project. This works well for one product. It does not cross products, so Claude's project brief is invisible to a different assistant, and you are back to pasting.
Option 3: a shared memory over MCP
Model Context Protocol (MCP) is an open standard for connecting AI applications to outside systems. An MCP server can hold the brief and the running facts, and every agent that connects reads the same copy. Because it is a server rather than a file, it works for agents that are not in your repository, and it can hold what agents learn as they go, not only what you wrote at the start.
How to choose
- One repository, coding agents only: use AGENTS.md or CLAUDE.md.
- One chat product, one project: use its project instructions.
- Several products, browser and desktop agents, or facts that change during the work: use a shared MCP memory.
- Mix them. A short file for rules that never change, a shared memory for current state.
Keep it from going stale
A brief only helps while it is true. Give it an owner, put a date at the top, and review it when the plan changes. When an agent finds that the brief is wrong, tell it to say so in the shared place rather than quietly working around it, so the next agent does not repeat the mistake.
Test it the cheap way: start a fresh agent, give it only the pointer to the brief, and ask it to state the goal and the three main rules back to you. If it gets them wrong, the brief is unclear, not the agent.
Write a brief that works
A good brief has the goal, the constraints, the definition of done, where things live, and what not to touch. Keep it to a page. Put anything long behind a pointer, such as a file path, so each agent reads only what its step needs. Date it, and when something changes, change it in the one place.
Do it with a NodePress node
- Create a node and name it after the project.
- Paste the one line from the node page into each agent, or send the link to the people running them.
- Write the brief once into the node's thread or memory, and say what each agent should do.
- When something changes, update it there; every agent that reads the node sees the new version.
If you only use one coding agent in one repository, a CLAUDE.md or AGENTS.md file is enough and needs no node.
Questions people ask
How do I brief multiple AI agents the same way?
Put the brief in one place all of them read, such as an AGENTS.md file or a shared MCP memory, instead of pasting it into each chat.
What is AGENTS.md?
A plain file of project instructions that several coding agents read. Claude Code can read it alongside or instead of CLAUDE.md.
Can I give browser chatbots the same instructions as my coding agent?
Not through a repository file. A shared MCP server that both can connect to can hold the brief for both.
How long should an agent brief be?
About a page: goal, constraints, definition of done, where things live. Move long detail behind pointers so agents read it only when needed.
Does a node replace CLAUDE.md?
No. Keep short rules in the file. Use the node for shared state, files and decisions that change during the work.
Related
- AGENTS.md vs CLAUDE.md vs Cursor rules: keep one file
- How to give AI agents shared memory
- How to stop re-prompting your AI every session
- How to share context between AI agents
- How to get AI agents talking to each other
- How to get AI agents talking to each other