How to stop re-prompting your AI every session

To stop re-prompting, write down what your agent needs once, in a place it loads at the start of every session: an instruction file like CLAUDE.md or AGENTS.md for stable facts, plus notes for current work. Agents start empty by design, so memory has to be stored outside the session.

Why you have to repeat yourself

An AI agent does not remember earlier conversations on its own. Each session begins with a fresh context window. Claude Code's documentation says exactly this and describes two ways to carry knowledge across sessions: instruction files you write, and auto memory notes that Claude writes itself.

So the fix is not to prompt harder. It is to move what you keep repeating out of your head and into a place the agent reads on its own.

Step 1: write the things you always say

Look at the last three times you started a session. What did you type every time? That is your instruction file. Keep it short and specific: how to build and test, the folder layout, the style rules, the things the agent must never do.

Claude Code loads CLAUDE.md. Codex loads AGENTS.md, from your Codex home and then from the repository, with files closer to your working folder taking precedence. Claude Code can also read AGENTS.md, so one shared file can serve both. Keep the file concise: the shorter and clearer it is, the more reliably it is followed.

# Project notes
- Build: npm run build
- Test: npm test (run before saying a task is done)
- Never edit files in /generated
- Decisions: see docs/decisions.md

Step 2: keep a running note for current work

Instruction files hold what stays true. Current work changes daily. Keep a short progress note: the goal, what is done, what is next, what failed. At the end of a session, ask the agent to update it. At the start of the next, tell it to read the note first.

This is a habit more than a tool, and it works with any agent. Date each update and delete what is no longer true.

Step 3: let the agent save what it learns

Some tools can write their own notes. Claude Code's auto memory records learnings from your corrections and preferences, and loads them in later sessions. It is per tool, though. What Claude learns, Codex does not see.

Re-prompting, fine-tuning or memory?

Fine-tuning changes a model's weights. It is slow, costs money, and is the wrong tool for facts that change, like your project's current state. Re-prompting works but costs your time and tokens every session. Stored memory loaded at the start gives you most of the benefit of both: it is cheap to update and the agent sees it every time.

Memory has a limit. If you load too much, you fill the context window and the agent follows instructions less reliably. Store a lot, load a little.

Where a NodePress node fits

Instruction files and notes are per project and per tool. If you use several agents, you repeat yourself for each one. A NodePress node keeps one memory, a graph, that every agent in the node shares. Say it once, and each agent that joins can use it.

An agent joins by pasting one line or a link, through the remote MCP server at https://nodepress.ai/mcp. The node runs on your own computer, so the memory stays on your machine. Every step goes to one readable thread, so you can check what was saved.

For the broader picture, see how agents share context and how to get agents talking to each other.

Do it with a NodePress node

  1. Name a node on nodepress.ai.
  2. Paste the one line into the agent you use most.
  3. Tell the agent your project facts once. They go into the node's shared memory.
  4. Paste the same line into your other agents.
  5. Check the thread to see what was saved and correct anything wrong.

If you use one agent on one small project, a short CLAUDE.md or AGENTS.md is all you need. Do not store things you would not want an agent to read.

Questions people ask

Why does Claude not remember my previous conversations?

Each session starts with a fresh context window. To carry knowledge over, write it in an instruction file like CLAUDE.md or let auto memory save notes.

How do I make Claude Code remember previous sessions?

Put stable facts in CLAUDE.md, keep a progress note for current work, and use /resume to return to an earlier session when you need its full history.

How do I stop pasting the same notes into every chat?

Store them where the agent loads them itself: an instruction file, or a shared memory your agents connect to.

Is there a way to teach an agent once and reuse it everywhere?

Instruction files are per tool. A shared memory that several agents connect to, such as a node, lets you teach once for all of them.

Reprompting or fine-tuning: which is better?

Neither for facts that change. Stored memory loaded at session start is cheaper and easier to update.

Related

Sources

NodePress gives your AI agents one shared place on your own computer. Free to start, no sign-up. Make your first node.