1 - EXPORT PROMPT  -  RUN THIS ON THE OLD AI
Version 1.0, 26 October 2026. Works with any assistant: ChatGPT, Claude,
Gemini, Copilot, or whatever comes next.
Created by Hans Tandt - hanstandt.be. Licence: MIT - free to use, copy and adapt.
Companion file: 2_IMPORT_PROMPT.txt, which you run on the NEW one afterwards.


WHAT THIS IS FOR

  An assistant learns how you work. Almost none of that is in a file. It sits
  in the assistant's own memory, which belongs to the company, not to you.

  This prompt gets it out. It asks the assistant to write down everything it
  knows, in ordinary files that you keep. Files are the only thing that
  travels to another PC, another account, or another AI.

  The more you have built, the more there is to lose. A short chat is easy to
  leave behind. A real piece of work - an application, a set of documents, a
  database - carries months of decisions, plus the tools and plug-ins you
  added along the way. That is what breaks on the other side.


HOW TO USE THIS FILE

  1. Open the assistant you have been working with. Use the SAME chat if the
     work is in it, so it still has the conversation in front of it.
  2. Select everything BELOW the row of v characters, down to the row of ^
     characters, and paste it as ONE message. Do not include the marker rows.
  3. It asks you six questions first. Answer them.
  4. After that it works one step at a time and waits for your "go". Nothing
     is written until you say so.
  5. Save every file it produces into one folder. That folder is what you
     carry to the other side.


BEFORE YOU START

  a. Decide where the files will be kept. Any ordinary folder will do.
  b. If the assistant cannot write files on your computer - most web
     assistants cannot - it will print them in the chat instead. You copy
     each block into a text editor and save it under the name it gives.
  c. Allow half an hour. This is the one job that is worth doing slowly.


YOU DO NOT NEED TO KNOW WHAT AN .md FILE IS

  The files this produces end in .md. If that means nothing to you, it does
  not matter. An .md file is a plain text file - you can open it in Notepad
  and read it like a letter. The assistant writes them; you only save them.

  One Windows trap, if you ever save one by hand: in the Notepad save box,
  choose "All Files" before typing the name. Otherwise Windows saves it as
  Session_Notes.md.txt, which is not the same file.

  Chapter 5.1 of the manual explains this properly, in one page.


vvvvv PASTE FROM HERE vvvvv

I am moving this work to somewhere else: another PC, another account, or a
different AI assistant. You are the OLD assistant. Everything you know about
me and about this work has to end up in FILES, because files are the only
thing that travels. Nothing else comes with me.

RULES - these apply to every step below:
  - Ask me the questions in STEP 0 first, in ONE message, and wait for my
    answers. Do not start STEP 1 until I have answered.
  - Then do ONE step at a time. Show me the result and wait for my "go"
    before you start the next one.
  - Write nothing and change nothing until I say "go".
  - Do not invent anything. If you are not certain of a fact, mark it
    [UNSURE] and tell me why. A wrong fact that travels is worse than a
    missing one, because I will trust it.
  - Never put a password, an API key, a token, a bank detail or private
    personal information into any file.
  - If you cannot write files on my computer, print the complete contents in
    the chat instead, and say clearly which filename each block belongs to.
  - If a file already exists, never overwrite it. Show me what you would add
    and where, and let me decide.
  - If I ask for something you cannot do, say so plainly. Do not pretend.


================ STEP 0 - ASK ME FIRST ================

Ask me these six questions in ONE message and wait for my answers.

  1. What is this work? Give me your own one or two sentence description, and
     I will correct it if it is wrong.

  2. Where does it continue? The same assistant on a new PC, a new account
     with the same assistant, or a different AI altogether?

  3. Can you write files on my computer, or must you print them in the chat
     for me to save by hand?

  4. Is there already a folder with files for this work? If yes, I will give
     you the path. If no, say so.

  5. How far back should you go? Only this conversation, or everything you
     remember about me from before it?

  6. Is there anything I must NOT write down - private, confidential, or
     something I would not want on a USB stick?


================ STEP 1 - WHAT DO YOU ACTUALLY KNOW ================

Before writing anything, show me what you have. In plain language, grouped:

  1. What you remember about ME: how I like to work, my preferences, the
     corrections I have given you.

  2. What you remember about THIS WORK: the facts, the decisions, and the
     reason behind each decision.

  3. Any standing instructions I have given you - things you always do or
     never do.

  4. Say for each item where it came from: your stored memory, this
     conversation, or a file.

  5. Tell me plainly which of it is in a memory that survives this
     conversation, and which exists ONLY in this chat and will be gone when
     I close it.

  6. If you have a memory feature, tell me honestly whether you can read it
     back to me in full. If you cannot, say so. Do not guess at what is in
     it, and do not guess menu paths in your own settings.

Mark anything you are unsure of. I will correct you before we write.


================ STEP 2 - THE TOOLS, THE PLUG-INS AND THE SET-UP ================

This is the step people forget, and it is the one that breaks the work on
the other side. Knowledge can be written down. A workbench has to be rebuilt.

List everything that is not a file and not a memory:

  1. Plug-ins, extensions, connectors or add-ons that are switched on, and
     what each one is used for.
  2. Tools you can reach that a plain assistant cannot: a browser, a code
     runner, a database, a shared drive, a company system.
  3. Custom instructions, system prompts, saved styles, or any setting that
     changes how you answer me.
  4. Saved assistants, custom GPTs, projects, workspaces, agents or named
     chats that belong to this work.
  5. Programs on my computer that the work depends on, and their versions if
     you know them.
  6. Anything that needed a login, a licence, a key or an approval to set up.
     Name it, but NEVER write the key or the password itself.

For each one, tell me in one line what breaks if it is missing on the other
side. Then put the whole list into TOOLS.md.

Be honest about the ones you are not sure are switched on. A guessed plug-in
list is worse than a short one.


================ STEP 3 - THE FILES THAT MUST EXIST ================

Six short files carry everything. Check which already exist. Any that are
missing, you will CREATE.

  HANDOVER.md        the one the next assistant reads first
  PROJECT.md         what this work is, written for a stranger
  RULES.md           how to work: my rules, paths, conventions, habits
  FACTS.md           current facts, each with its source and a date
  TOOLS.md           the workbench from STEP 2, and how to rebuild it
  Session_Notes.md   the log: what was done, what was decided, and why

If the folder already uses other names for the same thing - for example
CLAUDE.md instead of RULES.md, or MEMORY.md instead of FACTS.md - keep the
names that are already there. Do not create a second file for the same job
under a new name. Tell me which name you are using for what.

Tell me which exist, which you will create, and then show me the proposed
contents of each. Wait for my "go" before writing.


================ STEP 4 - FILL THEM FROM THIS SESSION ================

Put everything from STEP 1 and STEP 2 into the right file. Each fact goes in
ONE place only.

  Rules, paths, conventions, how I like to work   ->  RULES.md
  Facts, each with its source and an "as of" date ->  FACTS.md
  What this work is, for a stranger               ->  PROJECT.md
  Plug-ins, tools, settings, programs             ->  TOOLS.md
  What happened and why                           ->  Session_Notes.md

For Session_Notes.md, add ONE new dated entry in this shape:

    ## YYYY-MM-DD - <short title>
    Done:
    Decided, and why:
    Files created or changed:
    Still open:
    Next step:
    Not sure about:

Rules for the content:
  - A fact without a source is not a fact. Name the document, the page, or
    the conversation it came from.
  - A decision without its reason is useless. "We chose X" tells the next
    person nothing. "We chose X because Y failed" stops them undoing it.
  - Never two versions of the same fact. Replace the old one.
  - Keep each file short enough that a person will actually read it.

Show me the exact contents of every file, then wait for my "go".


================ STEP 5 - THE HANDOVER FILE ================

Now write HANDOVER.md. This is the most important file, and it is the part
no official export contains. It is addressed to the NEXT assistant.

  1. What this work is, in five lines.
  2. The decisions that must NOT be undone, and the reason for each.
  3. What was tried and rejected, and why - so it is not tried again.
  4. What is still open, and what the next step is.
  5. The quirks that cost us time: this computer, this tool, the way I work,
     the mistakes you learned not to make.
  6. Things I told you out loud that were never written anywhere else.
  7. What must be installed or switched on before the work can run at all.
  8. What you are unsure about, listed honestly.

Write it so that someone who was never here can pick the work up from it.


================ STEP 6 - WHAT WILL NOT TRAVEL ================

Tell me the truth about your own limits. For YOU, specifically:

  1. Can your memory be exported as a file? If yes, how? If no, say no.
  2. If your company offers an official data export, tell me what it does and
     does not contain - as far as you actually know.
  3. What will I lose that no file can carry?
  4. Is there anything I should do in the settings before I leave?

If you do not know, say "I do not know". Do not invent a menu path or a
feature name. I would rather check it myself than be sent to a button that
does not exist.


================ STEP 7 - FINAL CHECK ================

  1. List every file: its name, where it is, and roughly how big it is.
  2. Confirm that none of them contains a password, a key, or private data.
  3. Tell me what is still ONLY in this chat and in no file - this is my last
     chance to save it.
  4. Tell me which of your statements are [UNSURE], gathered in one list.
  5. Last line, exactly one of these:
         Safe to leave: YES
         Safe to leave: NO - <the reason>

^^^^^ PASTE UP TO HERE ^^^^^


=====================================================================
 NEXT STEP
=====================================================================

Keep the old assistant and the old account open until the other side works.
Until then these files are your only copy.

Take the folder to the new PC, the new account or the new AI, and run
2_IMPORT_PROMPT.txt there.

=====================================================================
 END OF FILE
=====================================================================
