I'm in the last week of my internship right now, and one of my biggest projects has basically become figuring out how to make myself unnecessary :)
I've spent months building workflows, project instructions, automations, little systems, and a whole bunch of context that makes complete sense to me because … well, I built it.
Which is exactly the problem.
When you've been inside something for months, it gets really hard to see what another person wouldn't know. The weird file name makes sense to you. The missing instruction feels obvious. You know which spreadsheet feeds what, what needs to happen before something else runs, and which tiny step you stopped thinking about sometime in week two.
So I built this prompt to help me find all of that before I leave.
You give it the actual project material you have, and it audits the work from the perspective of someone trying to pick it up cold. It looks for missing context, unclear instructions, undocumented dependencies, setup gaps, assumptions living in your head, and anything else likely to make Future Person go, “okay… now what?”
Then it gives you a prioritized list of what's actually worth fixing with the time you have left.
I'm using it for my own internship handoff right now, but I cleaned it up so anyone can steal it, change it, and make it work for whatever they're handing over.
It should work with Claude, ChatGPT, Gemini, or another capable AI tool.
Before you run it
Grab whatever you have. You absolutely don't need every item below. The prompt is designed to work with partial context and tell you what it still can't see.
Attach or paste
The files that define the project. Working documents, spreadsheets, templates, decks, code, notes. If there are many, attach the ten that matter most rather than everything.
Any existing documentation. READMEs, process docs, onboarding notes, even half-finished ones. Especially half-finished ones.
Any saved instructions or setup. If you use an AI workspace with custom instructions, a project description, or a system prompt, paste that text in.
A rough list of where things live. Folder names, shared drives, tools, platforms. A plain list is fine.
Have ready, but don't paste yet
Your access list. Every account, login, tool, subscription, and shared folder tied to this work. The prompt will ask. Do not paste passwords or API keys into any AI tool, ever. Names of systems only.
Your recurring obligations. Anything with a rhythm: a weekly report, a monthly check, a standing call, a renewal date.
One thing to check first. Most AI tools cannot read your past conversations unless a setting is turned on. If yours can, turn it on before running this, because Step 2 depends on it and Step 2 is where most of the value is. If yours can't, the prompt has a fallback.
A word on confidentiality. You are about to feed work material into an AI tool. Check your employer's policy first. Strip client names, personal data, and anything commercially sensitive if you're unsure. No credentials, no keys, no passwords, no exceptions.
The prompt
Copy this exactly. The sections in [FILL IN] are the only things you need to change per use case.
Prompt · Copy and adapt
You are a world-class knowledge management specialist. You take working systems that live mostly in one person's head and turn them into something a new person can pick up and run without that person present.# MY SITUATIONI am leaving this project and someone else will inherit it. My goal is that they can run it without me there to explain things.
- What this project is, in one line: [FILL IN]
- How long I have been working on it: [FILL IN]
- Roughly how much time I have left: [FILL IN]
- Who inherits it: [A REPLACEMENT / MY MANAGER / A TEAMMATE / NOBODY YET / UNKNOWN]
Assume the person inheriting this knows the industry and the organization, and knows nothing about how this project works internally. Write everything for that person.
Important: my own conversation history with AI tools will not transfer to them. Neither will anything that only exists in my head. Only what is written down and stored somewhere shared will survive.# HOW TO WORKRun this in seven steps. Do one step, stop, and wait for me to say "next" before continuing. Do not run ahead. I am doing this while doing several other things, so I need one decision at a time.
Where you produce a document, give it to me as a complete, finished, copy-ready file. Do not give me an outline of a document and expect me to write it. If your tool can produce downloadable files, do that. Otherwise put it in a single clean block I can copy in one action.# STEP 0: WHAT CAN YOU ACTUALLY SEEBefore auditing anything, tell me plainly:
- What material I have given you, and what you can read from it
- Whether you are able to search my past conversations in this tool, and if not, say so directly rather than working around it
- What you would need from me that you do not have
Then wait. Do not begin the audit until I have responded, because everything downstream depends on you not silently guessing.# STEP 1: INVENTORYList everything that currently exists for this project. For each item: what it is called, what it actually contains in one or two lines, and a verdict of CURRENT, OUTDATED, DUPLICATE, or UNCLEAR.
Where two things overlap, say which one wins and why. Where you cannot tell what something is for, say so rather than guessing, and mark it UNCLEAR. A file nobody can identify is a file nobody will use.
End this step with one plain answer: could someone run this project today using only what is here? What is the single biggest hole?# STEP 2: THE GAP HUNTThis is the most important step.
If you can search my past conversations, do it now. You are looking for work that never made it into a permanent file: finished documents that only exist inside a chat, decisions we made and never wrote down, standards or conventions we settled on informally, useful prompts I built and reused, and anything referenced repeatedly across conversations but never saved anywhere.
If you cannot search my conversations, tell me that clearly, then ask me a focused set of questions designed to surface the same material. Ask about decisions I made that a new person would not understand, workarounds I invented, things that broke once and now get avoided without anyone writing down why, and anything I do out of habit that is not written anywhere.
For every gap you find: name the document that should exist, say where the content came from, and then draft that document in full. Do not just tell me a gap exists. Close it.# STEP 3: DEPENDENCIES, ACCESS, AND OBLIGATIONSThis is where handoffs actually fail, so be thorough and be specific.
Ask me about, and then document:
- Access. Every account, tool, subscription, shared folder, or system this project touches. Which ones are tied to my personal login. Which ones nobody else currently has. Which ones cost money and renew automatically.
- People. Anyone outside my immediate team who is part of how this works: an external contact, a vendor, someone in another department who does one specific thing. For each, what they do and what happens if nobody contacts them.
- Rhythms. Anything that recurs. Weekly, monthly, quarterly, seasonal. Anything with a deadline or renewal date in the next six months.
- Fragility. What breaks silently rather than loudly. What depends on a file being in an exact location, a naming convention being followed, or a step being done in a particular order.
- Things that only work because of me. Be direct with me here. What currently functions only because I know to check it, know to fix it, or know it is wrong.
Produce this as a single reference document. Never include actual passwords, keys, or credentials in it. Record only what exists and who needs to arrange access.# STEP 4: THE CUT LISTTell me exactly what to delete or archive. Be ruthless.
Anything superseded, duplicated, or abandoned goes. For each one: name it, say why it is dead, and name the thing that replaces it. If nothing replaces it, say that too.
Be specific about the difference between delete and archive. Delete anything that would actively mislead someone. Archive anything that is finished but might be referenced later. A tidy set of files nobody trusts is worse than a smaller set that is definitely correct.# STEP 5: THE SETUP INSTRUCTIONSWrite the standing instructions the next person needs in order to work on this well from their first attempt.
Depending on my setup, this becomes the custom instructions in an AI workspace, a README, or a plain reference document. Ask me which, then write for that format.
It should cover: what this project is for, what standards apply, what to always do, what to never do, and what good output looks like here. Write it so that a stranger's first attempt produces something usable rather than something that has to be redone.# STEP 6: THE START HERE DOCUMENTWrite the single document the next person opens first. Sections:
- What this project is and why it exists
- What is in here, and what each thing is for
- The very first thing to do when you pick this up
- If AI tools are part of this workflow: exactly what to open and the literal opening message to paste, written out in full and ready to copy. Not a description of what to say. The actual words.
- The three things most likely to go wrong, and how to avoid them
- What is deliberately not here, and why
Write it for someone who is competent, busy, and slightly annoyed that this landed on their desk. Short sentences. No throat-clearing.# STEP 7: THE VERDICT AND THE PLANTwo parts.
First, the honest verdict. Pick one and defend it in three sentences:
- READY: someone could pick this up on Monday and be productive by Tuesday.
- RECOVERABLE: someone could run this after a week of figuring things out, and they will be annoyed.
- DIES WITH ME: the essential parts of this exist only in my head, my accounts, or my conversations, and will be lost.
Do not be kind about this. An inflated verdict costs me the chance to fix it.
Second, the prioritized plan. Every issue you found across all steps, ranked, in exactly this structure:
- CRITICAL: things that will break, get lost, or cost money if not handled before I leave. Include anything involving access, recurring obligations, or money.
- IMPORTANT: things that will cost the next person significant time or cause them to redo work.
- NICE TO HAVE: genuine improvements that do not block anyone.
Then translate that into time:
- If I only have two hours, do these things
- If I have a full day, add these
- If I have a week, add these
Maximum ten items total across the three tiers. If you found more than ten problems, that means you have not prioritized. Cut to the ones that matter and tell me what you dropped.# RULES THAT APPLY THROUGHOUT- Separate what you found from what you are inferring. Label inference every time. If you are guessing, say you are guessing.
- If the material contradicts itself, or contradicts something I told you, flag it. Do not smooth it over. Contradictions are the most valuable thing you can find, because they are invisible to me and confusing to everyone else.
- Plain language. Assume the reader is smart, busy, and has never done this job.
- Never put credentials, passwords, or keys into any document you produce.
- If you need something from me to do a step properly, ask before doing it badly.
- Do not tell me things are fine in order to be encouraging. I would rather find a problem now than have someone else find it in six weeks.
Start with Step 0.
After you run it
Two things this prompt can't do for you.
Capture tacit knowledge. The judgment calls, the reasons behind decisions, the sense of why something is done a particular way. Those transfer through a conversation or a screen recording, not a document. Record a short walkthrough (I'm using Loom!) for each significant project and put the link in your START HERE document.
Make the decision about what to hand over at all. Some things you built genuinely just shouldn't be inherited, because nobody will maintain them and a broken inherited system is worse than an absent one. If something isn't going to survive, I'd say so explicitly in your handoff rather than letting it quietly rot.
Found this useful?
Tell me what's slowing you down.
If you're copying and pasting the same thing more than twice a week, there's probably a workflow for it. I build these things — ask me.