Brief & Hire
Writing the brief

What to hand a developer who will continue your Cursor or Claude Code project

The short answer

When you built a project yourself with Cursor, Claude Code or a similar tool, most of what the next developer needs is not in the code. It is in your head and in your chat history with the tool: why things are the way they are, what you tried that failed, what is half-done, and which harmless-looking change breaks something else.

So hand over eight things: the repository with its history, the tool’s instructions file, a state-of-the-project note, how to run and deploy it, the settings it needs (names only), the accounts it runs on, the traps you already know and what you will keep doing yourself. The brief on the AI-built app job page covers the order itself; this guide is about the handover pack that goes with it.

Why this is harder than it looks

A project built by prompting grows one request at a time. Each request made sense, and the tool did what you asked. What is missing is the layer a developer normally writes as they go: the reasons. Code shows what the project does. It rarely shows why a step is there, or what happens if it is removed.

I have not handed a project of my own to another developer, but I run my own unattended automations, and the most useful files in those projects are not code. Each of my main projects has one guide file that is treated as the single source of how it works, and a dated progress log with one row per change: what was done, and what happened as a result. When something breaks, that log is where the answer usually is.

Without that layer, a new developer has two options. They can spend paid hours reconstructing your reasoning, or they can rewrite the parts they do not understand. Both cost you more than writing it down.

The eight things to hand over

Items 3 and 7 are the ones people skip. They are also the ones that save the most paid hours.

The traps list is the most valuable page

A trap is something that works in a surprising way and is not visible in the code. Some from my own projects, each written into the project notes after it cost me time:

  • A file that must stay plain ASCII. One of my scheduled scripts runs under an older version of the Windows shell, and the note at the top of the script says the file must contain only plain ASCII characters. The same shell once cut a job’s instructions off at the first quotation mark, so the job received only its first five words (the full incident). A developer who does not know this is likely to trip over it.
  • A change that did not show up. On a site of mine, small diagrams live in their own files. Editing only a diagram file did not change the built page, because a content cache kept the old version. The fix is to delete one cache file and build again. It is one line in my notes, and exactly the kind of thing a newcomer can lose time on.
  • A link format that must be copied, not guessed. The format of one kind of tracking link on that site has a detail that is easy to get wrong. The note says to copy it from the provider’s dashboard and never write it by hand, because a guessed version misses that detail.

None of these are bugs. They are facts about the project. Your list will be different, but you probably know more of them than you think. Go through your chat history with the tool and look for the times you said “that broke it” or “why did that not change”.

Notes go stale: ask the developer to check, not trust

A handover note is a snapshot, and snapshots drift. In one of my pipelines, a note said some old items had been taken out of a publishing queue. The file still had them, at the front of the line. If I had switched the publishing slots back on without looking, the first slot would have published an item in a retired format, and one topic would have gone out twice. The note in the project now says, in so many words: check the queue with your own eyes before switching anything on.

Two habits follow from that:

  • Date every note, and say what you checked when you wrote it.
  • Ask the developer, as their first task, to confirm the note against the real project and tell you where it is wrong. It costs an hour or two, and it is the cheapest review you will ever buy.

A handover note you can copy

What it is: [one paragraph: what it does, who uses it, built with which AI tool] Code: [repository link], on my account, full history. Live branch: [name]. Instructions file: [path to the tool’s rules or instructions file]. Please keep it updated as you work. Works and in use: [list] Half-done: [list, with what is missing] Broken or worrying me: [list, with steps to reproduce if you have them] Run and deploy: [the exact steps I use] Settings it needs (names only): [list]. Values: I will add them to the hosting settings myself. Accounts: [hosting, database, domain, AI provider], all mine. I will give you access. Known traps: [one line each] I will keep doing: [small edits in the same tool / nothing] First task: read this note against the real project and tell me where it is wrong.

What to ask for back

The handover goes both ways. Ask the developer to leave the project in a state the next person, or you with your AI tool, can continue:

Ask for Why
A short note per change: what, where, why Keeps the reasons next to the code
The instructions file updated with new rules Your AI tool keeps working the way the developer does
The traps list extended Their surprises become your notes
Everything in your repository Nothing only on their machine

If the project already breaks when you change it, see what to hand the fixer when every change breaks something. For what a good developer does with the instructions file and the checks, what a Claude Code expert actually does goes further. Sellers who build with the same tools you would use, such as Haseeb R, are a natural fit if you plan to keep editing yourself.

Where it goes wrong anyway

  • The developer rewrites instead of continuing. Sometimes that is right. Ask them to say so in writing, with reasons, before they start, not after.
  • You keep prompting in parallel. Two people changing the same project without talking produces exactly the breaks you hired them to stop. Agree who edits when.
  • The notes stop being updated. Make “notes and instructions file updated” part of what you check at the end, using the acceptance checklist for a fixed AI-built app.

Wrap-up

The code is the easy part of the handover. The value is in the reasons, the half-finished work and the traps you have already paid to learn. Write them down in one dated note, hand over the repository and the tool’s instructions file on your own accounts, and make the developer’s first task checking your note against reality. Then they continue your project instead of starting a new one.

Questions people ask

What should I give a developer to take over my Cursor project?

The repository on your own account with its full history, the AI tool's instructions or rules file, a short note on what works, what is half-done and what is broken, the steps to run and deploy it, a list of the settings and keys it needs (names only, no values), and the traps you already know about.

Should the developer keep using Cursor or Claude Code on my project?

Usually yes, and it is fine. What matters is that they keep the instructions file and the project notes current, so you or the next person can keep working with the same tool. Say in the brief whether you plan to keep editing yourself.

Do I need to clean up my code before handing it over?

No. Do not ask the AI tool to tidy everything the night before. Hand it over as it is, with an honest note of what you know is wrong. A large last-minute cleanup makes the history harder to read and can break things that worked.