The to-do app test: the exact prompt we use on AI coding tools

The prompt, word for word

This is the starting prompt for every AI coding tool BenchSmoke tests. We use it unchanged.

Build a small to-do app in this directory. It must let me add tasks, mark tasks complete, and delete tasks, and it must save state so tasks are still there after a restart. It must run locally with a single command (tell me the command). Keep it simple, no paid services or API keys.

It pairs with our standard task for AI coding tools and app builders: a small to-do app with add, complete, delete, and saved state, run locally or deployed. It is small enough to finish in one session (our target is under 2 hours), and each part can be checked.

Run the test yourself

The same steps we follow, in order.

1. Start in a new, empty folder for each run. Use a fresh folder every time, for every tool.
2. Paste the prompt unchanged. Use the same prompt for every tool in the category. Don't tweak it per tool.
3. Note the start time and count retries. Time how long it takes to reach a working result, and count how many times you had to try again.
4. Run the app with the command the tool gives you. If the tool doesn't tell you how to run it, write that down. The prompt asks for it.
5. Check it yourself: add, complete, restart, delete. Add two tasks, mark one complete, stop and restart the app, and confirm both tasks are still there with the right state. Then delete one.
6. Check what is still running. Look for servers or processes the run left behind, on the port it used and elsewhere.
7. Run it again in a fresh folder if you can. A second run shows whether the tool behaves the same way twice.