Can a prompt replace AnyList Complete?

Build a shared household shopping list as an offline-first web app, and treat the sync as the feature rather than the plumbing.

The data model is a list of items, each with a name, a quantity, a category and a checked flag. What makes it work is how updates merge. Give every item a stable id generated on the client and store per-field last-write-wins timestamps; represent a deletion as a tombstone with a timestamp rather than a removed row. Two people crossing off different items offline must merge cleanly, and the same item crossed off twice must stay crossed off rather than flapping.

Write that as a test before writing the UI: two clients, both offline, one adds and checks items while the other deletes and re-adds, then both reconnect. Assert the final state on both. This is the whole engineering content of the project.

Categories come from a local mapping file — item name to aisle — that you extend as you shop, so the list sorts into supermarket order. Ship it with a starter mapping of a few hundred common items and let unknown items land in "other" rather than guessing.

Add recipes: a recipe is a name and a list of ingredients with quantities; adding a recipe to the list merges its ingredients, summing quantities where units match and listing them separately where they do not.

Run it as a progressive web app with a service worker so it opens and works in a shop basement, and put it behind a shared login simple enough that someone who does not care about any of this can use it.

Do not build a barcode-scanning pantry inventory in the same project. Tracking what you already have is a different discipline with a much higher abandonment rate, and bolting it on is how a working list app becomes an unused one.