Pick a depth. Each prompt opens in your AI pre-loaded with the lesson. Click a row to preview the prompt.
AI builders are fast but they break quietly: a button that looks fine might do nothing, a form might accept garbage input, a counter might not update. Because you did not write the code, you cannot spot bugs by reading it — your quality gate is your own hands clicking through every screen like a suspicious stranger. This five-minute discipline is the single habit that separates an app that embarrasses you in front of a friend from one you are proud to send. Doing it deliberately now, with a written checklist, means you catch the quiet breakage before your first real user does.
Here is a real click-test on the habit tracker, done like an adversary trying to break it. Walk EVERY screen and click EVERY button: tap a habit — does it turn teal and does the counter go from 'Done: 2 / 5' to 'Done: 3 / 5'? Tap it again — does it un-check and drop back to 2? Tap the plus to add a habit named 'Meditate' — does it appear in the list and does the total change from 5 to 6? Now the adversarial part, SUBMIT SILLY INPUT: try adding a habit with a blank name — does it refuse, or add a nameless blank row (a bug)? Add a habit with a very long name like 'Do 100 pushups and then run five miles before breakfast every single day' — does the text wrap or overflow off the screen (a bug)? Add the same habit name twice — are duplicates allowed, and is that okay? Delete every habit one by one — does the counter show 'Done: 0 / 0' gracefully, or does it break with a divide-by-zero or a blank screen (a bug)? Write two lists as you go. WORKED: check/uncheck toggles color and counter, add and delete work, long names wrap correctly. DID NOT WORK: blank-name habit adds an empty row; deleting all habits shows a broken '0 / 0' with no friendly empty message. Those two failures become your next two change requests. The whole pass took under five minutes and caught two things the happy-path demo never showed.