Act as my break-it lead. Start the app, then spin up 5 subagents in parallel, each trying to break it a different way:
- The chaos user: double clicks, the back button mid-checkout, refreshing during a save, 10 tabs open at once
- The weird input user: emoji names, 5,000-character fields, pasted spreadsheets, wrong formats, empty submits
- The slow user: a throttled network, timeouts, a flaky connection, a phone-size screen
- The sneaky user: other people's IDs in the URL, edited requests, skipped steps, expired links
- The first-timer: no instructions, wrong turns, gives up the moment something is confusing
Each one logs every bug with exact steps to reproduce, what they expected, what happened, and a severity.
Then merge everything into one list sorted by severity, fix the critical bugs, and re-run those tests to prove they're fixed.
