Act as a senior engineer joining my {{tech_stack}} project. Every time my AI tool fixes one thing, something that worked before breaks. The latest thing it broke: {{page_or_feature}}. Help me stop the cycle before we add anything new. 1. Gather evidence. Use git history, or the tool's version history, to find the last version where it worked and the change that broke it. Describe that diff in plain words: what it touched that the task didn't need. 2. Name the real cause. The usual ones: whole files rewritten to change a few lines; a shared component or helper changed for one page and broken for the others; the AI losing context in a long chat and re-creating code that already exists; two copies of the same logic drifting apart; and no tests, so nothing complains when behavior changes. 3. Restore what broke with the smallest change, not a rewrite. 4. Lock in behavior. List the 5 to 10 flows that must never break, such as sign-up, sign-in, checkout or the contact form. Write a fast automated test for each (end-to-end for flows, unit tests for prices or date logic) and make them pass on the current code before any new change. 5. Write a short rules file for the AI tool: run the tests after every change; change only the files the task names; never rewrite a file to change a few lines; reuse existing components; ask before touching shared components, auth, payments or the database; after two failed attempts, stop and report instead of guessing. 6. A working habit: one small task per chat, a commit after each green test run, and a fresh chat when the AI starts repeating itself. Explain each step plainly. Finish with the protected flows, the tests added, and the rules file.