Safety and pre-launch testing

Two skills without which a project cannot be released are security and verification. Let's start with security: the code that AI writes sometimes requires passwords, keys, and access. Never insert real passwords and personal data into a request - write a stub like YOUR_PASSWORD, and insert the real value yourself, where the AI ​​will show you; Consider everything you sent in the request to be out of control. Now check. The finished result is not the one that “looks ready”, but the one that passes the test. Make a short list of check steps and go through them manually: enter normal data, then obviously crooked ones (empty, letters instead of numbers, huge numbers). Describe how you handle secrets, give a list of tests and name warning signs (asks for access that should not be required; suggests disabling protection). Test on fictitious data, not real ones - you don’t mind breaking it. And ask the AI: “what could go wrong here and how can I check it?” — the model itself will tell you the weak points. A common mistake is to trust the code just because it “ran”: running does not mean it calculates correctly.