Question
You're shipping a hand-rolled, fast C++ varint decoder for a networking protocol and asked an AI to help you test it adversarially. It proposed a fuzz target that feeds random bytes and asserts the decoder doesn't crash or read out of bounds (with ASan). That's useful but you want to be sure decoded values are correct, not just that it doesn't crash. Describe the testing approach you'd direct the AI to build, and the trap in verifying a decoder this way.
Treat the AI’s output as a draft to verify, not an answer to trust. Name the specific flaw and the input that triggers it, say how you’d catch it — tests, edge cases, reading critically — and how you’d re-prompt or decompose to get it right.
Vibe coding: describe the solution in plain language (or narrate it) and the coach grades your approach. Generating runnable code from your description is coming next.