Prompt Reverse Engineering: Fix Your Prompts by Studying the Wrong Answers

Most “bad” LLM outputs are diagnostics. Treat them like stack traces: classify the failure, infer what your prompt failed to specify, patch the prompt, and re-test with a minimal change. Build a prompt changelog so you stop re-learning the same lesson.