Continuous Prompt Evaluation: How We Use LLM Judges and Live Signals to Improve Kiro Agent Quality

Prompt behavior is difficult to validate exhaustively. A system prompt operates across combinations of models, tools, codebases, tasks, and users that no test...