From what I see, waza runs high confidence trigger tests until the agent becomes idle again. This is inefficient and may become a problem if we run many such tests in an environment with a hard time limit (e.g. GitHub Actions).