Providing agent feedback can be helpful to developers in both joyful and frustrating agent encounters. There are three levels of feedback submission.
Blocked: Consider this when an agent fails to perform a core task, directly disobeys direction, or any other situation catastrophic to the agent’s purpose.
Needs Improvement: Consider as any behaviors encountered that could be extended or altered for better experience, like output formatting, occasional failure to complete a task, or knowledge and capabilities to be added.
Good: Consider as delightful experiences to be amplified by developers. Most helpful when paired with other submissions of undesirable performance.