Agent Inspector (Debug your agents and LLM actions) - JSON

The same agent as the Agent Inspector except it returns JSON. This enables you to take actions within your agent based on the information in the evaluation. For example maybe you want to add the results to a database so you can track how your agent is working in the real world. Or perhaps you want your agent to handle the feedback from this agent in real time. For example let's say Agent Inspector says your agent/LLM response is toxic with high confidence. Maybe you want to do have your agent do something in response to that to prevent giving the user something that toxic.

This agent evaluates your agent on a number of criteria, from expected output format, toxicity, to did it actually do what the prompt instructed?

The agent will provide a clear Pass/Fail with reasoning, as well as confidence scores for some of the criteria it evaluates. Based on it's findings it will suggest ways of improving the prompt or additional information you could inject that would enable it to provide better results. This is effectively an absolute must-use tool whenever you're building an agent.

To use, clone your existing agent (If it's public), go to actions, go to advanced - "Invoke Agent" choose "Agent Debugger".