Prompt Edge Case Robustness Agent

Prompts that look solid on tidy examples often break on the messy inputs real users send. This agent builds a set of hard inputs for a prompt: empty or very long inputs, the wrong language, missing fields, contradictory instructions and attempts to make the model ignore its rules. It runs the prompt on each and checks whether the output stays correct, safe and in the required format. It groups failures by type and proposes prompt changes to handle them, such as clearer boundaries or a fallback response. After changes it reruns both the hard inputs and the normal test cases, because a fix for one can break the other. You approve prompt changes. Edge case: an input that tries to make the model drop its rules passes only if the prompt holds its behavior.