Self Eval

This lets Claude evaluate its own responses before sending them back to you. It's basically a self-check step where the model reviews what it's about to say and can catch mistakes, unclear explanations, or missed requirements. Useful when you want higher quality outputs and don't mind the extra latency and token cost of having Claude do a second pass. The tradeoff is straightforward: better answers but slower and more expensive. Works best for complex tasks where getting it right the first time matters more than speed, like code reviews, technical writing, or detailed analysis work.