It attacks the root causes of agent failure: bad parameters, missing validation, poor error handling, wrong tool selection, and lack of observability. By enriching tools with this metadata, agents make fewer mistakes and recover faster when issues occur.