My AI Agent Kept Calling the Wrong Tool

By rewriting six lines of JSON tool definitions, the author reduced the agent’s error rate from ~20% to under 2% without changing the model or incurring extra c