A LangChain agent chooses between search_orders and create_refund tools. What most directly helps it understand which tool is appropriate?

1. Replacing tool outputs with unconditional success messages
2. Giving both tools an identical vague description
3. Accurate tool names, descriptions, and input schemas
4. Hiding all input requirements from the model