When a model can call external tools — search engines, APIs, code interpreters, databases — a tool prompt provides the descriptions the model uses to decide whether and how to invoke each one. The quality of these descriptions directly affects whether the model uses tools appropriately: a vague tool description may cause the model to call it at the wrong time, miss it when it would help, or misinterpret the results. Writing good tool prompts requires understanding both the model’s reasoning process and the tool’s actual capabilities and limitations. For behavior architects, tool prompts are part of the broader context design challenge — the model can only be as good as the information it has about what resources it can use.