Define a validated function tool for an agent

Goal

Give an agent a tool with constrained, predictable inputs.

Steps

1. State the tool’s single task and forbidden side effects first. [openai-tools]
2. Expose a narrowly scoped function with @function_tool or @tool; use Python type hints/Pydantic BaseModel to generate and validate its argument schema. [openai-tools]
3. Use Field(ge=..., le=...) for numeric bounds and Literal[...] for choices; recheck business constraints at the handler boundary. [openai-tools]
4. Register the function only with agents that need it and document its visible description. [openai-tools]
5. Check the schema and rejection behavior with valid, missing-field, and out-of-range inputs. [openai-tools]

How to verify

Confirm the generated tool schema has the intended fields and invalid inputs trigger no side effects.

Troubleshooting

If the schema is too broad, split the task into narrower tools and validate again at the function boundary.