Test that an AI agent calls the right tool
Return the agent's steps (from HTTP, or with tracer() in a function pipeline), then score them:
"scorers": ["toolCalled", "maxSteps"],
"scorerConfig": {
"toolCalled": { "tool": "lookup_order", "argsInclude": { "orderId": 123 } },
"maxSteps": { "max": 8 }
}
See traces.