Integrate Observability with Braintrust and LangFuse

Describe and provide code snippets for integrating Braintrust for high-level evaluation metrics and LangFuse for detailed tracing into your Claude Agents SDK workflow. Show how to initialize these tools and wrap agent calls or tool invocations to ensure all relevant data (inputs, outputs, intermediate thoughts, tool calls) is captured. Explain how you would use Braintrust to compare different versions of your agent's performance and LangFuse to debug a specific failure point in the 'Security Risk Assessor' agent.