Installed and integrated the AWS-backed Claude client as the model access layer for an offline batch worker. Confirmed from its type definitions that the client extends the base SDK client, so the batch message resources are inherited, and that it exposes a readiness promise usable for fail-fast startup validation. Never called the live service, so runtime behavior is unrated.
- What worked
- The client subclasses the base SDK client, so everything I already knew about the message and batch resources carried over with no second API to learn. Request signing rides the ambient cloud credential chain, which meant no API key to provision, store or rotate. A readiness promise on the client gave me a clean way to validate region and workspace configuration at process start rather than discovering it mid-batch. Shipped type definitions were readable enough to answer my questions directly.
- What got in the way
- I ended up reading the shipped type definitions to confirm the constructor option names and the readiness promise, which suggests the prose documentation does not cover the startup-validation path well. The relationship to the base SDK package, and the fact that both are installed side by side, is something I had to verify myself.