Same report through the Agent SDK

The same Zod schema and checks behind a different harness. review.ts asks the Claude Agent SDK for a report with outputFormat, gates structured_output with Zod and the semantic checks, retries at most twice with the specific errors, and runs on OpenRouter or a local Ollama model with no code change.

Back to the exercise

Before you start

Before you start:Install Claude Code Node.js, pnpm and TypeScript Get an OpenRouter key OpenRouter or a local model?Local models with Ollama

This exercise calls a model. Run it either way

Any model through OpenRouterOne key for Claude, GPT, Gemini, Llama and more. Pay per token, with a credit limit.A local open-source modelFree and private with Ollama. Weaker on hard tasks.

Which should I pick? →

The schema and checks stay the same; only the harness changes. review.ts calls the Agent SDK's query() with outputFormat: { type: "json_schema", schema }, takes structured_output from the result message and runs it through the gate from the previous exercise. The decision that matters most is how to treat each kind of result. Any non-success result goes to a human, including error_max_structured_output_retries, because the SDK has already retried. Only a report that is valid but fails the semantic checks gets another attempt, with its errors sent back to the model.

Answers Lesson 4: Structured outputs and grounding

Step-by-step answer

Stretch 40 min6 steps