Run the title-review agent on the SDK

The lesson's title-review agent run for real on the Claude Agent SDK, over four synthetic official copies, on any OpenRouter model or a local Ollama model. Every number comes from the result message, the structured output is validated twice and checked against the register, and one session is resumed by ID.

Back to the exercise

Before you start

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

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 agent is the lesson's code, made runnable. buildOptions produces exactly the options from the lesson (tools: [], the registry tool in allowedTools, dontAsk, maxTurns: 8, maxBudgetUsd: 0.5, a draft-07 outputFormat) plus the two hooks. runOnce reads only the result message. query is a parameter, so the demo replays canned messages instead of calling a model. The SDK doesn't run on a Claude subscription, so the endpoint comes from a .env file pointing at OpenRouter or a local model through Ollama. The decision that matters most is what counts as success: subtype: "success", then a second Zod parse, then a check that every evidenceQuote really appears in the register, because a schema can make a field exist but can't make it true.

Answers Lesson 2: The Claude Agent SDK

Step-by-step answer

Stretch 60 min8 steps