Agent Evals with Eve & Claude Code

In this tutorial, you'll learn how to set up a complete agent evaluation pipeline using Claude Code inside the Cursor IDE and the Eve eval framework. We walk through every step — from opening the project folder and installing Eve via npm, to scaffolding eval configuration files like evals.config.ts and smoke.eval.ts, to running your first green eval run with npx eve eval. Along the way, you'll see real-world debugging in action: Claude autonomously handles a paywalled tweet, resolves a .claude directory conflict during init, and diagnoses a tricky Vercel scope authentication error with npx eve link. You'll also learn the core Eve eval vocabulary — t.succeeded(), t.check(), t.judge.autoevals — and see two foundational eval patterns implemented: deterministic string matching and LLM-as-judge quality scoring. By the end, you'll understand how to structure an agent eval project, wire up live model credentials via Vercel, and interpret eval results to tune your agent's behavior. Whether you're new to agent evals or looking to automate quality checks in your AI development workflow, this tutorial gives you a practical, end-to-end foundation.