eval-harness

To set up regression evals for Claude Code sessions, you need a defined evaluation structure that applies diverse graders across code changes and integration steps. This framework coordinates eval-driven development by formalizing the verification process.

Eval-driven development is a process that uses formal capability evaluations to guide AI development. It requires a defined eval structure and multiple graders, including code-based, model-based, and human graders, to rigorously verify code changes and prompts.

Yes, you can use a model grader alongside code-based and human graders. The framework supports multiple graders to provide robust assessment for capability, regression, and model-based evaluations across prompts and integration steps.

You can store evaluation baselines and reports using a centralized storage pattern. This governance approach tracks eval definitions and baselines over time, ensuring you can monitor progress and maintain rigorous verifications across sessions.

Yes, the evaluation framework operates without external dependencies. It coordinates eval-driven development natively within Claude Code sessions, requiring only a defined eval structure and diverse graders to function effectively across code changes.