ecc-eval-harness

Eval-driven development for Claude Code sessions requires a formal evaluation framework to define, implement tests, run checks, and generate reports. This Skill structures the workflow from defining capability and regression evaluations to reporting.

Eval-driven testing validates new features and prevents regressions in AI coding sessions. You need it when shipping capabilities with Claude Code to ensure reproducible checks using documented criteria, metrics, and artifact storage.

Claude Code evaluations support diverse grader types including code-based, model-based, and human graders. This flexible assessment approach allows you to measure reliability using metrics like pass@k and pass^k.

Preventing regressions in AI-assisted coding involves running regression evaluations alongside capability evals. This framework provides reproducible checks with documented criteria to prevent breakages before shipping features.

The evaluation framework supports reproducible checks with documented criteria, metrics, and artifact storage. It provides a structured end-to-end workflow from define to report to ensure reliable feature shipping.

AI coding reliability is measured using metrics like pass@k and pass^k within a formal evaluation framework. These metrics track progress and validate capabilities across multiple grader types during Claude Code sessions.