September 28, 2026
Claude Code Helps Build App Evaluations: Version 2.1.259 or Later Required
On Sep 28, Claude Code 2.1.259 received commands for building evaluations and tuning application settings. `/claude-api build-eval` creates a test suite, while `hillclimb` changes the configuration using train cases and checks the result against a separate test set.
ClaudeDevs
@claudedevs
Claude now helps build evaluations and improve apps based on their results. In this article, we explain how to design evaluations and use skills that let Claude Code improve your apps. https://claude.dev/blog/automating-eval-design-and-hillclimbing/
· 41.3K views
**Hillclimb used to require a ready-made evaluation.** Now Claude Code helps build one with `/claude-api build-eval`, while `/claude-api prompt-audit` checks prompts, skills, and tool descriptions in the working folder before launch.
A real-world example. In a support test, hillclimb raised results across 14 held-out tickets from 78.6% to 90.5% and cut the cost by roughly 5x. For production rollout, the documentation recommends around 50 cases and at least 5 runs for each configuration.
Paid Claude Code access starts with Pro at $17 per month. The Free plan does not include access.
The production rollout benchmark remains the same: around 50 cases and at least 5 runs for each configuration.
