Skip to content

Claude Code Adds claude plugin eval Command

Anthropic

The new command allows developers to generate test cases and benchmark output scores with and without their plugin enabled.

Key points

  • Running claude plugin eval init prompts Claude to draft test cases, pilot the suite, and estimate total run token costs.
  • Evaluation outputs include terminal score comparisons and an HTML report that can publish as a private artifact.
  • Developers can update Claude Code to access the feature via claude update.

Why it matters

Plugin authors can quantitatively test and measure the actual performance impact and token cost of their custom Claude Code skills.

Sources