A really convenient feature has been added to Claude Code, the AI coding tool for the terminal. It's the /goal command. While you previously had to press Enter and provide feedback at every step, you can now just set the final goal and step away entirely haha

The way it works is quite novel. For example, if you set a goal that all tests in a specific folder must pass, Claude modifies the code and then a secondary model, Haiku, steps in as a grader to verify the results directly. If there are any failures, it uses the grader's feedback as input to continuously loop and fix errors until everything passes.

# Claude Code 터미널에서 목표 설정하기
/goal test/auth 폴더 안의 모든 테스트가 오류 없이 통과해야 해

A monitoring feature has also been added for when you have multiple agent tasks running in the background. If you run the claude agents command in your terminal, you can manage active sessions at a glance, which is quite intuitive. If you've been tired of simple typing or repetitive test execution, definitely give this a try!