Anthropic just dropped a new /goal command for its CLI coding tool Claude Code. You tell Claude the end condition, and it'll keep working across multiple turns until the job's done. It works in interactive mode, with the -p flag, and via Remote Control. While executing, a terminal overlay shows elapsed time, round count, and token usage.
https://twitter.com/thsottiaux/status/2054082531432210765
The design is basically a carbon copy of OpenAI Codex's /goal feature. Codex's docs describe /goal as giving the model a long-term goal with a verifiable stop condition, letting it execute across multiple rounds — perfect for code migration, massive refactoring, deployment retries, experiments, and prototyping.
OpenAI Codex engineering lead Tibo Sottiaux then reshared a screenshot with a pointed comment: "Apprentice becomes master — Claude finally started copying Codex." He added, "But it can't keep up with GPT-5.5's acceleration."