Terminal-Universe

"Turning Agent Trajectories into Scalable Terminal Environments." The Qwen Team, with Tsinghua, observes that terminal-agent trajectories now exist at scale while executable environments, which agent post-training actually needs because each can be re-queried into many verifiable tasks with execution feedback, remain scarce. Terminal-Universe reconstructs environments from recorded trajectories by replaying their file operations, uses a completion agent to fill missing files and dependencies, and then synthesizes new verifiable tasks and continued interactions from the rebuilt environment for post-training. Project lead Zhenru Zhang; senior authors include An Yang, Fei Huang, and Dayiheng Liu.