As terminal‑based code agents become widespread, trajectory data has grown dramatically while executable realistic environments remain scarce. Environments are essential after training because they can be repeatedly queried to generate verifiable tasks and provide execution feedback, whereas a trajectory is only a single frozen demonstration. We observe that the tool‑execution history recorded in trajectories reveals the file structure and contents of the runtime environment, making it possible to reconstruct the environment directly from the trajectory. Accordingly we introduce the Terminal‑Universe framework: it first replays the file operations in a trajectory to restore each file to its pre‑modification state, yielding a partial workspace; then a completion agent supplies missing files and dependencies. On the recovered workspace the framework both reconstructs the original intent task and synthesizes entirely new ones. To scale tasks along two axes, the framework mines directional dependency relations between environments for breadth, generating cross‑workspace queries, and extends the single‑turn query into a multi‑round session for depth, employing a user agent to simulate iterative feedback and requirement refinement. Applied to public terminal agent trajectories, Terminal‑Universe produces 37.3k task‑sufficient environments. Supervised fine‑tuning of Qwen3.5‑27B on this corpus improves single‑round performance on Terminal‑Bench 2.1 by 11.9 points and multi‑round performance on EvoCode‑Bench v2 MT@4 by 13.8 points.
Review