Terminal-Universe Reconstructs Executable Environments From Agent Trajectories Instead of Generating Them From Scratch
Terminal-based code agents have produced trajectories at scale while realistic executable environments stay scarce, and environments are what post-training actually needs since each can be re-queried into many verifiable tasks while a trajectory is a single frozen demonstration. The insight is that a trajectory's tool-execution history exposes the structure and contents of the environment it ran in, so replaying recorded file operations restores each file to its pre-modification state, yielding a partial workspace a completion agent then fills in with missing files and dependencies. The framework scales tasks along breadth, by mining directional dependency relations between environments to synthesize cross-workspace queries spanning multiple codebases, and depth, by extending single-turn queries into longer interactions; it was the top paper on HuggingFace Daily Papers at 115 upvotes.
↳ Follow the thread