Large-Scale Terminal Agentic Trajectory Generation from Dockerized Environments

TL;DR

This paper introduces TerminalTraj, a scalable pipeline for generating high-quality, verified terminal trajectories from Dockerized environments, enabling improved training of agentic models for terminal tasks across diverse domains.

Contribution

The authors present TerminalTraj, a novel scalable pipeline that constructs Docker environments, generates task instances, and synthesizes verified trajectories, significantly advancing data quality and diversity for terminal-based agent training.

Findings

Curated 32,000 Docker images and generated over 50,700 verified trajectories.

Models trained on this data outperform baselines with up to 20% improvements.

TerminalTraj-32B achieves competitive performance with fewer than 100B parameters.

Abstract

Training agentic models for terminal-based tasks critically depends on high-quality terminal trajectories that capture realistic long-horizon interactions across diverse domains. However, constructing such data at scale remains challenging due to two key requirements: \textbf{\emph{Executability}}, since each instance requires a suitable and often distinct Docker environment; and \textbf{\emph{Verifiability}}, because heterogeneous task outputs preclude unified, standardized verification. To address these challenges, we propose \textbf{TerminalTraj}, a scalable pipeline that (i) filters high-quality repositories to construct Dockerized execution environments, (ii) generates Docker-aligned task instances, and (iii) synthesizes agent trajectories with executable validation code. Using TerminalTraj, we curate 32K Docker images and generate 50,733 verified terminal trajectories across eight domains. Models trained on this data with the Qwen2.5-Coder backbone achieve consistent performance improvements on TerminalBench (TB), with gains of up to 20\% on TB~1.0 and 10\% on TB~2.0 over their respective backbones. Notably, \textbf{TerminalTraj-32B} achieves strong performance among models with fewer than 100B parameters, reaching 35.30\% on TB~1.0 and 22.00\% on TB~2.0, and demonstrates improved test-time scaling behavior. All code and data are available at https://github.com/Wusiwei0410/TerminalTraj.