TermiGen: High-Fidelity Environment and Robust Trajectory Synthesis for Terminal Agents

TL;DR

TermiGen introduces a comprehensive pipeline for creating high-fidelity, verifiable environments and resilient trajectories, significantly improving open-weight LLMs' ability to execute complex terminal tasks by reducing hallucinations and enhancing error recovery.

Contribution

It presents a novel end-to-end method for synthesizing environments and trajectories with error correction, leading to state-of-the-art performance on terminal task benchmarks.

Findings

Achieved 31.3% pass rate on TerminalBench with TermiGen-Qwen2.5-Coder-32B.

Outperformed existing baselines and proprietary models like o4-mini.

Generated diverse, verifiable environments and error-rich trajectories for training.

Abstract

Executing complex terminal tasks remains a significant challenge for open-weight LLMs, constrained by two fundamental limitations. First, high-fidelity, executable training environments are scarce: environments synthesized from real-world repositories are not diverse and scalable, while trajectories synthesized by LLMs suffer from hallucinations. Second, standard instruction tuning uses expert trajectories that rarely exhibit simple mistakes common to smaller models. This creates a distributional mismatch, leaving student models ill-equipped to recover from their own runtime failures. To bridge these gaps, we introduce TermiGen, an end-to-end pipeline for synthesizing verifiable environments and resilient expert trajectories. Termi-Gen first generates functionally valid tasks and Docker containers via an iterative multi-agent refinement loop. Subsequently, we employ a Generator-Critic protocol that actively injects errors during trajectory collection, synthesizing data rich in error-correction cycles. Fine-tuned on this TermiGen-generated dataset, our TermiGen-Qwen2.5-Coder-32B achieves a 31.3% pass rate on TerminalBench. This establishes a new open-weights state-of-the-art, outperforming existing baselines and notably surpassing capable proprietary models such as o4-mini. Dataset is avaiable at https://github.com/ucsb-mlsec/terminal-bench-env.