CLI-Universe: Towards Verifiable Task Synthesis Engine for Terminal Agents

CLI-Universe is a new synthesis engine designed to generate high-quality, executable tasks for terminal agents, addressing the challenges posed by the lack of reliable training data. It utilizes a multi-dimensional capability taxonomy to create candidate tasks and employs a rigorous verification pipeline, discarding two-thirds of candidates to ensure the remaining tasks are genuinely challenging and verifiable. Fine-tuning the Qwen3-32B model on the newly established CLI-Universe-6K dataset achieved a 33.4% score on Terminal-Bench 2.0, setting a new state-of-the-art performance for models at or below 32B parameters, highlighting the effectiveness of structured data synthesis in improving model training efficiency.