Toward Scalable Terminal Task Synthesis via Skill Graphs

TL;DR

SkillSynth is a framework that constructs skill graphs to generate diverse, scalable terminal tasks for training autonomous command-line agents, improving trajectory diversity and agent capabilities.

Contribution

It introduces a scenario-mediated skill graph for controlled, large-scale terminal task synthesis, enhancing diversity and scalability in training data.

Findings

SkillSynth effectively controls trajectory diversity during task synthesis.

Synthesized tasks improve training outcomes for terminal agents.

Adopted synthesized tasks enhance agent capabilities in experiments.

Abstract

Terminal agents have demonstrated strong potential for autonomous command-line execution, yet their training remains constrained by the scarcity of high-quality and diverse execution trajectories. Existing approaches mitigate this bottleneck by synthesizing large-scale terminal task instances for trajectory sampling. However, they primarily focus on scaling the number of tasks while providing limited control over the diversity of execution trajectories that agents actually experience during training. In this paper, we present SkillSynth, an automated framework for terminal task synthesis built on a scenario-mediated skill graph. SkillSynth first constructs a large-scale skill graph, where scenarios serve as intermediate transition nodes that connect diverse command-line skills. It then samples paths from this graph as abstractions of real-world workflows, and uses a multi-agent harness to instantiate them into executable task instances. By grounding task synthesis in graph-sampled workflow paths, SkillSynth explicitly controls the diversity of minimal execution trajectories required to solve the synthesized tasks. Experiments on Terminal-Bench demonstrate the effectiveness of SkillSynth. Moreover, task instances synthesized by SkillSynth have been adopted to train Hy3 Preview, contributing to its enhanced agentic capabilities in terminal-based settings.