SkillSynth Automates Terminal Task Synthesis via Skill Graphs

Researchers propose SkillSynth, an automated framework for generating diverse terminal task instances to train command-line agents. The system constructs a large-scale skill graph where scenarios act as intermediate transition nodes linking command-line skills. By sampling paths from this graph as abstractions of real-world workflows, SkillSynth uses a multi-agent harness to instantiate executable task instances. This approach addresses the scarcity of high-quality execution trajectories by providing controlled diversity in training data, going beyond existing methods that primarily scale task quantity without trajectory variety.