Lets a main agent synthesize executable workflows to coordinate and verify sub-agents as a task unfolds. The paper reports 88.08% overall accuracy in Forced mode across six benchmarks.