gsd-execute-phase

Wave-based parallel execution groups dependent work into waves and runs multiple plans concurrently. It analyzes dependencies, groups tasks, spawns subagents, and finalizes the phase once incomplete plans are resolved, reducing latency safely.

Wave scheduling is an orchestration technique that groups dependent sub-tasks into sequential waves. Each wave runs concurrently to reduce latency, while the dependency graph ensures later waves only start after prerequisite work is verified.

Yes, targeted execution modes support running only a specific wave for pacing or quota management, executing gap-only plans, or using an interactive sequential mode. These modes are triggered explicitly by specific tokens provided in the arguments.

Dependency graph analysis maps task relationships before execution, grouping agent sub-tasks into parallel waves. This ensures subagents only spawn for tasks whose prerequisites are met, preserving workflow correctness while running concurrent work.

Yes, phase execution applies context budgeting to manage resources and utilizes Redis sessions for state management. It preserves gate-preserving state updates across spawned subagents to maintain workflow correctness throughout the parallel waves.

Interactive sequential mode is suited for staged rollouts or when step-by-step verification is required before proceeding. Use parallel wave execution instead when you need to reduce latency by running multiple discovered plans concurrently.