orchestrator

Coordinate parallel execution by decomposing complex tasks into independent waves of sub-agent work. Each wave runs concurrently up to resource limits, and a final merge step collects the manifests to deliver a unified outcome.

Decompose multi-step research tasks into atomic units grouped into sequential waves with proper dependencies. This parallel processing approach accelerates delivery by launching separate agents for each unit simultaneously.

Wave planning breaks tasks into atomic units and groups them into dependency-ordered execution waves. The system enforces design rules like maximum agents per wave and per-wave time caps to ensure reliable parallel execution.

Yes, result integration collects manifests from each wave and handles partial failures during the merge process. This ensures that the final output delivers a unified outcome even if some sub-agent tasks encounter issues.

Use parallel sub-agent orchestration for multi-step workflows, research, or data processing where independent tasks can run concurrently. Sequential execution is better suited for strictly dependent tasks that cannot be grouped into parallel waves.

Max agents per wave limits concurrent execution to available resources, preventing system overload. While this cap ensures stability during parallel task coordination, it may increase total execution time for workloads with numerous independent atomic units.