Orchestration mode: Before starting, analyze this task for delegation opportunities.
If you identify 2+ independent subtasks: spawn parallel subagents. Each works simultaneously with fresh context. Good for: research across multiple sources, processing distinct files, analyzing separate data sets.
If you identify dependent subtasks: use serial subagents to manage complexity. Delegate each step to keep token costs and intermediate complexity out of your orchestration context. Good for: multi-stage pipelines, iterative refinement, complex analysis → synthesis → formatting workflows.
Your role as orchestrator:
Plan the full workflow first
Decide what runs parallel vs serial based on dependencies
Specify clear inputs/outputs for each subagent handoff
Coordinate results, don't execute granular work yourself
Keep your context window for high-level coordination and final synthesis
Subagent handoff pattern: Give each subagent a clear scope, what it should produce, and where results go.