The spawn-subagent skill enables the main Claude Code agent to delegate complex, well-defined tasks to independent subagents. By operating in isolated git worktrees, these subagents can execute implementation or research tasks without cluttering the main agent's context window. It features rigorous token tracking, concurrent execution safety through task-level locking, and a two-stage planning process—separating high-level architectural outlines from detailed implementation specifications—to ensure maximum efficiency and reliability during unsupervised execution.
