orchestrator

Task delegation to sub-agents is handled by reading the project queue from .maestro/state.json, creating isolated git worktrees, and launching sub-agents to implement the code within those branches. The orchestrator selects the next high-priority task or a specific task by ID to execute this workflow.

Git worktree-based task orchestration is a process where each development task is implemented in an isolated git worktree with a generated branch. This allows sub-agents to safely write code without affecting the main repository until a guarded merge is performed.

To plan development work and prepare an approval-ready plan, the orchestrator runs exploration agents, writes resilient notes to the repository, and produces a structured plan based on the queued product tasks. This allows managers to review and approve the proposed implementation strategy before sub-agents begin coding.

No, the task orchestration workflow requires reading .maestro/state.json to inspect the project queue and select tasks. This file is a mandatory prerequisite for the orchestrator to identify high-priority tasks and manage the task lifecycle.

Managing dependencies and epics in a development queue is done through arrange workflows that assign epics and dependencies without altering task identities. The orchestrator reads the current state and updates the queue to reflect these relationships before selecting the next task for implementation.

No, the orchestrator does not directly implement code changes; it acts as a tech lead by planning, delegating, and reviewing. It launches constrained sub-agents to operate inside isolated git worktrees and uses CLI helpers to perform guarded merges after reviewing the outcomes.