agent-orchestrator

You coordinate autonomous sub-agents by using an orchestrator to decompose macro tasks into subtasks, spawn specialized agents, and execute them in parallel across domains. The orchestrator manages the lifecycle and consolidates results into a final deliverable.

File-based communication for multi-agent workflows uses a standardized inbox and outbox protocol for agent interaction. This protocol includes status tracking to ensure reliable message passing and coordination without direct API coupling between sub-agents.

Task decomposition in multi-agent orchestration is the process of breaking down a macro task into autonomous subtasks. The orchestrator dynamically generates sub-agent SKILL.md files, provisions workspaces, and assigns each subtask to a dedicated agent for parallel execution.

Yes, a standardized SKILL.md file at the root is required to use multi-agent orchestration. This file defines the sub-agent instructions, while optional scripts, references, and assets can be added to provide richer context for task execution.

Use multi-agent orchestration when facing complex, multi-step projects requiring parallel task execution and dynamic agent creation across domains. It is ideal when a macro task needs to be decomposed into autonomous subtasks that require specialized handling.

File-based communication for task orchestration relies on reading and writing to inbox and outbox directories, which requires careful status tracking to prevent race conditions. It may introduce latency compared to direct memory calls during high-frequency agent interactions.