agent-orchestrator

To coordinate autonomous sub-agents for parallel tasks, you orchestrate a workflow that decomposes a macro objective into independent work units, dispatches instructions to sub-agent workspaces via file-based inboxes, and consolidates their outbox deliverables into a final result.

File-based IPC for multi-agent communication uses an inbox and outbox directory structure to deliver instructions and collect deliverables, while a status.json file tracks each sub-agent's execution progress for reliable coordination.

Merging outputs from multiple sub-agents after task decomposition involves monitoring each agent's status.json file, validating their deliverables against explicit success criteria, resolving integration conflicts, and producing a consolidated summary.

Yes, autonomous delegation supports multi-stage workflows like research, analysis, writing, and code generation by dynamically generating SKILL.md files for specialized sub-agents and merging their independent outputs with conflict resolution.

No dependencies are required to run a multi-agent orchestrator workflow. It relies entirely on dynamic SKILL.md generation and native file-based inbox and outbox communication to coordinate sub-agents and consolidate deliverables.

Multi-agent orchestration for task decomposition should be avoided for simple, linear tasks that lack independent work units, as the overhead of generating sub-agent workspaces and managing file-based IPC outweighs the benefits of parallel execution.