agent-orchestrator
bởi aatmaan1
AI & LLMsTóm tắt
Meta-agent skill for orchestrating complex tasks.
Mô tả
Kỹ năng này làm gì
A meta-agent skill that decomposes complex tasks into parallelizable subtasks, spawns specialized sub-agents with dynamically generated SKILL.md files, and consolidates their outputs. Agents communicate through a file-based inbox/outbox protocol, with status.json tracking each agent's state from pending to completed or failed.
Parallel sub-agent execution cuts total task time compared to sequential single-agent workflows, particularly when subtasks have no shared dependencies.
Khi nào nên dùng
- Running parallel research agents across multiple independent topics
- Breaking a large codebase refactor into isolated agent-owned modules
- Generating multi-section reports with dedicated collector, analyst, writer, and reviewer agents
- Coordinating sequential data pipeline stages where one agent feeds the next
- Delegating documentation tasks across specialized agents with automatic output consolidation
Quy trình ví dụ
User asks: Create a comprehensive market analysis report
- 1
Decompose the task into four subtasks: data collection, analysis, writing, and review, with an explicit dependency chain
- 2
Generate a workspace and SKILL.md for each sub-agent using scripts/create_agent.py
- 3
Write task instructions to each agent's inbox/ directory and set status.json to pending
- 4
Spawn agents via the Task tool and poll status.json files until all report completed
- 5
Collect outputs from each outbox/, merge sections, and run scripts/dissolve_agents.py to archive workspaces
A consolidated market analysis report with a summary of each agent's deliverables and any issues encountered
Cài đặt
clawhub install aatmaan1/agent-orchestratorChạy lệnh CLI, hoặc dán prompt vào Claude Code, Cursor, Codex hay coding agent bạn đang dùng.