菜单
Coordinate planner and specialist agents to decompose, retrieve, and synthesize complex RAG answers.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
Use this skill when building, debugging, or improving Retrieval-Augmented Generation systems, including chunking, vector database selection, hybrid search, reranking, multimodal RAG, code documentation RAG, retrieval latency, and production RAG architecture.
Chunk nested documents into parent-child levels so retrieval can move from broad sections to fine-grained passages.
Use semantic boundaries and embedding similarity to chunk text for higher-relevance retrieval.
Route RAG chunking decisions across semantic, hierarchical, sliding-window, contextual-header, and framework-selection strategies.
Use overlapping windows to preserve context across chunk boundaries while controlling retrieval size.
Reduce retrieval latency with caching, batching, and index-level optimization.
基于 SOC 职业分类
| name | multi-agent-rag-orchestration |
| title | Multi-Agent RAG Orchestration |
| description | Coordinate planner and specialist agents to decompose, retrieve, and synthesize complex RAG answers. |
| allowed-tools | ["Read","Grep","Glob","Bash"] |
| category | rag-agents |
| tags | ["multi-agent","orchestration","planner","decomposition"] |
Multi-agent orchestration splits a complex question across specialized agents, such as a planner, one or more retrievers, and a synthesizer. Each agent owns a narrow job, and a coordinator sequences them and merges the results into a single grounded answer.
Some questions require several independent lookups, cross-source reasoning, or distinct skills that a single prompt handles poorly. Cramming everything into one agent produces long, unfocused prompts and answers that quietly drop parts of the question.
Have a planner break the question into subtasks that can be handled independently, with explicit dependencies where order matters.
Why: Clear decomposition prevents subtasks from overlapping or dropping parts of the original question.
Give each subtask to an agent with a single responsibility and only the tools it needs.
Why: Narrow scope keeps prompts short, decisions debuggable, and behavior predictable.
Merge partial answers in a synthesis step that resolves conflicts and preserves source citations.
Why: A dedicated synthesis step avoids contradictory fragments and keeps the final answer traceable.
For a worked pattern, see the Multi-Agent RAG example and the LangGraph multi-agent guide.
