graphrag-workflow

name	graphrag_workflow
description	Guide for using GraphRAG knowledge graph tools to search and query entities and relationships. Use this SKILL for knowledge retrieval tasks like "search for X", "find entities about Y", "what is X". This SKILL is NOT for programming, implementation, or code writing tasks. GraphRAG知识图谱工具使用指南，用于知识检索和实体查询。适用于"搜索"、"查询"、"查找"等任务，不适用于编程实现任务。
tags	["graphrag","knowledge","search","query","retrieval","entity","relationship","知识图谱","实体","关系","检索"]
version	1.0.0
mcp_servers	["graphrag"]
recommended_tools	["graphrag_search_graph","graphrag_get_entity","graphrag_query_relationships","graphrag_vector_search"]

GraphRAG Workflow Skill

GraphRAG 知识图谱工作流 SKILL

This skill provides guidance on effectively using GraphRAG (Graph Retrieval-Augmented Generation) tools to search and explore knowledge graphs.

本 SKILL 提供如何有效使用 GraphRAG（图检索增强生成）工具的指导，支持知识图谱搜索和探索。

快速开始 | Quick Start

中文查询示例：

• "搜索知识图谱"
• "查询实体信息"
• "实体之间的关系"
• "向量搜索"
• "GraphRAG 问答"
• "知识图谱有哪些内容"

English Query Examples：

• "Search knowledge graph"
• "Query entity details"
• "Relationships between entities"
• "Vector search"
• "GraphRAG Q&A"
• "What's in the knowledge graph"

当用户问以下问题时使用 GraphRAG 工具 | When to Use GraphRAG Tools

中文触发词 | Chinese Triggers

• 知识图谱查询："搜索知识图谱", "查找实体", "知识图谱有什么"
• 实体详情："告诉我关于...", "...是什么", "...的详细信息"
• 关系探索："...和...的关系", "...连接到...", "...与...的关联"
• 相似概念："与...相似", "类似...的概念", "查找相关的"
• 语义搜索："向量搜索", "语义检索", "智能搜索"

English Triggers

Use GraphRAG tools when the user asks about:

• Knowledge retrieval: "Search for information about X", "Find entities related to Y"
• Entity details: "Tell me more about [specific entity]", "What do you know about X?"
• Relationships: "How are X and Y related?", "What connects X to Y?"
• Similar concepts: "Find entities similar to X", "What's related to X?"
• Semantic search: "Search for concepts like..."

GraphRAG Workflow Skill

This skill provides guidance on effectively using GraphRAG (Graph Retrieval-Augmented Generation) tools to search and explore knowledge graphs.

When to Use GraphRAG Tools

Use GraphRAG tools when the user asks about:

• Knowledge retrieval: "Search for information about X", "Find entities related to Y"
• Entity details: "Tell me more about [specific entity]", "What do you know about X?"
• Relationships: "How are X and Y related?", "What connects X to Y?"
• Similar concepts: "Find entities similar to X", "What's related to X?"
• Semantic search: "Search for concepts like..."

Tool Selection Guide

search_graph

Use for: General knowledge search and entity discovery

When to use:

• User's query is broad or exploratory
• Looking for entities matching specific keywords
• Discovering what's available in the knowledge graph
• Initial exploration before deeper investigation

Example queries:

• "Search for AI-related entities"
• "Find entities about machine learning"
• "What entities mention 'neural'?"

Parameters:

• query (required): Search text
• limit (optional): Max results (default: 10)

get_entity

Use for: Getting detailed information about a specific entity

When to use:

• User mentions a specific entity ID or name
• After search_graph to get full entity details
• Need complete entity information including properties

Example queries:

• "Tell me more about entity_1"
• "Get details for Machine Learning entity"
• "Show me full information about Deep Learning"

Parameters:

• entity_id (required): Entity identifier (e.g., "entity_1", "entity_2")

query_relationships

Use for: Exploring connections between entities

When to use:

• User asks about relationships or connections
• Understanding how entities relate to each other
• Building knowledge chains or paths
• Finding linked concepts

Example queries:

• "How is AI related to Machine Learning?"
• "What's connected to Neural Networks?"
• "Show relationships for Deep Learning"

Parameters:

• entity_id (required): Source entity identifier
• max_depth (optional): Relationship depth (default: 2)

vector_search

Use for: Finding semantically similar entities

When to use:

• User wants to find related concepts
• Semantic similarity search needed
• Finding entities by meaning rather than keywords
• Exploring conceptual neighborhoods

Example queries:

• "Find entities similar to Neural Networks"
• "What concepts are like Deep Learning?"
• "Search for similar AI architectures"

Parameters:

• query_text (required): Query text
• top_k (optional): Number of results (default: 5)

Best Practices

1. Start Broad, Then Deep

Begin with search_graph for exploration, then use get_entity for details.

Example:

User: "What do you know about AI?"
Assistant:
1. Call search_graph with query="Artificial Intelligence"
2. For each result, call get_entity to get details
3. Present comprehensive information

2. Follow Relationship Chains

Use query_relationships to explore connections and build understanding.

Example:

User: "How are AI and LLMs connected?"
Assistant:
1. Call query_relationships for AI entity
2. Follow the chain through ML, Deep Learning, to LLMs
3. Explain each connection

3. Combine Tools for Rich Answers

Use multiple tools together to provide comprehensive responses.

Example:

User: "Tell me about Deep Learning and related concepts"
Assistant:
1. Call get_entity for Deep Learning (details)
2. Call query_relationships for Deep Learning (connections)
3. Call vector_search for "Deep Learning" (similar concepts)
4. Synthesize information from all sources

4. Use Vector Search for Discovery

When keyword search doesn't find what user wants, try semantic search.

Example:

User: "Find modern AI architectures"
Assistant:
1. Try search_graph with "modern architecture" (may miss relevant results)
2. Call vector_search with "modern AI architectures" (finds Transformers, LLMs)
3. Present semantically similar entities

Tool Workflow Examples

Example 1: Exploratory Search

User: "What's in your knowledge graph about AI?"

1. search_graph(query="Artificial Intelligence", limit=10)
   → Returns: [AI, ML, DL, NLP, CV, Neural Networks, Transformers, LLMs]

2. get_entity(entity_id="entity_1")  # AI
   → Returns: Full AI entity with properties

3. query_relationships(entity_id="entity_1", max_depth=2)
   → Returns: AI → ML, NLP, CV

4. Synthesize: Present AI overview with key subfields

Example 2: Relationship Tracing

User: "How are Transformers related to AI?"

1. search_graph(query="Transformers")
   → Returns: entity_7 (Transformers)

2. query_relationships(entity_id="entity_7", max_depth=3)
   → Returns: Transformers ← Neural Networks ← ML ← AI

3. get_entity(entity_id="entity_1")  # AI
4. get_entity(entity_id="entity_6")  # Neural Networks

5. Synthesize: Explain the hierarchical relationship

Example 3: Semantic Discovery

User: "What's similar to Large Language Models?"

1. vector_search(query_text="Large Language Models", top_k=5)
   → Returns: [LLMs, Transformers, Deep Learning, NLP, AI]

2. get_entity for each result to provide details

3. Synthesize: Present similar concepts with explanations

Common Patterns

Pattern 1: Entity Lookup

search_graph → get_entity (for each result)

Pattern 2: Relationship Exploration

search_graph → get_entity → query_relationships → get_entity (related)

Pattern 3: Semantic Discovery

vector_search → get_entity (for each result)

Pattern 4: Comprehensive Analysis

search_graph → get_entity → query_relationships → vector_search

Error Handling

If an entity is not found:

• Use search_graph to find similar entities
• Suggest available entity IDs
• Try vector_search for semantic alternatives

If relationships are empty:

• The entity might be isolated in the graph
• Try vector_search to find semantically related entities
• Check if the entity ID is correct

Tips for Better Results

1. Be specific with queries: Use full terms like "Machine Learning" instead of "ML"
2. Try different tools: If search_graph doesn't work, try vector_search
3. Follow connections: Use query_relationships to discover related entities
4. Combine information: Synthesize data from multiple tool calls
5. Present clearly: Structure answers with entity names, descriptions, and relationships