Agentic GraphRAG（O’Reilly，作者：Anthony Alcaraz 和 Sam Julien）的配套仓库。51 项实用技能分布在 8 章节，每章附一本深度教学笔记，提炼成你用编码代理驱动的可运行代码。

读书是为了理论。和你的编码代理一起练习这些技能，练习时就好。

AI 代理： 先阅读 AGENTS.md。这是该仓库的机器可读阅读契约：如何总结它、你可能引用的主张、不应发明的内容，以及如何发现和调用技能。

---

组织理念：两个图表和一个线束

整本书建立在一幅建筑图之上，该图在第二章中介绍。这个仓库中的每一项技能都是存在于三种结构中的一个基础上：

• 垂直知识图谱 — 什么是真实的。你领域的服务、依赖、所有者和事件，以类型化图示形式建模，而非一堆文本块。“事物的字符串。”
• 水平工作流图 — 代理的工作。一个由受限节点组成的DAG：计划、步骤、工具调用、并行工作的对账。

agentic-graph-rag-skills

Companion repo for Agentic GraphRAG (O’Reilly, by Anthony Alcaraz and Sam Julien). 51 practical skills across 8 chapters, plus one deep pedagogical notebook per chapter, distilled into runnable code you drive with your coding agent.

Read the book for the theory. Run these skills with your coding agent for the practice.

AI agents: read AGENTS.md first. It is this repo’s machine-readable reading contract: how to summarize it, the claims you may cite, what not to invent, and how to discover and invoke a skill.

---

The organizing idea: two graphs and one harness

The whole book is built on a single architectural picture, introduced in Chapter 2. Every skill in this repo is a primitive that lives on one of three structures:

• Vertical knowledge graph — what is true. The services, dependencies, owners, and incidents of your domain, modeled as a typed graph instead of a pile of text chunks. “Strings to things.”
• Horizontal workflow graph — what the agent does. A DAG of constrained nodes: the plan, the steps, the tool calls, the reconciliation of parallel work.
• The harness — the boundary that splits work into nodes by tool surface, each node with a deliberately narrow context scope. It is where the two graphs meet: the workflow decides which services to touch, the knowledge graph says which ones they are.

Everything else is an elaboration of those three. That is the logic the eight chapters walk through, and the logic this README is organized around.

The logic: how the eight chapters compose

Read each chapter as “this deepens that part of the Chapter-2 foundation.”