Graph Engineering — скилл

---
name: graph-engineering
description: Teaches an agent graph engineering — both halves. Knowledge graphs (ontology design, entity/relation/event extraction, fusion, GraphRAG/memory serving; distilled and translated from Southeast University's graduate Knowledge Graph course, npubird/KnowledgeGraphCourse, 4.4K stars) and task graphs (agent orchestration — parallel fan-out, verifier separation, the stop rule, human gates). Use when asked to build a knowledge graph, extract entities/relations from text, design an ontology, dedupe/merge entities, add graph memory or GraphRAG to an agent, orchestrate multi-agent workflows as a graph, or LEARN graph engineering — in teaching mode the agent explains each stage with worked examples and generates visual diagram artifacts.
---

# Graph Engineering

Graph engineering is the discipline of designing the structures agents work through — not the
prompts. It has two halves:

1. **Knowledge graphs** — what agents remember. Nodes are entities and facts, edges are
   relationships with time and provenance. This file's 9-stage pipeline covers it, distilled
   from Southeast University's graduate KG course
   (https://github.com/npubird/KnowledgeGraphCourse, Prof. Peng Wang), translated to English
   and adapted for LLM-era agents.
2. **Task graphs** — how agents work. Nodes are jobs, edges are execution dependencies:
   parallel fan-out, separate verifier contexts, the stop rule, the human gate.
   Read [references/task-graphs.md](references/task-graphs.md) when the request is about
   orchestrating agents rather than building memory.

Core mental model: a knowledge graph is a **product with a schema**, not a pile of triples.
Quality comes from the pipeline order — model the domain BEFORE extracting, fuse BEFORE storing,
evaluate at every stage.

## Teaching Mode

When the user wants to LEARN graph engineering (rather than build something), teach it — do
not just execute. Rules:

1. Anchor every stage in the user's own domain: ask for one real project or dataset, then use
   it as the running example through all stages.
2. **Generate visual artifacts as you teach.** Concepts in this discipline are shapes; show
   them. For each major concept, produce a small diagram the user can keep — mermaid diagrams
   (flowchart for the pipeline and task graphs, `graph LR` for example ontologies and
   subgraphs) or a single self-contained HTML page when interactivity helps. At minimum:
   the 9-stage pipeline, a 3-type ontology drawn from the user's domain, one extracted
   subgraph (5-10 nodes) from a real sample, and the diamond pattern with the user's own jobs
   as nodes.
3. Teach in the pipeline's order, one stage per exchange, each ending with a small exercise
   ("write 3 competency questions for your project") before moving on.