StackMap
Subscribe
Explore / graph-engineering
codejunkie99

graph-engineering

Graph engineering as a Claude skill: SEU's 9-stage knowledge-graph course (translated) plus task-graph orchestration patterns — with a teaching mode and nine paste-ready /kg workflows.

144 19 MITupdated 4 days ago
Curator's take

A discipline-in-a-repo: the thesis that agents are steered by topology — knowledge graphs for what they remember, task graphs for how they work — backed by an actual graduate curriculum (SEU's 4.4k-star course, translated from Chinese) rather than vibes. The teaching mode is the standout: /kg-tutor walks the whole pipeline drawing diagrams from YOUR domain, and eight chained workflows go from scoping to GraphRAG. NOT proven at scale: 144 stars and young — adopt it as excellent structured knowledge, not battle-tested tooling; the graphs you build with it still need the engineering the course itself teaches (fusion and provenance are where naive KG projects drown).

Mapped by ShipWithAI editors · links verified
README.md

Graph Engineering

The discipline of designing the structures AI 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. Ontology → extraction → fusion → serving.
  2. Task graphs — how agents work. Nodes are jobs, edges are execution dependencies. Parallel fan-out, separate verifiers, the stop rule, the human gate.

Prompt engineers steered the model's words. Loop engineers steered its iterations. Graph engineers steer its topology.

This repo turns Southeast University's graduate Knowledge Graph course (npubird/KnowledgeGraphCourse, 4.4K★, taught in Chinese since 2019) — plus the modern agent-orchestration research behind task graphs — into things you can actually use today.

What's inside

Path What it is
graph-engineering/ The skill. Hand it to your agent (Claude Code / any skill-compatible harness) — it learns the full 9-stage knowledge-graph pipeline, the task-graph patterns, and a teaching mode that explains every stage with diagrams drawn from your domain.
graph-engineering/references/ The distilled course: curriculum map (translated, with links to the original Chinese decks), modeling, extraction, fusion + GraphRAG, task graphs.
WORKFLOWS.md Nine paste-ready prompt blocks — a /kg-tutor that teaches you the whole course interactively, plus eight single-purpose tools (/kg-scope/kg-rag) that chain into a full build.
dist/graph-engineering.skill The packaged skill file.

Install (two commands)

git clone https://github.com/codejunkie99/graph-engineering.git
cp -r graph-engineering/graph-engineering ~/.claude/skills/

Then ask your agent to build ("build a knowledge graph from my docs") or to teach ("teach me graph engineering") — teaching mode walks the pipeline stage by stage with worked examples and generated diagrams, using your own project as the running example.

The 9-stage pipeline

flowchart LR
  A[1 scope] --> B[2 representation] --> C[3 ontology] --> D[4 entities]
  D --> E[5 relations] --> F[6 events] --> G[7 quality gate]
  G --> H[8 fusion] --> I[9 serve to LLMs]

Model the domain before extracting. Fuse before storing. Verify at every stage. A knowledge graph is a product with a schema, not a pile of triples.

The task-graph rules (the other half)

  • Delete fake edges: an arrow is real only when work flows through it.
  • The diamond: split → parallel workers → separate verifier contexts → one owned merge.
  • The stop rule (Google DeepMind × MIT, 180 configurations): teams win ~80% on work that splits; every team configuration loses on sequential work. The shape of the work decides.
  • The human gate: your approval sits exactly where a mistake is expensive to undo.

Details: references/task-graphs.md

Credits

The knowledge-graph half is an independent English distillation of 东南大学《知识图谱》研究生课程 (Southeast University's graduate Knowledge Graph course), Prof. Peng Wang — npubird/KnowledgeGraphCourse. All original lecture PDFs are in Chinese and remain in the original repo; none are redistributed here. Task-graph material draws on Google DeepMind × MIT's "Towards a Science of Scaling Agent Systems" and Anthropic's published multi-agent engineering work.

MIT licensed. Built

Continue your stack

What teams reach for next — and why each earns a place beside graph-engineering. Ranked by curator confidence.