[{"data":1,"prerenderedAt":4},["ShallowReactive",2],{"readme:graph-engineering":3},"\u003Ch1>Graph Engineering\u003C\u002Fh1>\n\u003Cp>\u003Cstrong>The discipline of designing the structures AI agents work through — not the prompts.\u003C\u002Fstrong>\u003C\u002Fp>\n\u003Cp>It has two halves:\u003C\u002Fp>\n\u003Col>\n\u003Cli>\u003Cstrong>Knowledge graphs\u003C\u002Fstrong> — what agents \u003Cem>remember\u003C\u002Fem>. Nodes are entities and facts, edges are\nrelationships with time and provenance. Ontology → extraction → fusion → serving.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Task graphs\u003C\u002Fstrong> — how agents \u003Cem>work\u003C\u002Fem>. Nodes are jobs, edges are execution dependencies.\nParallel fan-out, separate verifiers, the stop rule, the human gate.\u003C\u002Fli>\n\u003C\u002Fol>\n\u003Cp>Prompt engineers steered the model's words. Loop engineers steered its iterations.\nGraph engineers steer its \u003Cstrong>topology\u003C\u002Fstrong>.\u003C\u002Fp>\n\u003Cp>This repo turns Southeast University's graduate Knowledge Graph course\n(\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fnpubird\u002FKnowledgeGraphCourse\" rel=\"nofollow ugc noopener\">npubird\u002FKnowledgeGraphCourse\u003C\u002Fa>, 4.4K★,\ntaught in Chinese since 2019) — plus the modern agent-orchestration research behind\ntask graphs — into things you can actually use today.\u003C\u002Fp>\n\u003Ch2>What's inside\u003C\u002Fh2>\n\u003Ctable>\n\u003Cthead>\n\u003Ctr>\n\u003Cth>Path\u003C\u002Fth>\n\u003Cth>What it is\u003C\u002Fth>\n\u003C\u002Ftr>\n\u003C\u002Fthead>\n\u003Ctbody>\u003Ctr>\n\u003Ctd>\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fcodejunkie99\u002Fgraph-engineering\u002Fblob\u002FHEAD\u002Fgraph-engineering\u002F\" rel=\"nofollow ugc noopener\">\u003Ccode>graph-engineering\u002F\u003C\u002Fcode>\u003C\u002Fa>\u003C\u002Ftd>\n\u003Ctd>\u003Cstrong>The skill.\u003C\u002Fstrong> Hand it to your agent (Claude Code \u002F 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 \u003Cem>your\u003C\u002Fem> domain.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fcodejunkie99\u002Fgraph-engineering\u002Fblob\u002FHEAD\u002Fgraph-engineering\u002Freferences\u002F\" rel=\"nofollow ugc noopener\">\u003Ccode>graph-engineering\u002Freferences\u002F\u003C\u002Fcode>\u003C\u002Fa>\u003C\u002Ftd>\n\u003Ctd>The distilled course: \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fcodejunkie99\u002Fgraph-engineering\u002Fblob\u002FHEAD\u002Fgraph-engineering\u002Freferences\u002Fcurriculum.md\" rel=\"nofollow ugc noopener\">curriculum map\u003C\u002Fa> (translated, with links to the original Chinese decks), \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fcodejunkie99\u002Fgraph-engineering\u002Fblob\u002FHEAD\u002Fgraph-engineering\u002Freferences\u002Fmodeling.md\" rel=\"nofollow ugc noopener\">modeling\u003C\u002Fa>, \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fcodejunkie99\u002Fgraph-engineering\u002Fblob\u002FHEAD\u002Fgraph-engineering\u002Freferences\u002Fextraction.md\" rel=\"nofollow ugc noopener\">extraction\u003C\u002Fa>, \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fcodejunkie99\u002Fgraph-engineering\u002Fblob\u002FHEAD\u002Fgraph-engineering\u002Freferences\u002Ffusion-and-llm.md\" rel=\"nofollow ugc noopener\">fusion + GraphRAG\u003C\u002Fa>, \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fcodejunkie99\u002Fgraph-engineering\u002Fblob\u002FHEAD\u002Fgraph-engineering\u002Freferences\u002Ftask-graphs.md\" rel=\"nofollow ugc noopener\">task graphs\u003C\u002Fa>.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fcodejunkie99\u002Fgraph-engineering\u002Fblob\u002FHEAD\u002FWORKFLOWS.md\" rel=\"nofollow ugc noopener\">\u003Ccode>WORKFLOWS.md\u003C\u002Fcode>\u003C\u002Fa>\u003C\u002Ftd>\n\u003Ctd>Nine paste-ready prompt blocks — a \u003Ccode>\u002Fkg-tutor\u003C\u002Fcode> that teaches you the whole course interactively, plus eight single-purpose tools (\u003Ccode>\u002Fkg-scope\u003C\u002Fcode> → \u003Ccode>\u002Fkg-rag\u003C\u002Fcode>) that chain into a full build.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fcodejunkie99\u002Fgraph-engineering\u002Fblob\u002FHEAD\u002Fdist\u002F\" rel=\"nofollow ugc noopener\">\u003Ccode>dist\u002Fgraph-engineering.skill\u003C\u002Fcode>\u003C\u002Fa>\u003C\u002Ftd>\n\u003Ctd>The packaged skill file.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003C\u002Ftbody>\u003C\u002Ftable>\n\u003Ch2>Install (two commands)\u003C\u002Fh2>\n\u003Cpre>\u003Ccode class=\"language-bash\">git clone https:\u002F\u002Fgithub.com\u002Fcodejunkie99\u002Fgraph-engineering.git\ncp -r graph-engineering\u002Fgraph-engineering ~\u002F.claude\u002Fskills\u002F\n\u003C\u002Fcode>\u003C\u002Fpre>\n\u003Cp>Then ask your agent to \u003Cem>build\u003C\u002Fem> (\"build a knowledge graph from my docs\") or to \u003Cem>teach\u003C\u002Fem>\n(\"teach me graph engineering\") — teaching mode walks the pipeline stage by stage with\nworked examples and generated diagrams, using your own project as the running example.\u003C\u002Fp>\n\u003Ch2>The 9-stage pipeline\u003C\u002Fh2>\n\u003Cpre>\u003Ccode class=\"language-mermaid\">flowchart LR\n  A[1 scope] --> B[2 representation] --> C[3 ontology] --> D[4 entities]\n  D --> E[5 relations] --> F[6 events] --> G[7 quality gate]\n  G --> H[8 fusion] --> I[9 serve to LLMs]\n\u003C\u002Fcode>\u003C\u002Fpre>\n\u003Cp>Model the domain \u003Cstrong>before\u003C\u002Fstrong> extracting. Fuse \u003Cstrong>before\u003C\u002Fstrong> storing. Verify at every stage.\nA knowledge graph is a product with a schema, not a pile of triples.\u003C\u002Fp>\n\u003Ch2>The task-graph rules (the other half)\u003C\u002Fh2>\n\u003Cul>\n\u003Cli>Delete \u003Cstrong>fake edges\u003C\u002Fstrong>: an arrow is real only when work flows through it.\u003C\u002Fli>\n\u003Cli>The \u003Cstrong>diamond\u003C\u002Fstrong>: split → parallel workers → \u003Cem>separate\u003C\u002Fem> verifier contexts → one owned merge.\u003C\u002Fli>\n\u003Cli>The \u003Cstrong>stop rule\u003C\u002Fstrong> (Google DeepMind × MIT, 180 configurations): teams win ~80% on work that\nsplits; every team configuration loses on sequential work. The shape of the work decides.\u003C\u002Fli>\n\u003Cli>The \u003Cstrong>human gate\u003C\u002Fstrong>: your approval sits exactly where a mistake is expensive to undo.\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>Details: \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fcodejunkie99\u002Fgraph-engineering\u002Fblob\u002FHEAD\u002Fgraph-engineering\u002Freferences\u002Ftask-graphs.md\" rel=\"nofollow ugc noopener\">references\u002Ftask-graphs.md\u003C\u002Fa>\u003C\u002Fp>\n\u003Ch2>Credits\u003C\u002Fh2>\n\u003Cp>The knowledge-graph half is an independent English distillation of 东南大学《知识图谱》研究生课程\n(Southeast University's graduate Knowledge Graph course), Prof. Peng Wang —\n\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fnpubird\u002FKnowledgeGraphCourse\" rel=\"nofollow ugc noopener\">npubird\u002FKnowledgeGraphCourse\u003C\u002Fa>. All original\nlecture PDFs are in Chinese and remain in the original repo; none are redistributed here.\nTask-graph material draws on Google DeepMind × MIT's\n\u003Ca href=\"https:\u002F\u002Fresearch.google\u002Fblog\u002Ftowards-a-science-of-scaling-agent-systems-when-and-why-agent-systems-work\u002F\" rel=\"nofollow ugc noopener\">\"Towards a Science of Scaling Agent Systems\"\u003C\u002Fa>\nand Anthropic's published multi-agent engineering work.\u003C\u002Fp>\n\u003Cp>MIT licensed. Built\u003C\u002Fp>\n",1785169769198]