知脉 KnowPulse｜AI 智能知识图谱引擎

KnowPulse (知脉) by DAVYLAB validates subject-isolated knowledge graphs, traces prerequisite gaps, supports conservative learning diagnostics, and creates prerequisite-first learning paths for AI courses and education workflows.

DAVYLAB Knowledge Graph Skill

Trigger

Use for DAVYLAB Academy course dependency design, knowledge radar, diagnostic planning, prerequisite discovery, study paths, or curriculum graph validation. Do not use for medical, psychological or high-stakes assessment.

Procedure

1. Collect topics and dependencies from an owned or appropriately licensed source; request content provenance and educator-review status when used for instruction.
2. Validate input with node src/cli.js validate path/to/dataset.json. Reject validation errors; never silently drop cyclic or cross-subject edges.
3. For prerequisite questions call ancestors; for lesson recommendations call path (hard edges by default). Distinguish required (hard) from suggested (soft).
4. For answer-based screening, use createSession, applyAnswer, nextTopic, and summarize in src/engine.js; answers are provided as correct/wrong/skip by a separate verified evaluator. This skill does not evaluate free-form answers or generate trusted exam items.
5. Label every diagnostic result screening only — educator review required. Never infer prerequisite mastery solely from descendant performance.
6. Return structured JSON; the consumer app owns display, storage, login, access control, and data retention.

Commands

bash

node src/cli.js validate examples/ai-foundations.json
node src/cli.js ancestors examples/ai-foundations.json ai.agent
node src/cli.js path examples/ai-foundations.json ai.agent
node src/cli.js demo examples/ai-foundations.json ai.agent
node tests/engine.test.js

Data contract

• topics: array of {id:string,subject:string,title:string,level?:string,reviewStatus?:string}
• dependencies: array of {topicId:string,prerequisiteId:string,strength:"hard"|"soft",reason?:string}
• IDs must be globally unique; dependencies must refer to existing nodes and stay within a subject.
• buildGraph(topics, dependencies) returns indexes and topological ordering.
• ancestors(graph,id,{includeSoft}), learningPath(graph,targetIds,{includeSoft}) return ordered IDs.
• createSession(graph,targetIds,{confirm}), applyAnswer(graph,session,id,result), nextTopic(graph,session), summarize(graph,session) maintain an auditable immutable session.

Guardrails

• No original upstream question banks, curriculum mappings, texts or engine code included.
• Do not turn reported weak status into a clinical diagnosis or definitive pedagogical judgment.
• Never auto-publish recommendations without reviewed content and product-level QA.
• See docs/ACCEPTANCE.md and docs/P0-ANALYSIS.md.