Universal Knowledge Graph Builder

Problem

Technical documents, product notes, research papers, and code explanations often live across Markdown, PDF, and code files. Maintaining a manual index, summary, or relationship map can lag behind the content and makes it harder to answer questions such as where a concept appears. knowledge-graph-builder targets a local document directory and converts parseable content into a queryable, interactive knowledge graph rather than a static file listing.

Workflow

The skill checks the Python environment and LLM API configuration, then scans the document directory and estimates the token budget. Only after the user confirms the budget and parameters does it call the model to extract nodes and edges. The result is written to graph.json, then rendered into a single-file HTML page using an inline D3.js template that can be opened offline. When new documents are added later, it detects added, changed, and removed files, then performs an incremental update or full rebuild while reusing the existing configuration.

Boundaries

It depends on an external LLM API, so build time and cost scale with corpus size and model-call volume. 401, 429, proxy, or network errors require explicit configuration, concurrency, or network adjustments. It fits local knowledge organization and shared read-only exploration, not real-time synchronization, fine-grained permissions, or high-frequency collaborative editing.