Graphify

Install in one command

mkdir -p ~/.claude/skills/graphify && curl -fsSL https://eliteaiadvantage.com/skills/graphify/SKILL.md -o ~/.claude/skills/graphify/SKILL.md

Graphify takes any folder, code, docs, papers, tweets, screenshots, transcripts, and builds a persistent knowledge graph with community detection. It outputs three things: an interactive HTML visualization, GraphRAG-ready JSON, and a plain-language GRAPH_REPORT.md. Edges are tagged EXTRACTED, INFERRED, or AMBIGUOUS so you know what was found versus invented. Once built, the graph survives across sessions, query it weeks later, find paths between concepts, ask why two unrelated documents are connected.

It's built around the /raw folder workflow: drop everything in, let it grow, query it. Use it on a codebase you're new to (architecture before touching anything), a reading list, a research corpus, or your own dump folder. The cross-document surprise is the killer feature: community detection finds connections between concepts in different files you would never think to ask about directly.

• →Builds a persistent knowledge graph that survives across Claude sessions, no re-reading the corpus.
• →Tags every edge as EXTRACTED, INFERRED, or AMBIGUOUS so the audit trail is honest.
• →Surfaces cross-document connections via community detection that direct search would miss.
• →Exports to interactive HTML, Neo4j, GraphML, SVG, and an Obsidian vault, your tooling already supports it.
• →Supports incremental updates, video/audio transcription, and an MCP stdio server for agent access.
• →Gives you the structure of a corpus, not just a search interface over it.

• •Onboarding to a codebase you've never touched, see the architecture before editing anything.
• •A reading list of papers, tweets, and notes that needs a connection map.
• •A research corpus where citation graph + concept graph in one is the goal.
• •A personal /raw folder you've been dumping into for months.
• •Asking 'how is X connected to Y' across hundreds of files.
• •Generating a wiki-style index from a folder for downstream agent crawling.

Without this skill, Claude reads a folder file-by-file every session and forgets the structure when the context closes. Graphify reshapes the workflow into a persistent extract-and-cluster pass that produces durable artifacts (graph.json, HTML viz, plain-language report). Subsequent queries hit the graph instead of the raw files, fast, cheap, and structurally honest about what's a real edge versus an inference.