graphify

name	graphify
effort	low
description	Navigate the codebase using a pre-built knowledge graph instead of reading source files. Use before exploring unfamiliar code, tracing a bug, or understanding how two components connect. Reduces token cost by 10–466x vs naive file reading.
triggers	["graphify","knowledge graph","code graph","how does X work","find path between","explain node","codebase navigation","token efficient","graph query"]

Graphify - Code Knowledge Graph

Query a pre-built AST-based knowledge graph instead of reading raw source files. 22x average token reduction. No LLM tokens consumed to build or update the graph.

---

Setup (one-time, per project)

1. Install

pip install graphifyy
graphify install   # registers the /graphify skill in Claude Code

2. Build the graph

graphify update <project-root>
# e.g. graphify update /Users/yourname/cortextos

Outputs to <project-root>/graphify-out/:

• graph.json - queryable node/edge data
• GRAPH_REPORT.md - community hubs and unexpected connections
• cache/ - SHA256-based change detection for incremental updates

3. Install git hooks (auto-update on commit)

cd <project-root>
graphify hook install

After this, graph.json rebuilds automatically on every git commit and git checkout.

4. (Optional) Wire to Obsidian vault

mkdir -p <vault>/graphify/<project-name>
cp <project-root>/graphify-out/GRAPH_REPORT.md <vault>/graphify/<project-name>/
cp <project-root>/graphify-out/graph.json <vault>/graphify/<project-name>/

---

Usage

Query (replaces reading source files)

graphify query "how does <topic> work" \
  --graph <project-root>/graphify-out/graph.json \
  --budget 2000

Explain a node

graphify explain "<ClassName or FunctionName>" \
  --graph <project-root>/graphify-out/graph.json

Find the path between two concepts

graphify path "ComponentA" "ComponentB" \
  --graph <project-root>/graphify-out/graph.json

Benchmark token reduction

graphify benchmark <project-root>/graphify-out/graph.json

---

When to use

• Before exploring an unfamiliar subsystem
• Tracing how data flows from A to B
• Understanding what calls what before editing
• Any time you would otherwise read 3+ source files to answer one question

When not to use

• When you already know the exact file and line to edit
• For very small codebases (<20 files) where reading is faster
• When you need the actual source code to make an edit (use Read for that)

---

Deployment note: replace <project-root> with your actual project path. The graph persists across sessions - no rebuild needed unless the codebase changes significantly.