🕸️ Code Graph for Claude Code

Give your AI developer a map of your codebase before it touches anything.

Code Graph analyzes your project, builds a dependency graph, and lets Claude Code query it before every edit — so it always knows what depends on what.

Installation · How It Works · Commands · Screenshots · Languages

The Problem

AI coding agents are blind. They edit files without understanding the full architecture. Change a model? 15 views break. Refactor a service? The cache layer falls apart.

Code Graph fixes this. It builds a complete dependency map of your codebase and gives Claude Code targeted queries — so before it edits user_service.py, it already knows that 5 files depend on it and the change will cascade through 3 levels.

You: "Refactor the user service to use async"

Claude Code (without Code Graph):
  → Edits user_service.py ✅
  → Forgets to update user_views.py ❌
  → Breaks sync_tasks.py ❌
  → Cache layer still calls sync methods ❌

Claude Code (with Code Graph):
  → Queries: impact user_service.py
  → Sees: 5 files affected across 3 levels
  → Updates ALL dependent files ✅
  → Verifies cache layer compatibility ✅

⚡ Installation

Pick one. Copy → paste into terminal → done.

🔹 Option A: This project only

git clone https://github.com/EmperorKunDis/code-graph.git /tmp/code-graph && bash /tmp/code-graph/install.sh . && rm -rf /tmp/code-graph && echo "✅ Done — open Claude Code and run: /init-graph"

🔹 Option B: All projects, forever

git clone https://github.com/EmperorKunDis/code-graph.git /tmp/code-graph && bash /tmp/code-graph/global-setup.sh && rm -rf /tmp/code-graph && exec $SHELL

Then for any new project:

cd ~/my-project && code-graph-install .

Both options: after install, open Claude Code and type /init-graph to generate the graph.

What gets installed

your-project/
├── CLAUDE.md                        ← adds Code Graph instructions
└── .claude/
    ├── settings.json                ← auto-query hook before edits
    ├── commands/
    │   ├── init-graph.md            ← /init-graph command
    │   └── graph.md                 ← /graph <query> command
    └── skills/code-graph/
        ├── SKILL.md                 ← skill instructions
        └── scripts/
            ├── analyze_codebase.py  ← graph generator
            ├── query_graph.py       ← targeted queries (the key piece)
            └── generate_viewer.py   ← HTML visualization

Manual installation

If you prefer to install manually:

# Clone
git clone https://github.com/EmperorKunDis/code-graph.git

# Copy skill + commands into your project
mkdir -p /path/to/project/.claude/{commands,skills}
cp -r code-graph/claude-config/skills/code-graph /path/to/project/.claude/skills/
cp code-graph/claude-config/commands/*.md /path/to/project/.claude/commands/

# Add the CLAUDE.md section
cat code-graph/CLAUDE_MD_SECTION.md >> /path/to/project/CLAUDE.md

🔍 How It Works

1. Analyze → Build the graph

/init-graph

Scans every source file, parses AST (Python) or patterns (JS/TS/others), and detects:

Imports between files
Database reads/writes (ORM queries, raw SQL)
Cache operations (Redis, Django cache)
API calls (requests, fetch, axios)
Class inheritance chains
Route → handler mappings
Webhooks, events, signals
Task/job definitions (Celery, etc.)

Output: .code_graph.json — a complete graph of your codebase.

2. Query → Ask specific questions

Instead of reading the entire (huge) JSON, Claude Code runs targeted queries:

# Before editing a file — what's the risk?
python3 .claude/skills/code-graph/scripts/query_graph.py file views/users.py

# What breaks if I change this?
python3 .claude/skills/code-graph/scripts/query_graph.py impact models/user.py

# Who reads/writes this model?
python3 .claude/skills/code-graph/scripts/query_graph.py model User

Returns only the relevant nodes and edges — a few lines instead of thousands.

3. Protect → Auto-check before edits

The installed hook in settings.json makes Claude Code automatically query the graph before every file edit. No extra prompting needed.

📋 Commands

Slash Commands

Command	What it does
`/init-graph`	Generate/regenerate the dependency graph
`/graph overview`	Compact architecture overview
`/graph file <path>`	Check a file's risk level and connections
`/graph impact <path>`	Full cascade analysis — what breaks if changed
`/graph deps <path>`	What a file depends on
`/graph dependents <path>`	What depends on a file
`/graph model <Name>`	All readers/writers of a database model
`/graph hubs`	Most connected (riskiest) nodes
`/graph search <query>`	Find nodes by name or path
`/graph path <from> <to>`	Shortest connection between two files
`/graph risky-files`	Files ranked by change risk
`/graph dead-code`	Find potentially unused code
`/graph endpoint <path>`	Full request chain for an endpoint
`/graph cluster <path>`	Show connected component
`/graph stats`	Project statistics
`/graph report`	Full report — overview + risks + dead code + gaps in one call
`/graph changes <f1> <f2>`	Pre-change check for multiple files at once

Example: Impact Analysis

> /graph impact core/services/user_service.py

💥 Impact Analysis: user_service.py
   Risk: 🟡 MEDIUM — 8 connections, 3 dependents

  → Direct dependents (3 files):
    [test] test_users.py — tests/test_users.py
    [task] sync_tasks.py — core/tasks/sync_tasks.py
    [endpoint] user_views.py — api/views/user_views.py

  →→ 2nd-level impact (1 files):
    [endpoint] __init__.py — api/views/__init__.py

  →→→ 3rd-level impact (1 files):
    [router] urls.py — api/urls.py

  📊 Total affected: 5 files across 3 levels

Example: Model Usage

> /graph model User

🗄️  Model: User (core/models/user.py)
   Risk: 🔴 HIGH — 14 connections, 8 dependents

  📖 Read by (5):
    user_service.py [service]
    product_service.py [service]
    user_views.py [endpoint]
    ...

  ✏️  Written by (3):
    user_service.py [service]
    registration_view.py [endpoint]
    ...

Risk Levels

Level	Connections	Meaning
🟢 LOW	≤3	Safe to change freely
🟡 MEDIUM	4-10	Check dependents first
🔴 HIGH	10+	Run full impact analysis
⛔ CRITICAL	20+	Warn user, suggest incremental approach

📸 Screenshots

Interactive Graph Viewer

Generate a visual graph you can explore in your browser:

python3 .claude/skills/code-graph/scripts/generate_viewer.py .code_graph.json -o graph.html
open graph.html

Features:

Force-directed layout with physics simulation
Filter by node type (endpoints, models, services...)
Filter by edge type (imports, db_read, api_call...)
Search nodes by name
Double-click for detail panel with all connections
Minimap for navigation
Hover highlighting of connected nodes

🌐 Supported Languages

Language	Analysis Method	Detected Patterns
Python	AST parsing + patterns	Imports, classes, decorators, ORM, signals, Celery tasks
TypeScript	Pattern matching	ES6 imports, routes, Prisma/Mongoose, fetch/axios
JavaScript	Pattern matching	require/import, Express routes, DB operations
PHP	Pattern matching	use/require, namespaces
Ruby	Pattern matching	require, include
Go	Pattern matching	import statements
Java	Pattern matching	import statements
Rust	Pattern matching	use/mod statements
C#	Pattern matching	using statements
Vue/Svelte	Pattern matching	Component imports, script analysis

Node Types

Type	Color	Examples
`endpoint`	🔵 Cyan	API views, route handlers, controllers
`collection`	🔴 Red	Database models, schemas
`file`	🟢 Green	Source files, modules
`component`	🟢 Teal	Frontend components (React, Angular, Vue)
`service`	🟢 Dark Teal	Business logic, service layer
`task`	🟠 Orange	Background tasks, Celery, cron
`router`	🔵 Blue	URL configs, route definitions
`serializer`	🟠 Amber	Serializers, DTOs, data transformers
`middleware`	🔵 Indigo	Middleware components
`webhook`	🔴 Coral	Webhook handlers
`event`	🟣 Pink	Signals, event handlers
`external_api`	🟡 Yellow	External API integrations
`cache_key`	🟣 Magenta	Cache operations
`utility`	⚪ Gray	Helpers, utils
`test`	⚪ Dark Gray	Test files
`config`	🟤 Brown	Configuration files

⚙️ Configuration

Customize analyzer options

# Limit to specific languages
python3 .claude/skills/code-graph/scripts/analyze_codebase.py . -o .code_graph.json \
  --languages python,typescript

# Exclude directories
python3 .claude/skills/code-graph/scripts/analyze_codebase.py . -o .code_graph.json \
  --exclude migrations,fixtures,static,seeds

# Limit scan depth
python3 .claude/skills/code-graph/scripts/analyze_codebase.py . -o .code_graph.json \
  --max-depth 6

Auto-update with git hook

The installer doesn't set this up automatically, but you can add it:

cat > .git/hooks/post-commit << 'EOF'
#!/bin/bash
python3 .claude/skills/code-graph/scripts/analyze_codebase.py . -o .code_graph.json \
  --languages python,typescript \
  --exclude migrations,node_modules,static,media,dist,.angular,__pycache__
EOF
chmod +x .git/hooks/post-commit

Gitignore recommendation

# Generated viewer (large HTML file)
code_graph_viewer.html

# Keep .code_graph.json tracked — Claude Code needs it
# Do NOT gitignore .code_graph.json

🤝 Contributing

Contributions welcome! Some ideas:

More language analyzers — add full AST parsing for TypeScript, Go, etc.
Framework-specific detection — Next.js pages, FastAPI routes, Rails conventions
Graph diff — show what changed between two graph versions
VS Code extension — visualize the graph inline in your editor
MCP server — expose graph queries as an MCP tool

📄 License

MIT — do whatever you want with it.

Built by Praut s.r.o. — AI integration & business automation

If Code Graph saves you from a cascading bug, star the repo ⭐

Name		Name	Last commit message	Last commit date
Latest commit History 11 Commits
claude-config		claude-config
.gitignore		.gitignore
CHANGELOG.md		CHANGELOG.md
CLAUDE_MD_SECTION.md		CLAUDE_MD_SECTION.md
CONTRIBUTING.md		CONTRIBUTING.md
LICENSE		LICENSE
README.md		README.md
global-setup.sh		global-setup.sh
install.sh		install.sh
uninstall.sh		uninstall.sh

Navigation Menu

Repository files navigation

🕸️ Code Graph for Claude Code

The Problem

⚡ Installation

🔹 Option A: This project only

🔹 Option B: All projects, forever

What gets installed

Manual installation

🔍 How It Works

1. Analyze → Build the graph

2. Query → Ask specific questions

3. Protect → Auto-check before edits

📋 Commands

Slash Commands

Example: Impact Analysis

Example: Model Usage

Risk Levels

📸 Screenshots

Interactive Graph Viewer

🌐 Supported Languages

Node Types

⚙️ Configuration

Customize analyzer options

Auto-update with git hook

Gitignore recommendation

🤝 Contributing

📄 License

About

Releases

Packages

Contributors

Languages

Folders and files

Latest commit

History

Repository files navigation

🕸️ Code Graph for Claude Code

The Problem

⚡ Installation

🔹 Option A: This project only

🔹 Option B: All projects, forever

What gets installed

Manual installation

🔍 How It Works

1. Analyze → Build the graph

2. Query → Ask specific questions

3. Protect → Auto-check before edits

📋 Commands

Slash Commands

Example: Impact Analysis

Example: Model Usage

Risk Levels

📸 Screenshots

Interactive Graph Viewer

🌐 Supported Languages

Node Types

⚙️ Configuration

Customize analyzer options

Auto-update with git hook

Gitignore recommendation

🤝 Contributing

📄 License

About

Resources

Contributing

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages