CodeScope
A Claude Code plugin that gives AI deep, persistent understanding of your codebase.
Analyzes your project's conventions, dependencies, and risk zones so that every AI-generated change fits naturally into your existing code.
npx codescope init
"Every AI session starts from zero. CodeScope fixes that."
The Problem · How It Works · Install · Commands · MCP Tools
Install
As a Claude Code Plugin
/plugin marketplace add jwadhwa2259/codescope
/plugin install codescope@codescope
As a CLI
npx codescope init
The Problem
When you ask an AI to modify code in an existing project, it starts from zero every time. It doesn't know your naming conventions, your module boundaries, which files are high-risk, or how your imports are structured. The result: code that works in isolation but clashes with everything around it.
CodeScope fixes this by analyzing your codebase once, building a knowledge graph of how everything connects, and then injecting that context into every Claude Code interaction automatically.
How It Works
CodeScope runs as a Claude Code plugin with an MCP server. Once bootstrapped, it works invisibly through Claude Code's hook system:
- Before every file edit: Injects the file's conventions, danger zone status, and blast radius into Claude's context
- After every file edit: Validates the change against detected conventions and flags violations
- Before context compaction: Saves session state so nothing is lost when Claude's context window fills up
- On session resume: Restores context from the previous session
You don't need to do anything for this to work. It's invisible.
What You Get
| Feature | Description |
|---|
| Knowledge Graph | Full dependency map with community detection and centrality analysis |
| Convention Detection | Discovers coding patterns with adoption rates and confidence scores |
| Danger Zone Mapping | Identifies high-risk files by centrality and coupling |
| AI Readiness Score | Grades your codebase on convention coverage, type safety, test coverage, and import health |
| Blast Radius Analysis | BFS traversal shows everything affected by a change |
| Interactive Dashboard | Explore the graph, conventions, readiness trends, and blast radius in your browser |
| Autonomous Pipeline | Research, plan, execute, verify, evaluate, and self-correct -- with only two approval gates |
| Session Continuity | Pause and resume analysis workflows across Claude Code sessions |
| Pre-commit Enforcement | Optional convention checking on staged files |
Getting Started
Requirements
- Node.js 22 or later -- check with
node --version
- Claude Code -- the CLI, desktop app, or IDE extension
Step 1: Onboard your project
Open Claude Code in your project directory and run:
/codescope:onboard
This walks you through a short wizard:
- Detects your project -- languages, build commands, test commands, monorepo vs single repo
- Configures agent models -- which Claude models to use for different analysis tasks (defaults work fine)
- Sets workflow preferences -- how verbose you want the output, how strict convention enforcement should be
Step 2: Bootstrap (analyze your codebase)
/codescope:bootstrap
CodeScope scans your project, parses every source file into an AST, builds a knowledge graph, detects conventions, identifies danger zones, and computes an AI Readiness Score. For a typical project, this takes 1-5 minutes.
When it completes, you'll see a readiness report:
AI Readiness: B+ (82%)
| Dimension | Score |
|----------------------|-------|
| Convention Coverage | 87% |
| Type Safety | 91% |
| Test Coverage Proxy | 68% |
| Import Graph Health | 83% |
Step 3: You're done
That's it. CodeScope is now active. Every time Claude Code edits or writes a file, CodeScope automatically injects relevant context about that file's conventions, danger level, and blast radius.
Commands
The Main Workflow
/codescope:orient add a caching layer to the user service
This runs the full autonomous pipeline: