Codebase Analysis

Full-codebase static analysis. Powers the architecture-expert subagent and the get_architecture_knowledge tool. Ask about your project’s structure and get answers grounded in real analysis instead of the LLM guessing.

Quick start

CodeBuddy: Analyze Codebase & Answer Questions   (Cmd/Ctrl + Shift + 6)

Progress notification during the scan. Once done, ask things like:

• “What architectural patterns does this project use?”
• “Show me all API endpoints”
• “What are the most-imported modules?”
• “Are there any circular dependencies?”

Languages

Tree-sitter analysis for 7 languages:

Language	Extensions
JavaScript	.js, .jsx, .mjs, .cjs
TypeScript	.ts, .tsx, .mts, .cts
Python	.py
Java	.java
Go	.go
Rust	.rs
PHP	.php, .phtml

Plus config files (package.json, tsconfig.json, requirements.txt, pyproject.toml, go.mod, Cargo.toml, pom.xml, build.gradle, composer.json) and schema files (*.sql, schema.prisma) for dependency + model info.

What gets extracted

For each supported language, Tree-sitter pulls out classes / interfaces / structs / traits / enums (with methods, properties, decorators), functions (parameters, return types, export/async flags), imports + exports, React components (in JSX/TSX), and API endpoints via framework detection.

Framework endpoint detection:

Language	Frameworks
JavaScript / TypeScript	Express, NestJS, Fastify, Hono
Python	FastAPI, Flask, Django
Java	Spring (@GetMapping etc.), JAX-RS
Go	Gin, Chi, Echo, net/http
Rust	Actix, Axum, Rocket
PHP	Laravel, Symfony

NestJS decorators are parsed via the TypeScript compiler API for accurate parameter + return type extraction.

Architecture detection

ArchitectureDetector examines file-path patterns to identify layers and match patterns.

Layers detected: controllers, services, repositories, models, middleware, views, config, utils, tests.

Patterns matched with confidence scores:

Pattern	Criteria
Layered architecture	Controllers + services + repositories
MVC	Controllers + models + views
Microservices	Multiple service directories
Frontend SPA	Components + views directories
Full-stack web app	Pages + components + API routes
CLI tool	cmd/ or commands/ directories
Library / SDK	src/ with no framework indicators
REST API	API endpoints + no frontend components

Each detection includes confidence: “Layered Architecture (85% confidence)”.

Call graph

Import/dependency graph → extract:

• Node + edge counts (total modules + import relationships)
• Dependency hubs — most-imported modules (up to 8)
• Entry points — modules not imported by anything (up to 8)
• Circular dependencies — detected cycles (up to 5)

Middleware + auth

For web apps:

• Auth strategies (JWT, OAuth, session-based, API key)
• Middleware chain — ordered pipeline (up to 10)
• Auth flows — login, register, token refresh (up to 5)
• Error handlers — count + source files

Caching

Results cached in SQLite:

• Git state tracking — branch name + commit hash + diff hash stored with each analysis.
• Change detection — subsequent runs check git state; only re-analyze if significantly changed.
• Worker thread — analysis runs in a worker_threads Worker; UI stays responsive.
• File limit — up to 1000 files per run.
• Cancellation — long analyses can be cancelled via the progress notification.

Excluded dirs: node_modules, dist, build, .git, .codebuddy, coverage, .next, .nuxt.

Cache management

Command	Behavior
Show Cache Status	Panel with cache status, last analysis date, branch, file count, snapshot count, DB size
Clear Cache	Delete all snapshots (with confirmation)
Refresh Analysis	Force a fresh analysis regardless of cache state

Architecture-expert access

The get_architecture_knowledge tool accepts a section:

Section	Returns
overview	Project type, entry points, frameworks, file count
patterns	Architectural patterns with confidence scores
call-graph	Import graph stats, hubs, circular deps
middleware	Auth strategies, middleware chain, error handlers
endpoints	API routes with method + source file
models	Data models with property names
all	Everything (capped at 12 000 chars)

No data yet? The tool suggests running the analyze command first.

Branch-switch watcher

CodeBuddy watches .git/HEAD. When you switch branches:

Branch changed to feature/auth. Analyze now? [Analyze Now] · [Later]

Related

• Subagents — architecture-expert role
• Code indexing — orthogonal indexer (chunks + embeddings)