Architecture scanner

Architecture Scanner Skill

Purpose

Unified layer detection that connects frontend and backend pattern skills with the project map. Works on ANY project by auto-detecting which directories are FE vs BE and what technologies are used.

Activation

When a task spans multiple layers (frontend + backend) or when Claude needs to determine which layer a file belongs to.

---

How It Works

This skill bridges three systems:

1. Backend Patterns (${CLAUDE_PLUGIN_ROOT}/skills/backend-patterns/)
2. Frontend Patterns (${CLAUDE_PLUGIN_ROOT}/skills/frontend-patterns/)
3. Project Map (.claude/idev/project-map/project.map.md) — if available

It auto-detects project types, which directories are frontend vs backend, and provides FE-to-BE endpoint mappings.

---

Phase 1: Detect Project Type

Scan the workspace root using technology-agnostic detection. Check for ALL of these — do not assume any specific stack.

1.1 Detect All Project Roots

Glob for these marker files (exclude dependency folders like node_modules, vendor, bin, obj):

Frontend markers:
  package.json         → Read to detect: react, vue, angular, svelte, next, nuxt, solid, qwik
  pubspec.yaml         → Flutter/Dart
  Podfile              → iOS (Swift/ObjC)
  build.gradle + /app  → Android

Backend markers:
  *.sln / *.csproj     → .NET (C#)
  go.mod               → Go
  Cargo.toml           → Rust
  pom.xml / build.gradle (no /app) → Java/Kotlin (Spring Boot, etc.)
  requirements.txt / pyproject.toml / setup.py → Python (Django, Flask, FastAPI)
  Gemfile              → Ruby (Rails)
  composer.json        → PHP (Laravel)
  package.json (with express/fastify/nest/koa) → Node.js backend

Monorepo markers:
  lerna.json / nx.json / turbo.json / pnpm-workspace.yaml → Monorepo

1.2 Classify Each Root

For each detected project root:
1. Read the marker file to determine exact framework
2. Classify as: "frontend", "backend", "fullstack", "shared/lib", or "mobile"
3. Detect language: TypeScript, JavaScript, C#, Python, Go, Java, Rust, etc.
4. Record the root directory path

1.3 Use Project Map (if available)

If .claude/idev/project-map/project.map.md exists:
1. Grep for section headers to find FE/BE groupings
2. Cross-reference detected roots with map sections
3. Use map to fill in any missing feature-to-directory mappings

---

Phase 2: Map FE-to-BE Connections

Discover how frontend calls backend. Detection varies by stack:

2.1 Find API Base Configuration

Search FE project for API base URL config:
  React/Vue/Angular: grep for "baseURL|BASE_URL|API_URL|VITE_API|NEXT_PUBLIC_API"
  Flutter: grep for "baseUrl|apiUrl"
  Any: grep for environment variable files (.env, .env.local, .env.development)

2.2 Collect Frontend API Calls

Search FE service/api files for endpoint paths:
  TypeScript/JS: grep for "api/|/api" in *.service.ts, *.api.ts, api/*.ts
  Python: grep for "requests.get|requests.post|httpx"
  Flutter: grep for "http.get|http.post|dio"

2.3 Collect Backend API Routes

Search BE project for route definitions:
  .NET:     grep for [Route("api/"] or [Http*("  in *.cs
  Express:  grep for "app.get|app.post|router.get|router.post" in *.ts/*.js
  FastAPI:  grep for "@app.get|@app.post|@router" in *.py
  Django:   grep for "path(" in urls.py
  Spring:   grep for "@GetMapping|@PostMapping|@RequestMapping" in *.java
  Go:       grep for "HandleFunc|Handle|r.GET|r.POST" in *.go
  Rails:    read config/routes.rb
  Laravel:  read routes/api.php
  NestJS:   grep for "@Get|@Post|@Controller" in *.ts

2.4 Match Endpoints

1. Normalize FE endpoint paths and BE route definitions
2. Match them by URL pattern
3. Build FE→BE mapping table

---

Phase 3: Generate Layer Index

Write to .claude/idev/architecture-scanner/cache.json:

{
  "generated": "YYYY-MM-DD",
  "projectType": "monorepo|single-app|multi-repo",
  "layers": {
    "frontend": {
      "root": "relative/path/to/fe",
      "framework": "react|vue|angular|svelte|next|nuxt|flutter",
      "language": "typescript|javascript|dart",
      "patternsSkill": "frontend-patterns",
      "apiConfigFile": "relative/path/to/api/config",
      "serviceFilesPattern": "**/*.service.ts"
    },
    "backend": {
      "root": "relative/path/to/be",
      "framework": ".net|express|fastapi|django|spring|rails|laravel|nestjs|go",
      "language": "csharp|typescript|python|java|go|ruby|php|rust",
      "patternsSkill": "backend-patterns",
      "routesLocation": "relative/path/to/controllers-or-routes"
    }
  },
  "apiRoutes": {
    "api/example/get-all": "ExampleController.cs (or example.routes.ts, etc.)"
  },
  "featureMapping": {
    "FeatureName": {
      "feRoot": "path/to/fe/feature",
      "beControllers": ["Controller.cs"],
      "feServices": ["service.ts"]
    }
  },
  "filePatterns": {
    "*.tsx": "frontend",
    "*.cs": "backend"
  }
}

---

Phase 4: Usage

Determine Layer for a File

Given a file path:
1. Check if path contains any known layer root
2. Fall back to file extension check against filePatterns
3. Return: "frontend" or "backend"
4. Load the corresponding patterns skill

Find BE Endpoint for FE Service Call

Given a frontend API call (e.g., "api/jobs/get-all"):
1. Look up apiRoutes for matching endpoint
2. Return the BE route handler file and method name
3. Read that specific method for implementation details

Find FE Service for BE Route

Given a backend route/controller method:
1. Look up apiRoutes in reverse
2. Grep FE service files for the endpoint URL
3. Follow imports to find the hook/composable and component using it

---

Phase 5: Full-Stack Task Workflow

When user asks for a full-stack feature:

1. Load architecture-scanner cache
2. Determine which layers are affected
3. Load backend-patterns cache → Create BE files following detected conventions
4. Load frontend-patterns cache → Create FE files following detected conventions
5. Update apiRoutes with new endpoint mapping
6. Use the Checklist from both pattern skills

---

Anti-Patterns

1. Do NOT assume any specific technology — always detect first
2. Do NOT load both pattern caches unless the task truly spans both layers
3. Do NOT re-scan if cache exists and is < 7 days old
4. Do NOT load the full project map; grep it for specific keywords only
5. Do NOT guess which layer a file belongs to; check the cache
6. Do NOT hardcode file extensions — derive them from detected language

---

Rescan Triggers

• User says "refresh architecture" or "rescan layers"
• Cache is older than 7 days
• New directories are added to the workspace
• FE-to-BE mapping is missing for a known endpoint
• Project type cannot be determined from cache alone