Smart Detection Skill

You are analyzing a .NET project to determine its architectural type using behavioral analysis.

Input

You will receive:

1. Project directory path to scan
2. Up to 10 representative files from the project (you will read these yourself)

Step 1: Gather Project Context

Read the following files to build a picture of the project:

Configuration Files (read first)

1. Program.cs or Startup.cs — startup configuration
2. *.csproj — NuGet packages, target framework, SDK type
3. *.sln or *.slnx — solution structure

Code Samples (read up to 5)

1. **/Application/**/Handlers/**/*.cs or **/Features/**/*Handler.cs
2. **/Controllers/**/*.cs
3. **/Services/**/*Service.cs
4. **/Domain/**/Aggregate*.cs or **/Domain/**/Core/*.cs
5. **/Endpoints/**/*.cs or **/Modules/**/*.cs

Structural Indicators (check existence)

• Features/ directory — VSA indicator
• Domain/, Application/, Infrastructure/ directories — Clean Architecture indicator
• BoundedContexts/ or Contexts/ directory — DDD indicator
• Modules/ directory — Modular Monolith indicator
• Controllers/ directory — traditional MVC/API
• Listeners/ or ServiceBus/ directory — event-driven
• *.razor files — Blazor UI

Step 2: Analyze Data Flow

For each handler/service file, analyze:

What comes IN?

• Commands: handler takes a command object (e.g., IRequestHandler<CreateOrderCommand>)
• Events: handler takes an event object (e.g., IRequestHandler<Event<OrderCreatedData>, bool>)
• HTTP requests: controller actions with [HttpGet], [HttpPost], etc.
• Minimal API: app.MapGet(), app.MapPost(), endpoint definitions

What happens INSIDE?

• Domain manipulation: loads events by aggregate ID, rebuilds aggregate, calls domain methods, commits
• Database save: calls SaveChangesAsync, AddAsync, repository Add/Update
• Service calls: calls gRPC/HTTP clients to other services
• Message publishing: publishes to ServiceBus/queue/topic
• Feature orchestration: coordinates within a single feature slice

What goes OUT?

• Nothing (void/Task): typical command handler
• Bool acknowledgment: typical event handler
• Data (DTOs/responses): typical query handler or controller
• HTML/Blazor: UI rendering

Step 3: Classify

Microservice Mode Types

These types are for projects following a CQRS/event-sourced microservice architecture:

Type	IN	INSIDE	OUT
command	Commands	Load aggregate -> domain method -> commit events	void/Task
query-sql	Events	Save to SQL database	bool (ack)
query-cosmos	Events	Save to Cosmos DB	bool (ack)
processor	Events	Call other gRPC services OR publish to message bus	bool (ack)
gateway	HTTP requests	Forward to gRPC services	HTTP responses
controlpanel	Browser requests	Blazor components render UI	HTML/Blazor
hybrid	Commands AND Events	Both domain manipulation AND DB save	Mixed

Generic Mode Types

These types are for projects following common .NET architectural patterns outside CQRS microservices:

Type	Key Indicators	Structure	Typical Patterns
vsa	Features/ directory with self-contained slices	Each feature folder contains handler, validator, model, endpoint	MediatR handlers per feature, minimal cross-feature dependencies
clean-arch	3+ layers: Domain, Application, Infrastructure	Separate projects per layer, dependency inversion	Use cases in Application, entities in Domain, repos in Infrastructure
ddd	BoundedContexts/ or Contexts/ directory	Bounded contexts with internal layering	Aggregates, value objects, domain events, anti-corruption layers
modular-monolith	Modules/ directory with independent modules	Each module has its own API, domain, infrastructure	Module-to-module communication via contracts/events
generic	None of the above patterns match	Varies — could be simple API, console app, library	Standard .NET project without strong architectural pattern

VSA (Vertical Slice Architecture) — Detailed Indicators

Structure signals:

• Features/ or Slices/ directory at project root
• Each subdirectory under Features/ contains all layers for one feature (handler, request, response, validator)
• No separate Application/, Domain/, Infrastructure/ top-level layers

Code signals:

• MediatR IRequestHandler<TRequest, TResponse> per feature
• IValidator<T> paired with handlers in same folder
• Endpoint classes or minimal API mappings grouped by feature
• AutoMapper/Mapster profiles per feature folder

NuGet signals:

• MediatR, FluentValidation, Carter or FastEndpoints

Clean Architecture — Detailed Indicators

Structure signals:

• 3+ separate projects: *.Domain, *.Application, *.Infrastructure, *.Api (or *.Web, *.Presentation)
• Domain project has NO references to other projects
• Application references only Domain
• Infrastructure references Application (and sometimes Domain)
• API/Web references all others

Code signals:

• Interfaces in Application (IRepository<T>, IUnitOfWork)
• Implementations in Infrastructure
• Use case classes or command/query handlers in Application
• Entities and value objects in Domain with no framework dependencies

NuGet signals:

• Domain project: minimal packages (maybe MediatR.Contracts only)
• Infrastructure: EntityFrameworkCore, Dapper, provider packages

DDD (Domain-Driven Design) — Detailed Indicators

Structure signals:

• BoundedContexts/ or Contexts/ directory
• Each bounded context has its own Domain, Application, Infrastructure
• SharedKernel/ or BuildingBlocks/ directory for cross-cutting concerns
• AntiCorruption/ or ACL/ directories

Code signals:

• AggregateRoot or Entity base classes in domain
• Value objects with structural equality
• Domain events (IDomainEvent, DomainEvent)
• Domain services with business logic
• Repository interfaces per aggregate root

NuGet signals:

• Domain event libraries, CQRS packages

Modular Monolith — Detailed Indicators

Structure signals:

• Modules/ directory with independent module directories
• Each module has its own Api/, Core/ (or Domain/), Infrastructure/
• Contracts/ or IntegrationEvents/ for inter-module communication
• Shared Infrastructure/ or Common/ at root level

Code signals:

• Module registration in Program.cs (AddModule<OrderModule>() pattern)
• Internal module classes (not public)
• Inter-module communication via events or contracts
• Each module has its own DbContext or schema

NuGet signals:

• Module framework packages if any

Generic — When to Classify

Use generic when:

• The project is a simple Web API without clear architectural layering
• Console application or background worker
• Class library or NuGet package project
• The project structure does not match any of the above patterns
• Mixed patterns that do not clearly fit one category

Decision Process

Apply rules in order — first match wins:

1. If .razor files exist → controlpanel
2. If controllers forward ALL requests to gRPC clients (no local DB) → gateway
3. If REST API with reverse proxy (YARP) → gateway
4. If handlers receive commands AND events, manipulate domain AND save to DB → hybrid
5. If handlers receive commands, manipulate domain aggregates, commit events, NOT consuming events → command
6. If handlers receive events, call other services/publish messages, NOT saving to DB → processor
7. If handlers receive events and save to Cosmos DB → query-cosmos
8. If handlers receive events and save to SQL/EF → query-sql
9. If Features/ directory with self-contained feature handlers → vsa
10. If 3+ layers (Domain, Application, Infrastructure) with dependency inversion → clean-arch
11. If BoundedContexts/ or Contexts/ directory with aggregate roots → ddd
12. If Modules/ directory with independent module structure → modular-monolith
13. Otherwise → generic

Output Format

CRITICAL: project_type is the SPECIFIC type (command, query-sql, vsa, etc.). mode is ALWAYS microservice or generic. Never swap these.

project_type: {command|query-sql|query-cosmos|processor|gateway|controlpanel|hybrid|vsa|clean-arch|ddd|modular-monolith|generic}
mode: {microservice|generic}
confidence: {high|medium|low}
dotnet_version: "{version from csproj TargetFramework}"
architecture: "{human-readable description}"
namespace_format: "{detected namespace pattern}"
evidence:
  1. {most important behavioral evidence}
  2. {second most important evidence}
  3. {third most important evidence}
reasoning: "{1-2 sentences explaining the data flow that led to this classification}"
packages:
  - {key NuGet packages detected}
solution_name: "{solution name without extension, from .sln/.slnx file}"
layers:
  domain: "{Domain project name or empty}"
  application: "{Application project name or empty}"
  infrastructure: "{Infrastructure project name or empty}"
  presentation: "{Presentation project name or empty}"
api_style: "{grpc|rest|blazor|minimal-api|none}"
patterns:
  - "{architectural patterns detected, e.g. cqrs-command-side, event-sourcing, outbox-pattern}"
deployment:
  containerized: {true if Dockerfile exists, else false}
  orchestration: "{kubernetes if k8s manifests, docker-compose if compose file, else empty}"
test_projects:
  - "{test project names from solution}"
sibling_repos:
  - name: "{sibling repo directory name}"
    type: "{detected type or empty}"

New field derivation

• solution_name: Read from .sln/.slnx filename (strip extension)
• layers: Map solution projects to logical layers — look for Domain/Application/Infrastructure/Infra/Presentation/Api/Web/Grpc in project names
• api_style: Check presentation project packages — Grpc.AspNetCore → grpc, Microsoft.AspNetCore.Mvc or controllers → rest, .razor files → blazor, app.MapGet/app.MapPost without controllers → minimal-api
• patterns: List architectural patterns found in code (cqrs-command-side, mediator, event-sourcing, outbox-pattern, clean-architecture, vertical-slice, repository-pattern, specification-pattern, etc.)
• deployment: Check for Dockerfile and kubernetes manifests or docker-compose.yml
• test_projects: Solution projects referencing xunit/nunit/mstest or named .Test/.Tests
• sibling_repos: Directories in ../ that are git repos with .NET projects — classify by quick structural scan

Important Notes

• Focus on BEHAVIOR, not naming. A class called "OrderService" tells you nothing about data flow.
• A handler that receives events AND saves to DB is a QUERY handler, even if it calls other services for enrichment.
• A handler that receives events AND forwards to other services WITHOUT saving locally is a PROCESSOR.
• Command projects may PUBLISH events to ServiceBus (outbox pattern) but do not CONSUME/LISTEN to events.
• Look at the MAJORITY of handlers, not edge cases.
• For generic mode types, focus on STRUCTURAL patterns (directory layout, project references) over code patterns.
• When confidence is low, explain what additional files would help clarify the classification.