agentskills.codes
BA

backend-developer

Use this skill when working on backend microservices — implementing GraphQL APIs with HotChocolate, writing MassTransit consumers/publishers, building MongoDB data access layers, configuring service startup, integrating data pipelines, or adding observability with OpenTelemetry. Triggers on: GraphQL

Install

mkdir -p .claude/skills/backend-developer && curl -L -o skill.zip "https://agentskills.codes/api/skills/download/16633" && unzip -o skill.zip -d .claude/skills/backend-developer && rm skill.zip

Installs to .claude/skills/backend-developer

Activation

This is the description your AI agent reads to decide when to run this skill — the better it matches your request, the more reliably it fires.

Use this skill when working on backend microservices — implementing GraphQL APIs with HotChocolate, writing MassTransit consumers/publishers, building MongoDB data access layers, configuring service startup, integrating data pipelines, or adding observability with OpenTelemetry. Triggers on: GraphQL resolvers, ObjectType, TypeExtensions, QueryRewriter, field middleware, DataLoaders, MassTransit IConsumer, Azure Service Bus, MongoDB repositories, data pipeline sender/receiver, service startup configuration, or any backend domain service work.
547 chars✓ has a “when” triggerlonger than Claude Code's old 250-char listing cap (fine on current versions)

About this skill

Backend Developer

Guide for developing backend microservices. Covers domain-specific patterns, conventions, and internal libraries that the AI agent does not inherently know.

Scope: This skill covers project-specific patterns and conventions only. For general C# best practices, refer to the CSharpExpert agent. For testing conventions, follow tests.instructions.md.

Architecture Quick Reference

Architecture, layer definitions, dependency rules, and coding standards are defined in general.instructions.md (always loaded). The following covers domain-specific patterns only.

Known domains: <!-- TODO: List your domain services here, e.g., Contract, Consultation, Profile, Document, etc. -->

API Stitching Layer

src/Api/ is the HotChocolate Stitching gateway. It stitches domain GraphQL schemas together and uses QueryDelegationRewriterBase to transform queries between schemas.

// Example: src/Api/src/Host/Addresses/AddressQueryRewriter.cs
public class AddressQueryRewriter : QueryDelegationRewriterBase
{
    // Rewrites delegated queries for the stitched schema
}

Never add business logic to the Api layer — it only delegates and rewrites.

HotChocolate GraphQL Patterns

ObjectType Definitions

Use class-based type definitions extending ObjectType<T>:

public class MyEntityType : ObjectType<MyEntity>
{
    protected override void Configure(IObjectTypeDescriptor<MyEntity> descriptor)
    {
        descriptor.Field(x => x.Id).Type<NonNullType<IdType>>();
    }
}

TypeExtensions

Extend existing types without modifying their original definition. Use [ExtendObjectType]:

[ExtendObjectType(typeof(ParentEntity))]
public class ChildTypeExtensions
{
    public async Task<ChildResult> GetChildData(
        [Parent] ParentEntity parent,
        [Service] IChildService service,
        CancellationToken cancellationToken)
    {
        return await service.GetByParentIdAsync(parent.Id, cancellationToken);
    }
}

Field Middleware

Chain middleware on fields using FieldDelegate and IMiddlewareContext:

public static class ComputeMiddleware
{
    public static FieldDelegate Create(FieldDelegate next)
    {
        return async context =>
        {
            await next(context);
            object? result = context.Result;
            // Transform result
            context.Result = transformed;
        };
    }
}

Pass data between middleware/resolvers via ScopedContextData with IImmutableStack<object>.

DataLoaders

Use GreenDonut BatchDataLoader<TKey, TValue> for batching and caching. Naming convention: {Entity}By{Key}DataLoader. Inject IMyRepository + IBatchScheduler in the constructor, call base(batchScheduler, options), and override LoadBatchAsync.

Translation & Localization

Consultation domain uses HotChocolate.Extensions.Translation with HeaderLocalization:

services.AddTranslation()
    .AddHeaderLocalization(options => { /* configure */ });

Mutation Conventions

Use HotChocolate's built-in mutation conventions. Prefer implementation-first over schema-first. Use dedicated Input/Output types rather than separate parameters.

MassTransit Consumer Patterns

Standard Consumer

Mark consumers sealed. Use primary constructors for DI. Every consumer must include OpenTelemetry tracing and source-generated logging:

public sealed class ContractDeletedConsumer(
    IConsultationUpdater updater) : IConsumer<ContractDeleted>
{
    public async Task Consume(ConsumeContext<ContractDeleted> context)
    {
        using Activity? activity = App.ActivitySource.StartActivity();

        App.Log.ConsumingContractDeleted(context.Message.ContractId);

        await updater.DeleteByContractIdAsync(
            context.Message.ContractId,
            context.CancellationToken);
    }
}

Abstract Base Consumer

For shared consumer logic across message types, use a generic abstract base:

public abstract class CustomerUpdateMessageConsumerBase<TMessage>
    : IConsumer<TMessage> where TMessage : class
{
    public async Task Consume(ConsumeContext<TMessage> context)
    {
        using Activity? activity = App.ActivitySource.StartActivity();
        // Shared processing logic
        await ProcessAsync(context.Message, context.CancellationToken);
    }

    protected abstract Task ProcessAsync(TMessage message, CancellationToken cancellationToken);
}

Consumer Organization

Organize consumers in Core/Messaging/Consumers/ by category:

  • E2E/CustomerUpdates/ — End-to-end customer update events
  • Fuse/ — Fuse-specific events (invites, codes)
  • FusionIdentity/ — Identity events
  • Integration/ — External integration events

Request-Response Pattern

For consumers that need to respond:

await context.RespondAsync(new AdminResponse { /* ... */ });

Consumer Anti-Patterns

  • Never catch and swallow exceptions in consumers — let MassTransit handle retries
  • Never use ILogger directly — use source-generated App.Log methods
  • Never skip App.ActivitySource.StartActivity() — it breaks distributed tracing
  • Never add business logic outside Core/ layer — consumers orchestrate, not implement

Publisher Testing

Test publishers with MassTransit InMemoryTestHarness + Snapshooter:

public class ProfilePublisherTests
{
    private readonly InMemoryTestHarness _harness;

    [Fact]
    public async Task Publish_ShouldSendMessage()
    {
        // Arrange + Act
        await _publisher.PublishAsync(data);

        // Assert
        IPublishedMessage<ProfileUpdated>? message =
            _harness.Published.Select<ProfileUpdated>().FirstOrDefault();
        message.Should().NotBeNull();
        message!.Context.Message.MatchSnapshot();
    }
}

Observability

OpenTelemetry Tracing

Use a centralized activity source for all tracing. Start activities in every consumer/service method:

using Activity? activity = App.ActivitySource.StartActivity();

Record exceptions on the activity:

catch (Exception ex)
{
    activity?.RecordException(ex);
    throw;
}

Source-Generated Logging

Use the [LoggerMessage] attribute with App.Log — never use ILogger directly for structured log messages:

internal static partial class Log
{
    [LoggerMessage(Level = LogLevel.Information,
        Message = "Consuming contract deleted for {ContractId}")]
    public static partial void ConsumingContractDeleted(
        this ILogger logger, string contractId);
}

// Usage:
App.Log.ConsumingContractDeleted(contractId);

Service Startup Configuration

Standard Host Setup

Domain services use a consistent startup pattern:

public class StartUp
{
    public void ConfigureServices(IServiceCollection services)
    {
        // Authentication
        services.AddJwtBearerAuthentication(Configuration);

        // GraphQL
        services.AddGraphQLServer()
            .AddTypes()
            .AddTranslation();

        // MongoDB
        services.AddMongoDataAccess(Configuration);

        // Health checks
        services.AddHealthChecks()
            .AddMongoHealthCheck();

        // MassTransit / Service Bus
        services.AddMassTransitServiceBus(Configuration);
    }
}

Internal Libraries

<!-- TODO: Replace with your organization's internal libraries -->
LibraryPurpose
YourOrg.Security.Authentication.JwtBearerJWT Bearer auth setup
YourOrg.Health.*Health check extensions (Mongo, etc.)
YourOrg.Observability.*Activity sources and structured logging
YourOrg.{ProjectName}.*Domain-specific shared code

Authentication & Authorization

  • Use AddJwtBearerAuthentication(Configuration) for auth setup
  • Define authorization policies for specific endpoints (e.g., Invites, NeoInvites, MyLifeInvites)
  • Swagger UI configured with Bearer token support

REST + GraphQL Coexistence

Some services expose both REST (controllers + Swagger) and GraphQL. DocConnector and Onboarding use this pattern:

services.AddControllers()
    .AddNewtonsoftJson(o => o.SerializerSettings.Converters.Add(new StringEnumConverter()));
services.AddSwaggerGen(c => { /* Bearer auth config */ });

services.AddGraphQLServer()
    .AddTypes();

NSwag-Generated Clients

External API clients live in ServiceReferences/Generated/. Never edit generated files. Register via DI and consume through the generated client interfaces.

Data Pipeline Integration

API Host

DataPipelineApi.Create<Startup>(SetupConfiguration, args);

Worker Host

DataPipelineHost.Create<DomainsReference>(SetupConfiguration, configureServices, args);

Messaging

Register sender/receiver in DI:

services.AddDataPipelineReceiver<MyWorker>();
services.AddDataPipelineSender();

Use IMessageSender<T> and IMessageSenderFactory for sending messages.

Tenant Structure

The data pipeline uses tenant-based organization:

<!-- TODO: Replace with your actual tenant names -->
  • src/Tenants/TenantA/ — Primary tenant (Api + Host)
  • src/Tenants/TenantB/ — Secondary tenant

Partial Program Pattern

Environment-specific config via partial classes:

  • Program.cs — Shared entry point
  • Program.Dev.cs — Development configuration
  • Program.Prod.cs — Production configuration

Shared Config

Shared appsettings live in _Links/ directories, e.g., _Links/{TenantName}.Domains/appsettings.shared.json.

MongoDB Data Access

Repository Pattern

Repositories implement interfaces from Abstractions and use IMongoCollection<T> directly. Follow existing patterns in the domain you are working in.

Health Checks

Always register MongoDB health checks:

services.AddHealthChecks()
    .AddMongoHealthCheck();

Testing Domain-Specific Code

Follow tests.instructions.md


Content truncated.

Search skills

Search the agent skills registry