@polymorphism-tech/morph-spec 7.7.3 → 7.8.1

package/framework/standards/ai-agents/sweet-spot.md

@@ -1,109 +1,109 @@
-# Sweet Spot — When to use what in MAF
-
-> **Scope:** stacks=["dotnet"]
-> **Layer:** 0 (always-load via persona maf-expert)
-> **Keywords:** sweet spot, when to use agent, agent vs workflow, when not to use workflow, MAF decision, simple agent
-> **Read by Claude in:** plan (decide arquitetura da feature antes de quebrar em tasks) + implement (validar que a arquitetura ainda é a mais simples possível)
-
-**Verified against:** MAF Tier List research (2026-05-18) + Microsoft Learn docs. Last-verified: 2026-05-19.
-
----
-
-## The 4-feature rule
-
-```
-95% of production AI agents need ONLY:
-
-  1. Chat (the LLM call)
-  2. Structured Output (typed records, not free text)
-  3. Tools (typed functions the agent can call)
-  4. RAG Custom (pgvector tool, not hosted RAG)
-
-If your feature needs ONLY these 4 → use a single agent. Don't reach for workflows.
-```
-
-Apply this rule before deciding multi-agent or workflow. Most of the time the answer is "single agent".
-
----
-
-## Decision tree (Claude reads this in plan phase)
-
-```
-Step 1 — Does the task even need an LLM?
-├─ "Can a deterministic function solve this?" → YES → write the function, no agent
-└─ NO → continue
-
-Step 2 — Single agent vs multi-agent?
-├─ One topic + one decision → single agent (DEFAULT)
-├─ Multiple specialized topics with clear handoff → multi-agent
-│   └─ Prefer Agent-as-Tool pattern (one agent uses another as a tool)
-│   └─ See ai-agents-multi-agent-patterns
-└─ "I'm not sure" → start with single agent, escalate if needed
-
-Step 3 — Need a Workflow?
-├─ The feature has fan-in/out (many inputs aggregated) → maybe workflow
-│   └─ Confirm: can you write it as a foreach + LINQ aggregate?
-│       └─ YES → write the foreach, NO workflow
-│       └─ NO → workflow may help
-├─ Long-running with human-in-the-loop checkpoints → maybe workflow
-│   └─ See ai-agents-workflows (Onda 2)
-└─ Otherwise → NO workflow
-
-Step 4 — Which provider?
-├─ ALWAYS direct models — never a hosted agent service
-├─ Provider choice (OpenAI / Google / Anthropic / Ollama) lives in the
-│  project's Model Registry, not hardcoded → see ai-agents-providers-model-registry
-└─ Hosted services (Foundry Agent Service, Vertex AI hosted) → NOT used
-
-Step 5 — RAG approach?
-├─ Need control over chunking/embedding/storage? → RAG custom (pgvector)
-│   └─ See ai-agents-rag-custom-pgvector (Onda 2)
-└─ Don't reach for hosted RAG — loses control
-```
-
----
-
-## Anti-patterns (what Claude should NOT do)
-
-| Anti-pattern | Why it's wrong | What to do instead |
-|--------------|----------------|---------------------|
-| Reach for Workflow when a function does the job | "Tudo que Workflows fazem, código C# normal faz com menos complexidade" — pesquisa | Write the function. Wrap in a tool if an agent needs to call it. |
-| Use a hosted agent service | Hosted is "always 6-9 months behind" on models + removes control | Direct models always (OpenAI / Google / Anthropic / Ollama) |
-| Upload PDFs to a hosted RAG | Loses control of chunking + embedding + storage | Custom RAG with pgvector. See ai-agents-rag-custom-pgvector |
-| Add Memory/Session because "agents need state" | If not building a chatbot, don't add chat state | Skip. Add when a real chat UI exists. |
-| Use Semantic Kernel for new features | SK is superseded by MAF | Use Microsoft.Agents.AI. See ai-agents-setup |
-| Return free text and parse it downstream | LLM output is non-deterministic | Structured Output. See ai-agents-structured-output (Onda 2) |
-
----
-
-## When the 4-feature rule does NOT apply
-
-You may need to step outside the sweet spot when:
-- The feature genuinely fans-out (e.g., monthly report aggregating N items) → workflow with fan-in
-- The feature needs vision/audio/image-gen → modality-specific standards (see ai-agents-modalities-* — Onda 3)
-
-**Even in these cases:** the core of the agent still follows the 4-feature rule. Only the orchestration layer changes.
-
----
-
-## Checklist (verifiable by morph-eval)
-
-- [ ] Was each agent designed as single-agent BEFORE reaching for multi-agent? (Check `decisions.md`.)
-- [ ] If a Workflow is used, is fan-in/out or HITL checkpointing genuinely present? Or could `foreach + LINQ` replace it?
-- [ ] Are the 4 features (Chat, Structured Output, Tools, RAG Custom) all present where applicable?
-- [ ] Are hosted agent services avoided?
-
----
-
-## References
-
-- `ai-agents-setup` — packages and DI
-- `ai-agents-providers-model-registry` — provider-agnostic config
-- `ai-agents-structured-output` (Onda 2)
-- `ai-agents-multi-agent-patterns` (Onda 2)
-- `ai-agents-workflows` (Onda 2, reframed)
-- `ai-agents-rag-custom-pgvector` (Onda 2)
-
----
-
-*MORPH-SPEC by Polymorphism Tech — ai-agents/sweet-spot.md v1.0 (2026-05-19)*
+# Sweet Spot — When to use what in MAF
+
+> **Scope:** stacks=["dotnet"]
+> **Layer:** 0 (always-load via persona maf-expert)
+> **Keywords:** sweet spot, when to use agent, agent vs workflow, when not to use workflow, MAF decision, simple agent
+> **Read by Claude in:** plan (decide arquitetura da feature antes de quebrar em tasks) + implement (validar que a arquitetura ainda é a mais simples possível)
+
+**Verified against:** MAF Tier List research (2026-05-18) + Microsoft Learn docs. Last-verified: 2026-05-19.
+
+---
+
+## The 4-feature rule
+
+```
+95% of production AI agents need ONLY:
+
+  1. Chat (the LLM call)
+  2. Structured Output (typed records, not free text)
+  3. Tools (typed functions the agent can call)
+  4. RAG Custom (pgvector tool, not hosted RAG)
+
+If your feature needs ONLY these 4 → use a single agent. Don't reach for workflows.
+```
+
+Apply this rule before deciding multi-agent or workflow. Most of the time the answer is "single agent".
+
+---
+
+## Decision tree (Claude reads this in plan phase)
+
+```
+Step 1 — Does the task even need an LLM?
+├─ "Can a deterministic function solve this?" → YES → write the function, no agent
+└─ NO → continue
+
+Step 2 — Single agent vs multi-agent?
+├─ One topic + one decision → single agent (DEFAULT)
+├─ Multiple specialized topics with clear handoff → multi-agent
+│   └─ Prefer Agent-as-Tool pattern (one agent uses another as a tool)
+│   └─ See ai-agents-multi-agent-patterns
+└─ "I'm not sure" → start with single agent, escalate if needed
+
+Step 3 — Need a Workflow?
+├─ The feature has fan-in/out (many inputs aggregated) → maybe workflow
+│   └─ Confirm: can you write it as a foreach + LINQ aggregate?
+│       └─ YES → write the foreach, NO workflow
+│       └─ NO → workflow may help
+├─ Long-running with human-in-the-loop checkpoints → maybe workflow
+│   └─ See ai-agents-workflows (Onda 2)
+└─ Otherwise → NO workflow
+
+Step 4 — Which provider?
+├─ ALWAYS direct models — never a hosted agent service
+├─ Provider choice (OpenAI / Google / Anthropic / Ollama) lives in the
+│  project's Model Registry, not hardcoded → see ai-agents-providers-model-registry
+└─ Hosted services (Foundry Agent Service, Vertex AI hosted) → NOT used
+
+Step 5 — RAG approach?
+├─ Need control over chunking/embedding/storage? → RAG custom (pgvector)
+│   └─ See ai-agents-rag-custom-pgvector (Onda 2)
+└─ Don't reach for hosted RAG — loses control
+```
+
+---
+
+## Anti-patterns (what Claude should NOT do)
+
+| Anti-pattern | Why it's wrong | What to do instead |
+|--------------|----------------|---------------------|
+| Reach for Workflow when a function does the job | "Tudo que Workflows fazem, código C# normal faz com menos complexidade" — pesquisa | Write the function. Wrap in a tool if an agent needs to call it. |
+| Use a hosted agent service | Hosted is "always 6-9 months behind" on models + removes control | Direct models always (OpenAI / Google / Anthropic / Ollama) |
+| Upload PDFs to a hosted RAG | Loses control of chunking + embedding + storage | Custom RAG with pgvector. See ai-agents-rag-custom-pgvector |
+| Add Memory/Session because "agents need state" | If not building a chatbot, don't add chat state | Skip. Add when a real chat UI exists. |
+| Use Semantic Kernel for new features | SK is superseded by MAF | Use Microsoft.Agents.AI. See ai-agents-setup |
+| Return free text and parse it downstream | LLM output is non-deterministic | Structured Output. See ai-agents-structured-output (Onda 2) |
+
+---
+
+## When the 4-feature rule does NOT apply
+
+You may need to step outside the sweet spot when:
+- The feature genuinely fans-out (e.g., monthly report aggregating N items) → workflow with fan-in
+- The feature needs vision/audio/image-gen → modality-specific standards (see ai-agents-modalities-* — Onda 3)
+
+**Even in these cases:** the core of the agent still follows the 4-feature rule. Only the orchestration layer changes.
+
+---
+
+## Checklist (verifiable by morph-eval)
+
+- [ ] Was each agent designed as single-agent BEFORE reaching for multi-agent? (Check `decisions.md`.)
+- [ ] If a Workflow is used, is fan-in/out or HITL checkpointing genuinely present? Or could `foreach + LINQ` replace it?
+- [ ] Are the 4 features (Chat, Structured Output, Tools, RAG Custom) all present where applicable?
+- [ ] Are hosted agent services avoided?
+
+---
+
+## References
+
+- `ai-agents-setup` — packages and DI
+- `ai-agents-providers-model-registry` — provider-agnostic config
+- `ai-agents-structured-output` (Onda 2)
+- `ai-agents-multi-agent-patterns` (Onda 2)
+- `ai-agents-workflows` (Onda 2, reframed)
+- `ai-agents-rag-custom-pgvector` (Onda 2)
+
+---
+
+*MORPH-SPEC by Polymorphism Tech — ai-agents/sweet-spot.md v1.0 (2026-05-19)*

package/framework/standards/architecture/vertical-slice/vertical-slice.md

@@ -10,7 +10,8 @@
 
 Padrão arquitetural baseado em **KanaiyaKatarmal/VerticalSliceArchitectureTemplate** (.NET 10, Minimal APIs, EF Core 10).
 Cada feature é uma fatia vertical completa: endpoint → handler → validator → entity → DB.
-Sem MediatR, sem AggregateRoot, sem Domain Events, sem CQRS formal.
+Sem MediatR, sem AggregateRoot, sem Domain Events, sem CQRS formal, sem repository genérico
+(divergência intencional do template base: o handler usa `ApplicationDbContext` direto).
 
 ---
 
@@ -54,11 +55,6 @@ Sem MediatR, sem AggregateRoot, sem Domain Events, sem CQRS formal.
 ├── Pipelines/
 │   ├── ValidationDecorator.cs            # Decorator que executa validators
 │   └── LoggingDecorator.cs               # Decorator que loga request/failure
-├── Repository/
-│   ├── IRepository.cs                    # IRepository<T> genérico
-│   ├── Repository.cs                     # Repository<T> : IRepository<T>
-│   ├── IUnitOfWork.cs                    # IUnitOfWork
-│   └── UnitOfWork.cs                     # UnitOfWork
 └── Program.cs
 ```
 
@@ -190,8 +186,7 @@ public sealed record CreateBookRequest(string Title, string Author, string ISBN,
 public sealed record CreateBookResponse(Guid Id, string Title, string Author, string ISBN, decimal Price, int PublishedYear);
 
 public sealed class CreateBookHandler(
-    IRepository<Book> _bookRepo,
-    IUnitOfWork _unitOfWork) : IHandler<CreateBookRequest, Result<CreateBookResponse>>
+    ApplicationDbContext _db) : IHandler<CreateBookRequest, Result<CreateBookResponse>>
 {
     public async Task<Result<CreateBookResponse>> HandleAsync(
         CreateBookRequest command, CancellationToken cancellationToken)
@@ -205,8 +200,8 @@ public sealed class CreateBookHandler(
             Price = command.Price,
             PublishedYear = command.PublishedYear
         };
-        await _bookRepo.AddAsync(book, cancellationToken);
-        await _unitOfWork.CommitAsync(cancellationToken);
+        _db.Books.Add(book);
+        await _db.SaveChangesAsync(cancellationToken);
         return Result.Success(new CreateBookResponse(
             book.Id, book.Title, book.Author, book.ISBN, book.Price, book.PublishedYear));
     }
@@ -306,38 +301,41 @@ Ordem de execução (innermost-first): `LoggingDecorator → ValidationDecorator
 
 ---
 
-## Repository Pattern
+## Acesso a Dados — DbContext direto no slice
+
+O handler injeta `ApplicationDbContext` direto. Sobre EF Core um repository genérico é
+redundante: `DbSet<T>` já é repository, `DbContext` já é unit of work. **Nunca crie
+`IRepository<T>`/`IUnitOfWork`** — a auditoria de projetos reais mostrou que o par nasce
+registrado no DI e morre órfão, sem consumidores.
 
 ```csharp
-// Repository/IRepository.cs
-public interface IRepository<T> where T : class
-{
-    Task<IEnumerable<T>> GetAllAsync(CancellationToken cancellationToken = default);
-    Task<T?> GetByIdAsync(object id, CancellationToken cancellationToken = default);
-    Task<T> AddAsync(T entity, CancellationToken cancellationToken = default);
-    Task UpdateAsync(T entity, CancellationToken cancellationToken = default);
-    Task DeleteAsync(T entity, CancellationToken cancellationToken = default);
-    IEnumerable<T> ExecuteQuery(string query, DbParameter[] dbParams);
-    Task AddRangeAsync(IEnumerable<T> entities, CancellationToken cancellationToken = default);
-    void RemoveRange(IEnumerable<T> entities);
-}
+// CRUD por chave — direto no DbSet
+var book = await _db.Books.FindAsync([command.Id], cancellationToken);
+_db.Books.Add(book);
+await _db.SaveChangesAsync(cancellationToken);
+
+// Queries específicas — LINQ no slice (joins, projeção, paginação, filtros compostos)
+var page = await _db.Books
+    .Where(b => b.Author == command.Author)
+    .OrderBy(b => b.Title)
+    .Skip(command.Skip).Take(command.Take)
+    .Select(b => new BookDto(b.Id, b.Title))
+    .ToListAsync(cancellationToken);
 ```
 
-**Quando NÃO usar o repository:** o `IRepository<T>` genérico cobre CRUD por chave.
-Para queries com join, projeção, paginação ou filtros compostos, injete
-`ApplicationDbContext` direto no handler e escreva o LINQ no slice — não polua a
-interface genérica com métodos de um único slice, e não use `ExecuteQuery(string sql)`.
-Veja `backend/dotnet/vsa-handler-patterns.md` → "When to Use DbContext Directly".
+**Stores dedicados (exceção justificada):** quando um slice exige controle de concorrência
+(ex.: `SELECT ... FOR UPDATE`, retry de `DbUpdateConcurrencyException`) ou SQL específico
+que o LINQ não expressa, escreva um store hand-written e purpose-built (ex.: `OutboxStore`,
+`LeaseStore`) com métodos nomeados pela intenção — e registre a justificativa em
+`decisions.md`. Nunca um genérico por entidade.
+
+Veja `backend/dotnet/vsa-handler-patterns.md` → "Data Access: DbContext in the Slice".
 
 ---
 
 ## DI Registration (Program.cs)
 
 ```csharp
-// Repositories + Unit of Work
-builder.Services.AddScoped(typeof(IRepository<>), typeof(Repository<>));
-builder.Services.AddScoped(typeof(IUnitOfWork), typeof(UnitOfWork));
-
 // EF Core (SQLite/SqlServer/PostgreSQL)
 builder.Services.AddSQLDatabaseConfiguration(builder.Configuration);
 
@@ -385,8 +383,8 @@ app.MapApiEndpoints();
 4. **`result.Match()`** nos endpoints — nunca acessar `result.Value` sem checar `IsSuccess`.
 5. **IApiEndpoint `internal sealed`** — endpoints não são públicos.
 6. **`{Entity}Errors` compartilhado** por todas as operações da feature (1 arquivo por feature, não por slice).
-7. **Acesso a dados híbrido**: `IRepository<T>` injetado no handler para CRUD por chave; `ApplicationDbContext` injetado direto no slice para queries específicas (joins, projeção, paginação). Nunca uma Application Service layer entre handler e dados.
-8. **`IUnitOfWork.CommitAsync()`** em operações de escrita (Create, Update, Delete).
+7. **`ApplicationDbContext` direto no slice**: CRUD por chave via `DbSet<T>`; queries específicas (joins, projeção, paginação) via LINQ no handler. Nunca uma Application Service layer entre handler e dados; nunca repository genérico (`IRepository<T>`/`IUnitOfWork`).
+8. **`SaveChangesAsync()`** explícito em operações de escrita (Create, Update, Delete).
 9. **Sem validator em GetAll** — GetAllBooksHandler não tem validator (sem parâmetros para validar).
 
 ---
@@ -399,7 +397,8 @@ app.MapApiEndpoints();
 | `DomainEvent` | VSA não tem Domain Events — reações são handlers separados |
 | `MediatR` / `IRequest` / `IRequestHandler` | Substituído por `IHandler<TRequest, TResponse>` próprio |
 | DTOs compartilhados entre features | Cada slice define seus próprios records |
-| Application Service layer | Handler acessa Repository diretamente |
+| Application Service layer | Handler acessa o `ApplicationDbContext` diretamente |
+| Repository genérico (`IRepository<T>`/`IUnitOfWork`) | `DbSet<T>` já é repository, `DbContext` já é unit of work — vira código morto |
 | CQRS formal (Commands/Queries) | Handler simples, sem distinção Command/Query |
 | Shared Validators em Application layer | Validator fica no mesmo namespace do slice |
 

package/framework/standards/backend/api/validation.md

@@ -62,7 +62,7 @@ public enum OrderStatus
 ```csharp
 public async Task<Result> ProcessOrderAsync(Guid orderId, CancellationToken ct)
 {
-    var order = await _repository.GetByIdAsync(orderId, ct);
+    var order = await _db.Orders.FindAsync([orderId], ct);
     if (order == null) return Result.Failure("Order not found");
 
     // State validation (invalid states)
@@ -74,7 +74,7 @@ public async Task<Result> ProcessOrderAsync(Guid orderId, CancellationToken ct)
     if (order.RequiresPayment && !order.IsPaid) return Result.Failure("Payment required");
 
     order.MarkAsProcessing();
-    await _repository.UpdateAsync(order, ct);
+    await _db.SaveChangesAsync(ct);
     return Result.Success();
 }
 ```

package/framework/standards/backend/dotnet/scrutor.md

@@ -126,9 +126,7 @@ builder.Services.AddHandlersFromAssembly(typeof(Program).Assembly);
 The full DI registration sequence in VSA:
 
 ```csharp
-// 1. Infrastructure
-builder.Services.AddScoped(typeof(IRepository<>), typeof(Repository<>));
-builder.Services.AddScoped(typeof(IUnitOfWork), typeof(UnitOfWork));
+// 1. Infrastructure (DbContext)
 builder.Services.AddSQLDatabaseConfiguration(builder.Configuration);
 
 // 2. Endpoints (auto-discovery)

package/framework/standards/backend/dotnet/vsa-handler-patterns.md

@@ -11,7 +11,7 @@
 
 ## Overview
 
-Each feature in VSA is a self-contained slice with 5 files in a feature folder. Handlers implement `IHandler<TRequest, TResponse>` (not MediatR). Dependencies use primary constructors. The data path is `Handler -> Repository -> UnitOfWork` with no service layer in between.
+Each feature in VSA is a self-contained slice with 5 files in a feature folder. Handlers implement `IHandler<TRequest, TResponse>` (not MediatR). Dependencies use primary constructors. The data path is `Handler -> ApplicationDbContext` — no service layer, no generic repository in between.
 
 ---
 
@@ -71,8 +71,8 @@ public sealed record CreateBookResponse(
 namespace ProjectName.Features.BookFeature.CreateBook;
 
 using ProjectName.Abstractions;
+using ProjectName.Database;
 using ProjectName.Entities;
-using ProjectName.Repository;
 
 public sealed record CreateBookRequest(
     string Title, string Author, string ISBN, decimal Price, int PublishedYear);
@@ -82,8 +82,7 @@ public sealed record CreateBookResponse(
 
 // Primary constructor for DI — no field declarations needed
 public sealed class CreateBookHandler(
-    IRepository<Book> bookRepo,
-    IUnitOfWork unitOfWork) : IHandler<CreateBookRequest, Result<CreateBookResponse>>
+    ApplicationDbContext db) : IHandler<CreateBookRequest, Result<CreateBookResponse>>
 {
     public async Task<Result<CreateBookResponse>> HandleAsync(
         CreateBookRequest command, CancellationToken cancellationToken)
@@ -98,8 +97,8 @@ public sealed class CreateBookHandler(
             PublishedYear = command.PublishedYear
         };
 
-        await bookRepo.AddAsync(book, cancellationToken);
-        await unitOfWork.CommitAsync(cancellationToken);
+        db.Books.Add(book);
+        await db.SaveChangesAsync(cancellationToken);
 
         return Result.Success(new CreateBookResponse(
             book.Id, book.Title, book.Author, book.ISBN, book.Price, book.PublishedYear));
@@ -111,7 +110,7 @@ public sealed class CreateBookHandler(
 
 - **`sealed class`** — handlers are never inherited
 - **`sealed record`** — DTOs are immutable value types
-- **Primary constructors** — `(IRepository<Book> bookRepo, IUnitOfWork unitOfWork)` replaces field + constructor boilerplate
+- **Primary constructors** — `(ApplicationDbContext db)` replaces field + constructor boilerplate
 - **`Guid.CreateVersion7()`** — time-sortable, index-friendly UUID (not `Guid.NewGuid()`)
 - **`CancellationToken`** — passed through every async call
 - **`Result<T>` return** — never throw for expected failures
@@ -172,36 +171,35 @@ All endpoint classes implementing this interface are auto-discovered and mapped
 
 ---
 
-## Repository + UnitOfWork (No Service Layer)
+## Data Access: DbContext in the Slice (No Service Layer, No Generic Repository)
 
-Handlers call repositories directly. There is no service/manager layer:
+Handlers inject `ApplicationDbContext` directly. Over EF Core a generic repository is
+redundant — `DbSet<T>` **is** the repository, `DbContext` **is** the unit of work. A
+generic `IRepository<T>`/`IUnitOfWork` pair adds indirection with zero value and, in
+audited real projects, ends up registered in DI with zero consumers (dead code).
+
+CRUD by key goes straight to the `DbSet`:
 
 ```csharp
-public interface IRepository<T> where T : class
+public sealed class DeleteBookHandler(
+    ApplicationDbContext db) : IHandler<DeleteBookRequest, Result>
 {
-    Task<IEnumerable<T>> GetAllAsync(CancellationToken cancellationToken = default);
-    Task<T?> GetByIdAsync(object id, CancellationToken cancellationToken = default);
-    Task<T> AddAsync(T entity, CancellationToken cancellationToken = default);
-    Task UpdateAsync(T entity, CancellationToken cancellationToken = default);
-    Task DeleteAsync(T entity, CancellationToken cancellationToken = default);
-    Task AddRangeAsync(IEnumerable<T> entities, CancellationToken cancellationToken = default);
-    void RemoveRange(IEnumerable<T> entities);
-}
+    public async Task<Result> HandleAsync(
+        DeleteBookRequest command, CancellationToken cancellationToken)
+    {
+        var book = await db.Books.FindAsync([command.Id], cancellationToken);
+        if (book is null)
+            return BookErrors.NotFound(command.Id);
 
-public interface IUnitOfWork
-{
-    Task<int> CommitAsync(CancellationToken cancellationToken);
-    Task RollbackAsync(CancellationToken cancellationToken = default);
+        db.Books.Remove(book);
+        await db.SaveChangesAsync(cancellationToken);
+        return Result.Success();
+    }
 }
 ```
 
-**Pattern:** `Repository.AddAsync()` -> `UnitOfWork.CommitAsync()` — always commit explicitly.
-
-### When to Use DbContext Directly Instead
-
-`IRepository<T>` is the right tool for CRUD by key. It is **not** the right tool for
-slice-specific queries — joins, projections, pagination, composite filters. For those,
-inject `ApplicationDbContext` directly into the handler and write the LINQ in the slice:
+Slice-specific queries — joins, projections, pagination, composite filters — are LINQ
+written in the slice:
 
 ```csharp
 public sealed class SearchBooksHandler(
@@ -222,8 +220,16 @@ public sealed class SearchBooksHandler(
 }
 ```
 
-**Rule of thumb:** CRUD by key → `IRepository<T>` + `IUnitOfWork`. Anything that needs a
-custom query shape → `ApplicationDbContext` in the slice. Never reach for raw SQL
+### Dedicated Stores (the justified exception)
+
+When a slice needs concurrency control (`SELECT ... FOR UPDATE`, lease/outbox semantics,
+`DbUpdateConcurrencyException` retry loops) or SQL that LINQ cannot express, write a
+**hand-written, purpose-built store** with intention-named methods (e.g. `OutboxStore.ClaimBatchAsync`,
+`LeaseStore.TryAcquireAsync`) — and document the justification in `decisions.md`.
+Never a generic-per-entity abstraction.
+
+**Rule of thumb:** everything → `ApplicationDbContext` in the slice. Concurrency or
+specific SQL → dedicated hand-written store + `decisions.md` entry. Never reach for raw SQL
 (`FromSqlRaw`) unless a query genuinely cannot be expressed in LINQ — and document that
 in `decisions.md`.
 
@@ -251,7 +257,7 @@ public static class BookErrors
 Usage in handler:
 
 ```csharp
-var existing = await bookRepo.GetByIdAsync(command.BookId, cancellationToken);
+var existing = await db.Books.FindAsync([command.BookId], cancellationToken);
 if (existing is null)
     return BookErrors.NotFound(command.BookId);  // implicit conversion to Result<T>
 ```
@@ -264,9 +270,7 @@ if (existing is null)
 // 1. OpenAPI
 builder.Services.AddOpenApi();
 
-// 2. Infrastructure (Repository, UnitOfWork, DbContext)
-builder.Services.AddScoped(typeof(IRepository<>), typeof(Repository<>));
-builder.Services.AddScoped(typeof(IUnitOfWork), typeof(UnitOfWork));
+// 2. Infrastructure (DbContext)
 builder.Services.AddSQLDatabaseConfiguration(builder.Configuration);
 
 // 3. Endpoint auto-discovery
@@ -293,7 +297,8 @@ builder.Services.AddExceptionHandler<CustomExceptionHandler>()
 | Do NOT | Do Instead |
 |--------|------------|
 | Use MediatR `IRequest`/`Send()` | Use `IHandler<TReq, TRes>` resolved directly from DI |
-| Create a service/manager layer | Handler calls Repository + UnitOfWork directly |
+| Create a service/manager layer | Handler uses `ApplicationDbContext` directly |
+| Add a generic repository (`IRepository<T>`/`IUnitOfWork`) | `DbSet<T>` is the repository; `DbContext` is the unit of work |
 | Use `Guid.NewGuid()` | Use `Guid.CreateVersion7()` for time-sortable IDs |
 | Make handler classes non-sealed | Always `sealed class` |
 | Use mutable DTOs (class with setters) | Always `sealed record` with positional parameters |
@@ -302,7 +307,7 @@ builder.Services.AddExceptionHandler<CustomExceptionHandler>()
 | Share handlers across features | Each feature has its own handler, even if similar |
 | Use constructor + field for DI | Use primary constructors: `class Handler(IDep dep)` |
 | Group by technical concern (Controllers/, Services/) | Group by feature: `Features/BookFeature/CreateBook/` |
-| Force a slice-specific query through `IRepository<T>` | Inject `DbContext` and write LINQ in the slice |
+| Route data access through any indirection layer | Inject `DbContext` and write LINQ in the slice; dedicated store only with `decisions.md` justification |
 | Use raw SQL (`FromSqlRaw` / `ExecuteQuery(string)`) by default | Use LINQ; raw SQL only with a `decisions.md` justification |
 
 ---

package/framework/standards/data/nosql/cache/redis.md

@@ -32,7 +32,7 @@ builder.Services.AddStackExchangeRedisCache(opts =>
 public class CachedProductService : IProductService
 {
     private readonly IDistributedCache _cache;
-    private readonly IProductRepository _repository;
+    private readonly ApplicationDbContext _db;
 
     public async Task<Product?> GetAsync(Guid id, CancellationToken ct = default)
     {
@@ -44,7 +44,7 @@ public class CachedProductService : IProductService
             return JsonSerializer.Deserialize<Product>(cached);
 
         // Cache miss → read from DB
-        var product = await _repository.GetAsync(id, ct);
+        var product = await _db.Products.FindAsync([id], ct);
         if (product == null) return null;
 
         // Cache with TTL

package/framework/templates/CLAUDE.md.core.template

@@ -6,7 +6,7 @@
 
 VSA + .NET 10 + EF Core + Neon (Postgres) + Microsoft Agent Framework + Next.js (App Router) + Tailwind + Framer Motion + Shadcn + Hangfire (quando necessário).
 
-Padrões proibidos por padrão: `Guid.NewGuid()` (use `Guid.CreateVersion7()`), classes não-sealed sem justificativa, exceptions como fluxo de controle, camada de Service/Manager entre handler e dados, SQL cru sem justificativa. Acesso a dados é híbrido: `IRepository<T>` + `IUnitOfWork` para CRUD; `DbContext` direto no slice para queries específicas.
+Padrões proibidos por padrão: `Guid.NewGuid()` (use `Guid.CreateVersion7()`), classes não-sealed sem justificativa, exceptions como fluxo de controle, camada de Service/Manager entre handler e dados, repository genérico (`IRepository<T>`/`IUnitOfWork`), SQL cru sem justificativa. Acesso a dados: `ApplicationDbContext` direto no slice (`DbSet<T>` já é repository, `DbContext` já é unit of work); para concorrência ou SQL específico, store dedicado hand-written com justificativa em `decisions.md`.
 
 ## Harness