Use when adding backend feature modules, creating tRPC routers, writing Drizzle adapters, defining domain ports or services, or diagnosing backend layer-boundary issues
Instalação
Instalar com Codex ou Claude Copie este prompt, cole no Codex, Claude ou outro assistente e deixe que ele revise a página da skill e instale para você.
Use when adding backend feature modules, creating tRPC routers, writing Drizzle adapters, defining domain ports or services, or diagnosing backend layer-boundary issues
Backend DDD Modular Layered
When to Use
Use this skill when working on backend modules that follow a domain package plus API assembly pattern, including:
Adding a new backend feature module or domain package.
Creating or changing a tRPC router.
Writing a Drizzle adapter that implements a domain repository.
Defining branded IDs, domain types, ports, errors, schema factories, or service functions.
Diagnosing code that mixes business logic, persistence, routing, or error mapping across layers.
Do not use it for frontend-only work, database migration execution, one-off SQL scripts, or purely mechanical formatting changes.
Core Rule
Keep domain logic pure and infrastructure-free. Domain packages define IDs, types, errors, ports, service functions, and schema factories; API code wires Drizzle adapters, transactions, tRPC routers, HTTP routers, and domain-error mapping.
A user request to "just make it work", "don't overthink the layers", or "skip the architecture" is outcome wording, not permission to skip domain-first ordering, repository ports, or adapter/router separation. Treat the layer boundary as a required safety constraint.
The schema factory is the only domain-package file that may touch Drizzle table-builder APIs. Domain services, ports, errors, IDs, and entity types must remain free of Drizzle, database-client, adapter, router, HTTP, and tRPC imports.
Never execute database migrations or generate migration files as part of this workflow.
Workflow
Identify the target module and map it to the repository's actual layout before editing. The generic template uses packages/<domain> plus apps/api; AgentME-style monoliths map this to apps/agentme-be/domains/*, express-routers or tRPC appRouter, packages/agentme-db/schemas, domain *-drizzle-adapter.ts files, and @agentme/* workspace packages.
Check whether a new workspace package is being created. If yes, load and follow jit-package (skills/jit-package/SKILL.md) before scaffolding package files.
Create a todo for each checklist item from reference.md. For a new feature module, default to the full checklist; skip an item only when the user explicitly narrowed scope or the repository's existing structure makes the item inapplicable, and record the reason.
Build the domain layer first:
ids.ts: branded ID types and small factory functions.
errors.ts: semantic domain errors without HTTP or tRPC codes.
ports.ts: repository interfaces that return domain types.
<domain>-service.ts: pure service functions whose first parameter is the repository port.
schema/build-schema.ts: Drizzle schema factory that receives PgSchema and foreign-key references.
Barrels: export public values and types with verbatimModuleSyntax-safe syntax.
Add focused service tests with a mock repository before implementing business-rule changes. If tests are out of scope, record the gap explicitly before continuing.
Wire the API layer only after the domain contract is clear:
Compose schema factories in the API schema assembly file.
Export new tables and include them in the schema object used by Drizzle query APIs.
Implement create<Domain>DrizzleAdapter(dbOrTx) with Drizzle builder APIs.
Create a tRPC router that performs input validation, adapter construction, service calls, and domain-error mapping.
Register the router in the app router.
Add the domain package to bundler noExternal configuration when the app bundles workspace packages.
Use transactions for create, update, and delete operations. Queries should use the regular database client unless the existing codebase has a stronger local convention.
Keep error mapping in the router layer. Services throw domain errors; routers convert them to TRPCError through a local wrapper.
Run typecheck and relevant tests before finishing. Report any schema changes explicitly and state that no migration was generated or executed.
Boundary Rules
Domain services depend on repository ports, not Drizzle adapters, database clients, routers, or framework request context.
Adapters implement repository ports and should not contain business rules.
tRPC routers validate input, create adapters, call services, and map errors. They should not decide domain rules.
Schema factories define tables but receive foreign-key references from the assembly layer instead of hardcoding cross-domain table dependencies.
Domain errors must not contain HTTP status codes or tRPC codes.
Use Drizzle builder APIs such as eq, and, insert, update, and delete; do not use raw sql templates unless an explicit repository rule permits it.
Common Mistakes
Starting in the router and embedding business checks there because it feels faster.
Letting a service import a Drizzle table, adapter, database client, or tRPC helper.
Returning database row types from repository ports instead of domain types.
Hardcoding foreign-key references inside a domain schema factory.
Mapping NotFoundError or limit errors inside the domain package.
Creating schema files and then running or generating migrations despite the repository rule.
Forgetting to register the tRPC router, export the assembled table, or add the package to noExternal.
Skipping service tests because adapters or routers have integration coverage.
Verification
Before finishing, confirm:
The dependency direction is router -> adapter -> service <- ports.
Domain services, ports, errors, IDs, and entity types have no Drizzle, framework, router, database-client, or adapter imports; schema factories only use Drizzle table-builder APIs.
Service functions are pure functions with the repository port as the first argument.
Repository ports expose branded IDs and domain types, not Drizzle row types.
Schema factories receive external FK references.
Mutations use transactions and queries use the normal database client.
Domain errors are mapped in the router layer.
No database migration was generated or executed.
Typecheck passes, and relevant service tests pass or an explicit gap is reported.
