Nestjs 上下文验证

NestJS-specific implementation of DDD + Hexagonal + CQRS patterns. For the underlying theory (aggregates, domain events, architecture layers), see engineering-toolkit:engineering-foundations. This skill covers the NestJS-specific HOW. Before applying any topic, read its reference file in reference/.

Topics

Architecture Overview

+------------------------------------------------------------------+
|                     PRESENTATION LAYER                            |
|  apps/ (HTTP Controllers, DTOs, Request/Response handling)        |
+------------------------------------------------------------------+
|                     APPLICATION LAYER                             |
|  modules/ (Commands, Queries, Handlers, Events)                   |
+------------------------------------------------------------------+
|                       DOMAIN LAYER                                |
|  libs/common/domain/ (Entities, Value Objects, Enums)             |
+------------------------------------------------------------------+
|                    INFRASTRUCTURE LAYER                            |
|  libs/ (Repositories, External APIs, Database, Messaging)         |
+------------------------------------------------------------------+

Quick Decision Guide

Writing an endpoint?
  -> reference/api-design.md + reference/code-structure.md

Handling errors?
  -> reference/error-handling.md

Setting up config/env?
  -> reference/config.md

Adding authentication?
  -> reference/auth.md

Writing a database query?
  -> reference/typeorm-queries.md

Creating a migration?
  -> reference/typeorm-migrations.md

Implementing a domain model with events?
  -> reference/nestjs-domain-model.md

Adding logging?
  -> reference/logging.md

Gotchas

Claude-specific failure modes in NestJS codebases:

• Throwing HttpException from domain/application layer — Claude defaults to HTTP exceptions everywhere. Domain layer must throw domain-specific exceptions; the exception filter maps them to HTTP responses.
• Using process.env.X instead of ConfigService — Claude reaches for process.env out of habit. Always inject ConfigService and use .get().
• Putting auth logic in services — Claude tends to add if (!user.isAdmin) checks inside service methods. Auth belongs in guards; services receive already-validated context.
• Returning entities directly from controllers — Claude skips DTO mapping when "it's the same shape anyway." Always use response DTOs, even if they mirror the entity — the contract must be explicit.
• Circular imports between modules — Claude creates circular dependencies when wiring cross-module services. Use forwardRef() as last resort; prefer restructuring.
• Logging message-first instead of context-first — Claude writes logger.log('User created', { userId }) instead of logger.log({ userId }, 'User created'). Pino expects context object first.
• Raw SQL when QueryBuilder suffices — Claude jumps to raw SQL for anything beyond .find(). Follow the hierarchy: built-in methods > QueryBuilder > raw SQL.
• Forgetting to release QueryRunner — Must always release in a finally block. Claude sometimes puts release() only in the happy path.
• Importing from other module's internal paths — Use the module's public API (barrel exports), not deep ../other-module/internal/file imports.

Key Rules (Always Apply)

• Domain layer MUST NOT import HTTP exceptions — use domain exceptions, map in filters
• Never access process.env directly — use ConfigService
• Auth in guards, not in services — domain receives validated user context
• DTOs for all input/output — never expose entities directly
• Relative imports within modules — path aliases across modules
• Context-first logging — structured fields before message string
• Query hierarchy — built-in methods first, query builder second, raw SQL last resort
• Release query runners — always in a finally block

Source: anpham1925/claude-marketplace — distributed by TomeVault.