spring-data-jpa（弹簧数据 jpa）

Spring Data JPA Implementation

Purpose

Use this skill when the task is specifically about persistence design or implementation in a Spring Boot codebase. This skill adds value through aggregate-root guidance, query-pattern selection, CQRS read-model decisions, and the bundled repository and relationship templates.

Critical rules

• Never create repositories for every entity. Create repositories only for aggregate roots.
• Never rely on long derived query method names when the query has become non-trivial.
• Never use save() blindly when entity state transitions matter; understand persist versus merge behavior.
• Prefer projections or dedicated query services for read-heavy paths.
• Keep transaction boundaries in the service layer unless the existing architecture intentionally does otherwise.

Workflow

Step 1: Identify the persistence problem

Collect the minimum context first:

1. Is the type an aggregate root or an internal entity?
2. Is the task primarily read-side, write-side, or both?
3. Is the query simple lookup, filtered search, aggregation, projection, or dynamic criteria?
4. Is the path performance-sensitive?
5. Are there module-boundary or loose-coupling constraints that affect relationship modeling?

Step 2: Choose the implementation pattern

Use this table to decide what to load next.

Use this decision guide:

Step 3: Load the matching reference

Load only the references needed for the current task:

• references/query-patterns.md
• references/dto-projections.md
• references/custom-repositories.md
• references/cqrs-query-service.md
• references/relationships.md
• references/performance-guide.md

Step 4: Apply the matching asset

Use the bundled templates in assets/ instead of rebuilding the pattern from scratch:

• assets/query-repository.java
• assets/dto-projection.java
• assets/custom-repository.java
• assets/query-service.java
• assets/relationship-patterns.java

Step 5: Validate relationships and transaction boundaries

Before finalizing the change, check:

• repository exists only at the aggregate-root boundary
• lazy-loading behavior is intentional
• pagination or projections are used where row counts can grow
• @ManyToMany has not been introduced when a join entity is more appropriate
• read services use @Transactional(readOnly = true) where appropriate
• write operations stay in service-layer transactions

Step 6: Validate performance-sensitive paths

Read references/performance-guide.md when the task includes:

• N+1 risks
• fetch-plan problems
• unbounded queries
• batch operations
• heavy read views that should use projections

High-value patterns to prefer

Repository boundaries

• Aggregate roots get repositories.
• Internal child entities usually do not.

Query style

• Use derived query methods for simple lookups.
• Use @Query for joins, readable text blocks, or multiple filters.
• Use DTO projections when the response does not need entities.
• Use a CQRS query service when the read model differs from the write model.

Relationships

• Prefer @ManyToOne over @OneToMany when possible.
• Use IDs instead of entity references when loose coupling is more important than navigation.
• Treat @ManyToMany as a warning sign; prefer an explicit join entity.

Output format

When proposing or implementing a persistence change, return:

## Recommended pattern
- Pattern:
- Why:

## Files to change
- `path/to/file`

## References used
- `references/...`

## Risks to verify
- ...

When not to use this skill

• Generic SQL or database administration work outside Spring Data JPA
• Whole-project migration planning
• Broad project scaffolding that belongs in creating-springboot-projects