sba-marketplace

Spring Batch Architecture Agent (SBA)

⚠️ WORK IN PROGRESS: This project is actively being developed. APIs, file structures, and patterns may change.

A Claude Code plugin that guides you through designing and implementing enterprise-grade Spring Batch applications.

Why SBA?

Designing Spring Batch applications involves countless decisions:

• Which reader? JdbcCursorItemReader vs JdbcPagingItemReader vs JpaPagingItemReader?
• Chunk size? 100? 1000? 10000? Depends on your data and infrastructure.
• Fault tolerance? Skip? Retry? What exceptions? What limits?
• Parallelization? Multi-threaded? Partitioned? Remote chunking?
• Database optimization? Batch inserts? Fetch size? Connection pooling?

SBA encapsulates years of Spring Batch expertise into an AI agent that:

1. Asks the right questions to understand your requirements
2. Makes architecture decisions based on your volume, constraints, and tech stack
3. Applies proven patterns for your specific use case
4. Generates production-ready code following best practices
5. Reviews and optimizes the final implementation

The 5-Phase Workflow

┌─────────────┐     ┌──────────────┐     ┌──────────┐     ┌────────────────┐     ┌──────────┐
│  Discovery  │ ──▶ │ Architecture │ ──▶ │  Design  │ ──▶ │ Implementation │ ──▶ │  Review  │
└─────────────┘     └──────────────┘     └──────────┘     └────────────────┘     └──────────┘
      │                    │                  │                   │                    │
      ▼                    ▼                  ▼                   ▼                    ▼
  Requirements         Decisions          Detailed           Working              Optimized
   Gathered              Made              Design             Code                Solution

Phase	What Happens	Output
Discovery	Understand business requirements, data sources, volumes, constraints	Project context
Architecture	Choose patterns, persistence layer, parallelization strategy	8 ADRs, tech stack
Design	Detail data models, step configurations, error handling	Design document
Implementation	Generate job configs, readers, processors, writers, tests	Production code
Review	Validate against best practices, optimize performance	Final solution

Installation

Claude Code — Plugin (Recommended)

# 1. Register the marketplace
/plugin marketplace add zakariahere/sba

# 2. Install the plugin
/plugin install sba@sba-marketplace

Once installed, Claude automatically invokes the SBA agent when you mention Spring Batch. You can also trigger it explicitly:

/sba:sba migrate customer data from Oracle to PostgreSQL

Other Editors — NPM Package

For GitHub Copilot, Cursor, and other LLM-enabled editors:

# In your Spring Batch project directory
npx sba-init

# For GitHub Copilot
npx sba-init --type github

# For Cursor
npx sba-init --type cursor

Manual Copy

Copy the .claude/ directory from this repository into your project.

Usage

With the Plugin (Claude Code)

After installing the plugin:

1. Natural language — "Help me design a Spring Batch job for customer data migration"
2. Slash command — /sba:sba [optional description]
3. Agent panel — /agents → select sba

With sba-init (Other Editors)

1. Run npx sba-init in your project directory
2. Open your editor and invoke the SBA agent
3. Say: "Use the sba agent to help me design a Spring Batch job"

Quick Commands

During an SBA session:

sba status           Show current state and phase
sba next             Move to next phase
sba back             Return to previous phase
sba skip to {phase}  Jump to a specific phase
sba load skill {name} Load a specific skill
sba generate {artifact} Generate a specific artifact

Token-Efficient Design

Claude Code has context limits. SBA uses progressive loading:

• Only the current phase is loaded (~4–15k tokens per phase)
• Skills are loaded on-demand based on your tech choices
• Templates are loaded just-in-time during implementation

This keeps the agent responsive and focused while having access to a comprehensive knowledge base.

Supported Technologies

Persistence Layers

• JDBC — Direct cursor and paging readers, batch inserts (recommended for large volumes)
• MyBatis — XML and annotation-based mappers
• JPA/Hibernate — With batch optimization patterns

Databases

• PostgreSQL — With specific optimizations (COPY, advisory locks)
• Oracle — Including partitioning and parallel hints
• (MySQL, SQL Server coming soon)

Patterns & Skills

• Chunk-oriented processing
• Tasklet-based steps
• Partitioning (parallel processing)
• Fault tolerance (skip/retry/restart)
• Multi-threaded steps
• Conditional flows
• Job composition
• Listeners and monitoring

Plugin Structure