oopforge

OOPforge

AI ships the feature. OOPforge keeps the architecture.

Harness engineering that stops vibe coding from wrecking your backend.

Forge small. Compose forever. OOPforge defines OOP/DDD as a dialect your agent follows — skills are the grammar, hard rules are the lint, runnable examples/ are the reference, and install + commands are the runtime. A methodology pack plus agent harness, not a general agent framework.

It gives Claude Code, Codex CLI, Cursor, and compatible agents a clear way to design around domain models, aggregates, ports, adapters, and testable use cases before writing code.

Specialized for Java (Spring) and Python (FastAPI) — pick 3-tier (Controller/Service/Repository) or hexagonal/clean, with OpenAPI/Swagger built in.

English · 한국어

---

Quickstart

1. Install

bash -c "$(curl -fsSL https://raw.githubusercontent.com/LooSung/oopforge/main/scripts/setup/bootstrap.sh)"

2. Restart your agent

Restart Claude Code, Codex CLI, or your Cursor Agent CLI session so it picks up the new skills and commands.

Cursor: load the pack explicitly:

cursor-agent --plugin-dir ~/.oopforge

3. Run Craft

Entry point is Craft on every harness; only how you invoke it differs:

Harness	Invoke
Claude Code	/oopforge:craft <request> — registered slash command
Codex CLI	/skills → pick oopforge, then prompt without a leading / (Codex reserves / for its own commands)
Cursor Agent CLI	After --plugin-dir, use natural language (see Cursor setup)

Claude Code:

/oopforge:craft Add a single Email value object

Codex CLI (after /skills → oopforge):

Use OOPforge craft: Add a single Email value object

Cursor:

Use OOPforge craft: Add a single Email value object

---

Advanced Usage

For advisory requests, Craft recommends the smallest path without implementing (same on every harness).

Advanced users may ask Craft to start at a specific workflow stage such as Discovery, Design, Delivery Plan, Skeleton, Implement, Test, or Refactor.

Stack identifiers:

Stack	Architecture	When
java-spring-layered	3-tier (Controller/Service/Repository)	Small services, MVP
java-spring-hexagonal	Hexagonal (domain/application/adapter)	Complex domain, many adapters
python-fastapi-layered	3-tier (Router/Service/Repository)	Small services, MVP
python-fastapi-clean	Clean (domain/application/infrastructure/presentation)	Complex domain

All backend skeletons ship with OpenAPI/Swagger UI enabled by default (springdoc / FastAPI built-in).

Already installed? See Installation for manual setup, updates, and troubleshooting.

Harness guides: Claude Code · Codex · Cursor

---

How to use OOPforge

New to the workflow? Follow the step-by-step library loan guide:

Library loan walkthrough →
Discovery → Design → Skeleton → Implement (Java + Python) → Test

Guide index: EN · KO

Resource	Purpose
Library loan guide	Full tutorial — how to use OOPforge end to end
Examples index	Runnable proof — same calculator, 6 examples
calculator-java-layered · calculator-java-hexagonal · calculator-java-hexagonal-cqrs	Java 3-tier · hexagonal · hexagonal + CQRS
calculator-python-layered · calculator-python-hexagonal · calculator-python-hexagonal-cqrs	FastAPI 3-tier · hexagonal · hexagonal + CQRS
Reviewer checklist	Post-implement rule check

Each workflow stage ends with a human checkpoint — do not skip ahead.

---

Why OOPforge

OOPforge is a DDD / OOP specialized AI engineering pack — not a general agent framework. Think of it as harness engineering for an OOP dialect: skills encode grammar, hard rules act as lint, examples are reference implementations, and install/commands wire agents to the runtime.