# erupt-skill
English | [简体中文](README.zh.md)
A Claude Code Skill that generates a data admin backend from a single sentence.
Powered by the [erupt](https://www.erupt.xyz) low-code framework: say "build me a library admin backend" and get a complete, runnable admin system — login page, CRUD, search, import/export, and permission management out of the box.
**No dev environment required**: with no system JDK, the skill downloads OpenJDK 25 on demand (Adoptium official source, with a Tsinghua mirror fallback that's fast in China; cached in `~/.erupt-skill`), Maven is prepared automatically, project dependencies go through the Aliyun mirror, and an embedded H2 file database means no MySQL install. The repo itself stays ~1 MB, so cloning is instant. Non-developers can build a data admin backend for any domain with one sentence.
## Install
This skill follows the open [Agent Skills](https://agentskills.io) standard (SKILL.md), supported by most mainstream AI coding tools. Clone the repo into the skills directory of your tool:
| Tool | Skills directory | Notes |
|------|-----------------|-------|
| **Claude Code** | `~/.claude/skills/` (personal) or `.claude/skills/` (project) | Native support, auto-activates by description |
| **Codex CLI** | `~/.codex/skills/` (personal) or `.codex/skills/` (project) | Invoke with `$erupt-admin ...` or auto-activation |
| **Cursor** | `.cursor/skills/` or `.agents/skills/` (project) | See [Cursor Agent Skills docs](https://cursor.com/docs/skills) |
| **Trae** | `.agents/skills/` (project) | Supports the Agent Skills open standard |
| **CodeBuddy** | `.agents/skills/` (project) | Supports the Agent Skills open standard |
Example (Claude Code):
```bash
git clone https://github.com/plinian/erupt-skill.git ~/.claude/skills/erupt-admin
```
For any tool without native skill support, just clone this repo anywhere and tell the agent: *"Read SKILL.md in this repo and follow its workflow to build me a XXX admin backend."*
## Usage
Just say in Claude Code:
```
Build a library admin backend
Create a customer order management system with status filtering
Add a borrow-record module to the backend you just built
```
Then open http://localhost:8080 and log in with `erupt / erupt`.
## Directory Structure
```
SKILL.md # main skill workflow
reference/annotations.md # entity generation spec (decision tables + common pitfalls)
reference/erupt-*.md # official erupt reference docs (annotations, i18n, query, Cube, API, TPL)
template/ # runnable erupt project template (latest erupt release + Spring Boot 3.5 + H2)
└─ resources/public/ # app.js (eruptSiteConfig: title/logo/theme/hooks) · app.css (styles)
scripts/setup-env.sh # prepare JDK + Maven (system JDK → cached → download JDK: Adoptium official, Tsinghua mirror fallback)
scripts/compile.sh # compile-only check (catch syntax errors after code changes)
scripts/run.sh # one-command build and start
scripts/build-m2-seed.sh # optional: build a local dependency seed into vendor/m2 for fully-offline first builds
```
## Requirements
- macOS / Linux / Windows (on Windows run inside Git Bash, which Claude Code ships with)
- Network access (first run downloads Maven and project dependencies; with no system JDK it also downloads OpenJDK 25 once, then caches it)