Rierino Development

Rierino — Claude Code Plugin

Connect Claude Code to the Rierino low-code development platform. Drop this folder into your project and Claude gains full awareness of the platform — skills, agents, commands, and an authenticated MCP proxy — all with no build step.

Architecture

Claude Code
    │
    ├── reads skills/global/CONVENTIONS.md       (naming & data standards)
    ├── reads skills/global/RULES.md             (operational rules)
    ├── reads skills/<skill>/SKILL.md            (per-task instructions)
    │
    ├── runs /commands/rierino-status            (health check)
    ├── runs /commands/rierino-mcp               (list MCP capabilities)
    │
    └── calls MCP tools ──► whatever the Rierino server registers
                                  │
                        Token vault (AES-256-GCM)
                        Auth layer (JWT login + auto-refresh)
                                  │
                          Rierino MCP Server

The MCP server (proxy.js) is a transparent authenticated proxy — it forwards every JSON-RPC call straight to the Rierino server. Tools are discovered at runtime via tools/list; no tool definitions live in this plugin.

Tokens are encrypted in-memory with an ephemeral AES-256-GCM key. Claude only ever sees opaque masked handles — real JWTs never leave the proxy process.

---

Setup

1. Configure credentials

Security note: Do not store credentials in servers/rierino-mcp/.env or anywhere inside the plugin folder. Claude Code can read files within the project directory, which would expose your username and password in Claude's context. Use one of the two approaches below instead.

Option A — env block in .mcp.json (recommended)

Pass credentials directly in your .mcp.json (or claude_desktop_config.json). This file is read by the Claude Code harness to spawn the MCP server and is not passed to Claude as conversation context.

{
  "mcpServers": {
    "rierino": {
      "command": "node",
      "args": ["[PATH]/servers/rierino-mcp/proxy.js"],
      "env": {
        "RIERINO_BASE_URL": "https://your-instance.rierino.com",
        "RIERINO_USERNAME": "your-username",
        "RIERINO_PASSWORD": "your-password"
      }
    }
  }
}

Add .mcp.json to your .gitignore so credentials are never committed.

Option B — .env file outside the project

Store a .env file in a directory that is not inside your project tree (e.g. ~/.config/rierino/.env) and point cwd to that directory:

{
  "mcpServers": {
    "rierino": {
      "command": "node",
      "args": ["[PATH]/servers/rierino-mcp/proxy.js"],
      "cwd": "/home/you/.config/rierino"
    }
  }
}

proxy.js loads .env from its working directory, so this keeps credentials entirely outside any folder Claude can browse.

2. Project root .env

Create a .env file in your project root (next to .mcp.json) with:

RIERINO_UI_BASE_URL=https://your-instance.rierino.com

This is used by Claude to generate direct links to UI records (e.g. ${RIERINO_UI_BASE_URL}/app/design/common/ui?id={id}). This file contains no credentials — only a public base URL — so it is safe to keep inside the project.

Add .env to your .gitignore.

No npm install or build step — proxy.js has zero external dependencies.

3. Verify

> /rierino-status
> /rierino-mcp tools

---

What's included

rierino-claude-plugin/
├── .env                                    UI url (gitignore this)
├── .claude-plugin/
│   └── plugin.json                         Plugin manifest (v1.0.0)
│
├── servers/rierino-mcp/
│   └── proxy.js                            MCP stdio proxy (zero dependencies)
│
├── commands/
│   ├── rierino-status.md                  /rierino-status — health check
│   └── rierino-mcp.md                     /rierino-mcp — list MCP capabilities
│
├── agents/
│   ├── ai_agent_assistant.md              AI agents
│   ├── code_assistant.md                  Groovy & JavaScript scripting
│   ├── component_builder.md               React editor components
│   ├── drools_assistant.md                Drools rule (DRL) generation
│   ├── element_assistant.md               SYSTEM Element configurations
│   ├── jmespath_assistant.md              JMESPath expressions
│   ├── lister_builder.md                  React list/lister components
│   ├── query_builder.md                   Structured query definitions
│   ├── saga_assistant.md                  Saga flows (events, APIs, workflows)
│   ├── schema_builder.md                  JSON Schema (2020-12)
│   ├── template_assistant.md              Handlebars storefront templates
│   ├── test_data_generator.md             Mock JSON test data
│   ├── ui_assistant.md                    Screen/UI configuration
│   └── workflow_explainer.md              Plain-language saga flow explanations
│
└── skills/
    ├── global/
    │   ├── CONVENTIONS.md                 Naming & data-type standards
    │   └── RULES.md                       Operational rules (all skills)
    ├── ai_agent_assistant/SKILL.md
    ├── code_assistant/SKILL.md
    ├── component_builder/SKILL.md
    ├── drools_assistant/SKILL.md
    ├── element_assistant/SKILL.md
    ├── jmespath_assistant/
    │   ├── SKILL.md
    │   └── RIERINO_FUNCTIONS.md           122 custom JMESPath functions
    ├── lister_builder/SKILL.md
    ├── query_builder/SKILL.md
    ├── saga_assistant/SKILL.md
    ├── schema_builder/SKILL.md
    ├── template_assistant/
    │   ├── SKILL.md
    │   └── RIERINO_HELPERS.md             40+ Handlebars helpers
    ├── test_data_generator/SKILL.md
    ├── ui_assistant/SKILL.md
    └── workflow_explainer/SKILL.md

---

MCP tools

Tools are registered by the Rierino server itself and discovered dynamically at runtime. Run /rierino-mcp tools to see the current list.

The proxy forwards all JSON-RPC calls transparently — adding auth headers and token masking — without defining or restricting which tools are available.

---

Skills

All skills follow a common pattern:

• Read CONVENTIONS.md and RULES.md for global standards
• Read the skill's SKILL.md for task-specific instructions
• Read any references/ files before making API calls
• Return raw output only (no markdown fences unless specified)

Skill	Trigger / purpose
ai_agent_assistant	AI agent configurations
code_assistant	Groovy scripts and JavaScript event handlers
component_builder	React editor components for the page builder
drools_assistant	Drools Rule Language (DRL) rule files
element_assistant	SYSTEM Element configs for integrations
jmespath_assistant	JMESPath expressions for JSON transformation
lister_builder	React list/lister display components
query_builder	Structured query definitions (SIMPLE/AGGREGATION/PIPELINE)
saga_assistant	Saga flow definitions (events, APIs, workflows)
schema_builder	JSON Schema (2020-12) definitions
template_assistant	Handlebars templates for storefronts
test_data_generator	Realistic mock JSON test data
ui_assistant	Screen/UI configuration JSON (12-column grid, tabs)
workflow_explainer	Plain-language explanations of saga flows

---

Agents

Agents are autonomous and run with model: sonnet, maxTurns: 25. Each agent's system prompt delegates to its matching skill.

Agent	Purpose
ai_agent_assistant	Design AI agents with instructions, models
code_assistant	Generate and iterate on Groovy/JS scripts
component_builder	Build complete React editor components
drools_assistant	Write and validate DRL rules
element_assistant	Scaffold SYSTEM Elements from catalog
jmespath_assistant	Build and test JMESPath transformation pipelines
lister_builder	Build React list components with action handlers
query_builder	Compose queries with filters, aggregations, pipelines
saga_assistant	Design and iterate on saga flow configurations
schema_builder	Design schemas with proper structure and conventions
template_assistant	Build defensive, interactive Handlebars templates
test_data_generator	Generate sets of varied, realistic test records
ui_assistant	Generate full screen configurations from description
workflow_explainer	Explain saga flows in plain, human-readable language

---

Commands

Command	Description
/rierino-status	GET /actuator/health + Ping — confirms connectivity
/rierino-mcp	Calls tools/list, resources/list, etc. on the MCP server

---

Auth flow

Request → cached JWT valid? ──► yes → use it
                            └── no  → refresh token valid? ──► yes → refresh → retry
                                                             └── no  → basic auth login
401 response → clear tokens → re-auth → retry once

Tokens are encrypted with a randomly generated AES-256-GCM key on startup. Outbound responses have real JWTs replaced with opaque [TOKEN:handle] markers. Inbound requests from Claude have those markers swapped back before forwarding.

---

Customization checklist

• Set RIERINO_BASE_URL, RIERINO_USERNAME, RIERINO_PASSWORD in .mcp.json env block (or an out-of-project .env)
• Set RIERINO_UI_BASE_URL in the project-root .env
• Add .mcp.json and .env to .gitignore
• Set the proxy.js path in .mcp.json

---

Troubleshooting

Problem	Fix
Login failed (401)	Check RIERINO_USERNAME / RIERINO_PASSWORD in .mcp.json env block
python3: command not found	Use node -e or jq instead when parsing tokens in shell
Tools not appearing in Claude	Check .mcp.json path; run /rierino-mcp tools
Token errors after restart	Expected — ephemeral key regenerated each run, re-login auto