Intelligent Gemini CLI

A port of Intelligent Claude Code adapted for Gemini CLI, transforming it into an intelligent virtual development team with 14 specialized roles, behavioral framework guidance, and PRB-driven execution.

Installation

Prerequisites

• Node.js and npm
• Gemini CLI
• Ansible (for installation)

Quick Install

make install

This will:

1. Install all behavioral patterns to ~/.gemini/
2. Setup TOML commands in ~/.gemini/commands/
3. Configure settings.json
4. Initialize memory system

MCP Auto-Installation

The Intelligent Gemini CLI supports automatic installation of Multi-Context Prompts (MCPs) defined in mcp-manifest.json at the project root. These MCPs are automatically added to your ~/.gemini/settings.json file during installation.

To define MCPs for auto-installation, create an mcp-manifest.json file in your project's root directory with the following structure:

{
  "mcpServers": {
    "serverName": {
      "command": "path/to/server",
      "args": ["--arg1", "value1"],
      "env": {
        "API_KEY": "$MY_API_TOKEN"
      },
      "cwd": "./server-directory",
      "timeout": 30000,
      "trust": false
    }
  }
}

During the installation process (make install), the system will check for existing MCPs in ~/.gemini/settings.json and only install new ones, ensuring idempotency.

Makefile Parameters for Installation

The make install command supports the following optional parameters:

• MCP_CONFIG_PATH: Specifies an external path to a JSON file containing Model Context Protocol (MCP) server configurations for auto-installation. If provided, this file will be used instead of the default mcp-manifest.json in the project root. Usage: make install MCP_CONFIG_PATH=/path/to/your/mcp-config.json

• API_KEYS_FILE: Specifies a path to a file containing API keys (e.g., KEY=VALUE pairs, one per line) that should be substituted into the installed settings.json file. This allows for secure handling of sensitive credentials. Usage: make install API_KEYS_FILE=/path/to/your/api-keys.env

Uninstall

# Conservative (preserves user data)
make uninstall

# Complete removal
make uninstall FORCE=true

Features

Virtual Team System

• 14 Core Roles: @PM, @Architect, @Developer, @DevOps-Engineer, and more
• Dynamic Specialists: Auto-create domain experts for <70% capability matches
• Role Activation: Every @Role mention triggers appropriate specialist

PRB Engine

Product Requirement Blueprints with 5 complexity tiers:

• Nano (0-2): Trivial one-line changes
• Tiny (3-5): Simple single-file tasks
• Medium (6-15): Standard multi-file features
• Large (16-30): Complex features with sub-PRBs
• Mega (30+): System-wide changes

Memory System

• File-based storage in memory/ directory
• Topic-based organization
• Learning capture and retrieval
• Pattern recognition and application

Command System

TOML-based commands in ~/.gemini/commands/:

• /create-prb - Create new PRB
• /search-memory - Search learning patterns
• /analyze-complexity - Analyze work complexity
• /break-down-story - Convert stories to PRBs
• And more...

Project Structure

intelligent-gemini-cli/
├── GEMINI.md                 # Main context file
├── settings.json            # Configuration
├── Makefile                 # Installation system
├── ansible/                 # Ansible playbooks
├── src/
│   ├── behaviors/          # Behavioral patterns
│   ├── roles/             # Role definitions
│   ├── prb-templates/     # PRB templates
│   └── commands/          # TOML commands
├── memory/                 # Learning storage
├── prbs/                   # PRB management
│   ├── ready/             # Ready to execute
│   └── completed/         # Executed PRBs
└── stories/               # User stories

Usage

Basic Commands

# Create a PRB
/create-prb Implement user authentication

# Search memory
/search-memory authentication patterns

# Analyze complexity
/analyze-complexity Add OAuth2 integration

# Break down story
/break-down-story STORY-001

Virtual Team Interaction

# Direct role assignment
@Developer implement the login feature
@PM break down the authentication story
@Architect design the API structure

Configuration

Configuration for the Intelligent Gemini CLI is handled at two levels: global and project-specific.

Global Configuration (settings.json)

Global settings are stored in settings.json. These settings apply to all projects unless overridden by a project-level configuration. You can edit settings.json to configure:

• Autonomy level (L1/L2/L3)
• Git privacy settings
• Branch protection
• Directory paths
• Command locations

Project-Specific Configuration (config.md)

For project-specific settings, you can create a config.md file in the root of your project directory. Settings in this file will override the global settings from settings.json.

The format of the config.md file is as follows:

# Project Configuration

## Setting Name
key: value

## Another Setting
another_key: another_value

For example, to set the theme for a specific project, you can create a config.md file with the following content:

# Project Configuration

## Theme
theme: "light"

Based On

This is a port of Intelligent Claude Code adapted for Gemini CLI.

License

This project is licensed under the MIT License. See the LICENSE file for details.

---

Intelligent Gemini CLI - Virtual Team for Gemini