GemKit CLI

A command-line interface for working with Gemini and Claude AI agents, plans, and sessions.

Installation

npm install -g gemkit-cli

Usage

gk <command> [options]

Global Options

• --verbose - Enable verbose output
• --json - Output as JSON
• --help - Show help
• --version - Show version

Commands

Command	Description
gk init	Initialize GemKit in your project
gk new	Create new agents, plans, or skills
gk agent	Manage and spawn AI agents
gk team	Multi-agent team coordination
gk office	Agent Office visualization dashboard
gk session	View and manage sessions
gk plan	Work with execution plans
gk tokens	Analyze token usage and costs
gk config	Manage configuration
gk cache	Manage local cache
gk update	Update GemKit components
gk versions	Show available versions
gk doctor	Diagnose issues
gk extension	Manage extensions
gk catalog	Browse available kits
gk convert	Convert between formats
gk paste	Handle image and video pasting

Agent Commands

One-Shot Mode (Spawn)

Fire-and-forget agent execution that blocks until completion.

Command	Description
gk agent list	List all available agent profiles
gk agent info <name>	Show agent profile details
gk agent search "<task>"	Find best agent+skills for a task
gk agent spawn -p "<prompt>"	Spawn a sub-agent

Spawn Options:

• -a, --agent <name> - Agent profile name
• -s, --skills <list> - Comma-separated skills to inject
• -c, --context <files> - Context files (@file syntax)
• -m, --model <model> - Model override
• -t, --tools <list> - Comma-separated tools to auto-approve
• --cli <provider> - CLI provider: gemini (default) or claude
• --music - Play elevator music while waiting

Interactive Mode

Persistent AI session for multi-turn conversations with tool approval control.

Command	Description
gk agent start	Start interactive session
gk agent send "<prompt>"	Send prompt to session
gk agent wait [timeout]	Wait for completion (default: 120s)
gk agent pending	Check pending tool confirmations
gk agent exchange	Get structured JSON output
gk agent read [lines]	Read raw terminal output
gk agent status	Check session status
gk agent stop	Stop session

Start Options:

• -a, --agent <name> - Agent profile name
• -s, --skills <list> - Comma-separated skills
• -c, --context <files> - Context files (@file syntax)
• -m, --model <model> - Model override
• -t, --tools <list> - Comma-separated tools to allow
• --cli <provider> - CLI provider: gemini (default) or claude

CLI Providers:

• gemini - Uses Gemini CLI (default). Loads from .gemini/agents/
• claude - Uses Claude CLI. Loads from .claude/agents/
• Models and tools are automatically mapped between providers

Team Commands

Coordinate multiple AI agents working in parallel on complex tasks.

Team Management

Command	Description
gk team create <name>	Create a new team
gk team list	List all teams
gk team info [teamId]	Show team details
gk team kill [teamId]	Emergency shutdown
gk team cleanup	Clean up stale resources
gk team ports	Show port allocations
gk team reset	Delete all team data

Task Management

Command	Description
gk team task-create "<subject>"	Create a task
gk team task-claim <taskId>	Claim a task
gk team task-done <taskId>	Mark task completed
gk team tasks [teamId]	List all tasks

Agent Spawning

Command	Description
gk team start --name <agent>	Spawn agent as team member
gk team start --name <agent> -a <profile>	Spawn with explicit profile
gk team start --name <agent> --cli claude	Use Claude CLI

Messaging & Inbox

Command	Description
gk team send <agent> "<message>"	Send message to member
gk team broadcast "<message>"	Send to all members
gk team messages	View central inbox
gk team messages --pending	View pending items
gk team respond <msgId> --approve	Approve request
gk team respond --approve-all	Approve all pending

Agent Interaction

Command	Description
gk team exchange <agent>	Get structured output
gk team read <agent>	Read raw output

Agent Office Commands

Command	Description
gk office start	Start the web visualization dashboard
gk office status	Show current office state
gk office watch	Watch office state changes in real-time

Options:

• -p, --port <n> - Web server port (default: 3847)
• --no-open - Don't auto-open browser
• --json - Output as JSON

Quick Start

Single Agent (One-Shot)

# Initialize in your project
gk init

# Spawn an agent with a task
gk agent spawn -p "Help me write a function"

# Spawn with specific agent profile and skills
gk agent spawn -a researcher -s "frontend-design" -p "Build a dashboard"

# Spawn with Claude CLI instead of Gemini
gk agent spawn --cli claude -a researcher -p "Analyze the codebase"

# Search for the best agent for a task
gk agent search "implement user authentication"

Single Agent (Interactive)

# Start interactive session
gk agent start -a researcher -s research

# Send a prompt
gk agent send "Research JWT best practices"
gk agent wait

# Check for tool approvals
gk agent pending
gk agent send "2"  # Approve for session

# Get structured output
gk agent exchange

# Stop when done
gk agent stop

Multi-Agent Team

# Create a team
gk team create my-project --desc "Feature implementation"

# Create tasks
gk team task-create "Research API patterns" --desc "Investigate REST vs GraphQL"
gk team task-create "Implement backend" --desc "Build the API"

# Spawn agents
gk team start --name researcher-1 &
gk team start --name developer-1 &

# Wait for initialization, then assign work
gk team send researcher-1 "Claim and complete the research task"
gk team send developer-1 "Wait for research, then implement"

# Monitor and approve
gk team messages --pending
gk team respond --approve-all

# Check progress
gk team tasks
gk team exchange researcher-1

# Cleanup when done
gk team kill
gk team reset

Agent Office Dashboard

# Start the Agent Office dashboard
gk office start

# View session history
gk session list

Requirements

• Node.js >= 18

Links

• Website
• GitHub
• Issues

License

MIT