Agent Council

中文文档

A multi-model AI council CLI that provides consensus-driven decisions using Claude, Codex, and Gemini.

How It Works

Stage 1: Individual Responses
  All agents answer the question in parallel

Stage 2: Peer Rankings
  Each agent evaluates and ranks all responses

Stage 3: Chairman Synthesis
  A designated agent synthesizes the final answer based on all responses and rankings

Installation

# Install globally from npm
npm install -g agent-council

# Or clone and build locally
git clone https://github.com/mylukin/agent-council.git
cd agent-council
npm install
npm link

Claude Code Plugin

Use agent-council as a Claude Code plugin for seamless integration:

# Step 1: Install CLI globally
npm install -g agent-council

# Step 2: Add marketplace (in Claude Code)
/plugin marketplace add mylukin/agent-council

# Step 3: Install plugin
/plugin install agent-council

The plugin provides:

• council agent: Invoke with complex architectural decisions
• council-decision skill: Auto-triggered for design trade-offs

Prerequisites

The following CLI tools must be installed and configured:

• Claude Code (claude)
• Codex CLI (codex)
• Gemini CLI (gemini)

At least one agent must be available. The council works best with multiple agents.

Usage

REPL Mode (Interactive)

# Start interactive mode
agent-council

> What's the best database for real-time analytics?
# Council processes and responds...

> What about cost considerations?
# Follow-up question with conversation history...

Single Question Mode

agent-council "Your question here"

# Options
agent-council "Question" --chairman gemini  # Set chairman (default: gemini)
agent-council "Question" --timeout 60       # Per-agent timeout in seconds
agent-council "Question" --json             # Output as JSON

Slash Commands (REPL Mode)

Command	Description
/help	Show available commands
/agents	List available agents
/chairman [name]	Show or set chairman
/timeout [seconds]	Show or set timeout
/history	Show conversation history
/clear	Clear conversation history
/exit	Exit the REPL

Interactive Controls

During execution, use these keyboard controls:

Key	Action
1, 2, 3	Focus agent N
↑ / ↓	Navigate focus
k	Kill focused agent
ESC	Abort all agents
Ctrl+C	Quit

Development

npm run dev -- "Test question"  # Run without building
npm test                        # Run tests
npm run test:watch              # Run tests in watch mode

License

MIT