Agent Tower Plugin

Agent Tower Plugin for Claude Code

Multi-agent deliberation plugin for Claude Code. Orchestrate multiple AI coding assistants (Claude, Codex, Gemini) to get diverse perspectives on tasks.

Installation

Quick Install (Recommended)

From Marketplace

From GitHub

From Local Directory

Requirements

At least 2 of these CLI tools must be installed and accessible:

- Claude Code CLI: claude --version
- Codex CLI: codex --version
- Gemini CLI: gemini --version

Skills

/tower:council - Multi-Agent Council

Runs multiple agents in parallel, each providing an independent opinion. Before running, Claude analyzes your question and suggests relevant perspectives (personas) via an interactive prompt. Agents anonymously rank each other's responses, and a chairman synthesizes the final answer.

Options:
- --agents N: Use N agents (default: all available)
- --personas JSON: Custom personas as JSON array
- --no-personas: Disable persona assignment
- -v, --verbose: Show detailed progress

/tower:debate - Adversarial Debate

Two agents argue opposing positions on a binary decision. A judge evaluates and declares a winner.

Options:
- --rounds N: Number of argument rounds (default: 3)
- --pro-agent NAME: Agent for PRO position
- --con-agent NAME: Agent for CON position
- --judge-agent NAME: Agent to judge
- -v, --verbose: Show detailed progress

/tower:deliberate - Producer/Reviewer Loop

A producer generates a response, a reviewer provides feedback, and they iterate until consensus.

Options:
- --max-rounds N: Maximum rounds (default: 5)
- --threshold X: Consensus threshold 0.0-1.0 (default: 0.85)
- --producer NAME: Agent for producer role
- --reviewer NAME: Agent for reviewer role
- -v, --verbose: Show detailed progress

/tower:agents - List Available Agents

Check which agent…