About agent-tower-plugin
Agent Tower Plugin is a multi-agent collaboration skill for Claude Code, coordinating Claude, Codex, and Gemini to provide different perspectives on tasks. It offers parallel council, pro/con debate, and producer-reviewer workflows, with roles suggested to fit the task.
Key features
- 01Coordinate multiple coding agents
- 02Gather and synthesize parallel opinions
- 03Run pro/con debates with a judge
- 04Iterate toward consensus through review
- 05Suggest task-specific personas
Use cases
- 01Compare technology choices and trade-offs
- 02Evaluate strategies or complex questions
- 03Review code and refine it iteratively
Install
Copy the command and run it in a terminal.
npx skills add https:/After installing, type/agent-tower-pluginin Claude Code to confirm it loaded.
Bundled files
References and scripts installed along with SKILL.md.
requirements.txt.gitignoreREADME.md.claude-plugin/marketplace.jsonplugin.json.github/workflows/test.ymlcommands/agents.mddebate.mddeliberate.mdcouncil.mdscripts/run_council.pylist_agents.pyrun_debate.pyrun_deliberate.pyscripts/lib/__init__.pybase.pyclaude_backend.pycouncil.pycodex_backend.pyfeedback.pydebate_mode.pydeliberation_mode.pycouncil_mode.pygemini_backend.pypersonas.pyregistry.pyresponse.pytests/test_codex_command.py__init__.pytest_registry.pyDetails
From the project READMEAgent Tower Plugin
Multi-agent deliberation for Claude Code. Orchestrate multiple AI coding assistants (Claude, Codex, Gemini) to get diverse perspectives on tasks.
Interactive Configuration
Before running any multi-agent workflow, use AskUserQuestion to gather configuration preferences from the user. This ensures the user has control over:
- Number of agents/rounds
- Consensus thresholds
- Agent role assignments
- Other mode-specific options
Skip questions for any options the user explicitly provided in their command.
Available Skills
/tower:council
Multi-agent council with parallel opinions and synthesis
First, analyze the user's question and suggest relevant perspectives via AskUserQuestion. Then run multiple agents in parallel with the selected personas. Agents anonymously rank each other's responses, and a chairman synthesizes the final answer.
Best for: Evaluations, strategy decisions, comprehensive analysis, general knowledge questions
/tower:debate
Adversarial debate with pro/con agents
Two agents argue opposing positions on a binary decision. A judge evaluates the arguments and declares a winner with scores.
Best for: Binary decisions, trade-off analysis, technology choices
/tower:deliberate
Producer/reviewer consensus loop
A producer generates a response, a reviewer provides structured feedback, and they iterate until reaching consensus or hitting max rounds.
Best for: Code review, document refinement, iterative improvement
/tower:agents
List available agents
Shows which agent backends (Claude, Codex, Gemini) are currently available.
Agent Backends
| Agent | CLI | Default Model |
|---|---|---|
| claude | claude -p | opus |
| codex | codex exec | default |
| gemini | gemini | gemini-3-pro-preview |
Dynamic Personas
Analyze the user's question and suggest relevant personas. Use technical personas for code/architecture questions, generalist personas for general knowledge questions.
Technical: Security Analyst, Systems Architect, Code Quality Reviewer, DevOps Engineer, Data Engineer, Devil's Advocate
Generalist: Research Analyst, Local Expert, Critical Thinker, Practical Advisor
Business: Business Strategist, Product Manager, Financial Analyst, UX Designer
Custom personas can be passed via --personas '[{"name":"Expert Name","focus":"focus area"}]'
Examples
/tower:council "Evaluate this startup idea: AI-powered meal planning"
/tower:debate "Should we use microservices or a monolith?"
/tower:deliberate "Review the security of ~/GH/myproject"
/tower:agents