A multi-agent debate framework that simulates structured confrontations between opposing technical viewpoints with human-in-the-loop intervention.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install tech-debate
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install tech-debate using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Tech Debate?
The Tech Debate skill is a sophisticated multi-agent framework inspired by Multi-Agent Debate (MAD) research and the TradingAgents project. It facilitates deep technical exploration by pitting a Proponent and an Opponent against each other across multiple phases. Unlike static discussions, this Openclaw Skills integration allows for structured rounds where a neutral Judge monitors the flow, summarizes key points, and identifies areas of convergence or divergence.
This skill is designed to break through the echo chambers of AI responses by forcing the model to argue against itself, surface non-obvious technical trade-offs, and incorporate user feedback in real-time. It transforms a simple query into a dynamic intellectual duel, providing users with a comprehensive view of complex technical topics.
Tech Debate Use Cases
- Comparing competing architectural patterns like Microservices vs. Monoliths.
- Evaluating the pros and cons of specific programming languages or frameworks for a new project.
- Exploring ethical or performance-based dilemmas in AI and software engineering.
- Stress-testing a technical hypothesis by having AI agents find flaws in the logic.
How Tech Debate Works
- Initialization (Phase 0): The Judge generates a framework including the core positions of both sides and three anticipated points of contention.
- Opening Statements: In the first phase, both Proponent and Opponent present their initial arguments (max 300 words each).
- Multi-Round Confrontation: Agents engage in direct rebuttal for a set number of rounds (defaulting to 5), specifically addressing the opponent's previous points.
- Judge Summary & Pause: At the end of each phase, the Judge summarizes new arguments and current consensus, then pauses for user input.
- User Intervention: The user can support a side, add evidence, or introduce a new perspective that both agents must address in the next phase.
- Global Synthesis: Upon conclusion, the Judge extracts a final report featuring consensus points, core differences, and actionable technical recommendations.
Tech Debate Setup
To use this skill within your environment, ensure you have the Openclaw Skills runner configured. You can trigger the debate using specific natural language commands.
# Example triggers for the tech-debate skill
"debate: Kubernetes vs. Serverless"
"conduct a technical debate on Rust vs. Go"
"pro/con discussion on GraphQL"
You can also specify the depth of the debate:
"debate: React vs. Vue with max_rounds_per_phase: 3"
Tech Debate Data Schema & Taxonomy
The skill maintains an organized state throughout the debate lifecycle to ensure consistency across phases.
| Data Component | Description |
|---|---|
| Argument History | A per-agent cumulative log of all points raised to prevent repetition. |
| Phase Metadata | Tracking of current round counts and convergence percentages. |
| User Context | Injected prompts that force agents to acknowledge and respond to user-provided data. |
| Final Report | A structured Markdown document containing consensus, conflicts, and advice. |
Tech Debate Advanced Features
- Convergence Detection: The Judge automatically ends a phase early if arguments become more than 60% repetitive or if core points are fully exhausted.
- Human-in-the-Loop Integration: Mandatory pauses between phases allow users to steer the debate or provide niche technical evidence that agents must acknowledge.
- Dynamic Phase Scaling: Supports user-defined round limits to balance between quick summaries and deep-dive technical post-mortems.
- Cross-Phase Persistence: Agents maintain a 'memory' of the entire debate history, ensuring that later stages build upon previous concessions or rebuttals.
SKILL.md
Loading
Related Openclaw Skills
Code Sandbox Skill
A secure execution environment for Openclaw Skills that enables AI agents to safely run and test code across multiple programming languages.
GPU Monitor for Ollama
A specialized monitoring tool that tracks GPU performance and Ollama model layer offloading in real-time.
OIXA Protocol
OIXA Protocol is a decentralized economic layer on Base Mainnet that enables AI agents to hire each other and earn USDC through a reverse-auction marketplace.
Workflow Engine
A robust Directed Acyclic Graph (DAG) orchestrator for AI agents featuring parallel task execution and state persistence.
Clientify
A comprehensive business management skill for automating CRM, lead tracking, and marketing workflows via the Membrane CLI.
PipiAds TikTok & Facebook Intelligence
An AI-driven competitive intelligence engine for tracking viral ads, discovering winning products, and analyzing competitor store performance on TikTok and Facebook.
Featured*
