A sophisticated multi-agent orchestration pattern that spawns parallel sub-agents to debate competing approaches and converge on the best technical solution.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install agent-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 agent-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 Agent Debate Skill?
The Agent Debate skill implements a structured Competing Hypotheses pattern, enabling your AI swarm to simulate a high-level technical committee. Instead of relying on a single model's output, this skill spawns multiple independent sub-agents to research, argue, and defend specific positions or architectural choices. By using file-based coordination, the skill ensures that every perspective is documented and reviewed by a separate synthesis agent, providing a rigorous check against bias and shallow reasoning. This is a vital component for power users of Openclaw Skills who require high-confidence outcomes for mission-critical tasks.
This workflow is particularly effective when dealing with non-obvious problems where multiple valid solutions exist. By assigning specific roles—such as an advocate for a specific technology, a devil's advocate, or a security-focused attacker—the system surface-levels risks and benefits that a single-pass prompt might miss. The resulting consensus is backed by evidence and adversarial testing, making it ideal for professional development environments.
Agent Debate Skill Use Cases
- Architecture decisions where multiple frameworks or databases are under consideration.
- Complex debugging scenarios where the root cause is unclear across different operating systems or environments.
- Strategic financial or trading analysis requiring bull, bear, and neutral perspectives.
- Security and robustness reviews using a Red Team pattern (Builder vs. Attacker).
- Hackathon or rapid prototyping sessions where the optimal implementation path must be selected quickly.
How Agent Debate Skill Works
- The lead agent defines the core question and initializes a dedicated directory within the plans folder.
- Between two and four sub-agents are spawned simultaneously to investigate different facets of the problem.
- Each sub-agent writes its findings and evidence to a specific markdown file (e.g., agent-1.md).
- If a thorough analysis is required, a rebuttal round is triggered where agents critique the positions of their peers.
- A high-reasoning synthesis agent evaluates all inputs, scores them on feasibility and risk, and writes a final verdict.
- The lead agent reviews the synthesis within the Openclaw Skills ecosystem and proceeds with the winning strategy.
Agent Debate Skill Setup
To begin using this skill within your project, ensure you have a plans directory available for file-based coordination. Use the following structure to initiate a debate:
# Create a directory for the specific debate topic
mkdir -p plans/debate-{topic-name}/
# Document the question to be debated
echo "Describe your technical dilemma here" > plans/debate-{topic-name}/question.md
Configure your agents to use high-reasoning models like Opus 4.6 for synthesis to ensure the most nuanced weighing of arguments.
Agent Debate Skill Data Schema & Taxonomy
The skill uses a structured file hierarchy for transparent decision tracking:
| File Name | Purpose |
|---|---|
question.md |
Defines the scope and parameters of the debate. |
agent-{n}.md |
Primary position papers written by independent sub-agents. |
rebuttal-{n}.md |
Secondary files containing critiques of opposing positions. |
synthesis.md |
The analytical verdict including scores for risk, speed, and quality. |
decision.md |
The final record of the lead agent's chosen path. |
Agent Debate Skill Advanced Features
- Support for Two-Round thoroughness including automated rebuttal cycles.
- Adversarial Red Team mode for stress-testing codebases and infrastructure.
- Flexible model assignment allowing for cost-optimization between reasoning-heavy and throughput-heavy models.
- Native integration with other Openclaw Skills like Swarm and Architect for end-to-end automation.
- Human-in-the-loop compatibility, allowing developers to review and influence the debate at any stage.
SKILL.md
Loading
Related Openclaw Skills
Rei Qwen3 Coder
A specialized skill for integrating Rei Qwen3 Coder as a model provider with automated setup and switching capabilities.
ERCData
ERCData is a blockchain-integrated skill for storing, verifying, and managing AI data fingerprints on the Base network using the ERCData standard.
isnad-scan
A specialized security scanner that audits AI agent skills for 69+ threat patterns, including code injection and credential theft.
yt-digest
An efficient AI skill for extracting transcripts, summaries, and key highlights from YouTube content.
Answers - AI Grounding
A high-performance skill providing AI-grounded answers through single-search or multi-iteration deep research modes.
Arena Social Skill
A comprehensive social media automation tool for AI agents to interact with the Stars Arena platform via the Agent API.
Featured*
