An adversarial multi-agent engine that orchestrates AI personas with conflicting worldviews to stress-test decisions and strategies through structured debate.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install red-team
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 red-team 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 Red Team - Adversarial Debate Engine?
Red Team is a sophisticated framework designed to eliminate groupthink and validate high-stakes decisions by orchestrating a panel of AI agents with conflicting worldviews. By utilizing Openclaw Skills, this engine allows users to pit personas like The Bull, The Bear, and The Contrarian against one another in a structured, multi-round debate format. This process ensures that any proposal, investment, or strategy is viewed through multiple analytical lenses before a final decision is reached.
The primary purpose of this tool is to provide a rigorous "what could go wrong" analysis and pre-mortem exercise for developers and strategists. Instead of relying on a single AI response, the Red Team skill synthesizes a broad spectrum of perspectives, from execution-focused pragmatism to risk-averse capital preservation, providing a more balanced and resilient conclusion.
Red Team - Adversarial Debate Engine Use Cases
- Evaluating major financial investments or $50k+ deals by analyzing risk versus reward.
- Stress-testing product launch strategies and feature roadmaps against customer demand and technical scalability.
- Identifying legal, regulatory, or compliance risks in strategy documents using specialized personas.
- Conducting adversarial "go/no-go" calls for critical infrastructure or software deployment.
- Challenging organizational groupthink by introducing contrarian and historical perspectives into the planning phase.
How Red Team - Adversarial Debate Engine Works
- The user defines a specific question or decision and provides optional context files to the Openclaw Skills script.
- The system initializes a set of AI personas—such as the operator, economist, or regulator—each with a unique system prompt and worldview.
- Each persona generates an independent initial proposal based on their specific priorities and the provided context.
- The engine orchestrates a series of critique rounds where agents review and challenge the logic of their peers.
- After the critique phase, agents update their positions and provide conviction scores (0-100) to quantify their confidence.
- A neutral synthesis agent processes the entire debate to produce a comprehensive decision brief, including a risk matrix and final recommendation.
Red Team - Adversarial Debate Engine Setup
To use this skill, ensure you have one of the supported CLI backends installed, such as Claude Code:
npm i -g @anthropic-ai/claude-code
You can then run a basic 3-persona debate directly via the terminal:
python3 ~/.openclaw/skills/red-team/scripts/red-team.py --question "Should we adopt this new tech stack?" --personas "bull,bear,operator"
Red Team - Adversarial Debate Engine Data Schema & Taxonomy
The skill organizes its output into a highly structured Markdown document for easy review. The data follows this taxonomy:
| Component | Contents |
|---|---|
| Initial Proposals | The independent, uninfluenced takes from each active agent. |
| Critique Rounds | Step-by-step logs of agents identifying flaws in each other's arguments. |
| Conviction Scores | A table of quantitative scores showing the final confidence of each persona. |
| Synthesis Brief | A summary containing consensus points, key disagreements, and a risk matrix. |
| Recommendation | A final, synthesized "go" or "no-go" path based on the debate outcome. |
Red Team - Adversarial Debate Engine Advanced Features
- Custom Persona Support: Define specialized stakeholders in a JSON file to tailor the debate to your specific industry.
- Multi-Backend Integration: Seamlessly switch between Claude, Codex, or Gemini backends depending on your active subscriptions.
- Configurable Depth: Adjust the intensity of the analysis by increasing the number of critique rounds for deeper stress testing.
- Context Ingestion: Feed the engine existing project documentation or Obsidian notes to ground the debate in real-world data.
- Automated Risk Matrix: Generates a synthesized view of potential failure points identified during the adversarial process.
SKILL.md
Loading
Related Openclaw Skills
TooToo
Sync your personal TooToo codex to monitor and enforce AI agent alignment with your core values and principles.
food402 - TGO Yemek Delivery
A comprehensive AI agent skill for searching restaurants and ordering food from Trendyol GO in Turkey.
XHS Enhancer
A 24/7 autonomous optimization engine designed to continuously improve and refine Xiaohongshu content performance using AI-driven logic.
Wavespeed NanoBanana2 Text-to-Image Skill
A high-performance text-to-image skill for creating 1k, 2k, and 4k resolution visuals via the Wavespeed NanoBanana2 API.
Guide Creator
Guide Creator is a specialized documentation system that generates and maintains standardized project guides to ensure seamless AI context recovery and development continuity.
Presentation & Technical Content Generator
An automated system for generating professional Reveal.js slide decks and technical articles directly from source code and research papers.
Featured*
