A sophisticated dual-agent framework that decouples execution from quality assurance to ensure AI outputs remain faithful to requirements without context drift.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install duel-loop
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 duel-loop 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 Duel-Loop?
Duel-Loop is a specialized framework within Openclaw Skills designed to eliminate the 'emotional' and context-heavy bias that often accumulates in long AI conversations. By utilizing a twin-agent adversarial architecture, it separates the creation process from the verification process. The main agent serves strictly as an orchestrator, while isolated sub-agents handle execution and quality assurance (QA) in sterile environments.
This skill ensures that high-stakes outputs—such as protocol definitions, system rules, and technical strategies—are evaluated against a rigorous 5-dimension, 10-point scoring system. Only content that achieves a perfect score is delivered to the user, effectively replacing manual prompt engineering and human review with a self-evolving architectural solution.
Duel-Loop Use Cases
- Developing complex technical protocols or system rules requiring zero-drift accuracy.
- Automating the generation of high-quality documentation that needs independent verification.
- Creating implementation plans where boundary conditions and edge cases are critical.
- Preventing AI hallucination and bias in long-form dialogue sessions.
- Implementing automated 'Circuit Breakers' for quality control in AI workflows.
How Duel-Loop Works
- Dispatch: The main agent extracts requirements and spawns an isolated execution sub-agent to create the initial draft (v1.md).
- Review: A separate QA sub-agent evaluates the draft based on five dimensions: Root Cause Depth, Completeness, Domain Independence, Operability, and Side-effect Assessment.
- Accept/Reject: The main agent checks the score; a perfect 10/10 moves to approval, while anything lower triggers a rejection.
- Revision: Upon rejection, the feedback is passed back to a revision sub-agent to produce a v2.md, incrementing the version until quality is met.
- Circuit Break: If 10/10 quality isn't reached within 5 iterations, the loop breaks and prompts the user for manual intervention with a full diagnostic report.
Duel-Loop Setup
To integrate this skill into your Openclaw environment, ensure the following directory structure and template are present:
# Navigate to your skills directory
cd /root/.openclaw/workspace/skills/duel-loop
# Ensure the templates directory exists
mkdir -p templates
# The system will automatically read from:
# templates/qa-review-prompt.md
Trigger the workflow in your chat using keywords like "Run QA", "Review", or "Audit".
Duel-Loop Data Schema & Taxonomy
The skill manages data through a versioned draft system and template injection.
| File Path | Description |
|---|---|
skills/duel-loop/SKILL.md |
Core logic and protocol definition |
skills/duel-loop/templates/qa-review-prompt.md |
The blueprint for the QA sub-agent |
drafts/<name>-v<n>.md |
Iterative draft files stored in the workspace |
Metadata Taxonomy:
- {{DRAFT_PATH}}: Absolute path to the current draft.
- {{ROUND}}: Current iteration count (1-5).
- {{TASK_BRIEF}}: A 200-character summary of user requirements.
Duel-Loop Advanced Features
- Role Isolation: Sub-agents have zero visibility into the main session history, preventing context contamination.
- 5-Dimension Scoring: Quantitative analysis of root cause, boundaries, and operability.
- Automated Circuit Breakers: Prevents infinite loops and wasted tokens by halting after 5 failed quality gates.
- Multi-Agent Orchestration: Uses
sessions_spawnwith specific runtimes to maintain a clean execution environment.
SKILL.md
Loading
Related Openclaw Skills
Xquik X (Twitter) Scraper & Automation Skill
A comprehensive toolset for programmatic interaction with X (Twitter), enabling everything from real-time monitoring and bulk data extraction to AI-assisted content composition.
Xquik X (Twitter) Scraper & Automation
A comprehensive automation and data extraction skill for X (Twitter) providing 111 REST API endpoints and MCP tools for AI agents.
Xquik: Advanced X (Twitter) Data Scraper & Automation Skill
A comprehensive API integration for X (Twitter) enabling deep data extraction, automated account management, and real-time monitoring through Openclaw Skills.
Xquik Twitter Scraper & Automation
A comprehensive X (Twitter) integration suite providing 111 REST API endpoints and MCP tools for search, automation, and bulk data extraction.
Video Editing for Beginners
A cloud-powered AI skill that transforms raw video clips into polished, ready-to-post MP4 files using simple natural language commands.
OSINT Daily Brief
An automated intelligence gathering skill that generates structured daily reports on any target using web search, DNS, and Shodan data.
Featured*
