An AI agent skill that implements the Generator-Evaluator dual-agent architecture to eliminate self-evaluation bias through independent verification and automated feedback loops.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install double-agent
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 double-agent 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 DoubleAgent?
The DoubleAgent skill is a sophisticated framework designed to solve the problem of AI self-evaluation bias in software development. By separating the roles of creation and validation, this pattern ensures that generated code, UIs, and content are objectively tested against the original specification. It leverages the Openclaw Skills ecosystem to provide developers with a robust template for building multi-agent pipelines where an independent Evaluator agent validates the work of a Generator agent.
Instead of an agent grading its own homework, DoubleAgent introduces a dedicated Evaluator that uses tools like Playwright to interact with artifacts just like a human user would. This creates a high-fidelity feedback loop that significantly improves the reliability and quality of AI-generated outputs, ensuring they meet professional standards before deployment.
DoubleAgent Use Cases
- Designing AI-generated frontend UIs that require functional interaction testing through a browser.
- Implementing complex multi-step workflows such as multi-page forms or navigation flows.
- Validating API endpoint implementations against strict technical specifications and edge cases.
- Generating high-quality documentation or reports using text-based evaluation agents to ensure accuracy.
- Upgrading existing coder-tester workflows to use real-world interaction-based validation rather than just static analysis.
How DoubleAgent Works
- The user defines a concrete, observable Spec Contract that serves as the single source of truth for both agents.
- The Generator Agent produces an initial output, such as code or a UI component, based solely on the spec and any previous iteration history.
- The Evaluator Agent independently interacts with the output artifact using tool integrations like Playwright to verify functional requirements.
- The Evaluator calculates a score based on a weighted rubric covering dimensions like functional completeness and interaction quality.
- If the score is below the defined threshold, the system triggers an iteration loop, feeding the evaluator's structured feedback back to the generator for refinement.
- The process repeats for a set number of rounds or until the quality threshold is satisfied.
DoubleAgent Setup
To integrate this pattern into your workflow using Openclaw Skills, you need to configure your environment for dual-agent execution and install necessary interaction drivers.
# Clone the skill templates to your project
git clone https://github.com/openclaw/skills/double-agent.git
cd double-agent
# Install dependencies for the iteration loop and browser testing
pip install playwright
playwright install chromium
# Run the calibration utility to set your evaluator baseline
python scripts/calibrate_evaluator.py
DoubleAgent Data Schema & Taxonomy
The skill organizes its operation through a structured set of reference files, prompt templates, and execution scripts to maintain consistency across Openclaw Skills implementations.
| Component | File Path | Purpose |
|---|---|---|
| Architecture Patterns | references/architecture.md |
Defines the Generator-Evaluator relationship and spec templates. |
| Prompt Library | references/evaluator-prompts.md |
Contains calibrated prompts for different artifact types (UI, API, Code). |
| Control Logic | scripts/iteration_loop.py |
Manages the loop state, scoring trends, and strategy resets. |
| Calibration Tool | scripts/calibrate_evaluator.py |
Utility to prevent score drift by testing against known benchmarks. |
DoubleAgent Advanced Features
- Playwright MCP integration for real browser-based interaction, DOM verification, and screenshot capture.
- Automated calibration utilities to prevent score drift and grade inflation across different LLM models.
- Strategy reset triggers that force the Generator to switch architectural approaches if evaluation scores plateau.
- Weighted scoring rubrics that can be customized for specific domains like data science or creative copywriting.
- Structured JSON output for all evaluations, enabling seamless integration with automated CI/CD pipelines.
SKILL.md
Loading
Related Openclaw Skills
Title Replier
An automated utility that injects unique, personality-driven titles into AI assistant responses to enhance engagement and identity.
Capduck Intelligence Iran Briefing
A real-time geopolitical intelligence engine for tracking Iran-related conflicts and Trump policy signals via structured data feeds.
Birthday Video Maker
An AI-powered video automation skill that transforms fragmented photos and family clips into polished birthday tributes and social reels.
Social Intel Hub
A comprehensive social media intelligence center providing multi-platform scraping, structured data analysis, and AI-powered trend insights.
Fitness Video Editor
An AI-driven video production tool that automatically adds exercise labels, rep counters, and synchronized music to raw workout footage.
Tolk2Go Interpreter Booking API
A powerful API skill to find and book professional interpreters for phone, video, or on-site sessions across Western Europe.
Featured*
