​

Page Cloning

• Using ArchiveWeb.page to clone websites into WACZ files
• Hosting archives locally with the HUD page archives repository and CustomGym
• Uploading archives to app.hud.so for immediate cloud hosting
• Creating tasks that use these stable archived environments

​

Prerequisites

• HUD SDK installed
• Docker installed (for local hosting option)
• ArchiveWeb.page browser extension (for cloning pages)
• API keys for HUD and your chosen agent

​

Part 1: Cloning the Page

​

Installing ArchiveWeb.page

1. Install the Browser Extension:
   • Visit ArchiveWeb.page
   • Install the extension for Chrome/Chromium-based browsers
   • The extension icon will appear in your browser toolbar
2. Create a New Archive:
   • Click the ArchiveWeb.page extension icon
   • Click “Create New Collection”
   • Give your collection a descriptive name (e.g., “my-test-site”)

​

Capturing Web Pages

1. Start Archiving:
   • Click “Start” in the extension popup to begin an archiving session
   • Navigate to the website you want to clone
   • Interact with the site as your agent would (login, navigate through pages, fill forms)
   • All pages and resources will be captured automatically
2. Best Practices for Agent Testing:
   • Capture all relevant pages and states your agent will interact with
   • Include error pages and edge cases
   • If testing login flows, capture both logged-out and logged-in states
   • For form submissions, capture the form page and success/error pages
3. Stop and Download:
   • Click “Stop” in the extension when done capturing
   • Click “Download” to save your collection
   • Choose WACZ format (default)
   • Save with a meaningful filename (e.g., my-test-site.wacz)

​

Example: Cloning a Login Flow

1. Start archiving session
2. Visit https://example.com/login
3. Enter test credentials (e.g., testuser/password123)
4. Submit the form
5. Capture the dashboard/welcome page
6. Optionally capture logout flow
7. Stop and download as my-test-site.wacz

​

Part 2: Hosting the Website

​

Option 1: Local Hosting with CustomGym

​

Step 1: Clone the Page Archives Repository

git clone https://github.com/hud-evals/page-archives.git
cd page-archives

​

Step 2: Add Your Archive

1. Place your WACZ file:
   Copy

   Ask AI

   cp ~/Downloads/my-test-site.wacz archives/
   

2. Update archives/archive_list.json:
   Copy

   Ask AI

   {
     "archives": [
       {
         "name": "my-test-site",
         "displayName": "My Test Site Archive",
         "startPage": "https://example.com/login"  // Optional: default page to open
       }
       // ... other archives
     ]
   }
   

   Note: The name field must match your WACZ filename without the .wacz extension.

​

Step 3: Create a CustomGym for the Archive Server

from hud.types import CustomGym
from pathlib import Path

# Create a Dockerfile for the archive server
archive_server_dockerfile = """
FROM node:18-slim
WORKDIR /app
COPY . /app
RUN npm install
EXPOSE 3000
CMD ["npm", "run", "start"]
"""

# Save Dockerfile in the page-archives directory
with open("page-archives/Dockerfile", "w") as f:
    f.write(archive_server_dockerfile)

# Define the CustomGym
archive_server_gym = CustomGym(
    location="local",
    image_or_build_context=Path("./page-archives"),
    host_config={
        "port_bindings": {3000: 3000}  # Expose port 3000
    }
)

​

Step 4: Create Tasks Using the Archived Site

from hud import Task, run_job
from hud.agent import ClaudeAgent

# Task to test login flow on the archived site
login_task = Task(
    prompt="Log into the website using username 'testuser' and password 'password123'.",
    gym="hud-browser",  # Use browser to interact
    setup=[
        # Navigate to your archived site running locally
        ("goto", "http://localhost:3000/my-test-site")
    ],
    evaluate=("page_contains", "Welcome, testuser!")
)

​

Advanced: Query Parameters

# Open a specific page within the archive
specific_page_task = Task(
    prompt="Navigate to the user profile page",
    gym="hud-browser",
    setup=[
        ("goto", "http://localhost:3000/my-test-site?page=https%3A%2F%2Fexample.com%2Fprofile")
    ]
)

# Debug mode - shows full ReplayWeb.page UI
debug_task = Task(
    prompt="Explore the archive interface",
    gym="hud-browser",
    setup=[
        ("goto", "http://localhost:3000/my-test-site?debug=true")
    ]
)

​

Option 2: Cloud Hosting on app.hud.so

​

Step 1: Access Page Clone Feature

1. Go to app.hud.so
2. Click “Create” in the navigation
3. Select “Page Clone”

​

Step 2: Upload Your Archive

1. Click “Upload WACZ file”
2. Select your .wacz file created in Part 1
3. Provide a name for your cloned environment
4. Click “Create”

​

Step 3: Use the Hosted Archive

from hud import Task, run_job
from hud.agent import ClaudeAgent

# Task using the cloud-hosted archive
cloud_login_task = Task(
    prompt="Log into the website using username 'testuser' and password 'password123'.",
    gym="hud-browser",
    setup=[
        # Navigate to your cloud-hosted archive
        ("goto", "https://archives.hud.so/your-archive-id")
    ],
    evaluate=("page_contains", "Welcome, testuser!")
)

# Run evaluation
job = await run_job(
    agent_cls=ClaudeAgent,
    task_or_taskset=cloud_login_task,
    job_name="Cloud Archive Test"
)

​

Tips for Effective Page Cloning

1. Capture Complete Flows: Don’t just capture individual pages - capture entire user journeys
2. Include Resources: Ensure CSS, JavaScript, and images are properly captured
3. Test Your Archives: Always verify your archives work correctly before using them in evaluations
4. Document States: Keep notes on what states and pages are included in each archive
5. Update Regularly: Re-clone sites when significant changes occur

​

Key Takeaways

• ArchiveWeb.page makes it easy to create WACZ archives of any website
• Local hosting with CustomGym gives you full control and fast performance
• Cloud hosting on app.hud.so provides instant deployment without infrastructure
• Page cloning ensures consistent, reproducible testing environments for AI agents
• Archived sites eliminate external dependencies and enable offline testing