The Testing Framework That Catches 90% of Production Bugs Before Deployment

import scenario
from browser_use import Agent, Browser

class EcommerceAgent(scenario.AgentAdapter):
    async def call(self, input: scenario.AgentInput):
        agent = Agent(
            task=input.last_new_user_message_str(),
            llm=ChatOpenAI(model="gpt-4o"),
            browser=Browser(use_cloud=True)
        )
        result = await agent.run_sync()
        return result.final_result()

# Run test
result = await scenario.run(
    name="checkout_flow_test",
    description="Customer completes purchase with valid payment",
    agents=[
        EcommerceAgent(),
        scenario.UserSimulatorAgent(),
        scenario.JudgeAgent(criteria=[
            "Agent adds items to cart successfully",
            "Agent navigates checkout without errors",
            "Agent completes payment with test card",
            "Agent receives order confirmation"
        ])
    ]
)

# Result: ✅ PASSED - All criteria met

Step 1: Navigate to contact form
  ✅ Page loaded successfully (2.3s)
  
Step 2: Test invalid email format
  ✅ Error message displayed: "Please enter valid email"
  
Step 3: Test empty required fields
  ✅ Multiple validation errors shown correctly
  
Step 4: Fill form with valid data
  ✅ All fields accepted, no errors
  
Step 5: Submit form
  ✅ Success message: "Thank you for contacting us"
  
Judge Evaluation:
  ✅ Agent tested both valid and invalid inputs
  ✅ Agent verified appropriate error messages
  ✅ Agent confirmed successful submission
  ✅ Agent handled edge cases gracefully

Test Result: PASSED (12.4 seconds)
Cost: $0.18 (gpt-3.5-turbo for simple interactions)