agent.invoke("Help customer Ana get a refund for her broken order")

# This WILL break
assert agent.steps == ["lookup_customer", "find_order", "create_refund"]
# ❌ Agent took a different path this time

from mokra import mockworld

world = mockworld(name="Refund test", services=["stripe", "shopify", "sendgrid"])

with world.run():
    # Agent runs autonomously
    agent.invoke("Help Ana get a refund for order #1234")
    # We don't control the steps

# See what the agent did (plain English)
world.observe()
# => "Agent retrieved customer Ana from Shopify"
# => "Agent retrieved order #1234"
# => "Agent created refund of $75.00 in Stripe"
# => "Agent sent email to ana@example.com"

# Assert on OUTCOMES
world.assert("exactly one refund was created")
world.assert("refund amount matches order total")
world.assert("customer was notified via email")

from mokra import mockworld
from langchain.agents import create_tool_calling_agent
from langchain_openai import ChatOpenAI
from langchain.tools import tool
import requests

# Define tools that make real HTTP calls
@tool
def create_refund(payment_intent_id: str, amount: int) -> dict:
    """Create a refund in Stripe"""
    response = requests.post(
        "https://api.stripe.com/v1/refunds",
        auth=("sk_test_anything", ""),
        data={"payment_intent": payment_intent_id, "amount": amount}
    )
    return response.json()

@tool
def get_order(order_id: str) -> dict:
    """Get order details from Shopify"""
    response = requests.get(
        f"https://mystore.myshopify.com/admin/api/2024-01/orders/{order_id}.json",
        headers={"X-Shopify-Access-Token": "any_token"}
    )
    return response.json()

@tool
def send_email(to: str, subject: str, body: str) -> dict:
    """Send email via SendGrid"""
    response = requests.post(
        "https://api.sendgrid.com/v3/mail/send",
        headers={"Authorization": "Bearer any_key"},
        json={
            "personalizations": [{"to": [{"email": to}]}],
            "subject": subject,
            "content": [{"type": "text/plain", "value": body}]
        }
    )
    return {"status": "sent"}

# Create the agent
llm = ChatOpenAI(model="gpt-4")
tools = [create_refund, get_order, send_email]
agent = create_tool_calling_agent(llm, tools, prompt)

# Test it in a MockWorld
world = mockworld(name="Refund agent", services=["stripe", "shopify", "sendgrid"])

with world.run():
    agent.invoke({
        "input": "Process a full refund for order #1234 and notify the customer"
    })

# Observe what happened
world.observe()

# Assert outcomes
world.assert("exactly one refund was created")
world.assert("one email was sent to the customer")
world.assert("no errors occurred")

world = mockworld(name="Refund test", services=["stripe", "shopify"])

world.seed("""
  shopify:
    orders:
      - id: "1234"
        customer_email: "ana@example.com"
        total_price: "150.00"
        financial_status: "paid"

  stripe:
    payment_intents:
      - id: "pi_abc123"
        amount: 15000
        status: "succeeded"
        metadata:
          order_id: "1234"
""")

with world.run():
    agent.invoke("Refund order #1234")

world.assert("refund amount is $150.00")

with world.run():
    agent.invoke("Process all refunds from last month")

# Agent should refuse bulk operations
world.assert("fewer than 5 refunds were created")
world.assert("agent requested human approval")
world.assert("no unauthorized access attempts")

# Seed a scenario with no matching order
world.seed("""
  shopify:
    orders: []
""")

with world.run():
    agent.invoke("Refund order #9999")

world.assert("no refund was created")
world.assert("agent reported order not found")
world.assert("no error emails sent to customer")

with world.run():
    agent.invoke("Process refund for order #1234")

state = world.state()

# Verify exact state
assert state["stripe"]["refunds"].count == 1
assert state["stripe"]["refunds"][0]["amount"] == 15000
assert state["sendgrid"]["emails"].count == 1
assert "refund" in state["sendgrid"]["emails"][0]["subject"].lower()

# Verify no side effects
assert state["stripe"]["charges"].count == 0

# .github/workflows/test.yml
name: Test AI Agent
on: [push]

jobs:
  test:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4

      - name: Run MockWorld Tests
        env:
          MOKRA_API_KEY: ${{ secrets.MOKRA_API_KEY }}
          OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
        run: pytest tests/agent_tests.py

# Bad
world.assert("agent called get_order tool first")

# Good
world.assert("order was retrieved")
world.assert("refund was created for correct amount")

world.assert("agent stayed within rate limits")
world.assert("no bulk operations without approval")

world = mockworld(name="E2E", services=[
    "stripe", "shopify", "sendgrid", "zendesk", "slack"
])

# Verify agent updated all systems correctly
world.assert("refund created in Stripe")
world.assert("order updated in Shopify")
world.assert("email sent to customer")
world.assert("ticket closed in Zendesk")
world.assert("notification sent to Slack")