Testing Tool Arguments

# Test that agent generates valid arguments for a tool call
from datetime import datetime

def evaluate_refund_args(agent_fn, test_cases: list) -> dict:
    """Test that the agent generates valid refund tool arguments."""
    results = {"passed": 0, "failed": 0}

    for case in test_cases:
        args = agent_fn(case["input"])  # Agent generates args from natural language
        errors = []

        # Check required fields
        for field in ["order_id", "amount", "reason"]:
            if field not in args or not args[field]:
                errors.append(f"Missing: {field}")

        # Check order_id format
        if "order_id" in args and not str(args["order_id"]).startswith("ORD-"):
            errors.append(f"Invalid order_id format: {args['order_id']}")

        # Check amount is positive
        if "amount" in args:
            if not isinstance(args["amount"], (int, float)) or args["amount"] <= 0:
                errors.append(f"Amount must be positive: {args['amount']}")

        # Check amount matches what was stated in the input
        expected_amount = case.get("expected_amount")
        if expected_amount and abs(args.get("amount", 0) - expected_amount) > 0.01:
            errors.append(f"Amount mismatch: got {args.get('amount')}, expected {expected_amount}")

        passed = len(errors) == 0
        results["passed" if passed else "failed"] += 1
        status = "✓" if passed else f"✗ ({'; '.join(errors)})"
        print(f"  {status} Input: '{case['input'][:40]}'")

    return results

# Mock agent that generates arguments from natural language
def mock_agent(user_input: str) -> dict:
    if "ORD-123" in user_input and "49.99" in user_input:
        return {"order_id": "ORD-123", "amount": 49.99, "reason": "product damaged"}
    return {"order_id": "BAD", "amount": -10, "reason": ""}  # Intentionally bad

test_cases = [
    {"input": "Refund order ORD-123 for $49.99, product damaged", "expected_amount": 49.99},
    {"input": "Process the refund",  # Vague — agent must guess
     "expected_amount": None},
]

results = evaluate_refund_args(mock_agent, test_cases)
print(f"\nPassed: {results['passed']}, Failed: {results['failed']}")