Preventing repeated failing tool calls

Through observability you see a failing agent run called the same failing tool 5 times in a row. What is the appropriate action you can take to prevent this in the future?