Does anyone have suggestions on how to properly prompt a tool to be executed? The agent consistently tells the user it performed the action, but never invokes the tool that completes the action.