Monitor the inputs as well as the outputs. Trade count versus expectation, average holding period, realised slippage versus modelled, hit rate, and the distribution of position sizes all shift before the equity curve does, which makes them earlier warnings.
Set the rules in writing while calm: if realised slippage exceeds the model by 50% for a month, halve size; if the rolling-sharpe over six months falls below zero, review; if the drawdown exceeds 1.5 times the worst in backtest, stop and investigate. The value of writing it down is that the decision is made by a person who is not currently losing money.
Distinguish monitoring from second-guessing. The purpose is to detect breakage, not to trade the equity-curve. A strategy inside its expected drawdown distribution is behaving correctly even when it is unpleasant.
Related: live-vs-backtest-divergence, kill-switch, rolling-sharpe, alpha-decay