Your rule should not be "never kill civilians" or "kill target no matter what, ignoring civilian deaths" but "minimise civilian casualties in any possible manner".

Depends on your computing power.

For example, choosing "minimise civilian casualties in any possible manner" may encourage your opponent to take hostages they wouldn't take if you would precommit to "kill target no matter what, ignoring civilian deaths". If taking hostages makes crime relatively safe and profitable, this may encourage more wannabe criminals to take action. Thus, minimising the casualties in short term may increase the casualties in long term.

Also, it's important how much your actions are legible by your opponent, and how credible are your precommitments.

For example, if you choose the strategy "kill target no matter what, ignoring civilian deaths", but your opponent believes that you would follow the strategy if there are 10 hostages, but that you would probably change your mind if there are 10 000 hostages, well, you just motivated them to take 10 000 hostages.

(Then there are strategies to ruin your opponent's precommitment. Essentially, if your opponent precommits to "if X, then I do Y", your strategy is to do things that are very similar to X, but not completely X. You keep doing this, and while you technically didn't do X, only "X minus epsilon", so your opponent was not required to do Y, psychologically you weaken the credibility of their precommitment, because for most people it is difficult to believe that "X minus epsilon" doesn't bring the strong reaction Y, but X would.)