I'm sticking rigidly to a two-word maximum.
Some actions really do require more than two words to describe properly, and that's a limitation that I've struggled with; on this project, I've actually moved locations around and changed some aspects of puzzles to avoid ambiguities that would require longer commands. But I find that it's often possible to rework things to avoid those actions, and I also like being able to tell the player, "If you can't think of a way to explain what you want the PC to do with a VERB NOUN pair, it's not necessary in this game."
My design philosophy is that once the player has guessed the action that's necessary to progress, the game should give the player every benefit of the doubt. Situations where, for example, I couldn't (as a player) "BREAK LOCK", "USE CROWBAR", or "PRY LOCK", nor "BREAK LOCK WITH HAMMER" or "BREAK LOCK WITH ROCK", but I could only "BREAK LOCK WITH CROWBAR" feel unnecessarily coy to me, and I feel like that's the sort of puzzle writing that more complex commands can sometimes encourage. If I've already guessed that the way forward is to apply brute force to the lock, and I have a selection of sturdy implements to hand, including the "correct" one (another annoying issue in TA/IF), is it really necessary to make me try each one systematically?
I'm not accusing anyone here of that; I'm just explaining that I think that sticking to two-word commands helps me not to fall into those traps as an author.