You may be interested in this analysis on various recipe formats, along with the proposed "RxOL" notation for explicitly expressing recipes as such a tree based structure[0].
Personally I find RxOL hard to read & edit though, so I have been thinking about whether it would be doable able to infer such tree structures from CookLang's more readable recipe format[1][2].
Ingredients and times are already marked, so surely verbs & follow up mentions of ingredients can be extracted from the text without further annotation? Haven't tried doing it yet though.