It’s a kind of thing where there are more opinions than people, but if I was starting from the beginning, I’d probably take the following approach.
Say, a book lists 鉛筆 （えんぴつ） in the kanji form, and I don’t know the word and don’t know one or both kanji.
I would first learn the word without kanji.
Card 1: word in hiragana/furigana/audio → meaning.
This enables reading with furigana and listening, and builds a foundation for kanji and speaking.
If in 2 months I see 鉛筆 (with furigana) again and this time I remember the word, I might add it in kanji form.
Card 2: word in kanji → reading + general meaning (as vague or specific as you want)
Initially I would not make cards for isolated kanji, only analyze them and note down general meaning and 1 or 2 dominant readings. This info should be easily accessible (linked or embedded) when failing card 2.
This enables reading with sufficient context.
If it doesn’t seem to work, and similar kanji are being mixed up, add a kanji production card.
Card 3 (optional): kanji meaning+readings+example word → draw kanji or enumerate all components
This enables test-taking where you are quizzed on words in isolation.
I used to do Card 3 for pretty much every kanji in the past, but it took a lot of time disproportionate to benefit for reading, so I’ve stopped until/if I attempt to learn handwriting in the future.
All 3 card types on self-grade fail/hard/normal/easy system, not typing.
I would not personally recommend either kanji → English keyword, or English word → Japanese word cards, since I think they serve no purpose, but YMMV.
Specifically for English → Japanese direction, outside of basic concrete nouns (e.g. dog) nothing maps one to one. You work on your production and speaking capability by reading/listening/srs whole sentences and building predictive chains context → likely words one might want to use.