The reference solution is wrong here: it assumes "full caps" as in "more than 1 character from the beginning of a word is in upper case":

Super, RANDOMRandom CAPS Super,. seven Benny's Super, Benny's eight?

Expected: 'Repus, Modnarmodnar Spac Repus,. neves S\'ynneb Repus, S\'ynneb thgie?', instead got: 'Repus, MODNARModnar Spac Repus,. neves S\'ynneb Repus, S\'ynneb thgie?'

Super, superCAPS six Carl's. RANDOM Carl's Don't Random, no?
Expected: 'Repus, SPACrepus xis S\'lrac. Modnar S\'lrac T\'nod Modnar, on?', instead got: 'Repus, spacrEpus xis S\'lrac. Modnar S\'lrac T\'nod Modnar, on?'

It is also wrong in the case a word has an upper case letter in the middle:

Don't eightMoney Ex Ramses'. super Ramses' Benny's Benny's Random?
Expected: 'T\'nod yenoMthgie Xe Sesmar\'. repus Sesmar\' S\'ynneb S\'ynneb Modnar?', instead got: 'T\'nod yenomThgie Xe Sesmar\'. repus Sesmar\' S\'ynneb S\'ynneb Modnar?'

(And several more, I stopped trying to list out all of them)

• Node 14. should be enabled

• More sample tests on multiple special characters at end or start should be added

• There should be 100 random tests at least

• The description should specify what considers special characters

• Description should also be reformatted to make example tests clearer

• The random tests' expected and actual output should not be hidden

Needs Random Test.
And some edge cases

secretConverter('Dr. AaaaaB.Ccccc iOS co-worker ItOUCH macOS XMLHttpRequest')
secretConverter('real_pythoner_use_snake_case')