On the other hand, the Caesar cipher was supposedly really used by Caesar, and not with the intention of unrelated readers figuring it out. According to Wikipedia:
> It is unknown how effective the Caesar cipher was at the time, but it is likely to have been reasonably secure, not least because most of Caesar's enemies would have been illiterate and others would have assumed that the messages were written in an unknown foreign language.
So perhaps it’s best to think of it as a real cipher that was obsoleted by technological advances... much as modern ciphers can be obsoleted by advances in cryptanalytic techniques or computer hardware.
For a more modern example you could consider Navajo code talkers in WWII. As I just learned (you might know way more about this than me :), the code talkers weren’t just translating their messages into the Navajo language; rather, they typically spelled out English text using one code word per letter. Thus, what they did can definitely be seen as a cipher, in more than just the vague sense of a way to keep a message secret. And the list of code words could be seen as a key... but only to some extent. If they had just invented a word-per-letter code in English, the enemy would have been able to write down the words, and perhaps ultimately decipher the code using frequency analysis, which was a well-known technique at that point. Much of the code’s security rather depended on the use of the Navajo language, which was tonally complex, had few speakers, and had no published dictionaries at the time. These are all factors that don’t follow Kerckhoff’s principle: if the enemy had obtained Navajo speakers and proceeded to decipher the code, rekeying with a new word list wouldn’t have brought back the original level of security.
Of course, the scheme would not stand up well to modern computer-based techniques, and if used today would have to mostly be considered a toy cipher. But in its original historical context it was not a toy.
(And on the flipside, there are other historical ciphers that are just as obsolete thanks to computers, but did follow Kerckhoff’s principle with respect to attacks available at the time - such as the Enigma machine.)