FORTH.
One word says it all.
FORTH is a (really) stack-oriented language. It lives and breaths stacks.
FORTH is a small language and is easy to learn. It is centered around "words". There is a set of standard words but you almost immediately learn how to create your own words.
This language is completely dynamic. Every word lives in the dictionary. Your words can overwrite standard words. The "self modifying" aspect of this language is that it enables and encourages you to create "words" that do exactly what you want. A FORTH program is truly unique to the task at hand.
The fundamental "must learn" idea is the "threaded-interpreted language model". Words invoke words which invoke words... it is "kittens all the way down".
"Threaded Interpretive Languages: Their Design and Implementation"
by R. G. Loeliger is a must-read book. These ideas don't show up elsewhere.
FORTH can be implemented in 4K (that's kilobytes).
This is a language that "isn't a language". By that I mean it is one of the "clay-like languages, such as Lisp". Language shapes the way you think. Clay-like languages shape themselves to your thoughts.
Most languages suffer from an "impedance mismatch". They are like linking a firehose to a soda straw. Your "problem" has to move to the language. FORTH (and Lisp) simply shape themselves to the problem. They are "clay for the mind".
FORTH is a way of thinking. Indeed, "Thinking Forth" is another must-read book.
(http://thinking-forth.sourceforge.net/tf-kindle.pdf)
FORTH is a "must know" language.
