[ETA Epistemic status: publically, unelaborated conjecture; privately, fairly confident conjecture.]
Emotions are, in their essence, mental actions taken by entities in S1.
Emotions are coordinated, stereotyped modulations of multiple parts of the central and peripheral nervous system, enacted by entities in the unconscious mind. This can include de/activating certain muscle groups, up- or down-regulating breathing, digestion, and glycogenolysis, and de/activating parts of the cerebral cortex.
Emotions are not, in their essence, messages.
Emotions are not information encoded by one perspective, sent to another perspective, and then decoded to reconstruct a meaning. Rather, they could be viewed as one quasi-perspective (S1 or S1.5) directly taking mental actions, and in particular directly orchestrating thoughts and futher actions by de/activating parts of S2. It would be rather a stretch to say that, as you hammer a nail, your hand is transmitting a message to the mallet. The motive force indicated by an emotion, in its original and central function, is your S1 using your S2.
Some social emotions cause you to send messages to other people, but the emotions are not themselves messages.
Emotions are not, in their essence, goals.
"I want to feel good", "I want pleasure", "I want hedons", "I want to be content", "I want to feel accepted", "I want to feel like I'm doing something that matters". Feeling pleasure is an action taken by your goal-threads to modify your mind so that it'll keep doing what it just did. Emotions are actions taken to acheive goals; they are not goals. Seeking pleasure as a goal is goodharting.
Emotions can secondarily be used as information and goals.
Some of S2 can take emotions as information and as indicators. We have the experience of "receiving a message from S1" in the form of an emotion. I think this happens because of culturally transmitted mental technology for suppressing the effects of emotions on some parts of the brain. Those parts of the brain, call it S2.5, can maintain "objectivity" by exerting a homeostatic force on itself that counteracts actions taken by emotions. When an emotion happens, S2.5 views it as coming from outside of S2.5, and hence views it as a message from an external entity, namely S1. Prehistoric humans did not have this mental technology, and would always be "inside" their emotions, like "I'm not angry!!", except they wouldn't have a word for angry because once you get that word you can homeostat away its effects. When we're overcome by emotion, that's S2.5 losing its homeostatic battle. Some people have a very strong S2.5, which brings benefits and costs. You can train your S2.5 to be stronger, but that's not necessarily a good idea.
When S2.5 notices that an emotion is happening, it can take that emotion as information about what S1 wants, and about what S1.5 thinks and wants. S2.5 can take the emotion as an indication that what S2.5 has been doing is right or not right or should be changed in some particular way. S2.5 can pursue emotion as a goal, which is goodharting, but is okay in some cases; goodharting produces good results within some regimes of optimization power.