warning: huge block of text ahead
I was planning on it being an active-stative language, but I can't find any good resources on those languages, so really all I'm doing is fumbling around trying to mimic something I've only heard of. I like the way it works though, and I intend on keeping it the it is regardless of whether or not it counts as active-stative, ergative, nominative, or w/e.
As for how it works. It's a VSO language, though nouns can be moved to the beginning of the sentence due to focusing (thus, even SVO and OVS are possible, though VSO is the default).
As for nouns, the patien is marked with a preposition, but the agent is not as long as it either appears at the beginning of the sentence, or right after the verb (which, obviously is what happens 99% of the time, though there IS an agentive preposition just in case).
You don't have to specify an agent in every sentence, in fact the way you form a passive is to just omit the agent, nothing else is changed. (alternatively, you could just place the agent after the patien, this is one of the few cases where you could use an VOS order). I'm also thinking of the agentive marker doubling as a causative.
Some verbs do behave ergatively, normally though these are ditransitive verbs. The trigger for whether a noun counts as the agent or patien is which one is causing the event, and which one is affected by it. For example, 'to boil' behaves ergatively, because obviously water doesn't boil on its own, something causes it to boil.
I'll add to this, since I was kind of worried about space in the yahoo answers post:
The word for 'to boil' does behave ergatively, at least from our perspective. In the conlang itself, its just perceived as describing an event where an agent/cause isn't specified. Not all intransitive verbs behave this way of course. 'To run' behaves nominatively. The odd way that some verbs work also allows a lot of doubling-up in meanings. For example, the words for 'to knock over' and 'to fall over' are identical. The only difference is that when you're expressing 'to fall over', really you're just saying it was knocked over without specifying why.
As for sentences involving the concept of 'to be', I've been struggling with that for years, and is the primary reason why I've made so little progress on the conlang over the years, as evidenced by the fact that I have no vocabulary to give sample sentences. I've just been trying to figure out how this language should work.
I do know that the verb 'to be' in English is actually kinda broad, with many languages actually using several distinct verbs (note I use the word 'verb' loosely here). For example, Japanese has a distinct pair of verbs (differentiated by animacy) for indicating where something is located. And looking at a Maori reference grammar, they apparently have distinct verbs for things like 'to be inside of', and 'to be ontop of' and such (the location is marked with a generic locative preposition). Maori also has distinct ways of expressing 'to be'. Such as 'to be something specific, or A=B, and 'to be in a certain category', such as 'Clifford is a dog'.
The way my own conlang does the first sentence is actually rather odd. They actually use the same structure to form an inalienable possessive (more accurately, they use it when the possessed is actually a part of the possessor, such as 'my hand' or 'the door of the house'). Essentially it indicates that the two nouns are in fact one object. So if you were to say 'I am human' or 'my name is...' you would essentially use this indicate that you and the 'human' or named individual are one and the same. Its hard to explain...
But enough of that, the point is with 'to be' sentences normally the patien preposition is used. The agent would only be used if the person/thing chose to 'be' whatever they are. Really, the only case where the agentive would be used for the 'subject' is in sentences indicating location (assuming the person/thing chose to be where they are).
This honestly sounds like a fluid-S language to me, but I honestly don't know since information about these things is shockingly difficult to come back. I've been struggling for years just to find reference grammars for VSO languages.