The simplest things I can think of are cases where something blocks action. That could be something external, akrasia, laziness, uncertainty, who knows. I could recognize in my own case that I have chosen a course of action but not carried it out yet, and describe this as having an intention. If you tell me you have chosen a course of action, and I know you haven't acted, I could describe you as having an intention.
For an action actually carried out, we're in Quine-Duhem territory: there are any number of ways of describing the action, mixing reasons and causes and beliefs and preferences and circumstances. Presumably the only reason to bother parsing intention and friends here is to make better predictions than we can make just using the action itself.
For the blocked-action case, it seems like the baseline would be a single prediction: that the intended act will be carried out. (And then we can modify later as we lose patience and so on.) We can think of intention as the well-defined little box into which we put "what you will do unless ..." So an intention is a special kind of prediction. Duh.
One funny thing about this is that if it's a single prediction (instead of a whole complex with various confidences), we can reverse the association -- that is, treat something for which we have a single prediction as if it were an intention.
I think you can actually hear this in daily conversation. If the baseline expectation is same-as-me, I can just predict you'll continue to be same-as-me. (Single prediction, yay!) If it turns out you've not taken some action I would have, or in fact have, I assume you intend to (this is the previous paragraph's point).
"Got my jab yesterday. You get yours?"
"No."
"But you're going to right?"
It's the immediate fallback. If you're not just like me, something stopped you from being like me -- gosh, what could it have -- nope, doesn't matter what it was, no use spending calories on that, presumably things will get back on track soon and you'll be just like me again. I predict it, therefore you intend it.
We know from our own case how externalities can interfere with our actions, and when we're forced to consider someone's behavior diverging from our prediction, we'll reach for that first, and preserve the assumption that your intention is to be like us. To show you this chain of reasoning, movies have to use dialogue, and they do this all the time:
"How was your date with Marcus last night?"
"I didn't go."
"Oh, got called into work again?"
"No."
"Marcus couldn't make it?"
"No."
"????" unable even to form the thought that you decided not to go.
The point of all of this is that we might use the same-as-me strategy as the starting point for judgment because it's dirt cheap. Similarly we might use same-as-me predictions because they're dirt cheap, guaranteed to be degenerate non-branching decision trees. Non-branching trees we talk about as intentions, both for ourselves and for others. (This is consonant with current neuroscience, right? We act, for reasons we know not, and if needed bolt-on a retrodiction of that action and call it the intention we had when we acted.) Non-branching trees are cheaper, and we will resist giving them up even when surprised.
I keep emphasizing the same-as-me strategy because it does seem like the cheapest baseline available, but your (Goffmanesque?) scripts and part-playing are similar, right? Once I've stocked my toolkit with single-path predictions that can be quickly and cheaply selected, I'll insist on using them. And when you fail to say the lines I've assigned you, my immediate fallback will be assuming you intend to say them and something stopped you. Acknowledging that you diverged on purpose is the last thing I want to do, because then to predict you I'll have to engage in expensive research (i.e., talk to you, which is not so bad, talk is cheap, but in this case I'll also have to listen to you and that blows).