The approach I understand pemerton to use (please correct me if I'm wrong) is similar to Bob McKee's outline for generating "Story" in his eponymous book:
- establish what the aims of the party are; what stakes are they trying to win (in a "Story" sense, what is their immediate "dramatic need")
- put plausible obstacles in the way of them achieving that objective; these should be specific difficulties that require either one specific character or a combination or choice of characters to test a skill or ability.
- continue placing such obstacles until either (a) an insoluble obstacle (presented after 2 failures have already been made) is failed, or (b) sufficient obstacles are overcome to complete the challenge.
I don't know the book you refer to, but what you say is basically it, with one exception - the stake can actually change over the course of the challenge, as the complications that are introduced in response to the skill checks change the players' (and therefore the PCs') orientation towards the fiction.
Related to this - I'll often have a general sense of some of the key obstacles at the time when I frame the scene, but will tend to work out the details, plus add things in or change what I had planned in response to how the situation actually plays out in the hands of the players.
What is key - and what I think you have got right - is that the obstacles/complications are introduced drawing not just on extrapolation from previous ingame events, but in the interests of keeping the situation alive - until we get to the final check needed, when the definitive obstacle can be posed. This can be hard to GM, but I've found it produces rich and satisfying encounters with unexpected and plot-and-campaign-deepening resolutions.
And just for the hell of it, I'm going to post my favourite
Paul Czege quote which is probably the pithiest piece of advice I've used to help me run skill challenges (I've also drawn on Maelstrom Storytelling, HeroWars/Quest, and Burning Wheel):
There are two points to a scene - Point A, where the PCs start the scene, and Point B, where they end up. Most games let the players control some aspect of Point A, and then railroad the PCs to point B. Good narrativism will reverse that by letting the GM create a compelling Point A, and let the players dictate what Point B is (ie, there is no Point B prior to the scene beginning).
. . . [A]lthough roleplaying games typically feature scene transition, by "scene framing" we're talking about a subset of scene transition that features a different kind of intentionality. My personal inclination is to call the traditional method "scene extrapolation," because the details of the Point A of scenes initiated using the method are typically arrived at primarily by considering the physics of the game world, what has happened prior to the scene, and the unrevealed actions and aspirations of characters that only the GM knows about.
"Scene framing" is a very different mental process for me. Tim asked if scene transitions were delicate. They aren't. Delicacy is a trait I'd attach to "scene extrapolation," the idea being to make scene initiation seem an outgrowth of prior events, objective, unintentional, non-threatening, but not to the way I've come to frame scenes in games I've run recently. . . [W]hen I'm framing scenes, and I'm in the zone, I'm turning a freakin' firehose of adversity and situation on the character. It is not an objective outgrowth of prior events. It's intentional as all get out. . . I frame the character into the middle of conflicts I think will push and pull in ways that are interesting to me and to the player. I keep NPC personalities somewhat unfixed in my mind, allowing me to retroactively justify their behaviors in support of this. And like Scott's "Point A to Point B" model says, the outcome of the scene is not preconceived.
My 4e game is, I suspect, pretty light-hearted compared to Czege's game. But his technique - especially of eschewing extrapolation and "delicacy" in favour of deliberately pushing and pulling so as to make things interesting and force the players to make choices for their PCs - I think is as applicable to a light-hearted fantasy game as to something like My Life With Master.