The AI never decides. It only asks, gathers, or generates options.
This is a set of eight Claude skills covering the research and structure side of a design problem, from framing the problem through an adversarial critique of the solution. It stops before visual design starts. Every option a skill generates gets marked accepted or rejected, with the reasoning kept, because the rejected direction is usually the more interesting part of a case study, and it's the part most people leave out.
Structure first, opinions on request.
The pipeline moves from problem framing to accessibility constraints, research, interviews, direction, edge case testing, and a final adversarial review that's instructed to argue, not reassure. One more skill outside the sequence simulates a persona for a gut check, and is built to resist being quoted as if it were real research. Nothing is saved automatically, so a skill can only work from context it was actually given.
