genseq2 = genseq + an explicit structural-shape router. The model first classifies the task into one of six shapes (Pilot-and-Scale, Vertical-Slice-Growth, Organic-Expansion, Align-then-Prototype, Event-Flow, Envision-then-Provision), then draws on the ONE shape-matched worked example — or falls back to the principles if no shape fits. Same living-process theory as genseq; the worked-examples section adds the 'reference, not a template' preamble, the classify-then-route procedure, and one English exemplar per shape.
Invocation: explicit only — genseq2 fires only on /genseq2 (or when named); it does NOT auto-trigger. genseq remains the auto-triggering skill for sequencing tasks.
Empirical basis: routing to the single shape-matched example captured the example benefit (~+0.05 goldscore, replicated across two runs) while plain example-stacking diluted; the preamble is harm-reduction. See testforclaude/skillopt_mini/RESUME.md.
LICENSE / NOTICE: the theory section quotes Christopher Alexander's The Nature of Order (quoted for education, not covered by the MIT license) — see NOTICE.