Skeleton of Thought
Skeleton of thought is a reasoning pattern in which the model first produces a compact skeleton of the answer — a list of points or an outline — and then expands each skeleton point, often as independent sub-prompts running in parallel. On long outputs, parallel expansion reduces latency significantly. The trade-off is that the skeleton must be right: expansion cannot repair a bad outline, and parallel expansions do not share progressive context, so one section cannot react to what another one said. It suits list-like or section-independent content and is a poor fit for tightly argued prose where sections build on each other.
Example
Asked to write a 1,500-word guide on database indexing, a skeleton-of-thought system first outputs a 6-point outline: 1) what an index is, 2) B-tree vs hash, 3) when to add one, 4) covering indexes, 5) write cost, 6) monitoring. Points 1–6 are then expanded in 6 parallel model calls and stitched together. Total latency is roughly one expansion call instead of six sequential ones.
Put this into practice
Build polished, copy-ready prompts in under 60 seconds with SurePrompts.
Try SurePrompts