Skeleton-of-Thought (SoT)

Generate an outline first, then expand sections in parallel for speed and structure.

Skeleton-of-Thought prompting asks the model to first produce a high-level outline, then expand each point separately.

Why It Matters for Agents

SoT maps naturally onto parallelization: an orchestrator generates the skeleton, then distributes expansion tasks to workers (or parallel model calls).

When To Use

  • Long-form writing where structure matters.
  • Any task where the bottleneck is sequential generation latency.