Use when implementing or refactoring multi-step LLM workflows where outputs from one step feed later steps; especially useful for extraction, normalisation, generation, code refinement, and report synthesis.
