Context & Retrieval
Prompt Structure
Assembling a prompt from named parts (identity, task, context, constraints, output) so failures are diagnosable.
A prompt is the full set of instructions you hand the model on one call. Prompt structure means assembling that text from five named parts, identity, task, context, constraints, output format, instead of writing it as one undivided blob.
The reason is diagnosis. When a one-blob prompt fails, nothing says which part caused the failure, so you tweak the whole text and hope. Split into labelled sections, a failure points at one: an unclear task, a missing constraint, a loose output spec, and you rewrite that section alone. In a workflow with a fixed design, the constraints section is also where to tell the model to follow the supplied steps and invent no extra entities or state.
Structure fixes one message. Curating the window across a whole task is the same job at session scale.