Two decisions before an assistant writes anything

Which export should you paste?

Markdown, XML and JSON carry the same document - the canvas and its dimensions, every zone with its kind, its instruction and its semantic role, and the styling each one was given. They differ in how they read. Markdown is the shortest and the easiest to skim, XML names every attribute on every element, and JSON is the same facts arranged for a program rather than a reader.

The comparison walks one template through all three, shows where the same sentence ends up in each, and says which one suits what - a first conversation with an assistant, a document somebody will read, or something a script has to consume.

Markdown vs JSON vs XML

How to write the instructions

A template with a zone for every part of the composition only beats one long prompt if each zone says what belongs in it. That means naming the subject, fixing what can be counted - a word count, a number of rows, a length in seconds - setting the tone where the tone matters, and saying what must not change. The guide is about where those sentences belong: with the part they constrain, rather than in a paragraph at the top that everything has to be read out of.

It also covers what a brief cannot do. An instruction constrains the text an assistant writes; it does not decide the layout, because the layout is already decided - the zones, their positions and their sizes are the part of the template that no assistant has to guess.

Writing structured prompts

Where these fit

Neither guide is about the editor itself. How it works walks through the canvas, the zones and the export in order, the export reference lists what each format carries and what is deliberately left out of it, and the examples show three templates written out as zones, instructions and semantic roles - which is the shape these guides assume you have already seen.