Good text prompts almost always follow the same structure: role, task, context, output format, limits. Fill in what you know – the rest is simply left out. You can copy the result or publish it as a versioned prompt.
From which perspective does the model answer?
What exactly should be done? One to three sentences, imperative.
Audience, background, constraints.
As concrete as possible – the strongest lever there is.
How should it sound?
One or two input/output pairs. Often the biggest quality jump.
What explicitly must not happen.
Where the content gets inserted. Placeholders in double curly braces.
Publishing requires an account. Copying always works.