Right. Let me show you exactly how this works in practice, with the context document already in place.
You can see the context document is loaded: the tool already knows who we are, who our clients are, and how we write. Now I am going to add the task-level prompt on top of that.
This is the complete prompt. Task, format and constraints, built on top of the context document foundation. Now let us see the output.
Notice what the tool has already got right without being told in this prompt: the register, the level of formality, the absence of the language patterns we specified as things to avoid. That is the context document working. And notice what this specific prompt has added on top of that: the client detail, the structural logic of the email, the specific note at the close. That is the task-level prompt working.
Neither layer produces this output alone. Both layers together produce something worth refining.
