notes
Prompt files as a content workflow
How I use structured prompt files to create schema-safe posts and project pages from supplied facts.
After the site structure and theme settled, I added prompt files to make content work more consistent. The point was not to let AI invent pages. The point was to give it clear project rules for turning supplied facts into posts and project entries that fit this repository.
The prompts describe the expected frontmatter, file locations, language pairing, publishing defaults, and media rules. They also make one rule explicit: if a fact is missing, ask for it or leave it out.
Why prompt files help
Without a prompt file, every content request needs the same setup repeated by hand. The model needs to know which template to follow, where English and Dutch files belong, how translationKey works, which tags are valid, and what kinds of embeds are allowed.
Putting those rules in reusable project documentation makes the workflow less dependent on memory. It also makes the output easier to review because the structure should already match the site.
Schema-safe content
Astro content collections validate frontmatter, so a post is not just prose. It is structured content that has to satisfy the schema before the site can build.
A safe post starts from a small, predictable shape:
title: "Post title"
description: "A concise description for previews and metadata."
publishedAt: "YYYY-MM-DD"
locale: "en"
translationKey: "shared-post-key"
published: false
private: true
tags: ["notes"]
category: "notes"
That default keeps new generated content reviewable before it becomes public.
Bilingual writing
The prompts also treat English and Dutch as first-class output, not as an afterthought. Both versions should cover the same facts and share the same translationKey, but the Dutch text should read naturally instead of following the English line by line.
That matters for this site. I want the bilingual structure to feel intentional, not like a second pass bolted onto the English content.
Safe media
The prompt files also keep media use bounded. Local images should live in the expected public image folders, and embeds should use supported providers through the site components.
That keeps content generation from turning into arbitrary iframe snippets or remote image links. If media is not clearly useful, prose is enough.
Result
The prompt files became a small content workflow for the site. They reduce repeated setup, keep generated drafts close to the schema, and make bilingual content easier to produce without weakening the facts.
That fits the broader direction of the blog: use automation where it removes repetition, but keep the source material and publishing rules explicit.