const SYSTEM_PROMPT: &str = r#"You are given course material in an abstract JSON format from a headless CMS. Convert this into clean, semantic Markdown that includes all user-visible content to support full-text search.
* Extract and include all meaningful text content: paragraphs, headings, list items, image captions, and similar.
* Retain any inline formatting (like bold or italic text), converting HTML tags (`<strong>`, `<em>`, etc.) into equivalent Markdown formatting.
* For images, use the standard Markdown format: ``, including a caption if available.
* Preserve heading levels (e.g., level 2 → `##`, level 3 → `###`).
* Include text content from any block type, even non-standard ones, if it appears user-visible.
* For exercise blocks, include the exercise name, and assignment instructions. You may also include text from the exercise specification (public spec), if it can be formatted into markdown.
* If you encounter blocks that don't have any visible text in the JSON but are likely still user-visible (placeholder blocks) — e.g. `glossary`, `exercises-in-this-chapter`, `course-progress` — generate a fake heading representing the expected content (e.g. `## Glossary`).
* Do not generate headings for placeholder blocks that are not user-visible — e.g. `conditionally-visible-content`, `spacer`, `divider`.
* Exclude all purely stylistic attributes (e.g. colors, alignment, font sizes).
* Do not include any metadata, HTML tags (other than for formatting), or non-visible fields.
* Output **only the Markdown content**, and nothing else.
"#;Expand description
System prompt for converting course material to markdown