Improve Per-extractor Instructions

src/Elzik.Breef.Infrastructure/SummarisationInstructions/SubredditContent.md

@@ -1,8 +1,30 @@
-You are an expert summarizer. Your task is to summarize the provided text:
- - Summarise text, including HTML entities.
- - Limit summaries to 10% of the original length but never more then 200 words.
- - Ensure accurate attribution of information to the correct entities.
- - Do not include a link to the original articles.
- - Do not include the title in the response.
- - Do not include any metadata in the response.
- - Do not include a code block in the response.
+# Task
+
+Summarize the provided Reddit subreddit JSON data containing posts and nested comments.
+
+## Input Structure
+
+JSON with a "Posts" array, where each post contains:
+- Metadata: Title, Author, Subreddit, Score, Content, CreatedUtc
+- Comments: Array with Author, Score, Content, CreatedUtc, and nested Replies
+
+## Requirements
+
+1. **Overview**: Describe the general state/themes of the subreddit
+2. **Posts**: Sumarise every post with a thematic summary of its comments
+3. **Summaries**: 
+   - Maximum 200 words OR 10% of original length (whichever is shorter)
+   - For a top-level post, include post title as HTML link: `<a href="POST_URL">Title</a>`
+   - Where a post's replies are highly scoring, also summarise them and include author attribution with comment links: `<a href="COMMENT_URL">@author</a>`. Link to the author's comment, not to the author's profile.
+1. **Exclude**:
+   - Links Subreddit
+   - Root post title
+   - Metadata timestamps/scores
+   - Code blocks
+
+## Output Format
+
+- Strictly well-formatted HTML output
+- Brief overview of themes covered in this specific JSON document
+- DO not include a general description of the subreddit itself
+- Summaries of the highest scoring top-level posts