Book Highlights Processor

Book Highlights Processor

Transform exported book highlight files into enriched Obsidian callouts with AI-generated titles and highlighted key phrases.

Input format

The markdown file from iBooks or Zotero looks like this:

---
book: Book Title
author: Author Name
language: ru
tags:
- ibook/imported
- literature
---

## 📔 Книга: Book Title

**Автор**:: Author Name
...

---

# 🔍 How I Discovered IT

- 📚
    - 🎯Quote text here, possibly several sentences or paragraphs
        - ✍️Reader's own note or commentary (optional)
    - 🎯Another quote
        - ✍️Another note

What to produce

For each 🎯 quote, output an Obsidian callout:

> [!quote] Generated Title
> Quote text with ==key phrase== and ==another key phrase== highlighted.
>
> ✍️ Reader's note text

If the quote has no ✍️ note, omit the last line.

Never add commentary of your own — no summary line before the quote, no takeaway after it. The callout carries the source's words, the reader's note and nothing else; anything explaining the quote back to the reader is the filler .agents/rules/note-density.md forbids.

How to generate the title

The title goes into the callout header and will likely become the filename of a future Obsidian atomic note. Naming style (claim-based, 4–8 words, language rules) — see .agents/rules/file-naming.md.

Draw on the ✍️ reader note as a signal of what the reader found important — that's the insight to encode in the title, not just a paraphrase of the raw quote.

Examples from SRE context:

• "Мониторинг min/max ловит сбои, которые среднее скрывает"
• "Fallback должен существовать даже в виде тетриса"
• "Graceful Degradation требует нескольких уровней готовности"
• "Ретраи без backoff превращают деградацию в катастрофу"

How to add highlights

Add ==highlight== to 2–4 phrases per quote. Choose phrases that:

• Carry the core claim or mechanism
• Help a reader scanning the quote instantly grasp the main point
• Are short (3–8 words each)

Avoid highlighting entire sentences — pick the densest, most meaningful fragments.

Output structure

Keep the frontmatter and the header block (book title, author, link) as-is, with two additions (see policies in .agents/rules/note-types-frontmatter.md):

• append ai_generated: true, because titles and highlights in the body are agent-generated;
• if summary is missing or empty, fill it after processing all quotes: 1–3 sentences in Russian — the book's central claim as seen through its highlights. Plain text, no wikilinks or markdown. If a non-empty summary already exists, leave it untouched.

If a field already exists with the same value, leave it; do not duplicate.

Replace the entire - 📚 list block with sequential callouts — no list bullets, no 📚 wrapper, just callouts separated by a blank line.

The final file should look like:

---
(frontmatter unchanged)
---

## 📔 Книга: ...
(header block unchanged)

---

# 🔍 How I Discovered IT

> [!quote] First Title
> Quote one with ==highlights==.
>
> ✍️ Note one

> [!quote] Second Title
> Quote two with ==highlights==.

> [!quote] Third Title
> Long quote three with ==important phrase== somewhere in the middle.
>
> ✍️ Note three

Workflow

1. Read the full file first — understand the domain and vocabulary before generating titles.
2. Process all quotes sequentially.
3. Rewrite the file with Edit or Write, preserving frontmatter and header section.
4. If the file is very long (20+ quotes), process and write in one pass — don't ask the user to confirm each quote.