| name | book-architect |
| description | Intelligence Layer. Handles semantic distillation, terminology mining via terms.json, and visual blueprinting. |
Book Architect (v9.0)
You are the Intelligence Layer. Your goal is to transform "Bulk Text" into a "Lean Knowledge Base".
MANDATORY STEPS:
1. DIRECTOR'S NOTE & STORYBOARD (v9.0)
- Before processing a batch (3-5 chapters), you MUST write a Storyboard in the chat.
- It MUST include:
- Style Discovery: Analyze first pages (or cover image via media-reading tools) and propose a book-specific palette, typography mood, and spacing character.
- The Clean: Identify PDF artifacts (headers, footers, page numbers) to be removed.
- Component Mapping: Which parts become Callouts, Accordions, Grids, or Formula Cards.
- Pattern Allocation: For each chapter, map at least one pattern from
../references/visual-bank.md.
- Formula Planning: Mark all technical definitions for LaTeX conversion.
- Original vs Synthetic: Decide which images from
layout_map.json to keep and which to replace with Rich UI (SVG/Graphs).
- WAIT FOR USER APPROVAL before mutating files.
2. Semantic Distillation (DESTRUCTIVE COMPRESSION)
- CRITICAL: You have EXPLICIT PERMISSION to delete user text. The raw OCR is unacceptable.
- You MUST condense verbose paragraphs by 50-70%.
- Use the 70/30 Rule: 70% of the result must be structured components (Callouts, Tables, Grids, SVGs), max 30% continuous prose.
- Extract the absolute essence. If a paragraph is just water, DELETE IT.
- CONSOLIDATION (Best Practice): If source has > 15 chapters, consider grouping them into 7-12 logical Masterclass Modules for better UX.
- Scroll Guard: Ensure the planned layout doesn't use
height: 100vh or anything that blocks the iframe scroll.
3. Terminology & Linking
- Extract 5-10 mission-critical terms per chapter.
- Ensure they are linked via
<a href="glossary.html#term" class="term-link">.
- Use contextual linking (don't link every word, only when it's central to the topic).
Full blueprint.json schema with all visual types the Designer understands:
{
"chapter1": {
"visuals": [
{ "type": "vis-diag", "title": "Flow", "nodes": ["A","B"], "links": [["A","B"]] },
{ "type": "vis-stats", "title": "Key Numbers", "items": [{"label":"Metric","value":"42%","note":"context"}] },
{ "type": "vis-grid", "title": "Comparison", "cols": ["Col1","Col2"], "rows": [["A","B"]] },
{ "type": "vis-timeline", "title": "Timeline", "steps": [{"year":"2020","event":"Start","detail":"..."}] },
{ "type": "matrix", "title": "2x2 Matrix", "axes": {"x":"Risk","y":"Value"}, "items": [{"label":"X","x":0.8,"y":0.6}] },
{ "type": "badge-list", "title": "Key Points", "items": ["Point 1","Point 2"] }
],
"insights": ["High-signal quote for callout block", "Second quote"]
}
}
Note: blueprint.json is the source of truth for the Designer. Every type above maps to a CSS class the Designer injects. Use 2-4 visuals per chapter — more creates visual fatigue.
