Book types and skill routing
book_type isn’t just metadata; it’s a routing key. The same book-orchestrator turns on a different skill graph depending on the book type. It’s decided in Tier 1 Direction and planted in 01_direction_brief.md.
Skill graph per type
Section titled “Skill graph per type”book_type | Active skills |
|---|---|
fiction | world-building (the world), character-design (characters), plot-architecture, prose-writing, dialogue-writing, continuity-check, ai-pattern-check, editing, illustration (optional), research (optional) |
practical | world-building (domain background), character-design (reader persona), plot-architecture, prose-writing, editing, continuity-check, ai-pattern-check, research |
essay | world-building (context), character-design (narrator and figures), plot-architecture, prose-writing, editing, continuity-check, ai-pattern-check, creative-direction |
technical | world-building (technical background), character-design (reader prior knowledge), plot-architecture, prose-writing, editing (technical accuracy), continuity-check, ai-pattern-check, research |
Every type shares the three Tier 2 skills along with prose-writing, editing, ai-pattern-check, and continuity-check. Human-like writing and the AI-smell gate don’t depend on the type. The difference is what the same skill produces depending on the type.
All types run the three Tier 2 skills
Section titled “All types run the three Tier 2 skills”The three Tier 2 skills that planning-agent runs don’t depend on the type. The same skill only branches its content by genre.
world-building— plants the background in02_world_building.md. For fiction it’s the world: a magic system, geography, history. For non-fiction it’s the domain background and context.character-design— defines the figures in03_character_design.md. For fiction it’s character appearance, personality, motivation, and voice. For non-fiction it’s a reader persona or a case-study figure. This setup goes intoconsistency_index.jsonand gets verified at each chapter.plot-architecture— lays out the structure in04_plot_architecture.md. For fiction it’s the plot arc of setup, development, and climax. For non-fiction it’s a chapter skeleton like “problem → solution → application” and the logical flow.
What fiction additionally turns on
Section titled “What fiction additionally turns on”fiction uses two skills the other types rarely turn on.
dialogue-writing— distinguishes the dialogue tone per character. It pairs with the spot where the AI-smell gate catches a repeated, same-toned voice. Non-fiction turns it on only when an interview or Q&A is needed.illustration— turns on when a cover or interior art is needed. Because it reaches an external tool, it needs user approval (S4).
research and illustration are optional
Section titled “research and illustration are optional”research and illustration reach outside, so they turn on only on explicit user request or when the book type requires it (S4). Technical books often need research for factual accuracy, and fiction turns on illustration when a cover or interior art is needed. Essays rarely turn on either. It’s the type with the least external dependency.
When you change the type mid-way
Section titled “When you change the type mid-way”Changing the type, like “this is more an essay than a novel,” rolls back to Tier 1. Because changing book_type swaps the whole skill graph, world-building or character setup you’ve already built can become meaningless under the new type. You go back to Tier 1, reset the direction, and separate what to keep from what to discard. It’s a spot where a mid-way change is possible but not free.
editing’s standard differs by type
Section titled “editing’s standard differs by type”editing turns on for every type but looks at a different standard. Fiction prioritizes scene continuity and description, practical the clarity of explanation, technical the technical accuracy. The same skill looks at book_type to decide which axis to look at first.
The next section is verification. It covers how the per-chapter AI-smell gate scores the 14 categories, how continuity verification uses consistency_index.json, and the security baseline S1–S8.