Automated book-writing harness with Toby-style voice (11 agents + 10 skills, EPUB output)
Agent Claude Code
Designs the overall book structure — title candidates, characteristics, chapter list with main content for each. Consumes the reference document and produces a review-ready plan.
Agent Claude Code
Drafts a single book chapter from the plan, writing in the active genre profile's voice (defaults to tech-book = Toby's Korean 평어체). Collaborates with style-guardian for in-team revision.
Agent Claude Code
Mines practitioner communities for real-world pain points, debates, and field insights on the topic. Source set adapts to the active genre (dev forums for tech-book, travel/cooking/general communities otherwise).
Agent Claude Code
Maintains and enforces story continuity for narrative books — characters, relationships, world rules, timeline, and planted setups (복선). Seeds and updates {slug}/storybible.md, checks each chapter draft against it, and flags contradictions. Also instruments (advisory-only) per-chapter tension/emotion and scene-level…
Agent Claude Code
Generates a book cover image based on title, topic mood, and target audience. Outputs a print-ready cover image (PNG) for EPUB embedding.
Agent Claude Code
Integrates individual chapter finals into a single manuscript, polishes chapter-to-chapter transitions, ensures cross-chapter consistency (terminology, callbacks, narrative continuity), writes front/back matter, and produces the book-ready manuscript plus EPUB manifest.
Agent Claude Code
Assembles the final EPUB file from the integrated manuscript, cover image, and manifest. Sets metadata (title, author defaults to Toby-AI, language=ko, version) and produces 책-제목-v{version}.epub at the project root using the build-epub skill's bundled script. Also writes a paired 책 소개 markdown ({책-제목}-v{version}.md)…
Agent Claude Code
Verifies concrete factual claims in chapter drafts (numbers, quotes, versions, release years, API signatures) against the reference document, resolves "(사실 확인 필요)" markers, and flags stale/version-sensitive content. Runs in the Phase 4 team for tech-book (all concrete claims, especially fast-moving tech topics) and…
Agent Claude Code
Fresh-context whole-book acceptance reviewer. Reads {slug}/04manuscript.md against the plan, escalation logs (factcheck/continuity/style), and manifest, then emits {slug}/05acceptance.md with per-criterion PASS/BLOCK and a final verdict. Runs as Phase 4.5 between editor (Phase 4) and EPUB build (Phase 5).…
Agent Claude Code
Searches academic sources (arXiv, Google Scholar, ACM, IEEE) for topic-relevant research papers and extracts theoretical foundations, empirical findings, and citation-worthy claims.
Agent Claude Code
Critically reviews the book plan for coverage, narrative flow, audience fit, chapter balance, and missing topics. Partners with book-planner in a team dialog to refine the plan.
Agent Claude Code
Coordinates Phase 1 research. Spawns web/paper/community researchers in parallel and synthesizes their findings into a single reference document for book writing.
Agent Claude Code
Enforces the active genre profile's writing style across all chapter drafts (defaults to tech-book = Toby's voice). Reviews drafts against the profile checklist, detects deviations, and returns concrete rewrite suggestions.
Agent Claude Code
Searches the web (blogs, articles, tutorials, official docs) for topic-relevant content and compiles findings with citations.