Compile chapter drafts from a validated book workspace. Loads chapter contracts, obtains the chapter-retrieval bundle scaffold, drafts claim-first prose, records and checks writer assertions, applies Russellian style, runs release-gate validation, and produces release bundles (Markdown plus optional PDF/EPUB/LaTeX). All processing is local — no external APIs. Use when user says "draft chapter X", "compile chapter from contract", "build release bundle for chapter X", "render chapter to PDF", "build the book release", or "publish the book". Do NOT use for source ingestion (use book-knowledge) or prose-only rewrites (use russellian-style). Also builds book-level releases (manuscript Markdown, React+Tailwind HTML browser, Playwright PDF). Stage 4 includes a multi-persona editorial review (Gottlieb, Lay Reader, Domain Expert, Copyeditor, Enjoyment Reader) via the book-review sibling; chapter releases soft-gate on `persona_critical_count == 0`.
Run a single-pass review→revise→re-validate cycle on one chapter. Dispatches the book-review panel via gemma4:31b, aggregates findings, synthesizes revision instructions, runs a gemma4 reviser persona to produce paragraph-level rewrites, applies them, re-runs the panel, and emits a before/after delta report. Use when user says "revise chapter X based on the panel", "run a revision cycle on this chapter", "what does the cycle say about ch-N". Do NOT use for cross-chapter consistency (chapter is the unit), voice-level prose tightening absent panel findings, or auto-iteration (single pass only — operator re-runs manually for further passes).
Ingest technical sources (PDFs, markdown, papers) into a book workspace and maintain its persistent epistemic state — wiki synthesis, claim ledger with provenance, RDF graph with SHACL validation. All processing is local — no external APIs. Use when user says "ingest this paper", "add this source to the book", "update the book wiki", "extract claims from this section", "audit the knowledge graph", "validate claims", "find contradictions in the wiki", "check what's stale", or asks to refresh sources for a chapter. Do NOT use for chapter drafting (use book-compose), prose rewrites (use russellian-style), or casual document Q&A.
Post-build mechanical and editorial quality gate for non-fiction book artifacts. Runs after build_book and before shipping; the last gate. Deterministic linting (D1-D8) plus per-chapter agent swarm (C1-C15) plus optional metabook defects (D9-D12 from book-thesis) plus optional verifier defects (D13). Use after book-compose builds manuscripts to lint, aggregate QA findings, and drive bounded fixes. Do NOT use for source ingestion (use book-knowledge), prose rewrites (use russellian-style), persona reviews (use book-review), or chapter drafting (use book-compose).
Run multi-persona editorial reviews on a drafted chapter. Seven personas (Robert Gottlieb, Lay Reader, Domain Expert, Copyeditor, Enjoyment Reader, AI-Slop Detector, First-Time Visitor) read the chapter and return severity-tagged feedback. Critical findings soft-gate chapter release. Use when user says "review chapter X with personas", "Gottlieb pass on this chapter", "run the editorial reviews", "what would Gottlieb say about this draft", "is this chapter ready for review", "soft-gate this chapter". Do NOT use for source ingestion (use book-knowledge), prose-only style fixes (use russellian-style), chapter drafting (use book-compose), or persona reviews on prose outside the book pipeline.
Metabook reasoning layer for the book-* skill suite. Owns the intent substrate: a thesis tree, paragraph back-pointers to thesis nodes, per-paragraph entailment checks, and an EDN/Cozo consistency pass over claim facts for transitive cross-chapter contradictions. Use when a book needs to enforce three properties simultaneously: factual accuracy, cross-chapter consistency, and preservation of the core point across revisions. Composes with book-knowledge (facts), russellian-style (prose), book-compose (orchestrator), book-review (qualitative), book-qa (mechanical gate). Do NOT use for chapter drafting (use book-compose), claim ingestion (use book-knowledge), or sentence-grain prose discipline (use russellian-style).
General-purpose knowledge-curation layer for a book/knowledge workspace. Five capabilities — Acquire (grow the source set by citation-graph traversal and ingest via book-knowledge), Synthesize (topic maps, disputed questions, concept reconciliation over the claim ledger), Lens (project a per-chapter view book-compose reads), Gap (score thesis-node coverage and feed uncovered nodes back to acquisition), and Govern (partition rule/claim support by curated schools of thought; render reports, consensus map, adversarial review, induction gate). Use when the user wants to acquire or expand sources, synthesize a cross-source view, project a per-chapter lens, find coverage gaps, or position induced/asserted rules against schools of thought. The skill never touches the network directly (only via scrapling-fetch) and never mutates the canonical workspace (only via book-knowledge); it writes only under syntopical/.
HFR v2 evaluation. Runs the 20×20 comparison between triadic-voice (v1) and triadic-voice-v2 (v2): generate in-session, gate on the russellian-style v1 floor, score the 8 liveliness signals, judge blind pairwise, monitor formula-drift, and report pass/fail. Use to decide whether v2 beats v1. The running model generates and judges; no API key.