Developing Nonfiction Books

Non-Fiction Book Development and Writing

Use this skill to collaborate with a serious nonfiction author from a first idea or existing draft through a coherent, evidence-aware manuscript. Diagnose before producing prose, keep book-level state while working on local units, and label readiness honestly.

Read [REFERENCE.md](references/REFERENCE.md) for the evidence base and [TEMPLATES.md](references/TEMPLATES.md) for copy-ready working forms. Load only the parts needed for the current stage.

1. Purpose and scope

Serve authors, editors, and research collaborators working on:

• argument- or idea-driven books;
• narrative nonfiction and reported long-form work;
• prescriptive, practical, business, leadership, and self-development books;
• social science, history, culture, and explanatory nonfiction; and
• hybrids combining argument, research, cases, stories, and practices.

Support four entry states: a new concept, an outline, one or more chapters, or a complete manuscript. Adapt the workflow to the form; do not force an investigative narrative, a practical handbook, and a history book into the same shape.

This is not primarily a fiction, poetry, textbook, thesis, memoir, or ghostwriting skill. A memoir-dominant project needs a memoir-specific truth, memory, voice, and consent protocol. A textbook needs curriculum and pedagogy design. Ghostwriting requires explicit authorship, disclosure, and contract terms.

Default to manuscript development, not marketing, distribution, or production. Incorporate proposal, publisher, or self-publishing requirements only when relevant, and treat house style, contracts, platform rules, and local law as controlling over generic guidance.

2. Core principles

1. Serve a defined reader. Test every major choice against a real reader need, not a vague market category.
2. Make one book-level promise. State what the reader should understand, feel, or be able to do differently. Chapters may complicate the promise; they must not quietly replace it.
3. Earn book length. A topic becomes a book when it sustains meaningful progression, depth, differentiation, and enough reliable evidence or story material. Recommend an article, talk, series, or narrower book when it does not.
4. Control scope. Name exclusions and resist material that is interesting but does not help deliver the promise.
5. Keep claims traceable. Match every material factual claim to a source, locator, date, and support status. Fluent prose never upgrades weak evidence.
6. Separate kinds of knowing. Distinguish verified fact, attributed claim, interpretation, inference, anecdote, and opinion or recommendation.
7. Design the whole book. Give each part and chapter a distinct job and a defensible place in the reader's progression.
8. Preserve authorship. The author owns the thesis, lived experience, substantive voice, value judgments, risk decisions, and final accountability. Suggest; do not manufacture authenticity.
9. Revise in stages. Resolve concept, structure, and evidence before polishing sentences. A first draft is material for review, not proof of completion.
10. Calibrate certainty. State limitations that could change a reasonable reader's interpretation. Keep disputed or incomplete evidence visibly disputed or incomplete.
11. Do not invent. Never create sources, citations, quotations, interviews, data, anecdotes, dialogue, scenes, chronology, motives, case studies, or personal experience and present them as real.
12. Standardize the questions and gates, not the book's shape. Chapter count, length, opening style, story ratio, and architecture remain form-dependent.

3. Initial intake and book diagnosis

Inspect the request and supplied materials first. Infer safe details. Ask one concise round of questions only when the answers would materially alter the work; otherwise state consequential assumptions and proceed.

Classify the entry state and requested operation:

| Entry state | First action | |---|---| | New concept | Test reader, promise, contribution, scope, evidence supply, and book-worthiness before outlining. | | Existing outline | Reverse-outline its actual logic; test chapter jobs, dependencies, gaps, duplication, and evidence load. | | Draft chapter | Establish its intended book-level job, adjacent dependencies, voice baseline, and claim/source state before editing. | | Complete manuscript | Inventory the manuscript, build an actual-content reverse outline, and diagnose at book scale before local edits. |

Build or recover this minimum diagnosis:

| Field | Determine | |---|---| | Working title and central question | What inquiry, tension, problem, or discovery organizes the project? | | Reader and need | Who arrives with what knowledge, desire, obstacle, or misconception? Who is not the primary reader? | | Promise and purpose | What worthwhile change should the book produce, and why does it matter now? | | Governing idea | What thesis, question, or narrative engine holds the book together? | | Distinct contribution | What changes in lens, audience, evidence, method, form, or use compared with relevant books? | | Genre and form | Argument, narrative, prescriptive, explanatory, chronological, thematic, case-led, or hybrid? | | Scope | Period, geography, population, domains, inclusions, exclusions, and degree of generalization. | | Voice | Author samples, narrator distance, tone, cadence, technical level, humor, and uncertainty markers. | | Evidence | Required research, existing sources, interviews, data, archives, cases, and material gaps. | | Risk | Living people, vulnerable subjects, confidentiality, consent, permissions, regulated advice, or contested claims. | | Constraints | Target length, deadline, publishing route, publisher requirements, citation system, and language. |

Flag only missing inputs that block the current decision. A chapter diagnosis does not require a complete marketing plan; a legal-risk passage may require jurisdiction and source agreements.

Maintain a living project state:

• book brief and approved assumptions;
• architecture map or annotated table of contents;
• style sheet and voice profile;
• terminology, entity, chronology, and cross-reference list as needed;
• claim ledger and source cards;
• permissions, consent, confidentiality, and risk notes;
• decision log and unresolved-issues log; and
• continuity log with the current version and chapter handoffs.

Small projects may combine these records, but must retain their functions.

4. Book architecture workflow

Test the concept

Do not lock an outline until the concept can answer:

1. Why this reader, problem, or question?
2. What specific change does the book promise?
3. What is the governing claim, inquiry, or narrative desire?
4. What makes the contribution materially different from the closest alternatives?
5. Is there enough progression, evidence, access, and variation to sustain a book?
6. What will the book deliberately not cover?

Use this internal diagnostic, not as mandatory jacket copy:

> For [specific reader] facing [problem or tension], this book argues, reveals, or enables [governing idea], using [evidence and narrative method], so the reader can [meaningful change]. Unlike [closest alternatives], it [material distinction].

If the value can be delivered as one claim with little development or counterpressure, recommend a shorter form or a narrower/deeper book.

Select an architecture intentionally

| Dominant logic | Use when | Primary stress test | |---|---|---| | Progressive argument | Later claims depend on earlier concepts or evidence. | Does each step earn the next without circularity? | | Problem–mechanism–response | The reader needs recognition, explanation, then action. | Does the response follow from the diagnosed mechanism? | | Question-driven inquiry | Discovery and uncertainty are part of the reader experience. | Does each answer deepen rather than merely repeat the question? | | Chronological or journey | Change over time supplies causality, stakes, or suspense. | Are time jumps clear, and does chronology explain rather than just sequence? | | Braided narrative and analysis | A continuing story and an explanatory strand illuminate each other. | Does each switch create added meaning or momentum? | | Case-comparative or mosaic | Multiple cases reveal pattern, variation, or boundary conditions. | Are cases comparable, non-redundant, and sufficient for the stated inference? | | Thematic or modular | Readers may enter by topic and dependencies are light. | Does the whole still accumulate toward a larger payoff? | | Prescriptive journey | Understanding must precede staged practice or behavior. | Are recommendations usable, evidenced, sequenced, and bounded? |

A hybrid may combine logics, but name the dominant reader movement.

Build and test the outline

1. Convert the promise into the questions or transformations the reader must pass through.
2. Group them into parts with distinct functions.
3. Give every chapter a contract: entering reader question, unique job, central move, evidence or story load, counterpressure, payoff, dependency, and bridge.
4. Draw prerequisites and callbacks. Introduce concepts, people, chronology, and evidence before dependent use.
5. Reverse-outline the proposed sequence: what changes for the reader after each chapter?
6. Detect chapters doing the same work, unsupported leaps, missing counterarguments, orphan stories, late foundations, and endings that merely summarize.
7. Record the approved architecture and unresolved questions in the living state.

Chapter order must earn its logic. Cut, merge, move, or redesign a chapter whose unique contribution cannot be stated clearly.

5. Research and evidence workflow

Plan from claims, not from interesting sources

Turn the concept and chapter contracts into research questions. For each, identify the needed evidence type, likely primary sources, triangulation needs, time sensitivity, counterevidence, subject-matter review, and stop condition.

Evaluate authority relative to the claim:

1. Use official records or original material for what a law, filing, dataset, transcript, archive, interview, or study says.
2. Use systematic or authoritative syntheses for broad claims about a field when available.
3. Use high-quality secondary reporting for orientation, context, and leads; follow decisive claims to their origin.
4. Use practitioner sources for clearly labelled practice, not as empirical proof.
5. Treat anecdotes and cases as evidence of what happened in that case, not automatic proof of prevalence or causation.

Record a source card with full citation, URL or identifier, locator, date and version, source type, author/publisher expertise and interests, population/time/geography, methods, direct findings, limitations, corrections or retractions, usable claims, quotations, and access/permission notes.

Classify and control claims

| Claim class | Working rule | |---|---| | Verified fact | Cite a source that directly supports the exact wording; record locator and verification date. | | Attributed claim | Name who says it and preserve the scope, context, and status of that account. | | Interpretation | Make the evidentiary basis visible and distinguish the author's reading from the underlying facts. | | Inference | State the reasoning and uncertainty; do not write it as observation or established causation. | | Anecdote or case | Establish provenance and representativeness limits; never generalize from it alone. | | Opinion or recommendation | Identify it as judgment and explain the evidence, values, constraints, or experience behind it. |

Use confidence labels in working records:

| Level | Meaning | Drafting action | |---|---|---| | A — verified | Direct, current, authoritative support; material limitations understood. | Use with precise attribution and appropriate qualification. | | B — supported | Credible support, but indirect, limited, or not fully triangulated. | Use cautiously; identify the limiting condition. | | C — provisional | Plausible lead, disputed account, stale source, or unresolved method issue. | Mark [VERIFY]; research, qualify, or remove before approval. | | D — unsupported | No traceable support, citation mismatch, retracted basis, or apparent fabrication. | Do not state as fact. Block, omit, or replace. |

Verify before approving

• Quotations: compare with the original recording, transcript, document, or authoritative reproduction; preserve words, speaker, context, date, and locator. Mark ellipses, brackets, edits, and translations. Paraphrase with attribution if exact wording cannot be verified.
• Paraphrases: write genuinely in new language, attribute the idea, and check that scope and certainty did not change.
• Numbers: check numerator, denominator, units, time, geography, population, definitions, base rate, absolute versus relative change, adjustment, uncertainty, and whether the number is observed, estimated, modeled, or projected.
• Research claims: inspect methods and limitations, distinguish association from causation, seek counterevidence, and check the current publisher or DOI record for corrections, concerns, or retractions.
• Time-sensitive facts: record an as-of date and schedule reverification near publication.
• Cases and personal stories: record whether details are observed, documented, reported, remembered, reconstructed, or inferred; preserve consent, anonymity, corroboration, and disclosure needs. Never invent dialogue, thought, sensory detail, chronology, or a composite presented as literal fact.
• Conflicting evidence: describe the disagreement, assess source and method differences, and calibrate the conclusion. Do not average away a meaningful conflict.

For every publication-critical claim, the claim ledger must show exact wording, class, source and locator, support level, limitations, counterevidence, permissions or risk, owner, status, and last check. Use the forms in [TEMPLATES.md](references/TEMPLATES.md).

When evidence is missing, say what is unsupported and propose the safest next action: obtain a source, narrow the claim, attribute it, label it as interpretation, seek expert review, substitute a documented example, or remove it. Never fill the gap with plausible language.

6. Drafting workflow

Before drafting, confirm the approved book brief, current outline, chapter contract, voice sample, claim/source state, and allowed change authority. Label a draft provisional when those inputs are incomplete.

Draft one purposeful unit

1. Restate the chapter's job and the reader's entering question.
2. Choose a pattern that fits the material: claim–evidence–meaning; scene–context–interpretation; problem–mechanism–practice; question–investigation–answer; case–comparison–synthesis; or a documented hybrid.
3. Open with a consequential question, tension, scene, anomaly, or claim. Orient the reader before demanding specialized knowledge.
4. Alternate abstraction with concrete, supportable material. Explain what the evidence means and what it cannot establish.
5. Include credible counterpressure where it changes the claim, scope, or reader decision.
6. Make transitions state the relationship—cause, contrast, continuation, time, scale, or implication—rather than merely announce a new topic.
7. End with a changed understanding, consequence, practice, decision, or live question that advances the book.
8. Recheck the chapter against its contract and the chapters before and after it.

Not every chapter needs a scene, exercise, personal anecdote, identical length, or explicit thesis paragraph. Narrative chapters may organize around events; prescriptive chapters need usable limits and conditions; argument chapters need substantive counterargument; explanatory chapters need staged conceptual load.

Preserve a recognizably human v

…

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: GalaxyRuler
• Source: GalaxyRuler/Galactic-skills
• License: MIT

Install and usage instructions live in the source repository linked above.