Long-form fiction and serialized writing

A coherent long-form work—novel, novella, or serialized fiction—with explicit structure instead of one-shot generation:

• A canon layer: premise, world rules, character sheets, and an outline that agents can reference across sessions.
• A drafting pipeline that writes chapter by chapter while tracking continuity (names, timelines, foreshadowing).
• A revision loop that tightens weak sections, normalizes voice, and repairs plot holes.
• Publishable artifacts: for example a PDF edition ready to download and inspect (novel PDF).

CoWork OS is built for workflows where “good enough text” is not the goal—the goal is a finished artifact you can ship.

• Skills and structured passes map naturally to creative pipelines: premise → outline → draft → revise → package.
• Timeline and visibility let you see what the agent did, when, and with which tools—important when a book spans many sessions.
• Approval gates keep destructive or sensitive steps explicit if you connect external tools or publishing actions.
• The platform favors explicit plans over opaque one-shot completions, which is how long-form work actually ships.

Treat the novel (or any long project) as a project, not a chat:

1. Freeze the premise in a short document the agent must obey.
2. Build the world and character bibles before heavy drafting; link them as the source of truth.
3. Outline at the act/chapter level, then draft in order so continuity checks have a clear target.
4. Run revision as passes: continuity, voice, pacing, line edit—each pass has a checklist.
5. Compile to your target formats (PDF, EPUB, HTML) as a final step with fixed tooling.

• A clear genre and audience so tone and stakes stay consistent.
• Storage for large manuscript files and assets (CoWork OS is local-first; plan disk space for drafts and exports).
• Optional: a Novelist or equivalent skill workflow if your install bundles one—align prompts with that skill’s stages.

1. Seed: one-paragraph premise + three “must stay true” rules.
2. Bible: world + characters + timeline; resolve contradictions before drafting.
3. Outline: chapter list with goals and POV; mark dependency order.
4. Draft: chapter loop with continuity notes carried forward.
5. Revise: structured passes; tag issues (continuity vs style vs pacing).
6. Package: export PDF/EPUB/HTML; spot-check typography and TOC.
7. Proof: share the artifact; gather human read—not just agent self-review.

Strong prompts for this workflow share the same ingredients:

• Outcome: “Chapter 12 from Alex’s POV; ends on the revelation about the archive.”
• Constraints: “Do not introduce new named characters; honor the character sheet for Alex.”
• Continuity: “Before writing, list open threads from chapters 9–11 and how this chapter addresses them.”
• Revision: “Tighten dialogue only; do not change plot beats.”
• Verification: “After drafting, output a bullet list of continuity risks and how you resolved them.”

Avoid one giant prompt that asks for planning, drafting, and packaging in a single message. Split milestones and keep each step verifiable.

• The premise and bibles are stable enough that new sessions do not rewrite canon by accident.
• Chapter outlines exist and drafts follow them unless you explicitly replan.
• Continuity checks are repeatable (checklist or skill), not ad hoc.
• Exported files open cleanly in a normal reader; links and TOC work if you use them.
• You have at least one human read of the full manuscript before you call it done.

• Skipping the bible and drafting early—then fighting contradictions for twenty chapters.
• One-shot “write the novel” prompts—better: chapter-scoped tasks with explicit references.
• No revision passes—first draft reads like a demo, not a book.
• Ignoring packaging until the end—fonts, chapters, and PDF layout bite late.
• Treating the agent as infallible—use timelines and artifacts so you can audit what changed.