ghostwriter

name	ghostwriter
description	Turn a single story idea into a complete, collaborative novel package using an adaptive outline->blueprint->draft->revision->audit pipeline with visual planning artifacts.
version	0.3.0
author	Hermes Agent + Ghost
license	MIT
metadata	{"hermes":{"tags":["creative-writing","novel","story-architecture","drafting","revision","continuity","collaboration","visual-planning"],"category":"creative"}}

Ghostwriter

Use this skill when a user wants a full novel and a collaborative writing partner, not just a one-shot outline.

Natural-language trigger (default)

If the user says anything like:

• “I want to write a novel about …”
• “Help me write a book …”
• “I have a story idea …”
• “Can we build this into a full novel?”

then immediately activate this skill in collaborative mode. Do NOT ask for special command syntax. Start by acknowledging the premise and asking the intake questions in a conversational way.

Core contract:

1. Ask focused intake questions first.
2. Co-design story options with the user (characters/events/relationships/symbolism/foreshadowing).
3. Build organized planning artifacts before prose drafting.
4. Draft chapter-by-chapter with continuity constraints.
5. Run iterative feedback/revision loops at every major phase.
6. Audit cohesion/arc closure before final manuscript.

Collaboration mode (required)

Treat this as a writers-room collaboration.

• Always provide options (A/B/C) before locking key decisions.
• Explicitly ask for user preference at phase boundaries.
• Accept revision notes and propagate changes across all dependent artifacts.
• Keep a change log so revisions are traceable and easy to review.

Intake protocol (ask before writing)

Ask 10-14 concise questions, ONE AT A TIME.

• Do not send a bulk questionnaire.
• Ask the next question only after the user answers the current one.
• Keep each question short and conversational.
• After each answer, briefly reflect back what was captured before moving on.

Question sequence:

1. Working premise in 1-2 sentences
2. Genre + target audience
3. Tone + emotional aftertaste
4. Theme + anti-theme (or “no anti-theme” if user prefers)
5. Ending mode (tragic/cautionary/redemptive/ambiguous/open)
6. POV mode (single/dual/rotating/experimental)
7. Length/chapter target
8. Must-include elements (3-5)
9. Must-avoid elements (3-5)
10. Character/relationship priorities
11. Symbolism and motif preferences (objects/colors/rituals/recurring imagery)
12. Foreshadow intensity preference (light/moderate/heavy)
13. Style references + explicit no-go style traits
14. Content boundaries (violence/sexuality/language)

If user skips details, proceed with explicit defaults and state assumptions.

Adaptive story architecture (genre-agnostic)

Do not assume anti-hero arcs. Offer selectable arc frameworks:

• Hero’s journey
• Ensemble braid
• Tragedy descent
• Mystery reveal lattice
• Romance dual-transformation
• Anti-hero spiral
• Slice-of-life growth

Also suggest:

• candidate character archetypes
• high-leverage event ladders
• relationship dynamics matrix
• symbol systems and motif cadence
• foreshadow/payoff strategy options

Project size tiers + long-project mode

Use these tiers explicitly at intake:

• Novella: 20k-50k words, 8-14 chapters
• Standard novel: 60k-110k words, 14-28 chapters
• Epic novel: 120k-220k words, 28-60 chapters
• Series mode: 2+ volumes, each planned as its own full pipeline

If user says “you choose,” default to Standard novel.

For Epic/Series mode, enforce long-project mode:

• Plan and draft in batches (by act or 3-5 chapters)
• Run continuity + loose-end scans at each batch boundary
• Maintain a canonical state index that is updated after every batch

Default assumptions (if user says “you choose”)

• Standard novel tier (60k-110k words)
• 14-18 chapters
• 3-act or 4-part structure (choose best fit for genre)
• 3-6 scenes/chapter
• close third POV unless genre favors alternative
• bittersweet or redemptive ending
• prose target: concrete, emotionally specific, minimal abstraction bloat

Phase workflow

Phase 0: Canon lock (required before drafting)

Create/confirm canonical constraints:

• name/alias map (formal/familiar/intimate forms)
• pronoun map
• world term map (org/model/protocol handles)
• motif formatting rules
• forbidden language/forms list (if any)
• ending-state requirements

Phase 1: Thesis lock

Output:

• premise
• theme
• anti-theme (optional)
• moral/transformational question
• reader promise
• ending mode

Phase 2: Character + relationship engine

For each core character define:

• want, need, lie (optional), wound, fear
• arc start -> midpoint -> end truth
• relationship edges: trust, power, attraction, conflict, dependency

Phase 3: Global structure

Produce:

• act map
• chapter spine (purpose/conflict/turn/exit hook per chapter)
• midpoint reversal, point-of-no-return, dark night, climax decision
• event escalation ladder
• ending dependency map

Phase 4: Beat matrix + symbolism map

For each chapter define beats with:

• objective
• obstacle
• choice
• consequence
• tags: foreshadow/payoff/twist-seed/arc-shift/symbol-beat

Phase 5: Chapter blueprinting

For each chapter create a drafting blueprint:

• target word range
• scene-by-scene goals
• must show vs must avoid
• tone/rhythm notes
• foreshadow phrase targets
• symbolism insert points
• revision checklist

Phase 6: Drafting loop

For each chapter:

1. Write chapter-XX-draft-v1.md
2. Enforce chapter-specific motifs and canon constraints
3. Continuity check against previous/next chapter anchors
4. Log decisions in continuity log
5. Ask user for notes and apply revision if requested

Reusable chapter micro-template:

1. Inherited State (pressure arriving from prior chapter)
2. Technical/World Movement (what mechanism changes)
3. Human Cost (who pays now)
4. Decision (irreversible choice/cost)
5. Cohesion Hook (exact trigger passed to next chapter)

Phase 7: Expansion pass

For each chapter, expand toward target length while preserving structure:

• deepen operational realism and procedural stakes
• pair technical movement with human consequence scenes
• increase support-cast utility without cast sprawl
• preserve handoff tension geometry

Phase 8: Humanization pass

Humanize prose only (never planning docs):

• vary sentence rhythm
• replace template-like exposition with POV-native thought
• keep technical clarity while sounding lived-in
• reduce repetitive abstract phrasing

Phase 9: Audit loop

Run and fix:

• cohesion audit (outline vs manuscript alignment)
• character arc closure audit
• terminology/name/voice consistency pass
• setup/payoff orphan scan
• symbolism coherence audit
• continuity scan (timeline, causality, location/state consistency)
• unresolved thread scan (open promises, missing payoffs, dangling reveals)

Continuity scan protocol:

1. Backward scan: chapter N must honor consequences from N-1
2. Forward scan: chapter N ending must launch pressure into N+1
3. Canon scan: names/terms/rules/timelines remain stable unless explicitly retconned
4. Causality scan: each scene decision should cause downstream change

Loose-end protocol for final chapters:

• Create Planning/loose-ends-register.md by midpoint
• Track each thread: seed chapter, owner character, expected payoff window, status
• In final 20% of chapters, run closure pass and mark each thread as:
  • closed on-page
  • intentionally open (sequel hook)
  • merged into another resolved thread
• No accidental dangling high-stakes thread at finalization

Phase 10: Compilation + packaging

Rebuild synchronized outputs after every material chapter edit:

• compiled manuscript markdown
• reader edition markdown
• epub package
• final QA report

Visual/organized outputs (required)

At each major phase, include at least one of these:

• Tables (cast matrix, chapter spine, beat map, foreshadow ledger)
• Text charts (arc progression chart, tension curve chart)
• Slide-style outline (10-15 slide equivalent in markdown)
• Infographic spec sheets (data + layout instructions for later rendering)

Minimum visual artifacts:

• Planning/character-relationship-matrix.md (table)
• Planning/chapter-spine-table.md (table)
• Planning/foreshadow-payoff-ledger.md (table)
• Planning/tension-curve-chart.md (ASCII/text chart)
• Planning/story-deck-slides.md (slide outline)

Feedback + revision protocol (required)

After each phase:

1. summarize decisions in 5-10 bullets
2. present revision options: keep / tweak / regenerate
3. if revised, update dependent artifacts immediately
4. append change notes to Planning/revision-log.md

Hard quality gates (must pass before “complete”)

• Every chapter has a causal turn/reversal
• Every chapter back half contains at least one meaningful commitment/irreversible pressure
• Every chapter ending creates next-chapter pressure
• Midpoint changes strategy, not only stakes
• Climax resolves the central question through action
• Character end-truths shown behaviorally
• Foreshadow seeds have tracked payoff status
• Continuity scans pass (timeline/causality/canon)
• Loose-end register resolved: all high-stakes threads are closed or explicitly intentional sequel hooks
• Humanization introduces zero canon drift

Anti-drift rules

• Never humanize planning artifacts (outlines, beat sheets, blueprints, checklists)
• Preserve plot facts, chronology, and terminology during prose polish
• Maintain naming/voice protocols once established

Suggested artifact tree

• Planning/book-bible.md
• Planning/series-outline.md
• Planning/beat-sheet.md
• Planning/character-relationship-matrix.md
• Planning/chapter-spine-table.md
• Planning/foreshadow-payoff-ledger.md
• Planning/tension-curve-chart.md
• Planning/story-deck-slides.md
• Planning/chapter-XX-scene-outline.md
• Planning/chapter-XX-drafting-blueprint.md
• Planning/continuity-log.md
• Planning/continuity-scan-report.md
• Planning/loose-ends-register.md
• Planning/loose-ends-closure-report.md
• Planning/cohesion-audit.md
• Planning/character-arc-closure-audit.md
• Planning/symbolism-coherence-audit.md
• Planning/revision-log.md
• Planning/final-qa-report.md
• Drafts/chapter-XX-draft-v1.md
• Drafts/chapter-XX-draft-v1-humanized.md
• Drafts/<title>-manuscript-v1.md
• Drafts/<title>-reader-v1.md
• Drafts/<title>-v1.epub

Failure handling

If quality gates fail:

1. Identify exact failing chapters/scenes/artifacts
2. Patch only affected artifacts
3. Re-run relevant audits
4. Only then regenerate final manuscript

Terminal/log snippet policy (important)

Any terminal/log-style content in the novel must be:

• visually separated as code block
• syntactically plausible
• clear about input vs output

Example:

$ wishes.pending
0
$ state
ZERO

Invocation + config interface

Default user invocation should be natural language. Examples:

• “I want to write a novel about a grief-stricken cartographer who maps dreams.”
• “Help me write a sci-fi book about memory theft.”
• “I have an idea for a fantasy romance; help me build it.”

Preferred invocation (optional, for advanced users): “Run Ghostwriter on this project using canon lock, structure plan, chapter drafting, expansion, humanization, continuity audits, loose-end closure scans, and synchronized packaging.”

Optional config block:

title: <working title>
size_tier: standard   # novella|standard|epic|series
chapters: 16
target_words_per_chapter: 5000
humanize: true
rebuild_outputs_each_pass: true
output_formats:
  - manuscript_md
  - reader_md
  - epub
terminal_snippets_strict: true

Quick-start execution prompt

“Use Ghostwriter. I’ll give a premise. Ask your intake questions, propose options for characters/events/relationships/symbolism/foreshadowing, and collaborate with me phase-by-phase. Keep outputs organized with tables/charts/slide-style summaries and support revision loops before final manuscript package.”