A writing workflow plugin for Cowork (also compatible with Claude Code) that maintains coherence across novel-length fiction. Provides structured continuity tracking, thread management, foreshadowing systems, and living documents that grow with your manuscript.
What this is: A methodology plugin. It doesn't generate plots or characters for you — it gives the AI agent the workflow discipline to maintain a 100k+ word manuscript without losing track of what happened in chapter 3 by the time you're drafting chapter 30. Genre-neutral, no fantasy/sci-fi specifics baked in.
Long-form fiction breaks without external memory. An AI agent can't hold an entire novel in context simultaneously, so structured documents become the memory system. This plugin implements a workflow loop drawn from a methodology that has produced 116k words of published fiction (in a more automated pipeline form — here it's packaged for interactive, chapter-by-chapter co-writing):
Plan → Draft → Log → Verify → Repeat
Every chapter gets an outline before drafting. Every draft triggers updates to whatever tracking documents the project currently has — continuity always, plus the scene log, glossary, thread tracker, and foreshadowing checklist once those exist. The documents grow with the manuscript, not before it: the project starts with a small core and adds tracking as the work produces a real need for it (see Living Documents below).
If a session ever dies between drafting a chapter and logging it, the tracking docs fall out of sync with the manuscript silently. You don't need to manage this by hand — /wrap and /review both open by reconciling the newest drafts against the tracking docs and repairing any gap before doing anything else. If you're unsure the last session finished cleanly, run /review before drafting more.
The tracking documents grow with the manuscript, not before it. /start-project creates only a small core — a pitch, a characters doc, and a continuity doc (plus a world or outline doc if you discussed enough to fill one). Everything else is proposed later, when the work produces a concrete need:
- A scene log after the first chapter is drafted, so nothing gets lost between sessions.
- A glossary once you've invented enough names and terms that you're double-checking spellings.
- A thread tracker once three or more subplots are running and one is at risk of going dark.
- A foreshadowing checklist once you're planting deliberate seeds that need to pay off.
- A voice doc and style guide after the first few chapters, once the narrator's voice and prose patterns have emerged (or a voice doc upfront via
/voice-session).
The reason for holding off is simple: an empty template document is a lie. An empty glossary just reminds you that you haven't written enough yet. Each doc gets proposed with a specific example from your own manuscript — "we've invented fifteen terms and I had to re-check the spelling of one" — and you decide whether to create it. /write-chapter proposes these mid-draft; /wrap takes the wider-angle view at session's end.
At series length, two of these documents split for scale rather than staying strictly singular: the scene log becomes one per book, and a compact "canon so far" summary is distilled from the full continuity doc to serve as the working reference (the full continuity doc remains the authority). This happens only when the documents have genuinely outgrown the context window — not by default.
Add the marketplace, then install the plugin:
/plugin marketplace add third-order-labs/claude-plugins
/plugin install longform@third-order-labs-plugins
/plugin marketplace add third-order-labs/claude-plugins
/plugin install longform@third-order-labs-plugins
/start-project
This walks you through defining your concept — pitch, main character, tone, scope — through conversation, then writes down just enough to start chapter one: a pitch, a characters doc, and a continuity doc (plus a world or outline doc only if the conversation produced real material for one). It's a conversation, not a questionnaire, and it deliberately doesn't build tracking infrastructure you haven't earned a reason for yet. The rest of the documents emerge as the manuscript grows.
/plan-chapter
Creates a detailed outline with scene breakdowns, lived-in targets, thread assignments, and foreshadowing seeds.
/write-chapter
Drafts the chapter, then updates whatever tracking documents exist — continuity always, plus the scene log, glossary, threads, and foreshadowing once those have been created. It also proposes new tracking docs when the manuscript has earned them.
/review
Reconciles the tracking docs against your drafts, then audits whatever tracking exists — continuity, thread health, foreshadowing, glossary — across a recent window of chapters. Catches drift before it becomes a problem. (Reading every drafted word is a separate, explicit /review deep pass.)
/wrap
Summarizes progress, verifies all documents are current, and previews the next session's work.
Interactive project setup. Figures out the story with you through conversation, then writes down only what's needed to start writing chapter one — it doesn't front-load infrastructure the project hasn't earned yet.
Always created: the working directories, AGENTS.md (with a one-line CLAUDE.md pointing to it, so Claude Code auto-loads the workflow rules), and three core docs — 00-pitch.md, 03-characters.md, and 07-continuity.md.
Created only if the conversation produced real material for them: 02-world.md, 04-outline.md, 01-north-star.md. If you didn't discuss enough to fill one, it isn't created as an empty template — it emerges later.
Not created at start: glossary, scene log, thread tracker, foreshadowing checklist, style guide, voice doc, chapter-outline template. These are living documents that get proposed by /write-chapter and /wrap when the manuscript produces a concrete need for them.
/start-project
/start-project "noir detective novel set in 1940s LA"
Creates a detailed chapter outline before drafting. Checks thread health, foreshadowing needs, and recent scene log for context. Produces a spec the agent can execute against.
/plan-chapter
/plan-chapter next
/plan-chapter 12
/plan-chapter "The Crossing"
The core workflow loop. Drafts from the outline, then automatically updates scene log, continuity, glossary, threads, and foreshadowing checklist. Reports what was written and what was updated.
/write-chapter
/write-chapter next
/write-chapter 12
Health check across all tracking documents. Scans for continuity contradictions, stale threads, unplanted foreshadowing seeds, scene log gaps, and glossary inconsistencies. Produces a severity-ranked report.
/review # Full audit
/review recent # Last 5 chapters
/review threads # Thread health only
/review continuity # Continuity check only
Clean session boundary. Verifies all living documents are current, summarizes what was accomplished, flags documents needing attention, and previews the next session's work.
/wrap
Conversational voice capture session. Produces the authorial consciousness layer — what makes this novel sound like it was written by one mind. Two modes: full session for the narrator's worldview, or character-focused session for individual POV character profiles.
/voice-session
/voice-session character "Elena"
| Skill | Description |
|---|---|
long-form-methodology |
Core philosophy: why structure enables coherence at scale, document hierarchy, workflow loop, living documents, decision rules |
continuity-management |
Scene log format, canon tracking, conflict resolution (canon always wins), cumulative tracking, glossary as entity resolver |
narrative-craft |
Thread tracking (I/D/P/S markers), thread health monitoring, foreshadowing as deliberate system, arc pacing, multi-book continuity |
scene-craft |
Lived-in texture, texture beats, supporting character motives, consequence glue, dialogue craft, POV discipline, violence/humor/magic presentation |
authorial-voice |
Voice in long-form fiction: authorial consciousness layer, character voice profiles, two-layer capture system, novel-specific anti-patterns |
project-scaffolding |
Full template library for all planning documents, directory conventions, file naming rules |
- Run
/start-project— define your concept through conversation - Review the core documents it created in
docs/ - Run
/plan-chapterto outline chapter 1 - Run
/write-chapterto draft it — it'll propose new tracking docs (scene log, glossary, and so on) as the manuscript earns them - Continue the plan → write cycle
- Run
/reviewto audit all tracking documents - Fix any critical continuity issues before drafting more
- Address stale threads in the next chapter plan
- Plant any overdue foreshadowing seeds
- Run
/plan-chapter nextto outline the next chapter - Run
/write-chapterto draft it - Repeat for additional chapters
- Run
/wrapto close the session cleanly
- Set up the project as a series in
/start-project - Keep continuity, the thread tracker, and the glossary cumulative across all volumes — canon, arcs, and names are what tie a series together, so they stay single documents
- At series length, two docs split for scale: keep a scene log per book (the per-volume ledger you rarely need at long range), and distill a compact "canon so far" summary from the full continuity doc as the working reference that loads before drafting (the full continuity doc stays the authority). Reach for these only when the docs have actually outgrown the context window, not preemptively
- Use
/reviewat the end of each volume to check series-level thread health, and plant foreshadowing seeds for future volumes during current drafting
/start-project creates a small core. The rest are living documents that emerge later, proposed by /write-chapter and /wrap when the manuscript produces a concrete need for them. This is the full structure a mature project grows into — a new project does not start with all of it:
your-novel/
├── AGENTS.md # Agent workflow rules [created at start]
├── CLAUDE.md # One-line pointer to AGENTS.md [created at start]
├── docs/
│ ├── 00-pitch.md # Hook, synopsis, core contrasts [created at start]
│ ├── 03-characters.md # Cast with arcs [created at start]
│ ├── 07-continuity.md # Canon facts (highest precedence) [created at start]
│ ├── 01-north-star.md # Guardrails and open decisions [at start, if discussed]
│ ├── 02-world.md # Setting, geography, society [at start, if discussed]
│ ├── 04-outline.md # Act structure and beats [at start, if discussed]
│ ├── 06-scene-log.md # Running ledger of what happened [emerges: after ch. 1]
│ ├── 05-glossary.md # Names, terms, places [emerges: ~10+ terms]
│ ├── 13-threads.md # Thread tracking (I/D/P/S) [emerges: 3+ subplots]
│ ├── 19-foreshadowing-checklist.md # Seeds and payoff targets [emerges: deliberate seeds]
│ ├── 08-voice.md # Authorial consciousness + voices [emerges: ch. 3-5 or /voice-session]
│ ├── 25-style-guide.md # Prose register, POV, tone [emerges: ch. 3-5]
│ └── 09-chapter-outline-template.md # Reusable chapter outline format [emerges: from first outlines]
├── outlines/
│ └── book-one/ # Chapter outlines
├── draft/
│ └── book-one/ # Draft chapters
└── manuscript/ # Final assembled manuscript
Why it works this way: an empty template document is a lie — an empty glossary just reminds you that you haven't written enough yet. A glossary created at chapter 5, when you've invented fifteen terms and can't remember how you spelled one, is useful. Each tracking doc is proposed with a concrete example from your own manuscript, and you decide whether to create it. See Living Documents.
When documents conflict, resolve in this order (highest precedence first):
- Continuity (
07-continuity.md) — canon facts that must not be broken - Voice doc (
08-voice.md) — authorial consciousness + character profiles - Outline (
04-outline.md) — the structural plan - Threads (
13-threads.md) — active narrative threads - Scene log (
06-scene-log.md) — what actually happened - Style guide (
25-style-guide.md) — prose mechanics - Pitch (
00-pitch.md) — the original vision
Voice sits above the outline because a scene can deviate from the structural plan and still work, but a scene that violates the authorial consciousness feels wrong regardless of whether it hits its plot beats. It sits below continuity because facts are facts — you can't break canon for the sake of voice.
longform-plugin/
├── .claude-plugin/plugin.json
├── README.md
├── commands/
│ ├── start-project.md
│ ├── plan-chapter.md
│ ├── write-chapter.md
│ ├── review.md
│ ├── wrap.md
│ └── voice-session.md
└── skills/
├── authorial-voice/SKILL.md
├── long-form-methodology/SKILL.md
├── continuity-management/SKILL.md
├── narrative-craft/SKILL.md
├── scene-craft/SKILL.md
└── project-scaffolding/SKILL.md