book-draft
Skill40 starsChanged 2 months agoApache-2.0View on GitHub
Phase 3: Drafting. Voice lock on a sample chapter, then sequential full-context chapter drafting with draft-critique-revise, bible ledger updates, and periodic continuity checks.
In this repository23 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- dashboard/CLAUDE.mdCLAUDE.md
- .cursor/rules/velith-editing.mdcCursor rule
- .cursor/rules/velith-genres.mdcCursor rule
- .cursor/rules/velith-pipeline.mdcCursor rule
- book-academicSkill
- book-editSkill
- book-fictionSkill
- book-gameSkill
- book-genre-creatorSkill
- book-ideationSkill
- book-illustrateSkill
- book-initSkill
- book-nonfictionSkill
- book-outlineSkill
- book-poetrySkill
- book-publishSkill
- book-screenplaySkill
- book-statusSkill
- book-technicalSkill
- book-visualsSkill
- loomSkill
The file
We split this skill into the rules it gives an agent
8 rules. About 0 are common, 2 are distinctive, and 6 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
6
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 11
Read ${CLAUDE_PLUGIN_ROOT}/skills/loom/quality-bar.md, then PRD.md, STYLE.md, outline.md, bible.md.
- Very specific: line 17
If STYLE.md's ## Voice Lock says pending:
- Very specific: line 20
Run scene-generator (fiction) then chapter-writer on that chapter exactly as in normal drafting.
- Very specific: line 31
Technical, reference, academic: read the dependency DAG in outline.md.
- Distinctive: line 33
Never run two chapter-writers on the same chapter.
- Distinctive: line 41
Verify the writer's self-reported scores.
- Very specific: line 42
If the chapter spec lists figures, the writer has placed !alt references with captions at the right spots.
- Very specific: line 74
All chapters in outline.md exist as drafts.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Phase 3: Drafting
Goal: a complete manuscript in one voice, internally consistent, each chapter already revised once against the quality bar. Drafting done this way needs an editor; drafting done to word count needs a rewrite.
Read ${CLAUDE_PLUGIN_ROOT}/skills/loom/quality-bar.md, then PRD.md, STYLE.md, outline.md, bible.md.
Refuse to draft if outline.md lacks ## Approval. Point the author to /book-outline.
Step 0: Voice lock (once per book)
If STYLE.md's ## Voice Lock says pending:
- Choose the sample chapter: chapter 1 by default, or the chapter the author names. For fiction, the architect may recommend a mid-book chapter if chapter 1 is atypical (prologue, framing device).
- Run
scene-generator(fiction) thenchapter-writeron that chapter exactly as in normal drafting. - Present the chapter to the author with the writer's own critique scores. Ask: is this the voice? What is off? Required stop.
- On approval, record in
STYLE.mdunder## Voice Lock: date, chapter, the author's notes, and 2-3 quoted passages from the approved chapter that best represent the voice. Chapter-writer reads these on every later chapter. - On rejection, revise the fingerprint per the author's notes, regenerate, present again. Up to three rounds, then ask the author to supply or edit a sample directly.
--voice-lock forces this step even if a lock exists (voice change mid-book).
Step 1: Order and concurrency
Narrative genres (fiction, screenplay, game, memoir, narrative nonfiction, poetry sequences): strictly sequential. Chapter N requires chapter N-1 saved with status: draft or later and a bible ledger entry.
Technical, reference, academic: read the dependency DAG in outline.md. Chapters with no unsatisfied prerequisites and no mutual references may draft concurrently, at most 3. Each concurrent writer still reads all completed chapters.
Never run two chapter-writers on the same chapter.
Step 2: Per-chapter loop
For each chapter to draft (skip existing unless --force):
- Fiction only: invoke
scene-generator→drafts/ch{NN}-scenes.md(a plan, not prose). - Invoke
chapter-writerwith the chapter number. It reads everything, drafts, critiques, revises, writesdrafts/ch{NN}-{slug}.md, appends the chapter ledger inbible.md, and writes.velith/critiques/ch{NN}.md. - Verify the writer's self-reported scores. If any axis < 6 after its revision, run it again with the critique as additional input before moving on.
- If the chapter spec lists figures, the writer has placed
references with captions at the right spots. Invokefigure-engineerfor those ids now (technical and nonfiction) so the figure exists while the chapter is fresh; illustrations wait for the look lock in/book-visuals. node ${CLAUDE_PLUGIN_ROOT}/velith.mjs scan [project-dir]so the dashboard updates.
Step 3: Continuity checks
Invoke continuity-editor after chapters 3, then every 4 chapters, then after the final chapter. It reads the whole manuscript against the bible. Minor issues are fixed in place; Critical and Major issues are reported and, for narrative genres, must be resolved before drafting continues (a contradiction in chapter 5 propagates into 6 through 20).
Chapter file format
---
chapter: 7
title: "..."
slug: "..."
word_target: 4000
words: 4120
status: draft # draft | edited | final
revision: 1
critique_score: {voice: 7, structure: 7, depth: 8, specificity: 7, reader: 7}
created: 2026-09-04
updated: 2026-09-04
---
# Chapter title
Body...Korean projects count characters excluding whitespace; count_unit: chars in PRD.
Gate
All chapters in outline.md exist as drafts. Every chapter has a critique with all five axes ≥ 6. Bible ledger has an entry per chapter. Last continuity report has no unresolved Critical issues. Voice lock recorded.
Resuming
/book-draft with no chapter argument resumes from the first missing chapter. /book-draft 12 drafts chapter 12 only (refuses for narrative genres if 11 is missing). /book-draft 12 --force redrafts 12 and warns that later chapters may need continuity review.
Post-completion
node ${CLAUDE_PLUGIN_ROOT}/velith.mjs scan [project-dir]Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

