agentleFS
epicsagasepicsagas/Velith/skills/book-draft/SKILL.md

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.

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
  1. Very specific:

    Read ${CLAUDE_PLUGIN_ROOT}/skills/loom/quality-bar.md, then PRD.md, STYLE.md, outline.md, bible.md.

    line 11
  2. Very specific:

    If STYLE.md's ## Voice Lock says pending:

    line 17
  3. Very specific:

    Run scene-generator (fiction) then chapter-writer on that chapter exactly as in normal drafting.

    line 20
  4. Very specific:

    Technical, reference, academic: read the dependency DAG in outline.md.

    line 31
  5. Distinctive:

    Never run two chapter-writers on the same chapter.

    line 33
  6. Distinctive:

    Verify the writer's self-reported scores.

    line 41
  7. Very specific:

    If the chapter spec lists figures, the writer has placed !alt references with captions at the right spots.

    line 42
  8. Very specific:

    All chapters in outline.md exist as drafts.

    line 74

≈ “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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

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:

  1. 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).
  2. Run scene-generator (fiction) then chapter-writer on that chapter exactly as in normal drafting.
  3. Present the chapter to the author with the writer's own critique scores. Ask: is this the voice? What is off? Required stop.
  4. On approval, record in STYLE.md under ## 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.
  5. 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):

  1. Fiction only: invoke scene-generator → drafts/ch{NN}-scenes.md (a plan, not prose).
  2. Invoke chapter-writer with the chapter number. It reads everything, drafts, critiques, revises, writes drafts/ch{NN}-{slug}.md, appends the chapter ledger in bible.md, and writes .velith/critiques/ch{NN}.md.
  3. 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.
  4. If the chapter spec lists figures, the writer has placed ![alt](../visuals/figures/{id}.svg) references with captions at the right spots. Invoke figure-engineer for those ids now (technical and nonfiction) so the figure exists while the chapter is fresh; illustrations wait for the look lock in /book-visuals.
  5. 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]

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.