book-genesis
Skill159 starsChanged 19 days agoMITView on GitHub
Use when someone wants to write a complete book from an idea, even a one-line one, or to continue, audit, revise or package a book in progress. Book Genesis plans the book, drafts every chapter in a consistent voice, has blind simulated readers and a read-only auditor critique it, revises in bounded passes, and delivers a commercial-length manuscript with a scored report and an editorial package (logline, blurb, synopsis, query letter, cover brief). The author agrees at each step or lets it run to the end. Fiction and memoir first; nonfiction works but is not calibrated.
2 other projects use a version of this.
Adopted by
Most recently taken up by
- jodyjdc/ultimate-best-seller-studiotheir own version, first seen Nov 8, 2026
- felipelobomotta-blip/book-genesis-v4word for word, first seen Oct 26, 2026
Other versions2 projects use a version of this.
Word for word · 1
- felipelobomotta-blip/book-genesis-v4★ 159100% same
Edited · 0
None found.
Wrote their own version · 1
- jodyjdc/ultimate-best-seller-studio★ 2same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository4 other files it gives its agents.
- AGENTS.mdAGENTS.md
- beta-readerSkill
- editorial-packageSkill
- literary-agent-panelSkill
The file
We split this skill into the rules it gives an agent
6 rules. About 0 are common, 3 are distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 16
Read references/pipeline/plugin-onboarding.md first.
- Very specific: line 35
Read references/pipeline/host-contract.md.
- Very specific: line 36
Read references/prompts/orchestrator.md.
- Distinctive: line 102
Never skip Phase 4, and never score before it.
- Distinctive: line 103
Write the book and every message to the author in the book's language.
- Distinctive: line 105
Never claim a book will sell, is a bestseller, or is ready to publish without a human decision.
≈ “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
Book Genesis
Take an idea to a complete, commercial-length book inside the author's own agent: planned, drafted, read blind, audited, revised and packaged, with every decision saved to files.
This recovery build restores the 5.0 creative approach: draft with a compact story packet and fewer constraints, then audit and repair the prose. Preserve voice, asymmetry and consequential choices. Execution safeguards and file verification stay with the orchestrator.
The goal is a book written with the measured patterns of books readers finish, audited the way a demanding editor would audit it, and packaged to sell. It is a goal, not a promise: no workflow can guarantee sales.
Start here
Read references/pipeline/plugin-onboarding.md first. It routes new ideas,
resume, review-only, status and export requests, and keeps the author in a
simple conversation while you handle the technical steps.
For an autonomous Codex plugin request, follow the onboarding's controlled Codex start and run the packaged preparation bridge before creating any book records. Do not manually construct install receipts, project configuration or native commands. Once started, monitor that controller; skip the host-owned steps 1–3 below. The plugin routes the conversation; the shared controller owns the long-running book.
The optional checkout launcher can also run an autonomous Codex project through
scripts/autonomous_book.py. In that mode a deterministic controller owns state,
phase prerequisites, revision budgets and delivery; the native model receives
only a scoped role packet and returns its artifact. Do not start a second
orchestrator or edit the controller's .book-genesis/workflow.json. The shared
phase, writing, criticism and export rules below still apply. Existing guided
projects retain their original host-owned workflow and snapshot.
- Read
references/pipeline/host-contract.md. It covers starting, resuming, checking in with the author, and recovering from failures. - Read
references/prompts/orchestrator.md. It is the main loop. - Open or create the author's book folder. If
PROJECT_STATE.yamlexists, resume from it; otherwise start at Phase 0 with whatever idea the author gave.
Resolve every references/... path from this skill's folder. The author's files live in the book folder:
<book>/
PROJECT_STATE.yaml ASSUMPTIONS.md RUN_REPORT.md
artifacts/ plan, audit, score and package files (00 to 13)
manuscript/chapters/ chapter-01.md, chapter-02.md, ...
evaluations/ panel verdicts, continuity checks, revision passes
delivery/ proofread manuscript and exports
work/ chapter briefs, draft attempts, critic packetsThe eight phases
Load only the active phase's prompt and the references references/pipeline/manifest.yaml lists for it.
Those references are the orchestrator's working set, not a critic packet. Never forward the whole list to another role. Assemble each role's packet from the allowed material in references/scoring/evaluator-protocol.md.
| Phase | Prompt |
| Phase 0: Intake | references/prompts/intake.md |
| Phase 1: Foundation | references/prompts/foundation.md |
| Phase 2: Architecture | references/prompts/architecture.md |
| Phase 3: Drafting | references/prompts/drafting.md |
| Phase 4: Adversarial Audit | references/prompts/adversarial-audit.md |
| Phase 5: Revision Loop | references/prompts/revision-loop.md |
| Phase 6: Final Score | references/scoring/genesis-score.md |
| Phase 7: Editorial Package | references/prompts/editorial-package.md |
references/pipeline/phases.md is the one-screen overview.
What makes the criticism worth reading
- Critics are blind. Readers see only prose; the auditor sees the plan but never targets or earlier scores; the writer never sees the rubric.
references/scoring/evaluator-protocol.mdsets who sees what. - Independence is measured and shown. Critics run in isolated subagents or in a second installed tool from another model family when the host allows it (
references/hosts.md). Every verdict carries its grade. - Four readers decide, not one. Primary, hostile, adjacent-genre and casual readers, built per book (
references/roles/panel.md,references/roles/blind-reader.md). - Two gates, bounded. The panel decides whether the book holds readers; the rubric (
references/scoring/rubric-criteria.md) decides what to fix. The orchestrator alone reads the gate policy inreferences/scoring/genesis-score.md. Three passes per gate, then an honest stop. - Measured patterns, named sources.
references/patterns/index.mdpoints to prose, structure and emotion patterns from named books and studies, and to machine-prose tells in English and Brazilian Portuguese.
Specialists
Load when the phase calls for them:
references/specialists/research.md: comps and market in the book's own languagereferences/specialists/narrative-foundation.md: characters, theme, voice and the rhythm contractreferences/specialists/prose-craft.md: openings, chapter endings, dialogue, revision movesreferences/specialists/continuity.md: the ledger and the five-chapter checksreferences/specialists/revision-editor.md: ticket-driven, targeted revisionreferences/specialists/production.md: proofreading and exportscripts/export_book.py: canonical manuscript and EPUB export with receipt validation; optional DOCX/PDF proofs use already-installed libraries onlyscripts/check_progress.py: read-only reconciliation of saved chapter numbers, prose counts and progress state before resuming or claiming complete deliveryreferences/specialists/causality.md: a small causal contract and middle-book review when the story depends on a mechanism or mysteryreferences/pipeline/evidence.md: original reader receipts, current-text coverage and delivery statusscripts/verify_book.py: read-only verification of current reader evidence and source-bound deliverables; it never grades prose or approves publicationscripts/prepare_rubric.py: freezes all visible manuscript passages with source hashes, so critics can cite passage IDs without recopying prosescripts/validate_rubric.py: read-only checks of rubric coverage, all ten dimensions and source evidence; invalid or stale evidence cannot support a passing scorescripts/check_correction.py: freezes narrowly factual corrections for an independent canon check and verifies original native evidence before guarded promotion; it does not invent preference votes or gradesscripts/decide_comparison.py: translates four validated anonymous votes using the frozen mapping; ambiguous ties retain the accepted chapter without inventing defect countsreferences/specialists/series.md: only when the book belongs to a series
Rules that never bend
- Save important decisions to files; keep
PROJECT_STATE.yamltrue to what exists. - Never skip Phase 4, and never score before it.
- Write the book and every message to the author in the book's language.
- A book is "complete" when it meets its length contract (or a declared short form), every planned chapter exists, and the package exists. Quality gates are reported beside that label, never hidden.
- Never claim a book will sell, is a bestseller, or is ready to publish without a human decision.
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.

