Book Genesis

Book Genesis Universal Core

Book Genesis is Felipe's portable book-production pipeline for Claude Code, Codex, Kimi Code, and other file-aware agents. This folder is the canonical workflow. Platform adapters install the same files without changing editorial rules. OpenClaw and Hermes Agent can load this same package. Resolve references/ relative to this skill directory and skills// against sibling installed skills. The repository Python helper is optional; when it is unavailable, use the host agent's file tools to maintain the documented project tree.

Use this skill for:

• fiction, memoir, nonfiction, or hybrid book projects
• turning a rough idea into a structured book
• drafting full manuscripts chapter by chapter
• auditing AI-written prose and narrative structure
• producing Genesis Score evaluations
• creating synopses, cover briefs, launch/editorial packages, and formatting handoff notes

Do not use the legacy V4 17-phase workflow as the default. It is copied only for historical reference in references/legacy-v4-book-genesis.md.

Core Rule

Persist important decisions to files. A Book Genesis run should leave a durable project tree, not just chat output.

Commercial-length is a hard contract, not a preference. A manuscript cannot be called complete, final-scored as publication-ready, or packaged for a bestseller/public launch unless it either:

• meets the genre/audience target_floor_words recorded in PROJECT_STATE.yaml, or
• is explicitly positioned from intake as a novella, short novel, or serial installment.

If no explicit short-form positioning exists, default to full-length commercial book standards.

Default project layout:

/
  PROJECT_STATE.yaml
  ASSUMPTIONS.md
  artifacts/
  manuscript/
    chapters/
  evaluations/
  delivery/

Canonical Pipeline

Load references/pipeline/manifest.yaml before starting or advancing phases.

1. Phase 0: Intake
2. Phase 1: Foundation
3. Phase 2: Architecture
4. Phase 3: Drafting
5. Phase 4: Adversarial Audit
6. Phase 5: Literary Barrier Revision Loop
7. Phase 6: Final Score
8. Phase 7: Editorial Package

Use references/pipeline/phases.md for the compact phase overview.

Phase References

Read only the prompt for the active phase:

• Intake: references/prompts/intake.md
• Foundation: references/prompts/foundation.md
• Architecture: references/prompts/architecture.md
• Drafting: references/prompts/drafting.md
• Adversarial Audit: references/prompts/adversarial-audit.md
• Literary Barrier Revision Loop: references/prompts/literary-barrier-loop.md
• Final Score: references/scoring/genesis-score-codex.md
• Editorial Package: references/prompts/editorial-package.md

Before any independent or final score, read references/scoring/evaluator-protocol.md and record the achieved independence grade.

references/prompts/orchestrator.md contains the portable orchestration rules.

Operating Loop

Read references/pipeline/host-contract.md at start and resume. It defines installed-path resolution, native-host capabilities, progress, checkpoint recovery, and bounded retries. Use the bundled references/pipeline/project-state.yaml when initializing without the optional Python helper.

1. Identify or create the project directory.
2. Read PROJECT_STATE.yaml if it exists. If not, initialize it from the manifest phases and user idea.
3. Read ASSUMPTIONS.md if it exists. If not, create it and mark inferred assumptions clearly.
4. Confirm or infer the commercial-length contract before architecture or drafting:

• target_range_words
• target_floor_words
• target_ceiling_words
• positioning: full-length novel, novella, short novel, serial, nonfiction, etc.
• chapter_count_planned
• average_words_per_chapter_planned

1. Load the current phase prompt and produce only that phase's required outputs.
2. Update PROJECT_STATE.yaml after every phase, chapter block, audit, or score.

• record cumulative manuscript word count
• record planned vs actual word count by chapter/block
• record whether the length gate is PASS, FLAG, or BLOCK

1. Do not skip Phase 4. Audit before final scoring.
2. When drafting, write in chapter files under manuscript/chapters/ and keep the state synchronized.
3. When user feedback changes direction, record it in project files before continuing.
4. If the user asks to break a quality threshold such as 8.5, activate the Literary Barrier loop after the adversarial audit. Respect the host contract's retry and no-progress bounds; stop at a saved checkpoint when the threshold cannot be supported, rather than looping until an evaluator agrees.
5. If the manuscript is below target_floor_words, run expansion/re-architecture before any final score or editorial package claims publication readiness.

Quality Policy

• Prefer fewer constraints during drafting; evaluate and repair afterward.
• Separate drafting, audit, scoring, and editorial judgment in the workflow.
• Keep writer, revision editor, and final evaluator in isolated contexts when the runtime supports subagents.
• Use the Genesis Score floor principle: the book is only as strong as its weakest major dimension.
• Treat internal literary scores as inflated until calibrated. Apply the score calibration rules in the scoring reference before claiming a threshold was reached.
• Do not optimize prose to satisfy rubrics while drafting. The writer writes; the critic and editor operate after text exists.
• For Portuguese books, write artifacts and prose in Portuguese unless the user requests otherwise.
• If a task is only editing, scoring, or packaging an existing manuscript, start from the matching phase instead of forcing a full restart.
• Do not let literary quality hide product incompleteness. A brilliant 46k-word draft is not a full-length adult thriller unless the project explicitly chose short-form positioning.
• Bestseller/public-launch readiness requires both craft strength and product fit: length, package, target reader, cover/copy, metadata, and launch plan.

Complementary Skills

Use these only when relevant and already available:

• copy-editing for prose-level cleanup
• book-swarm-panel for MiroFish-style simulated reader swarms, niche-risk scouting, public-opinion tests, heatmaps, interviews, and revision tickets
• humanizer for less synthetic phrasing
• launch-strategy or content-strategy for go-to-market assets when separately installed
• imagegen or cover-specific workflows for cover ideation

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: felipelobomotta-blip
• Source: felipelobomotta-blip/book-genesis-studio
• License: MIT
• Homepage: https://github.com/felipelobomotta-blip/book-genesis-v4

Install and usage instructions live in the source repository linked above.