Book Genesis

Your creativity. Your agent. Your book.

An open-source collection of writing skills for the AI agent you already use. Install into Claude Code, Codex, DeepSeek Harness, OpenCode, Cursor, GitHub Copilot, Qwen Code, Pi, Windsurf, Antigravity, Gemini CLI, Kimi Code, OpenClaw, or Hermes Agent. A shared directory target is also available.

Bring an idea. Book Genesis gives your agent a workflow for developing it into a manuscript: direction, characters, outline, chapters, editorial review, revision, and a publishing package. You keep the project files and creative decisions.

MIT license · Installation guide · Compatibility evidence · Launch kit · Project casebook

New: six more install targets. DeepSeek Harness, Cursor, GitHub Copilot, Qwen Code, Pi, and Windsurf receive the same 15-skill bundle. There are now 15 installation targets: 14 named hosts plus Shared. Installation and file integrity are tested; native writing acceptance is tracked separately in the compatibility table.

Back to the skills

This repository has returned to its portable-skills foundation, based on commit c974486. The host agent handles models, authentication, tools, progress, and permissions. Book Genesis supplies the writing workflow and reference material.

The later standalone app and model-orchestration experiment is preserved in Git history. See what was restored and verified. Existing app book folders should be preserved; automatic migration of their state to the older skill workflow has not been verified.

Install into your agent

You need Git and Python 3.10+ for this installer, plus the agent you want to use. The installer only copies skills and their references. It does not start a model, install the host agent, or collect API keys.

git clone https://github.com/felipelobomotta-blip/book-genesis-v4.git
cd book-genesis-v4
python runner/cli.py verify-suite

Choose one target:

python runner/cli.py install claude
python runner/cli.py install codex
python runner/cli.py install kimi
python runner/cli.py install openclaw
python runner/cli.py install hermes
python runner/cli.py install opencode
python runner/cli.py install antigravity
python runner/cli.py install gemini
python runner/cli.py install deepseek
python runner/cli.py install cursor
python runner/cli.py install copilot
python runner/cli.py install qwen
python runner/cli.py install pi
python runner/cli.py install windsurf

PowerShell and shell shortcuts are also included:

.\install.ps1 -Target hermes

bash install.sh openclaw

Use --dry-run to preview an installation. Existing modified skills block replacement; an explicit --force backs them up first. See the installation guide for profiles, custom destinations, remote agents, and updates. Copying just SKILL.md is insufficient: the reference folders are part of the workflow.

Check the installed package before starting (use your chosen target):

python runner/cli.py verify-install opencode

This checks every skill and reference against your checkout. Confirm discovery in the host too: OpenCode exposes opencode debug skill; Gemini CLI exposes gemini skills list; Hermes exposes hermes skills list. See the compatibility evidence and host checks.

Give it an idea

Open a new session in your agent with a writable folder for your book, then say:

Use the book-genesis skill. My idea is a mystery about a retired train dispatcher who finds a farewell letter inside a station clock. Help me choose the direction, then develop the outline and write the book. Save the chapters and project state so we can continue later. Write in English.

In hosts that expose skill commands, use /book-genesis. You can also ask for book-bestseller-studio when you want the broader research, editorial, positioning, and launch workflow.

The agent asks for missing creative decisions and works within its own context and quota limits. If interrupted, return to the same project folder and ask it to read PROJECT_STATE.yaml, inspect the saved chapters, and continue from the unfinished step.

How it works

1. Direction: clarify the reader, premise, language, form, and intended length.
2. Foundation and outline: develop the characters or argument, voice, and chapter structure.
3. Writing: draft chapter blocks and preserve progress in files.
4. Editorial review: audit the manuscript, identify specific weaknesses, and revise their responsible sections.
5. Delivery: prepare the synopsis, positioning, cover brief, and editorial handoff.

What you install

The same 15-skill suite goes to every supported target. It includes the universal core, narrative foundation, prose craft, book research, editing, reader simulations, manuscript management, series planning, and editorial packaging.

| Host | Default destination | | --- | --- | | Claude Code | ~/.claude/skills/ | | Codex | ~/.codex/skills/ | | Kimi Code | ~/.kimi-code/skills/ | | OpenClaw | ~/.openclaw/skills/ | | Hermes Agent | ~/.hermes/skills/ | | OpenCode | ~/.config/opencode/skills/ | | Antigravity | ~/.gemini/config/skills/ | | Gemini CLI | ~/.gemini/skills/ | | DeepSeek Harness | ~/.dsh/skills/ | | Cursor | ~/.cursor/skills/ | | GitHub Copilot | ~/.copilot/skills/ | | Qwen Code | ~/.qwen/skills/ | | Pi | ~/.pi/agent/skills/ | | Windsurf / Cascade | ~/.codeium/windsurf/skills/ | | Shared Agent Skills | ~/.agents/skills/ via install shared |

Host-specific home variables and --dest override these locations. OpenCode also respects XDG_CONFIG_HOME; its explicit OPENCODE_CONFIG_DIR takes priority. The OpenClaw and Hermes destinations follow their official OpenClaw and Hermes skill-directory conventions. Installer compatibility is tested separately from live writing in those hosts.

DeepSeek Harness respects DSH_HOME; Pi respects PI_CODING_AGENT_DIR. See the six new host setup notes and official sources. The package is MIT licensed. Model usage may have a cost or quota under your host account.

The work behind it

The casebook preserves earlier experiments, including The Source Code, Protocolo Não Encontrado, and Age of Aquarius. Case notes distinguish planned work, reported manuscript progress, and public artifacts. Private manuscripts and historical model scores are not independent product benchmarks.

The idea is simple: creativity should be the starting point. People should be able to explore a book with the tools they already have. Writing quality still depends on the idea, model, direction, and editorial work; a literal bestseller cannot be guaranteed.

The architecture audit records the reliability fixes and remaining limits.

Development

The optional Python helper installs the suite, scaffolds projects, validates files, and prepares phase packets. It does not call a model. You do not need to keep it running while your agent writes.

python -m unittest discover -s tests -v

Contributing · Security · Changelog · Issues

Created by Felipe Lobo. MIT licensed.