modoojunko/awesome-novel-agent

awesome‑novel‑agent

What it is – A multi‑platform skill that turns large‑language‑model agents (Claude Code, OpenCode, Reasonix, Codex, ZCode, DeepSeek Harness) into a collaborative novel‑writing assistant. It provides a set of nine specialised agents (outline planner, prompt crafter, writer, anti‑AI filter, reviewer, updater, etc.) and a knowledge‑base that together guide a user from genre selection to finished manuscript.

Key capabilities

• One‑command setup – Tell your AI tool “install awesome‑novel‑agent” and it runs install.sh to place the skill in the appropriate user‑level directory for the selected platform.
• Project initialization – init.py creates a complete novel workspace (folders for story index, settings, volumes, chapters, prompts, archives, memory, knowledge, and platform‑specific agent definitions).
• Agent‑driven workflow – A top‑level novel‑agent orchestrates the sub‑agents through the stages: setup → outline → draft → anti‑AI → optional review → archive. The user only confirms or edits at each step.
• Style distillation – Import sample texts; the style‑distiller extracts quantitative style parameters (sentence length, dialogue ratio, adjective density) and injects them into prompts, continuously refining the model’s voice.
• Anti‑AI filtering – A dedicated anti‑ai agent runs a multi‑stage gate that detects and removes common “machine‑y” phrasing, scoring the result.
• Dynamic memory – All interactions, preferences, and rule updates are stored in per‑project memory/ files; frequently used rules are promoted to permanent memory.
• Genre presets – 24 built‑in genre profiles (xianxia, urban, mystery, sci‑fi, etc.) supply default world‑building, character archetypes, and pacing templates.
• Two collaboration modes – Step‑by‑step (default, asks for confirmation) or full‑authorisation (“you have full authority”) for hands‑off generation.

Supported platforms

Platform	Install path (user‑level)	Project‑level folder
Claude Code	~/.claude/skills/awesome-novel/	.claude/
OpenCode	~/.config/opencode/skills/awesome-novel/	.opencode/
Reasonix	(project‑level only) .<project>/.reasonix/skills/	.reasonix/
Codex	~/.codex/skills/awesome-novel/	.codex/
ZCode	~/.zcode/skills/awesome-novel/	.zcode/
DeepSeek Harness	~/.dsh/skills/awesome-novel/	.dsh/

Typical usage

1. Install – In the chosen AI console say 帮我安装 awesome-novel-agent. The tool runs ./install.sh <platform>.
2. Create a novel project – Navigate to an empty folder and invoke the skill (/awesome-novel or @novel-agent). The agent walks you through genre, world‑view, characters, and style.
3. Write – Use commands like 写下一章, 帮我继续写, or solo (full‑auto). The agents generate outlines, prompts, drafts, run the anti‑AI filter, optionally review, and archive the finished chapter.
4. Iterate – At any time you can modify settings (改一下世界观里的…) or import your own style samples.

Installation details

• Requires Python 3.9+ (macOS ships 3.9; 3.11+ recommended). OpenCode/Codex/ZCode/DeepSeek also need pyyaml (pip install pyyaml).
• Run ./install.sh <platform> (or install.ps1 on Windows). The script checks the Python version and dependencies before copying the skill.
• For Reasonix the skill is deployed per project via init.py rather than the global installer.

Project layout (after init)

<novel>/
├─ story.md                # high‑level index
├─ settings/…              # world, style, genre, timeline, characters
├─ volumes/                # volume‑level outline
├─ chapters/               # chapter‑level outline
├─ prompts/                # generated prompts
├─ archives/               # final chapters
├─ .agent/                 # status & task files
├─ .<platform>/            # agents, knowledge, memory for the chosen platform
└─ ...                     # optional sandbox, samples, etc.

Only the directory matching the selected platform is created.

License – GPL 3.0 (free for personal use; commercial use requires author permission).

Community – A QQ discussion group (ID 1006050538) is provided for support, feedback, and sharing of generated novels.

Contributing – See CONTRIBUTING.md. Contributions can be:

• Bug reports / feature requests via GitHub issues.
• Adding new genre profiles or anti‑AI rule examples.
• Improving agent definitions (agents/*.md) or skill SOPs.
• Code changes following the project’s Python/markdown conventions.

---

This summary is based solely on the repository’s README; no additional features have been inferred.