zenstory-ai/oh-story-claudecode

Oh Story – AI‑assisted web‑novel writing skill pack

What it is – A collection of 13 “skills” that turn a programming‑agent (Claude Code, Codex CLI, Google Antigravity, OpenCode, ZCode, OpenClaw, Reasonix, etc.) into a full‑stack novel‑writing assistant. The package handles the whole pipeline for Chinese web‑novels:

1. 扫榜 – scrape market rankings to pick profitable topics.
2. 拆文 – de‑construct best‑selling works into reusable plot modules.
3. 写作 – generate outlines, detailed chapter plans, and the prose itself.
4. 去 AI 味 – lint the output to remove stereotypical AI phrasing.
5. 封面 – produce a cover image via a built‑in image model.

All of this runs on the model already used by your agent – no GPU or extra model deployment is required. The only extra computation is a set of local Node/Python scripts that enforce deterministic checks and manage project files.

---

Quick start / Installation

# Install globally (available in any folder)
npx skills add zenstory-ai/oh-story-claudecode -y -g
# or install locally to the current project
npx skills add zenstory-ai/oh-story-claudecode -y

After the skill is added, run the bootstrap command in the root of a writing project:

/story-setup   # or $story-setup for Codex

This creates the required directory layout (设定/, 大纲/, 正文/, 追踪/ …) and registers the seven specialised agents and eight automation hooks.

---

Core workflow (mermaid diagram simplified)

flowchart LR
  A[Setup] --> B[扫榜]
  B --> C[拆文]
  C --> D[写作]
  D --> E[去 AI 味]
  E --> F[封面]

Long‑form novels follow the 扫榜 → 拆文 → 长篇写作 path; short‑form works use the analogous short‑form steps.

---

Main commands (skills)

Skill	Trigger	What it does
story-setup	/story-setup or $story-setup	Deploys the agents, hooks and file structure for the current project
story-long-write	/story-long-write	Generates outlines, character sheets and chapter prose for a long novel
story-short-write	/story-short-write	Same as above but for short‑story projects
story-deslop	/story-deslop	Lint pass that removes known AI‑style sentence patterns, punctuation issues, and clichés
story-cover	/story-cover	Calls a GPT‑Image‑2 model (or fallback API) to create a cover image based on the book’s title/genre
story-long-analyze / story-short-analyze	/story‑long‑analyze / /story‑short‑analyze	De‑constructs a reference work into plot modules, scores pacing, hooks, etc.
story-import	/story-import	Turns an existing manuscript into the standard project layout for continuation
story-dashboard	/story dashboard	Starts a local web UI (127.0.0.1) to browse the file tree, view demos, edit markdown, and run commands

---

How it keeps long‑form consistency

• File‑system memory – All settings, outlines, chapter drafts, and a JSON “tracking state” live on disk. The agent only reads the tiny slice it needs for the current chapter, avoiding context overflow.
• Guard hook – guard-outline-before-prose.sh blocks prose generation until a detailed outline for that chapter exists.
• Consistency checker – After prose is written, a consistency‑checker agent validates that foreshadowed plot points, character states, and “author‑truth vs. reader‑known” timelines match.
• Deterministic hooks – 7 agents (architect, writer, consistency, character designer, researcher, explorer, extractor) and 8 hooks enforce quality without stochastic surprises.

---

Example output (excerpt)

## 当前章节
- 第 21 章 – 场景：火箭军文工团办公室

## 活跃伏笔
- F054｜谭守义已给邻市地址，江晨定次日上门听故事｜埋第20章｜第22章｜高
- F057｜谭守义有个讲了五十年没人听全过的故事，内容未揭示｜埋第21章｜第22章｜高

The demo folder contains full‑length novel files, a de‑construction report for the classic 盘龙, and a generated cover image.

---

When to use it

• You already work with a Claude‑Code or similar coding‑agent and want to add novel‑writing capabilities without spinning up a separate LLM server.
• You need deterministic, repeatable output for long serial web‑novels (tens of thousands of words) where context size is a bottleneck.
• You want an automated “AI‑flavor removal” step to make the prose feel more human before publishing.

---

Limitations / Gotchas

• The cover generator relies on a GPT‑Image‑2 model; if your agent does not have access to it you’ll need an external API key.
• All checks are local scripts; they do not guarantee that external AI‑detectors (e.g., Zhuque) will label the text as human‑written, but they substantially reduce stereotypical AI phrasing.
• The skill set is tied to the ClawHub distribution mechanism – you must install via npx skills … or issue the natural‑language install command.

---

Community & Support

• Telegram group: https://t.me/ohstoryclaudecode
• GitHub Discussions & Issues for troubleshooting, feature requests, and sharing output examples.
• Contribute new skills, market‑data updates, or additional knowledge‑base entries via the usual pull‑request workflow.

---

TL;DR

Oh Story turns a coding‑agent into a turnkey Chinese web‑novel author: it scrapes market trends, de‑constructs best‑sellers, builds outlines, writes chapters, lints away AI‑style prose, and even creates a cover – all without extra GPUs. Install with npx skills add …, run /story-setup, and start issuing /story‑long‑write or /story‑short‑write commands.