OH Story — All-in-One Skill Pack for Chinese Web Fiction
Covers trend scanning, story breakdown, writing, de-AI-ifying, and cover art for Chinese web novels, adapted for multiple AI coding tools.
- Source repo
- zenstory-ai/oh-story-claudecode
- Stars
- ★ 7.7k
- Last updated
- 1d ago
- License
- MIT
- Primary language
- Python
- FA score
- 75/100 · Good
At a glance
- How it runs
- Works with
- Portable with changesCodex · Claude CodeChatGPT (Partial support)
- Cost
- Free software; you pay for model usage
- Setup effort
- Medium · a few setup steps
- You'll need
- Typical use
- Long-form web novelists: deploy via story-setup, scan trends with story-long-scan, break down comparator books with story-long-analyze, and write chapters with story-long-write.
- Not a fit if
- English-only writers who can't work from Chinese documentation and examples
- Users on platforms like ZCode 3.3.4 that lack custom agent support but expect full multi-agent collaboration
- Source review
- 75/100 · Good
What does this agent do, and when should you use it?
oh-story-claudecode is a skill pack for Chinese web fiction writing, including 13 skills covering long-form and short-form workflows: market trend scanning, story breakdown, drafting, AI-flavor removal, and cover generation, plus reverse-importing existing novels into a writable project. It orchestrates 7 specialized agents (story-architect, narrative-writer, consistency-checker, etc.) and 8 automated hooks for session management, outline guarding, and post-write checks. The pack uses a file-system-based project structure for settings, outlines, prose, and tracking state, and supports Claude Code, OpenCode, ZCode, OpenClaw, Codex CLI, and Reasonix, with solo fallback for environments without custom agents. Its core methodology breaks down writing into trendy-scanning, breakdown, and commercial writing, emphasizing emotional satisfaction and modular reuse.
Executes trend scans (story-long-scan/story-short-scan) for platforms like Qidian, Tomato, Zhihu; analyzes stories (story-long-analyze/story-short-analyze) to output structured breakdowns with plot nodes, emotional modules, and writing techniques; writes chapters (story-long-write/story-short-write) based on outlines and comparator books, using a single-authority tracking state for continuity; removes AI flavor (story-deslop) via linting and external detection; imports existing novels (story-import) by reverse-engineering outlines, settings, and tracking; generates covers (story-cover) via GPT-Image-2; reviews (story-review) with multi-agent scoring. All skills trigger via natural language or slash commands, deposit configuration into project .claude/agents/ or .codex/agents/, and use hooks to enforce outline-before-prose workflow.
- Long-form web novelists: deploy via story-setup, scan trends with story-long-scan, break down comparator books with story-long-analyze, and write chapters with story-long-write.
- Short-form creators: use story-short-scan for hot topics, story-short-analyze for viral breakdowns, and story-short-write for new drafts.
- Authors with existing serials: reverse-import published chapters with story-import to continue seamlessly.
- Writers seeking to de-AI their text: apply story-deslop for deterministic lint and external detection.
- Authors needing cover art: generate covers with story-cover.
- Users on Codex, OpenCode, or other CLIs: deploy with $story-setup and use skills accordingly.
How do you install or deploy this agent?
Option 1: Tell your AI coding tool 'Install this skill https://github.com/worldwonderer/oh-story-claudecode'. Option 2: Run npx skills add worldwonderer/oh-story-claudecode -y -g for global install (omit -g for current directory). Repeat the same command to upgrade. If Windows shows ENOENT but ends with 'Done!', rerun the command to fix.
How do you use this agent?
After installation, run /story-setup (or $story-setup) in your writing project root to deploy hooks, agents, and AGENTS.md, then start a new session. Use slash commands like /story-long-write or natural language phrases like '帮我开书' to trigger skills. For existing novels, run /story-import first. Content stays local; the dashboard only listens on 127.0.0.1.
What are this agent's strengths and limitations?
- Comprehensive workflow coverage from trend scanning to cover generation, reducing tool-switching overhead.
- Modular breakdown and comparator learning lets you reuse proven patterns.
- Multi-platform support with natural language and slash command triggers.
- File-system-based state management enables audit trails and version control.
- Specialized agent roles (architect, writer, consistency checker) streamline complex writing tasks.
- Deployment varies by CLI; some platforms (e.g., ZCode 3.3.4) lack custom agent support, degrading functionality.
- Updates require re-running /story-setup and starting a new session, which can be disruptive.
- Features like browser-cdp rely on network and may hit anti-scraping measures.
- Documentation and examples are primarily in Chinese, which may hinder non-Chinese speakers.
How does this agent compare with similar options?
No explicit competitors are named in the source. Compared to generic writing assistants or pure prompt templates, this pack offers structured breakdown and project-level engineering.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| OH Story — All-in-One Skill Pack for Chinese Web Fiction This agent | 75 · Good | Agent plugin / skillFree + model costs | ★ 7.7k | 1d ago | Python | Codex · Claude Code |
| AIWriteX Content Studio | 43 · Major gaps | Desktop appFree + model costs | ★ 2.1k | 1mo ago | Python | — |
| KrillinAI Video Localization | 62 · Some gaps | Desktop appFree + model costs | ★ 14k | today | TypeScript | OpenAI API |
| InkOS Story Creation Workspace | 50 · Major gaps | CLIFree + model costs | ★ 10k | today | TypeScript | Claude Code · OpenAI API · Claude API |
How does FollowAgents rate this agent?
Why each dimension lost points
Evidence shows: least privilege is respected, hooks are mostly non-blocking, dashboard listens on 127.0.0.1, data not uploaded; user confirmation: installation requires user action, deployment requires /story-setup, hooks require trust; data flow transparency: README clearly describes data flow, dashboard local; sensitive data handling: novel content not uploaded, but API key handling not specified; dependency security: few dependencies, but no audit; external effects: hooks modify project files, but require deployment; rollback: UPGRADING.md provides migration, but no automatic rollback; source attribution: README acknowledges third-party tools, but not all sources. Deductions: sensitive data handling and dependency security insufficient, rollback only documentation.
Evidence shows: internal consistency good, README matches skills; dependency availability: few dependencies, but no lock file; failure messages: README mentions Windows install errors, but no comprehensive error handling docs. Deductions: dependency locking and error handling docs insufficient.
Evidence shows: clear target users (web fiction authors), scenarios cover long/short full workflow; capability boundaries: README describes each skill, but no explicit limits; trigger precision: multiple trigger methods, but no conflict handling; environment fit: supports multiple CLIs and platforms, with adaptation notes. Deductions: capability boundaries and trigger conflict handling insufficient.
Evidence shows: clear information architecture, README has full TOC; install notes detailed, multiple methods; naming stability: skill names consistent, but no naming convention; examples and FAQ: demos and common issues; known limitations: some platform limitations mentioned, but not comprehensive; license: MIT clear; versioning and changelog: CHANGELOG.md and version notes; maintenance responsibility: maintainer not specified. Deductions: naming convention, known limitations, maintenance responsibility insufficient.
Evidence shows: high output usability, structured outputs and demos; marginal value: complete writing workflow, clear value; cost benefit: free and open source, but requires environment setup. Deductions: cost benefit not quantified.
Evidence shows: claims traceable, demos and docs; cross-source corroboration: external tools referenced, but no independent verification; fact/inference separation: README distinguishes facts and inferences. Deductions: cross-source corroboration insufficient.
- Publisher identity unverified; review code before installation.
- Hooks modify project files; back up before deployment.
- Dependencies not locked; supply chain risk.
- Sensitive data handling unclear; avoid storing secrets in project.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
v0.9.0 v0.9.0 — 资料按写书时刻重组
Release notes excerpt
升级后在写作项目根重跑 /story-setup（Codex 用 $story-setup），再新开会话。本版 agents_version: 36：助手模板、项目指令和参考资料包一起换了，不重跑的话旧助手会去找已经删掉的资料；书稿、设定、追踪和作者记忆不用迁移。 v0.9.0 把 13 个技能的参考资料按「作者在写书的哪一步」重新整理，每一步有一份流程说明和至多一份方法资料，重复、过时和没人读的资料合并后删除。 - 排纲：出卷纲、出细纲分「主会话怎么做」和「排纲方法」两部分，九份重复的方法资料并入对应文件；作者只说「这章没劲」之类的症状时，先诊断再回炉。 - 写正文：写手的技法规则和去 AI 味方法不再互相打架。对比试写测不到退步，检测器必须修的命中减半。 - 短篇：平台、导语、字数、付费点只在一处定，导语规则按七猫热门短篇定；平台统一叫「知乎盐言」。 - 导入：长篇卷纲改用单元卡，导入后用 /story-long-write 写第N章 续写；之前导入的长篇要重写旧章时，先把那一卷的旧单元改写成单元卡。 - 拆文与扫榜：旧拆文成果只补缺口，不整本重拆；同一页上的两份相同榜单只算一份，选题结论会推荐要拆的对标书。 …View on GitHub ↗ -
v0.8.6 v0.8.6：三层去味与全技能精修
Release notes excerpt
> 本版 agents_version: 35、setup_skill_version: 1.3.3。已有项目更新技能包后，在项目根重跑 /story-setup（Codex 用 $story-setup），再新开会话；助手模板、项目指令与参考资料包需要一起更新。书稿、设定、追踪和作者记忆数据无需迁移。 v0.8.6 把去 AI 味扩展为表达、叙事用笔与故事因果三层诊断，并精修全部技能的工作流说明。长期积累的词汇、句式和样本继续保留，作者记忆集中由长短篇写作管理。 Changed - 去味分别检查三层问题。 未限定范围的独立去味完整检查表达、详略与故事因果；先治理电报体和短碎句，也检查长句中的逐项播报。词频不再决定是否检查后两层，七个 Gate 保留为定点范围接口，不再按轻中重或固定删减率改稿。旧词表64个核心词项、重要句式分支与复合原例全部保留，命中后回到语境判断。 - 改稿与审稿同步执行边界。 表达、已有承接和详略可在普通去味中修复；改变动机、事件、结果或揭露顺序时，区分设计建议与已授权重写。只检测保持正文不变，定点改稿不整理范围外标点。审稿先复核检测提示的实际阅读损失，再按影响定级；功能性原句的保留仍遵守作者选择和既有白名单。 …View on GitHub ↗ -
v0.8.5 v0.8.5 — 写正文能带上的作者习惯约多一倍，扫榜加书库新书
Release notes excerpt
> 技能文件有改动，agents_version 与助手模板都没变（仍是 34）：更新技能包、新开会话即可，不用重跑 /story-setup。 v0.8.5 让写正文能带上的作者习惯约多一倍，没带上的会在汇报里点名；起点、七猫扫榜加上最近冒头的书库新书。 Added - 扫起点、七猫时能看最近冒头的新书：两边各多了一份「书库新书」——最近三天还在更新、30 万字以内的连载书，七猫按点击、起点按人气，翻页取前几页（默认 3 页，最多 10 页），名次就是点击或人气排名，汇总时算进「新书榜」那一列。七猫这份不用开浏览器，题材会补成和排行榜一样的大类；起点这份要开浏览器，页面上加密的字数会解出来，签约、收费从作品页补；新书的总推荐数值小，单独列出，不和月票榜、畅销榜混算热度。说「扫七猫新书」也会进长篇扫榜。 Changed - 写正文时能带上的作者习惯大约多一倍：记下的写作习惯，原来大半篇幅花在编号、分类这些格式上，写手实际读到的不到一半，十几条就开始漏。现在按写手真正读到的文字计算篇幅，同样的篇幅能装下约一倍的习惯（每条 40 字左右时从 8 条到 15 条）。 Fixed …View on GitHub ↗