Sumeru Writing skill
须弥写作 (Sumeru Writing)
Eight skills for Chinese web novels: topic, outline, chapters, logic review, polish and export for publishing platforms
- 232 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The review and polish stages change chapter files in place, and the previous version is copied to `.sumeru/write/original/`
- Sub-agents at the review stage run in parallel and burn token limits fast
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/sumeru-writingDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The review and polish stages change chapter files in place, and the previous version is copied to `.sumeru/write/original/`
- Sub-agents at the review stage run in parallel and burn token limits fast
This is third-party code. Review the repository files before installing.
What it does
Sumeru Writing is a set of eight skills for writing a web novel (the Chinese online fiction genre) in Claude Code and OpenCode, from topic to finished manuscript. The main skill `sumeru-worldbuilder` runs the whole pipeline: an optional reference-book analysis, topic choice, an outline with a detailed chapter breakdown, writing, logic review, style polish and file release. Each stage can be called on its own: `/sumeru-topic`, `/sumeru-outline`, `/sumeru-write`, `/sumeru-review`, `/sumeru-polish`, `/sumeru-finalize` and `/sumeru-deconstruct`. Numbers and checks come from scripts on the Python standard library: `plan-chapters.py` counts chapters, `chapter-word-counter.py` counts length, `validate-timeline.py` catches time running backwards and age contradictions, and `build.py` assembles md, txt, html and epub for Qidian, Fanqie, Jinjiang and Zongheng. Working data lives in the `.sumeru/` folder, so after an interruption you can resume from the same stage.
Who it is for. Authors of Chinese web novels who write long texts together with an agent and want planning, plot control and platform export in one place.
Good fit when
- You only have a genre and a couple of keywords and need a topic, a plot and a chapter list: `/sumeru-worldbuilder`
- The novel runs to hundreds of thousands of characters, and you must track chronology, character ages and names
- A manuscript has to be prepared for Qidian, Fanqie, Jinjiang or Zongheng as md, txt, html or epub
- Work was interrupted, and you want to continue the novel from the last stage using the data in `.sumeru/`
- You want the structure of a popular book as a model: the analysis yields a template and pacing parameters without retelling the text
Not a fit when
- The novel is not in Chinese: the genre references, blocklists and rules are written for Chinese web fiction
- You want a close rewrite of someone else book: the analysis gives only a structural template and checks overlap, allowing up to 12 characters in a row
- You need a short story or a text of another genre: the pipeline is built for long form from 200 thousand characters
Example request
/sumeru-worldbuilder 都市 "重生+投资+创业" 长篇 精品文Limitations
The README, commands and references are in Chinese. Topic demand estimates come from the model own knowledge, and the skill has no live platform charts. The scripts need Python 3 with the standard library. The review and polish stages change chapter files in place, and the previous version is kept in `.sumeru/write/original/`. The blocklists of names, characters and places help avoid collisions but do not replace a legal check before publication. Parallel sub-agents at the review stage spend many tokens.
How to disable. Delete the eight `sumeru-*` folders from the agent skills directory (`.claude/skills/` or the OpenCode directory). Working data stays in `.sumeru/`, and finished chapters in `chapters/` and `output/` in your project; delete them by hand.
FAQ
How do I run the whole pipeline at once?
With `/sumeru-worldbuilder <genre> "<keywords>"`, for example `/sumeru-worldbuilder 玄幻 "废柴逆袭+系统流"`. You can add a title, length, style and tone, or the words for resume and skip to continue or skip stages.
What lengths are supported?
Four formats: short (about 200 thousand characters, 40 to 70 chapters), medium (200 to 500 thousand), long (500 thousand to 1 million) and extra long (over 1 million characters, more than 340 chapters). A base chapter is 3000 characters, and `plan-chapters.py` counts the chapters.
How does the skill catch plot contradictions?
The `sumeru-review` stage pulls events into a structured timeline, and `validate-timeline.py` checks time running backwards, character ages, overlapping intervals and cause references. Sub-agents then check chapter details, three chapters each.
What does the reference-book analysis do?
The `sumeru-deconstruct` skill analyzes a book and yields the structure, pacing and template parameters. The `check-derivative.py` script verifies that overlap with the original is no longer than 12 characters and that the report holds no proper names from the author.
Where are results saved?
Intermediate data lives in `.sumeru/`, chapters in `chapters/` and the platform files in `output/` of the working directory. `python3 skills/sumeru-finalize/scripts/build.py --check-only` checks the manuscript without exporting.
Discussion
No questions or reviews yet.
Related
SkillsA skill for multi-volume novels: 30-plus protocol files keep world, characters and style consistent from chapter to chapter
A Claude Code plugin for long web novels: volume planning, chapter writing and review, memory of characters and plot, a state dashboard
A skill helps Codex plan Chinese web fiction, revise scenes and check character actions against earlier chapters
A set of 13 skills for writing web fiction inside a coding agent: rank scanning, hit deconstruction, outline and prose, AI-flavor cleanup, cover art
A skill for original Chinese short fiction: you pick a plot strategy and outline first, then it writes and runs quality checks
15 skills for Codex that cover Xiaohongshu content from account research and topic choice to title, draft and pre-publish QA
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| Sumeru Writing (this card) | no | 232 | |
| Webnovel Writer | no | 7,939 | It also helps write web novels with an agent; Sumeru is split into eight skills with gates between stages, script-based timeline validation and export for platforms. |
| Chinese-WebNovel-Skill | no | 1,006 | A skill for the same genre; Sumeru adds separate stages for topic choice, reference analysis, logic review and platform compatibility. |
README in short
The README in Chinese lists features, the eight-module structure and a table of lengths, then gives installation, command examples with parameters for each stage, a description of the scripts and a folder layout for data. The end covers compliance: blocklists and naming rules.
README badge
Original SKILL.md
---
name: sumeru-worldbuilder
description: 网文/小说全流程创作世界构建师，适用于用户说"我想写小说"、"写一本网文"、"从零开始创作小说"、"帮我写本XX类型的小说"、"我要写本小说"、"给我整个小说创作流程"、"自动写小说"、"小说创作一站式服务"、"帮我完成一本小说"、"我只有创意怎么写小说"等整体小说写作需求，协调选题→大纲→写作→审查→润色→导出全流程，每阶段有可执行验收门禁，章节级操作使用子Agent并行（每Agent最多3章）
type: skill
---
## 网文创作世界构建师 (worldbuilder)
### 触发关键词
我想写小说、写一本网文、从零开始创作小说、帮我写本XX类型的小说、我要写本小说、给我整个小说创作流程、自动写小说、小说创作一站式服务、帮我完成一本小说、我只有创意怎么写小说、从零开始写网文、小说全流程创作
### 核心功能
worldbuilder 是网文创作的一站式主控技能，负责统筹协调从创意萌芽到作品完稿的完整创作链路，构建完整统一的故事世界。通过智能编排各专项技能的调用顺序与参数传递，实现自动化、流水线式的网文创作体验：
0. **可选前置·拆书**：如果用户提供了参考书/对标作品，先调用 `sumeru-deconstruct` 输出结构模板与节奏参数（只拆结构，不抄文本）
1. **选题策划**：调用 `sumeru-topic` 进行市场分析、选题定位，生成核心创意与卖点
2. **大纲设计**：调用 `sumeru-outline` 构建完整世界观、人物设定、分卷大纲与**完整章节细纲**
3. **内容创作**：调用 `sumeru-write` 按大纲进行分章节内容撰写，保持风格统一（**必须完成所有章节后才进入下一阶段**）
4. **逻辑审查**：调用 `sumeru-review` 对**所有已完成章节**进行全面审查，校验时间线一致性、人物行为逻辑、前后剧情连贯性
5. **内容润色**：调用 `sumeru-polish` 对**所有已审查章节**进行文笔优化、细节丰满、节奏调整
6. **完稿校验**：调用 `sumeru-finalize` 对**所有已润色章节**完成错别字检查、标点规范、逻辑漏洞最终排查与作品打包
### 全局约束
两项全局硬约束统一定义在 AGENTS.md，本 Skill 不重复展开：
- **子Agent并行**：每个子Agent最多负责 3 个章节（写作/审查/润色/校验/细纲均适用）
- **自动备份**：review / polish 修改 `chapters/` 前，必须备份到 `.sumeru/write/original/`
worldbuilder 的调度责任只有三条：
1. 调用子技能时显式传递“每Agent最多3章”参数
2. 在阶段检查点执行 AGENTS.md 定义的「阶段门禁规则」
3. 门禁未通过时阻断流程，不得“先往下走再说”
### Skill 协调流程
worldbuilder 负责以下数据流转和协调工作：
```
用户需求
↓