Back to home

Compare

Comparing: Juejin Developer Builds a Skills Pipeline to Fix AI Novel Continuity & AI 写网文卡在「前后矛盾」——掘金开发者用 Skills 搭出可续写流水线

AEN
SkillsAnthropicObsidian·

Juejin Developer Builds a Skills Pipeline to Fix AI Novel Continuity

What this is

This week, a post on Juejin showed where AI writing of long-form web novels truly gets stuck: it is not prose quality, but characters abruptly acting out of character (OOC) midway through the story. Developer “Xiao Feichang (小肥肠)” merged a previously built n8n workflow and a short-story Skill into a single workbench, then added Obsidian for local storage to create a button-based creative pipeline.

Skills is a concept Anthropic introduced alongside Claude. In simple terms, it is a set of prompts and resources that can be called with a single click. The novel-writing Skills do three things: continue writing in the styles found in an existing reference knowledge base; automatically generate the next chapter from the outline and character profiles; and check narrative continuity between chapters—in other words, prevent characters from changing from one chapter to the next. Every completed chapter is saved directly to Obsidian, and before the next chapter begins, the AI first reads the previous chapter. This is the blunt method it uses to resolve contradictions, and it is currently the most reliable one.

Industry view

Supporters argue that the real value of this workflow lies not in a technological breakthrough, but in moving AI from a “chat window” to “pipeline buttons.” This directly addresses individual creators’ need to combine scattered actions into a one-click flow; the workbench interface is the key design that allows Skills-based tools to be put into practice.

There are two strands of opposition, however. The first comes from web-fiction platforms and veteran editors: continuity checks can fix contradictions on the surface, but paid-chapter retention depends on designing anticipation and controlling emotional pacing, which AI still cannot grasp. The second comes from the developer community: the author himself acknowledges that, after generating dozens of chapters in batches, Token costs and the workload of manual repairs begin to erode AI’s efficiency gains. Therefore, the real capability boundary of this pipeline today is “help you produce a draft, and you revise it,” not “write the whole thing for you.”

Impact on regular people

For AI tool developers: wrapping scattered Prompt operations, APIs, and knowledge bases into a visual, clickable workbench is the main form AI tools are taking this year to lower the barrier for nontechnical users. It is not limited to fiction.

For independent professionals: solo content producers in self-media, consulting, and operations are using similar Skills to build their own content pipelines, further lowering the barrier to the “one-person company + AI Skills” model.

For the web-fiction consumer market: when AI tools can consistently produce manuscripts good enough to sign, the supply side will face a shock—the barrier to securing top-tier contracts will rise, while opportunities for mid- and lower-tier creators may open because of the cost advantage.

Source: juejin.cn
BZH
SkillsAnthropicObsidian·

AI 写网文卡在「前后矛盾」——掘金开发者用 Skills 搭出可续写流水线

这是什么

这周掘金上一个帖子展示了 AI 写长篇网文真正卡住的地方:不是文笔,是「写到中途主角性格突变 OOC」(Out of Character,脱离人设)——开发者「小肥肠」把之前做过的 n8n 版本和短篇 Skill 合并进同一个工作台,再搭上 Obsidian 做本地存储,造出一套按钮式的创作流水线。

Skills 是 Anthropic 给 Claude 配套推出的概念,简单说就是「可一键调用的提示词 + 资源包」。这套小说 Skills 含三件事:参考知识库里已有文风做续写、按大纲和角色设定自动生成下一章、检测章节间的剧情连贯性(也就是「防止换章换人」)。每一章写完直接存进 Obsidian,下一章开始前 AI 先读上一章——这是它解决前后矛盾的笨办法,也是目前最稳的办法。

行业怎么看

支持者认为这套工作流的真正价值不是技术突破,而是把 AI 从「聊天窗口」推进到「流水线按钮」,正好踩中个人创作者要把零散操作合并成一键的痛点;工作台界面是 Skills 类工具能落地的关键设计。

但有两层反对声音。一层来自网文平台和老编辑:连续性检测能修文字表面的矛盾,但追订率靠的是「期待感设计」「情绪节奏」,这些 AI 还摸不到;另一层来自开发者圈——作者自己承认,批量写几十章以后,Token 成本加上人工修补的工作量开始吃掉 AI 提效。所以这条流水线今天真实的能力边界是「帮你出一稿,你来改」,不是「替你写完」。

对普通人的影响

对 AI 工具开发者:把零散 Prompt、API、知识库封装成可视化可点的工作台,是今年 AI 工具降低非技术用户门槛的主要形态,不只适用于小说场景。

对个人职场:自媒体、咨询、运营这些「单兵作战」的内容生产者,正在用同类 Skills 搭自己的内容流水线,「一人公司 + AI Skills」的组合门槛被进一步压低。

对网文消费市场:当 AI 工具能稳定产出可签约水平的稿件,供给侧会出现一次冲击——头部签约门槛被抬高,腰部和底部的口子反而可能因为成本优势被打开。

Source: juejin.cn