ZenStory AI

Practical guide · 实用指南

Continue an unfinished novel with AI: import 10–20 chapters into Oh Story

已有 10–20 章怎么让 AI 接着写：导入 Oh Story，接续人物设定与前史

Source checked 2026-09-12 · 源码核对日期

If you already have 10–20 chapters, do not ask an empty chat to ‘continue in the same style.’ Use story-import to reconstruct the manuscript as an Oh Story writing project, review its inferences, then request the next chapter. Import preserves the source prose while building settings, outlines and continuity records. It cannot recover unwritten intent, guarantee a perfect voice match or settle every ambiguity; the manuscript remains the evidence.

已经写了 10–20 章，不要在空白聊天里只说“照这个风格续写”。先用 story-import 把书稿重建成 Oh Story 写作工程，审阅反推结果，再点名写下一章。导入会保留原文，同时建立设定、大纲与连贯性记录；它不能恢复未写进正文的意图，也不保证无损复刻文风或补对所有歧义。书稿证据始终优先于模型猜测。

Oh StorySource on GitHub

Before you start · 开始之前

Steps · 操作步骤

  1. Choose import, not deconstruction Use story-import for a continuable project: settings, outlines, normalized prose and tracking. Use story-long-analyze only for a deconstruction library and report. Import analyzes your manuscript, but that analysis is a data source, not the deliverable or an external comparison title.
  2. Prepare a clean source boundary Use one .txt/.md file, a directory with sortable chapter filenames, or pasted text. Exclude advertisements, author notes and alternate drafts unless they are evidence. State the title, genre, platform, completion status and last complete chapter. Confirm the detected chapter range and long/short classification.
  3. Set up first, then import For Claude Code’s npx path, run /story-setup in the workspace, refresh the session, then run /story-import and provide the path. Direct import is allowed, but missing project infrastructure can mean setup first or a serial fallback. Do not mix this spelling with namespaced marketplace-plugin commands.
  4. Keep author decisions with the author Confirm continuable project versus analysis only, the last complete chapter, and treatment of a partial ending. If the text has no volume markers, review candidate divisions before volume outlines are finalized. Unsupported fields should stay [待补充] or become continuity risks—not confident-looking filler.
  5. Review the reconstructed handoff Confirm 正文/ has every expected chapter and unchanged prose. Then inspect 设定/角色/, 大纲/ and 追踪/上下文.md. Current identities, locations, goals, relationships, knowledge, unresolved threads and the next-chapter promise must match evidence through the last complete chapter. Separate author-only truth from character and reader knowledge.
  6. Restore intent, then request prose Resolve important [待补充] items or leave them uncertain. State the ending to carry forward, viewpoint and voice constraints, facts not yet revealable, and the next chapter’s job. Correct the plan first if reconstruction points the story the wrong way. This restores intentions that prose could not reveal.
  7. Name the next chapter and keep one authority From the imported book directory, use /story-long-write 写第21章 (replace 21 with N+1) or a scoped daily request. A bare command only diagnoses the project. Revise an old chapter through its revision route; do not hand-edit derived tracking views. A current project with 追踪/_tracking-state.json normally continues without full re-import. If you expand the imported source range, the documented path is another full import rather than appending history to old snapshots. Back up any work added since the first import before rebuilding.
  1. 先选导入续写，不要只做拆文 要得到可续写的设定、大纲、正文与追踪状态，就用 story-import；只要拆解资料和报告，才用 story-long-analyze。导入会分析书稿，但拆文库是重建数据源，不是最终工程，也不会把自己的书自动登记成外部对标。
  2. 把导入范围整理干净 使用一个 .txt/.md、文件名可排序的章节目录，或直接粘贴文本。广告、作者的话和不同版本草稿若不是故事证据，就移出导入集。说明书名、题材、平台、完本状态与最后完整章，并确认系统识别的章节范围和长短篇分类。
  3. 先部署环境，再导入 按 Claude Code 的 npx 路径，在工作区运行 /story-setup，刷新会话，再运行 /story-import 并提供路径。直接 import 也可以，但基础设施缺失时会选择先 setup 或串行降级。不要与 marketplace 插件的命名空间命令混用。
  4. 作者决定不能外包 确认目标是可续写工程还是仅分析、最后完整章号，以及残稿如何处理。原文没有卷标时，先审阅候选卷界，再定卷纲。正文无法证明的字段保留 [待补充] 或写入连贯性风险，不为表格整齐而补猜。
  5. 正文前先审阅交接状态 确认 正文/ 章节齐全且原文未被改写；再看 设定/角色/、大纲/ 和 追踪/上下文.md。人物身份、位置、目标、关系、本人已知、未结事项与下一章承诺，都要能由最后完整章以前的内容证明，并分开作者真相、角色认知与读者已知。
  6. 补回作者意图，再要正文 解决关键 [待补充]，或明确保留为未知。说明上一章结尾怎么承接、视角与语言习惯、暂不能揭露的事实，以及下一章任务。正文从未呈现的意图由你补回；若反推规划方向错了，先修规划，不要硬续。
  7. 点名下一章，只保留一个权威工程 进入导入后的书名目录，用 /story-long-write 写第21章（把 21 换成 N+1），或给出范围清楚的日更请求。裸调用只诊断状态。旧章变化影响连续性时走大修流程，不要手改派生追踪视图。已有 追踪/_tracking-state.json 的项目通常直接续写；扩大原文导入范围时，文档要求重新完整导入，而不是向旧快照追加历史；重建前先另存备份，保留首次导入后新增的工作。

Example · 示例

Suggested requests—not recorded or tested model output. Assume chapters 1–20 are complete and chapter 21 is a separate partial draft.

/story-import
Import /Users/me/novels/harbor-radio/complete-chapters.md as a continuable long-form project. Chapters 1–20 are complete; do not treat the chapter-21 fragment as fact. Preserve the prose. Before migration, show the detected range, unresolved facts and candidate volume boundaries.

After reviewing the project:

/story-long-write 写第21章
Continue chapter 20’s final scene in close third person on Lin Zhou. Keep the restrained dialogue rhythm. The reader knows about the red notebook; Chen Yu does not. Do not reveal who changed the broadcast log. Make Lin Zhou choose whether to repair the receiver before the tide returns. Write only chapter 21 and report changed project records.
以下是建议请求，不是已记录或已测试的模型输出。假设第 1–20 章完整，第 21 章残稿另存。

/story-import
把 /Users/me/novels/海港收音机/完整章节.md 导入为可续写长篇工程。第 1–20 章完整；第 21 章残稿暂不视为事实。保留原文。迁移前展示章节范围、待确认事实和候选卷界。

审阅工程后：

/story-long-write 写第21章
承接第 20 章最后场景，继续林舟的第三人称限知视角，保持克制的对话节奏。读者知道红色笔记本，陈雨不知道；不要揭示谁改了广播日志。本章让林舟在涨潮前决定是否修接收机。只写第 21 章，并说明更新了哪些项目记录。

Expected files · 预期文件

Verify the result · 验证结果与边界

Sources and version notes · 来源与版本说明