huangziyuan-general/dsh-novel-forge 预览 preview

huangziyuan-general/dsh-novel-forge

DSH 小说锻炉:把 AI 长篇写作通病变成代码强制的硬约束(事实账本/上下文包/阶段门禁/零费用去AI味扫描/确定性审计/提案制修订)。Novel-writing guardrails plugin for DeepSeek Harness (DSH).

Project Overview项目介绍

This is a purpose-built native plugin for DeepSeek Harness (DSH) focused on AI-assisted long-form web novel writing. It follows the core design principle of replacing prompt-based self-discipline from the model with hard code-level constraints to fix common recurring issues that plague AI long-form writing. The plugin adds a dedicated “Forge” tab to DSH’s right sidebar, and exposes over 20 specialized novel_* tools that the DSH agent can call directly during interactive writing sessions. It also supports exporting its full toolset as a stdio MCP server that works with external clients like Claude Desktop and Cursor.

It is targeted at AI-assisted writers who create long-form serialized web novels for major Chinese platforms, and it enforces a strict 9-stage creation workflow to keep the project organized. Each stage has clear entry requirements that are checked automatically by code, and the plugin will explicitly list any missing required content before you can move to the next step. All new chapter writes require prior approval of the chapter outline, and every draft automatically passes through multiple gatekeeping checks before it can be saved to your project. Checks include repetition detection, word count validation, content sensitivity scans, and platform-specific rule audits.

This plugin is released under the open-source MIT license, and it is free to use and modify for any purpose. To install it via DSH’s plugin manager, you need a working DSH installation and Node.js version 22 or higher; versions 22.19+ enable full local search functionality, while older versions automatically fall back to a compatible mode that does not break core features. Uninstalling the plugin will never delete your existing novel manuscript data, which is stored in your DSH session workspace and can be imported directly into Git or Obsidian for backup and management. Most detection features are heuristic, and there are a small number of documented edge cases where false negatives may occur.

这是专为DeepSeek Harness(DSH)打造的原生小说创作插件,核心思路是用代码层硬约束解决AI长篇写作的各类常见通病,替代依赖模型自觉的提示词方案。插件为DSH添加了独立的右侧「锻炉」面板,同时向Agent暴露了20余个以novel_开头的专用工具,覆盖从项目初始化到完稿审稿的全流程,也可作为MCP服务器供Claude Desktop、Cursor等其他客户端使用。

它面向需要创作长篇网文的AI写作者,遵循固定的九阶段创作流程,从立意、设定、人物到大纲、细纲、正文逐步推进,每个阶段都由代码检查入场条件,缺项会直接列出提示用户补全。写章前必须批准细纲才能解锁创作,生成正文后会自动经过重复检测、字数检查、敏感内容自查、平台规则审核等多道门禁,不符合要求会直接拒绝保存落盘。

该插件采用MIT许可证开源,完全免费供用户使用,安装要求DSH环境正常运行,Node.js版本不低于22,版本22.19及以上可启用完整的本地检索功能,低版本会自动降级不影响核心功能。卸载插件不会删除用户已创建的书稿数据,所有内容存储在DSH会话工作区,可直接导入Git或Obsidian进行管理,目前已知部分启发式检测存在少量边界限制。

Pre-install check安装前体检Compatibility · Security兼容性 · 安全性 1 note1 项提示
  • 15 stars - an early-stage project星标 15,属于早期项目
DSH walks through these 9 checksDSH 会逐条核对这 9 项

Compatibility兼容性

  • DSH, Node, OS and profile requirementsDSH 版本 / Node 版本 / 操作系统 / profile 是否满足要求
  • External dependencies and runtimes (Electron / Python / Docker, ...)外部依赖与运行时(Electron / Python / Docker 等)是否齐备
  • Conflicts with installed plugins: command names, skill / tool names, ports, duplicate MCP registration与已装插件是否冲突:命令名、skill / tool 重名、端口占用、重复 MCP 注册

Security安全性

  • Repo matches the facts registered here; archived or abandoned?仓库是否与页面登记一致,是否归档或长期停更
  • Safety of preinstall / install / postinstall and install.sh / setup.ps1preinstall / install / postinstall 与 install.sh、setup.ps1 是否安全
  • curl|bash, download-then-execute, obfuscation, unrelated domains → stop immediatelycurl|bash、下载即执行、混淆代码、无关域名 → 立刻停止
  • Typosquatting or unmaintained packages among the new dependencies新增依赖里有没有 typosquatting 或无人维护的包
  • Requested permissions vs. what the feature actually needs申请了哪些权限、是否超出功能所需(filesystem / network / shell / clipboard)
  • Any sudo / admin requirement, plus uninstall and rollback是否要求 sudo / 管理员权限,以及卸载与回滚方式

Anything uncertain must be marked unknown with a note on how to confirm it. This site's signal screen is a static snapshot, not a security audit.拿不准的必须标「未知」并说明要我怎么确认。本站的信号筛查是静态快照,不能替代安全审计。

Or use CLI install (for developers)或使用命令行安装(适合开发者)

CLI Install命令行安装

dsh plugin --profile web add github:huangziyuan-general/dsh-novel-forge

把 huangziyuan-general/dsh-novel-forge 加入你的 DSH 配置(web profile)即可启用。

READMEREADME

dsh-novel-forge · 小说锻炉

release license tests

DeepSeek Harness(DSH)的小说创作插件。设计主线只有一条:

代码强制 > 提示词自觉。 凡是"靠模型自觉"的约束都会失效—— 一致性、门禁、审计必须是工具层/数据层的硬约束。

AI 长篇写作的通病不是玄学,每一个都有对应的工程解法。本插件把其中"有明确对错"的部分做进代码:

AI 写作通病 本插件的硬约束 落点
设定/战力崩坏 事实账本:同章改值=冲突拒绝;世界书关键词自动注入 novel_ledger / novel_worldbook
模型"看不见前文" 写前简报按预算组装上下文包(细纲→人物卡→账本→伏笔→世界书→上章结尾) novel_briefing
口头跳阶段 阶段门禁:细纲未批准,novel_write_chapter 直接拒绝 novel_outline approve
跨章复读/水文 8 字 shingle 与前文重复检测,超阈值拒绝保存 novel_write_chapter 机审
章节字数失控 机审门槛默认 2000–4000 字(网文连载单章标准,目标 3000);写前简报带「本章字数目标」段 novel_write_chapter / novel_briefing
章末无钩子 章末钩子启发式(问句/悬念/省略/感叹),缺了给警告 novel_audit
AI 味 六维结构性扫描:模板句/库存词密度/情绪直写/句式模板/段落节奏方差/信息稀释(纯本地,零模型费用) novel_noai_scan
偷偷覆盖正文 提案制:修订走提案→用户确认→生成新版本,旧版永不覆盖 novel_propose
审稿放水 机审与模型审分离:审计产出确定性数字,模型审稿必须引用证据 novel_audit
悬念提前泄底 场景契约:隐藏人物档案不进上下文(连账本/摘要里的痕迹也擦掉),正文出现其名=内容门禁拦下 novel_scene
写偏/漏写细纲 细纲契约段(必写场景 / 禁止偏离项)由代码算覆盖率与偏离度,命中禁项拒绝落盘 novel_write_chapter
多角色千人一腔 语言基因卡结构化注入(句长/逻辑/口头禅/绝不说/小动作/语域),审稿核对禁忌词 novel_character voice
长篇上下文爆炸 契约在场时只注入本章出场人物的卡,世界书按白名单取 novel_briefing
设定没定就想写大纲 九阶段状态机:每阶段入场条件由代码判定,缺什么直接列出来 novel_project phase
同一章反复写不对 熔断:同章连续驳回 3 次即拒写(不是提醒),改完细纲/契约自动解除 novel_write_chapter
不合平台口味 平台审稿两张表:起点看结构/章末钩子/移动端段长,番茄看前 1000 字爽点/打脸/憋屈时长 novel_audit platform
踩平台红线 敏感自查七类,命中给行号与改法 novel_audit censor
内部工序占满主对话 旁路直调:润色/校对/打标/起草走独立流,不占主对话、不写会话记录;结果只出提案 lib/engine.js
批量起草绕过门禁 并发生成、串行提交:每章仍过同一道门禁,单章失败不回滚整批 draft-batch
写到百章后"看不见前文" 本地检索索引:按记忆碎片把段落找回来(中文手工二元切分,零依赖) novel_search

界面一览

右侧栏「锻炉」面板实机截图(演示书目为演示新建的虚构内容):

项目列表
详情 · 基本信息
章节听书
阅读卡

Showing the opening section of the README — the full document lives in the repository以上为 README 开头摘要,完整文档在仓库内 · View the full README on GitHub →在 GitHub 查看完整 README →

← 上一个 Prev dsh-ui-skin-loader 下一个 Next dsh-smart-charts →