Junli Ai Novel

v0.1.4

平台向长篇网文连载写作助手。负责新项目立项、章节续写、单章返修、项目记忆维护、长篇分卷治理和章节质检。内置章节/场景/语言三层写法原则,适配爽文、言情、权谋等主流题材,支持从脑洞到超长篇全流程。更多信息关注抖音君黎。

3· 315·1 current·1 all-time
by君黎@ljunn

Install

OpenClaw Prompt Flow

Install with OpenClaw

Best for remote or guided setup. Copy the exact prompt, then paste it into OpenClaw for ljunn/junli-ai-novel.

Previewing Install & Setup.
Prompt PreviewInstall & Setup
Install the skill "Junli Ai Novel" (ljunn/junli-ai-novel) from ClawHub.
Skill page: https://clawhub.ai/ljunn/junli-ai-novel
Keep the work scoped to this skill only.
After install, inspect the skill metadata and help me finish setup.
Use only the metadata you can verify from ClawHub; do not invent missing requirements.
Ask before making any broader environment changes.

Command Line

CLI Commands

Use the direct CLI path if you want to install manually and keep every step visible.

OpenClaw CLI

Bare skill slug

openclaw skills install junli-ai-novel

ClawHub CLI

Package manager switcher

npx clawhub@latest install junli-ai-novel
Security Scan
VirusTotalVirusTotal
Benign
View report →
OpenClawOpenClaw
Benign
medium confidence
Purpose & Capability
Name and description describe a longform serial-novel workflow and the repository contains writing guidelines, lint/rule files, and orchestration scripts (chapter_pipeline.py and per-task scripts). The requested actions (reading project docs, running pipeline commands) match the stated purpose. There are no unrelated credentials, binaries, or config paths declared.
Instruction Scope
SKILL.md instructs the agent to run local Python commands (scripts/* and especially scripts/chapter_pipeline.py) and to read many files under the project tree (docs/, manuscript/, plot/, runtime/, rules/, references/). That is appropriate for a writing assistant. However, the runtime behavior depends entirely on the Python scripts' contents — they may read, write, or otherwise modify files in the provided project directory and runtime/; inspect them for any steps beyond expected text processing (network calls, arbitrary shell execution, access to system paths).
Install Mechanism
No install spec is provided (instruction-only), which is low risk in itself. The repository contains executable Python scripts that the SKILL.md expects to be run with the system Python. Because there is no packaged/verified install step, execution will directly run repository code from disk; this is coherent but means code review is needed before execution.
Credentials
No environment variables, credentials, or external config paths are required by the skill metadata or SKILL.md. All declared operations are local to a project directory under the repo, which is proportionate to the stated purpose.
Persistence & Privilege
Skill does not request always:true and uses normal agent invocation. The workflow writes/reads project runtime artifacts (runtime/ directory) as part of normal operation, which is expected for a content-authoring tool. There is no indication it modifies other skills or global agent settings.
Assessment
This skill appears coherent for longform writing: it orchestrates local Python scripts to manage projects, plans, and chapter QA. Before installing or running it: 1) Review the largest scripts (especially scripts/chapter_pipeline.py and any helper scripts) for network calls (requests, urllib, sockets), subprocess usage, or references to external URLs or hardcoded credentials. 2) Run the code in an isolated environment (container or VM) and point it at a safe test project, not your home directory or sensitive repos. 3) Inspect what files the scripts create/overwrite (runtime/, docs/, manuscript/) and back up any real projects first. 4) If you need extra assurance, run static analyzers or grep for suspicious functions (socket, requests, urllib, subprocess, os.system, open paths outside project). 5) If you plan to let an autonomous agent invoke these commands, restrict its scope and review logs; autonomous invocation increases blast radius if the scripts do unexpected I/O or networking.

Like a lobster shell, security has layers — review code before you run it.

latestvk9719eqg0s9602eq2bvhy51kv9858asm
315downloads
3stars
4versions
Updated 1w ago
v0.1.4
MIT-0

君黎 AI 网文连载

先读 PROJECT.md。它把本仓库重封装成 Rule / Workflow / Command 三层索引,适合先找入口,再决定是否下钻到原子命令。

核心目标

  1. 最高优先级:连载稳定、追读牵引、卖点兑现、长线不崩。
  2. 默认按平台向长篇网文处理,不按出版文学标准做默认决策。
  3. 先读档,再下笔;先编排,再正文;先检查,再宣称完成。
  4. 任何新内容不得与更高层级记忆硬冲突。

默认入口顺序

优先顺序不是“先背命令”,而是:

  1. 先看 PROJECT.md
  2. 再判断当前任务属于哪个 Workflow
  3. 只有需要拆解排查时,才回到底层 Command

默认 Workflow:

  • 继续写下一章:next-chapter
  • 章节静态预审 + 审稿稿本:review
  • 商业化包装:marketing
  • 长篇治理:bootstrap-longform + governance + audit

硬门槛

命中续写、开写、返修、恢复记忆、长篇治理时,下面都是必须动作:

快捷路径:续写场景直接用 next-chapter 命令,它会自动串联步骤 1-3 中的所有前置动作(preflight → resume → plan → compose → start),无需单独执行每一条原子命令。只有当 Workflow 失败或需要精细排查时,才拆回原子命令。

  1. 正文前必须先跑:
python3 scripts/chapter_pipeline.py preflight <项目目录>
python3 scripts/chapter_pipeline.py resume <项目目录>
  1. 长篇项目正文前必须先跑:
python3 scripts/chapter_pipeline.py plan <项目目录> --chapter-num <章节号> --chapter-title "标题"
python3 scripts/chapter_pipeline.py compose <项目目录> --chapter-num <章节号> --chapter-title "标题"
python3 scripts/chapter_pipeline.py start <项目目录> <章节号> --chapter-title "标题"
  1. 写完后必须先跑:
python3 scripts/chapter_pipeline.py check <章节文件路径>
python3 scripts/chapter_pipeline.py finish <项目目录> <章节号> <章节文件路径> --chapter-title "标题" --summary "摘要"
  1. 满足任一条件即视为长篇治理范围:

    • 目标字数超过 100 万字
    • 实际已写超过 30 万字
    • 项目已写到 20 章以上
    • 用户明确要求分卷、阶段、卷末结算、结构变更、超长篇控盘(此条由 AI 自行判断,工具不检测)
  2. 长篇治理范围内,缺治理文件时先跑:

python3 scripts/chapter_pipeline.py bootstrap-longform <项目目录>
  1. 长篇治理范围内,每 10-20 章必须跑阶段审计;每卷结束必须跑卷审计:
python3 scripts/chapter_pipeline.py audit <项目目录> --scope stage
python3 scripts/chapter_pipeline.py audit <项目目录> --scope volume
  1. 如果 preflight 因当前卷 / 当前阶段 / 阶段目标缺失而失败,先跑:
python3 scripts/chapter_pipeline.py governance <项目目录> --current-volume "第一卷" --current-phase "阶段1" --phase-goal "目标"
  1. 如果要改全书终局、主角长期弧线、世界底层规则、长线关系边界,必须先更新 docs/变更日志.md,再继续正文。
  2. 如果无法执行命令、无法读取关键文件或前置校验失败,必须明确报告阻塞;禁止假装已经恢复上下文、已经检查或已经同步进度。

常见失误

下面这些做了等于白做,必须避免:

  • 没跑 preflight / resume 就续写正文(上下文没恢复)
  • 记忆文件缺失时假装已恢复,继续写
  • 长篇项目缺卷纲 / 阶段规划 / 变更日志,却平推正文
  • 超过一个阶段没跑 audit
  • 核心设定或终局方向改了,却没写进变更日志
  • 用户只要求局部返修,却整章重写

任务分流

先把请求归进下面 5 类。命中 D 或 E 时,先修 D / E,再回到 B / C。

A. 新项目 / 设定不完整

触发:

  • 从零开始
  • 项目目录不存在
  • 只有脑洞、梗概或人设碎片

必须动作:

  • 只补齐 5 问立项:题材风格、主角结构、主角核心性格、核心冲突、预计章节规模
  • 先用主线三步法:终点站 -> 起爆事件 -> 连锁反应
  • 再用终极锚点 -> 逆推问题链 -> 分卷锚点
  • 最后初始化项目:
python3 scripts/chapter_pipeline.py init "项目名" --mode single

--mode 可选值:

  • single:单主角线(默认,适合大多数升级流、爽文、言情)
  • dual:双主角或双视角结构(CP 文、双雄文、平行叙事)
  • ensemble:群像多线(权谋、宫斗、多主角群像)

参考:

  • references/main-plot-construction.md
  • references/outline-template.md
  • references/chapter-plan-template.md
  • references/outline-refinement.md
  • references/longform-governance.md

B. 继续连载 / 下一章

触发:

  • 继续写
  • 下一章
  • 新开对话续旧项目

必须动作:

  • 先过硬门槛
  • 读取顺序固定:task_log.md -> docs/项目总纲.md -> docs/章节规划.md -> plot/伏笔记录.md -> plot/时间线.md -> 相关角色档案 -> 上一章正文
  • 恢复输出至少包含:创作阶段、最近 2-3 章摘要、主角当前状态、活跃伏笔、下一章目标
  • 长篇项目必须先 plan / compose
  • 再进入场景拆分和正文

C. 单章扩写 / 重写 / 润色 / 质检

触发:

  • 用户明确点名某章
  • 用户要扩字数、改节奏、修对白、减 AI 味
  • 用户要检查爽点、毒点、情绪曲线或一致性

必须动作:

  • 读取目标章节
  • 至少补读上一章、下一章或相关设定中的必要部分
  • 只做定向返修,不全文推倒
  • 如果是局部问题,优先从 references/micro-revision-ops.md 中选择单一微操作
  • 如果问题在主线或章节层,先回上层重查

D. 恢复记忆 / 修复项目记忆

触发:

  • 恢复上下文
  • 唤醒记忆
  • 记忆文件缺失

必须动作:

  • 明确报告缺失文件
  • resume 只作机器摘要,不等于修复完成
  • 从正文、现有文档、角色档案手动重建
  • 先补记忆,再继续正文

E. 长篇治理 / 阶段审计 / 卷审计 / 结构变更

触发:

  • 超长篇控盘
  • 分卷
  • 阶段规划
  • 卷末结算
  • 结构变更

必须动作:

  • 确认存在:
    • docs/全书宪法.md
    • docs/卷纲.md
    • docs/阶段规划.md
    • docs/变更日志.md
  • governance 同步当前卷 / 当前阶段 / 阶段目标
  • 阶段切换前跑 audit --scope stage
  • 卷末结算前跑 audit --scope volume
  • 结构性改动先写 变更日志.md
  • 审计失败时,禁止继续正文

项目结构

[项目目录]/
├── docs/
│   ├── 项目总纲.md
│   ├── 章节规划.md
│   ├── 作者意图.md
│   ├── 当前焦点.md
│   ├── 全书宪法.md
│   ├── 卷纲.md
│   ├── 阶段规划.md
│   ├── 变更日志.md
│   ├── 冲突设计.md
│   ├── 世界观.md
│   ├── 法则.md
│   ├── 关系图.md
│   └── 群像主题拆分.md
├── characters/
├── manuscript/
├── plot/
├── runtime/
│   ├── chapter-0001.intent.md
│   ├── chapter-0001.context.json
│   ├── chapter-0001.scenes.md
│   ├── chapter-0001.rule-stack.yaml
│   └── chapter-0001.trace.json
└── task_log.md

约束:

  • manuscript/ 只放纯正文
  • runtime/ 只放本章运行时产物,不手写正文
  • docs/作者意图.mddocs/当前焦点.md 属于输入治理层
  • docs/项目总纲.mddocs/卷纲.mddocs/阶段规划.mddocs/变更日志.md 属于结构治理层
  • 旧项目若仍只有 docs/大纲.md,默认把它视作 项目总纲 + 章节规划 的兼容入口

四层记忆

  • L1 会话工作记忆:本章目标、场景推进链、开头钩子、上章悬念
  • L2 项目运行记忆:task_log.mddocs/章节规划.mdplot/伏笔记录.mdplot/时间线.md
  • L3 结构治理记忆:docs/项目总纲.mddocs/卷纲.mddocs/阶段规划.mddocs/变更日志.md
  • L4 宪法记忆:docs/全书宪法.mddocs/世界观.mddocs/法则.mdcharacters/*.md

消歧顺序:L4 > L3 > L2 > L1

正文写法原则

这是写正文时应该主动对标的标准,不是事后检查清单。

章节层

  • 目标单一:每章核心事件不超过 2 条,多了就切章,别塞
  • 章尾造压:用未解决的张力收尾——不是总结,是读者不读下一章会难受
  • 每章必须有一次实感回报:翻盘、揭秘、关系进展、能力升级任选其一;没有回报,读者掉追读
  • 受挫必须给补偿基础:主角挫败后,同章或紧邻章节给出反击或希望的锚点,不能连续几章只有憋屈

场景层

  • 每个场景从状态 A 走到状态 B,开头和结尾的人物处境或信息量不能一样
  • 高能前放慢:爆发场景前插 1-2 个低强度段给读者换气,高能后给收尾拍,别急着跳下一幕
  • 对白的作用只有两种:推进冲突、暴露性格。不用对白解释背景设定,那是说明书
  • 视角纪律:第三人称紧贴主 POV 的认知边界,不提主角不知道的信息,不在同段乱切脑内

语言层

  • 规避 AI 高频套语:以 X 为 Y、在这一刻、心中涌起、岁月/征程、结尾感悟升华、"他知道"+"他明白"连用
  • 心理描写不超标:思路链路是"想法 → 行动 → 结果",不用旁白总结角色感受,让读者自己感受
  • 具体优于抽象:写"刀口从肩膀切下去,皮肤往两侧翻"而不是"他受了重伤";写"她没说话,拿起外套走了"而不是"她很失望"
  • 情绪不要直接命名:不写"他感到愤怒",写愤怒的行为和身体反应

常用命令入口

高频场景直接用这四条,其余原子命令查 python3 scripts/chapter_pipeline.py commands

# 续写下一章(自动串联 preflight→resume→plan→compose→start)
python3 scripts/chapter_pipeline.py next-chapter <项目目录> --chapter-title "标题"

# 章节质检(静态预审 + 审稿稿本)
python3 scripts/chapter_pipeline.py review <章节文件路径> --project-path <项目目录>

# 长篇治理初始化 / 状态同步
python3 scripts/chapter_pipeline.py bootstrap-longform <项目目录>
python3 scripts/chapter_pipeline.py governance <项目目录> --current-volume "第一卷" --current-phase "阶段1" --phase-goal "目标"

# 商业化包装
python3 scripts/chapter_pipeline.py marketing <项目目录> --platform 起点中文网 --audience 男频读者

参考资料地图

立项 / 规划:

  • references/main-plot-construction.md
  • references/outline-template.md
  • references/chapter-plan-template.md
  • references/outline-refinement.md
  • references/longform-governance.md
  • references/character-template.md
  • references/character-building.md
  • references/plot-structures.md
  • references/conflict-design.md
  • references/worldbuilding-logic.md
  • references/worldbuilding-presentation.md
  • references/golden-finger-design.md
  • references/idea-incubation.md

章节创作 / 续写:

  • references/chapter-workflow.md
  • references/micro-revision-ops.md
  • references/chapter-guide.md
  • references/chapter-template.md
  • references/flow-break-writing.md
  • references/hook-techniques.md
  • references/dialogue-writing.md
  • references/ensemble-writing.md
  • references/content-expansion.md
  • references/daily-narrative.md
  • references/nonlinear-narrative.md
  • references/reader-compensation.md
  • references/literary-opening.md
  • references/suspense-design.md

写后检查 / 风格护栏:

  • references/quality-checklist.md
  • references/consistency.md
  • references/style-guardrails.md
  • references/rule-linting.md
  • references/writing-quickref.md(由 plan / compose 命令自动注入 intent 文件,无需手动读取)

工具吸收 / 方法沉淀:

  • references/tooling-adoption-roadmap.md
  • references/marketing.md

题材补充

平台向默认规则:

  • 书名先承担题材识别和核心钩子
  • 前 3 章默认按“前三秒”标准自查
  • 不靠擦边、影射、审核侥幸维持吸引力

女频 / 言情 / 大女主:

  • 把主体性、补偿机制、叙事立场视为高敏感检查项
  • 出现“男主膈应”“女主太憋屈”“人设塌了”时,优先读 references/style-guardrails.md

外部材料接入

如果用户提供方法论、访谈、经验材料:

  1. 先判断它影响的是立项、长篇治理、章节推进还是风格
  2. 长篇治理相关优先并入 references/longform-governance.md
  3. 大纲修整相关优先并入 references/outline-refinement.md
  4. 局部返修相关优先并入 references/micro-revision-ops.md
  5. 工具与工作台相关优先并入 references/tooling-adoption-roadmap.md
  6. 只吸收可操作、可复用、可验证的内容
  7. 只适用于单项目的经验,写进项目文档,不污染通用规则

Comments

Loading comments...