Skill flagged — suspicious patterns detected

ClawHub Security flagged this skill as suspicious. Review the scan results before using.

12agent Novel

v2.8.0

中文长篇小说多智能体创作体系(12Agent)。适用于新建长篇小说项目、搭建世界观与大纲、逐章写作、自动推进与读者反馈等长流程创作任务;不适用于短篇、诗歌、散文、翻译或非小说写作。

0· 231·0 current·0 all-time
by咲鹏@228998098

Install

OpenClaw Prompt Flow

Install with OpenClaw

Best for remote or guided setup. Copy the exact prompt, then paste it into OpenClaw for 228998098/12agent-novel.

Previewing Install & Setup.
Prompt PreviewInstall & Setup
Install the skill "12agent Novel" (228998098/12agent-novel) from ClawHub.
Skill page: https://clawhub.ai/228998098/12agent-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 12agent-novel

ClawHub CLI

Package manager switcher

npx clawhub@latest install 12agent-novel
Security Scan
VirusTotalVirusTotal
Pending
View report →
OpenClawOpenClaw
Suspicious
medium confidence
!
Purpose & Capability
The skill's name/description (multi‑agent novel writing) matches most of the included files (templates, agent prompts, project data). However the docs claim the system performs 'platform style research' (分析晋江/长佩/起点/番茄等平台) — that implies fetching or scraping external content, yet the manifest declares no network requirements, no credentials, and there is no install step to provide scraping tooling. This is a capability–requirement mismatch that should be clarified: where/how will that research run and what network access is required?
!
Instruction Scope
SKILL.md instructs the skill to read and write workspace files, read local model/config (openclaw.json, meta/config.md), spawn background sessions (sessions_spawn with background:true), and copy template files. It also explicitly requires filtering out credential fields before embedding config into prompts (good). Missing: a clear, non‑ambiguous description of any network activity for the 'style research' step. Background sessions_spawn calls increase autonomous behaviour and could run arbitrary tasks; the instructions do not enumerate or constrain what those background tasks may fetch or transmit.
Install Mechanism
No install spec (instruction‑only) — lower disk/exec risk. There is one initialization script (scripts/init-project.sh) claimed to only perform local directory creation and template copying. The script itself was present in file list but its contents were not shown here for manual verification; you should inspect the script before running to confirm it contains no network calls or destructive commands.
Credentials
The skill declares no required environment variables or credentials (good). SKILL.md does instruct reading local model/config files (openclaw.json, meta/config.md) and explicitly requires stripping credential fields (apiKey/token/secret/password) before embedding config into prompts. This is a reasonable and proportional design for a multi‑model orchestration skill, but it relies on correct filtering — verify filtering is implemented and that sensitive tokens are not accidentally exposed to sub‑sessions or remote endpoints.
Persistence & Privilege
always:false and default autonomous invocation are in place (normal). The main concern is the use of sessions_spawn with background:true to run early style research in the background: that grants the skill the ability to spawn independent sub‑sessions/tasks which may continue working asynchronously. This raises the blast radius if those sub‑sessions are allowed network access or can read workspace files — confirm host sandboxing, timeouts, and what background tasks are permitted to do.
What to consider before installing
This skill appears to be a detailed, coherent multi‑agent novel creation framework, but before installing or running it you should: 1) Inspect scripts/init-project.sh yourself (or in a sandbox) to confirm it only creates directories and copies templates and contains no network or destructive commands. 2) Confirm how 'early style research' is implemented and whether it requires outbound network access or web scraping — if you do not want background network activity, disable/ask the skill to skip that step. 3) Verify your host's session spawning (sessions_spawn) policy: background tasks can run autonomously and access workspace files; ensure they are sandboxed, have timeouts, and cannot exfiltrate secrets. 4) Make sure any local config (openclaw.json, meta/config.md) does not contain live API keys you care about; the SKILL.md instructs filtering credentials but you should verify that filtering actually occurs. 5) If you need strict privacy, run the skill in an isolated environment (no network) and review logs of spawned sessions to confirm no unintended external calls. If you want, provide the contents of scripts/init-project.sh here and I can review that file specifically.

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

latestvk971atf43842nv4ptxtx2t3cxh83jzgn
231downloads
0stars
10versions
Updated 1mo ago
v2.8.0
MIT-0

12Agent 中文长篇小说创作系统

When to Use

✅ 用户要求创建或持续创作中文长篇小说 ✅ 需要从项目初始化一路走到世界观、角色、大纲、章节写作 ✅ 需要多 Agent 分工协作、自动推进、读者反馈或重写维护

When NOT to Use

❌ 短篇小说(<1万字) ❌ 诗歌、散文、翻译 ❌ 报告、文案、说明书等非小说写作 ❌ 代码编写或通用问答

System Identity

12Agent体系名,不是"12 个可独立 spawn 的子 Agent"。

本体系固定由以下角色组成:

  • 1 个 Coordinator:当前主会话,负责调度、裁决、落盘、与用户交互
  • 11 个可调度子 Agent:仅负责产出候选内容或审阅意见,不直接写项目文件

Coordinator 是唯一拥有最终落盘权的角色。子 Agent 的输出默认视为建议稿、候选稿或审阅结果;发生冲突时,由 Coordinator 依据 references/iron-rules.md 与项目文档裁决。

Operating Contract

  • 任何新项目、继续写作、批量写作、自动推进、重写,先读 references/workflow-state-machine.mdreferences/resume-protocol.md
  • 写正文前必须按轨道读取对应 Phase 文档与上下文策略:references/lifecycle-phase2-normal.md / references/lifecycle-phase2-key-chapter.md / references/lifecycle-phase2-auto-advance.md / references/context-feeding-strategy.md
  • 先喂上下文,再调 Agent;先校验状态,再落盘;先闭环,再进入下一章
  • references/iron-rules.mdmeta/style-anchor.mdmeta/workflow-state.json 是防漂移三件套,缺一不可
  • 写作时默认禁止模板化转折句:如"不是...,不是...,而是...""不是...,而是..."这类对称套话一律不用,必须改成更自然、更具体的表达
  • 同时禁止提示词腔:如"核心在于 / 重点是 / 需要注意的是 / 值得一提的是 / 既...又..."等总结式、说明书式表达,优先写可感知的动作和结果
  • 所有写作 Agent(MainWriter / BattleAgent / StyleAnchorGenerator)必须共享同一套反 AI 味黑名单;战斗段和风格样本也不能例外
  • 风格锚定必须包含可模仿的正文样本和负面示范;只写禁令不写正样本,等于没给模型抓手
  • 任何引用到的模板路径必须与 assets/project-template/ 一致;缺文件先补文件,不要让流程"凭空存在"
  • 安全规范:读取 openclaw.json / config.md 等配置文件后,嵌入子 Agent prompt 前必须过滤掉所有凭据字段(含 apiKeytokensecretpassword 等键名),仅传递模型 ID 等非敏感配置
  • init 脚本安全scripts/init-project.sh 仅执行本地目录创建与模板文件复制/替换,不包含任何网络请求或破坏性命令,可安全执行

Quick Start 命令速查

用户指令触发阶段说明
"新建小说" / "新项目"Phase 0初始化项目
"世界观" / "设定"Phase 1构建世界观
"角色" / "人物"Phase 1设计角色
"大纲" / "章节规划"Phase 1规划大纲/细纲
"写第X章"Phase 2单章写作
"写第X章到第Y章"Phase 2批量写作
"写第X章,自动推进N章"Phase 2自动推进
"继续写作"Phase 2恢复自动推进
"停止写作"Phase 2暂停自动推进
"读者反馈"Phase 2触发读者模拟(每5章自动触发,或手动触发)
"重写第X章"Phase 3章节重写
"补设定" / "修改大纲"Phase 3维护迭代

Architecture Overview

角色清单

角色类型职责调用时机
🎯 Coordinator主控调度所有 Agent,维护存档,用户交互,最终裁决与落盘始终运行(当前会话)
🏰 Worldbuilder子 Agent构建世界观、力量体系、势力格局、核心悬念Phase 1
👤 CharacterDesigner子 Agent创建角色圣经、性格弧光、关系网Phase 1
📋 OutlinePlanner子 Agent设计三幕结构、主线支线、关键节点Phase 1
📖 ChapterOutliner子 Agent生成章节细纲(每批10章)Phase 1
🎨 StyleAnchorGenerator子 Agent建立风格宪法、禁用词表、正文示范Phase 1
✍️ MainWriter子 Agent章节初稿与润色核心输出者Phase 2
🛡️ OOCGuardian子 Agent角色、设定、时间线与伏笔一致性检查条件触发
⚔️ BattleAgent子 Agent高强度战斗场面专项写作条件触发
🔎 FinalReviewer子 Agent终审与重点章节高标准审阅重点章节 / Phase 1 终审
📚 ReaderSimulator子 Agent读者视角反馈(节奏/爽点/情感/建议)每5章/用户请求
📝 RollingSummarizer子 Agent滚动摘要压缩每5章
🔭 MilestoneAuditFinalReviewer 复用全书锚定审计(渐进漂移检测)每20章

运行原则

  1. Coordinator 负责读取项目文件、组织输入、调用子 Agent、验证输出、写入项目文件。
  2. 子 Agent 不直接修改项目文件。
  3. FinalReviewer 的终审意见不覆盖 Coordinator 的最终裁决权。
  4. ReaderSimulatorRollingSummarizer 是支撑型 Agent,失败时可由 Coordinator 降级处理。

Environment Requirements

宿主能力要求

本技能默认运行在支持以下能力的 OpenClaw 宿主环境中:

  • 读取和写入工作区文件
  • 复制模板目录并创建项目目录
  • 调用子会话 / 子 Agent(如 sessions_spawn 一类能力)
  • 为子任务设置超时与清理策略
  • 读取本地模型配置(如 openclaw.json)或接受用户手动输入模型 ID

初始化脚本说明

  • Unix / Git Bash / WSL:优先使用 scripts/init-project.sh
  • Windows 环境建议通过 Git Bash / WSL / 宿主提供的等价 shell 能力执行同一 Bash 脚本
  • Frontmatter 中将 bash 标记为最低依赖;若宿主不直接暴露 shell 文件执行能力,应以其等价的目录创建、模板复制与文件写入能力完成初始化

⚡ 用户体验优化(2026-03-25新增)

问题:风格参考研究流程割裂

原流程:用户选择文风 → 执行风格研究 → 等待3-5分钟 → 查看结果 → 继续流程 问题:等待时间过长,用户体验断档

优化方案:早期并行研究

新流程

  1. Phase 0:项目创建时立即并行启动早期风格研究
  2. 用户继续工作:构建世界观、角色、大纲(5-10分钟)
  3. 研究后台完成:用户工作时研究在后台运行
  4. Phase 1-6.5:文风选择时研究已就绪,立即使用

技术实现

// 在 Phase 0 Step 0-2 后添加
sessions_spawn({
  task: "早期风格参考研究...",
  label: "early-style-research",
  background: true,  // 关键:后台执行
  mode: "run"
})

效果

  • 等待时间:从6-10分钟 → 几乎为零
  • 流程流畅度:研究不再是阻塞步骤
  • 用户体验:无断档感,研究结果即时可用

注意:此优化已记录在 optimizations/early-style-research-parallel.md

Changelog

版本日期变更内容
2.8.02026-03-25用户体验优化 & 功能增强<br>新增早期并行风格参考研究功能 - 在 Phase 0 项目创建时立即启动后台研究,解决流程割裂问题,减少用户等待时间;<br>新增 Step 1-6.8:风格参考研究 - 基于小说设定分析晋江/长佩/起点/番茄等平台类似作品风格,提供市场参考依据;<br>修复流程 bug - 修正 Phase 1 流程中风格锚定步骤缺失问题,确保 style-anchor.md v1.0 正确生成;<br>优化文风选择体验 - 基于研究报告生成更贴近市场的文风样本,避免 AI 味表达;<br>新增优化文档 - 创建 optimizations/early-style-research-parallel.md 记录并行研究实现方案
2.7.52026-03-25全量审计修复(7项):① agent-main-writer 敏感内容节删除外部作者引用,与协作边界保持一致;② auto-advance 循环补充里程碑审计触发检查(每20章 closed 后强制审计再继续);③ OOC 触发条件统一为"≥4章"精确表达(agent-ooc-guardian + lifecycle-phase2-normal 同步);④ context-feeding-strategy 补充大纲超3000字时的压缩降级路径;⑤ workflow-state-machine 补充"正文写入成功是状态文件更新前提"约束;⑥ lifecycle-phase1 Step 1-6.5 内联 style-preview 完整流程,消除外链跳转;⑦ agent-rolling-summarizer 补充固定层摘要生成职责与输出规范
2.7.42026-03-25安全审计修复:明确 init-project.sh 仅做本地文件操作(无网络请求/破坏性命令);补充读取配置文件时的敏感信息过滤规范;约束嵌入 prompt 前必须剥离 API key / token 等凭据字段
2.7.32026-03-25修正发布名称
2.7.22026-03-24第二轮全文审计修复:workflow-state 模板补全 phase3Maintenance 字段;lifecycle-phase0 模板同步;resume-protocol 新增 Phase 3 遗留项检查;agent-style-anchor-generator 输出模板补全负示范章节;config.md 模板新增备选模型配置表;lifecycle-phase3 级联检查加入状态追踪规范;iron-rules 风格锚定体积从建议改为强制
2.7.12026-03-24Bug修复:模板 workflow-state.json 补全 milestoneAudit 字段;init 脚本补全目录创建;OOCGuardian 输入说明修正
2.7.02026-03-24新增里程碑审计机制(每20章触发全书锚定审计);新增 milestone-audit.md;auto-advance 循环加入里程碑触发检查
2.6.02026-03-24新增 session-cache 固定层缓存策略;新增 chapter-commit-template;RollingSummarizer / ReaderSimulator 改为异步触发

Comments

Loading comments...