📦 Wanxiang — 万象
v1.0.0Scroll Tools万象绘卷工具与安全分卷 – 包含实用工具、安全防御、拆书融合、小说分析
0· 16·0 当前·0 累计
by 下载技能包
最后更新
2026/4/19
安全扫描
OpenClaw
可疑
medium confidence该技能的内容符合写作/工具 + 安全指导目的，但它明确指示创建/修改代理工作区配置文件（AGENTS.md / SOUL.md / IDENTITY.md），并要求 AI “内化”规则——这是一个持久化/权限步骤，安装前需仔细审查。
评估建议
This skill is mostly writing tools plus a security chapter and can be useful for creative work and guidance on prompt-injection defenses. Before installing or following its 'deployment' steps, consider: 1) Do NOT blindly write the suggested AGENTS.md / SOUL.md / IDENTITY.md into your agent workspace — review their full text and understand consequences. 2) Backup ~/.openclaw/workspace/ (or equivalent) before making any changes. 3) Prefer applying rules manually or in a sandboxed test agent first;...详细分析 ▾
ℹ 用途与能力
The name/description (writing tools + security) align with the provided files (character tools, novel tools, advanced techniques, and an extensive security chapter). Asking users/agents to load security guidance and mode-isolation rules is reasonable for a 'security' subpackage. However, the skill goes beyond passive documentation: several files instruct writing persistent persona/config files into the agent workspace (~/.openclaw/workspace/), which is a stronger privilege than typical instruction-only writing tools.
⚠ 指令范围
SKILL.md plus the referenced docs explicitly direct editing/creating files under ~/.openclaw/workspace/ (AGENTS.md, SOUL.md, IDENTITY.md) and instruct the AI to load and 'internalize' those rules. The instructions reference system paths and a deployment procedure (nano commands). That grants the skill the capability to change agent behavior and persist constraints — scope creep relative to a pure authoring toolkit.
✓ 安装机制
This is an instruction-only skill with no install spec, no downloaded code, no required binaries, and no requested environment variables. From an installation-supply perspective there is low supply-chain risk.
ℹ 凭证需求
The skill requests no environment variables or external credentials, which is proportionate. However, it requests access to the agent workspace (via explicit file paths and edit instructions) to deploy 'highest-priority' persona files — effectively requesting write access to local agent configuration. That access is sensitive even though no secrets are requested.
⚠ 持久化与权限
The docs instruct creating permanent agent configuration files that are described as '最高优先级/不可覆盖' (highest priority, not overridable) and say AI will read them on startup. That gives the skill a pathway to persistently alter agent behavior across sessions and to influence future agent decisions. While the content is framed as defensive, persisting new personality/config files is a high-privilege operation and should be treated cautiously.
安全有层次，运行前请审查代码。
运行时依赖
无特殊依赖
版本
latestv1.0.02026/4/19
首次发布，提供实用工具、安全防御和新颖分析/模块。 - 新增角色与实用脚本，附带高级用法提示。 - 引入防御机制：提示注入防护、模式隔离、攻击检测及 AI 使用指南。 - 包含书籍拆解、多源创意写作及配套素材管理的小说分析系统。 - 专为配合主 wanxiang-scroll 技能使用而设计。
● 无害
安装命令
点击复制官方npx clawhub@latest install wanxiang-scroll-tools
镜像加速npx clawhub@latest install wanxiang-scroll-tools --registry https://cn.longxiaskill.com镜像同步中
技能文档
# 万象绘卷 - 工具与安全分卷 > 实用工具 + 安全防御 + 拆书分析 | v1.0.0 ## 包含内容 ### 第八章 - 工具脚本使用指南 - 角色工具 - 小说工具 - 高级技巧 ### 第九章 - 安全与防御机制 - 提示词注入防御 - 模式隔离机制 - 攻击检测 - AI使用指南 ### 第十一章 - 拆书融合创作 - 拆书方法论 - 多源融合创作 - 小说档案馆 ### 第十二章 - 小说档案馆 - 小说分析系统 - 素材管理 ## 使用方式 配合主技能包 wanxiang-scroll 使用： `` clawhub install wanxiang-scroll clawhub install wanxiang-scroll-tools `` --- 万象绘卷分卷系列 - 工具与安全