Skill flagged — suspicious patterns detected

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

Memory Palace

v1.0.0

> 基于 Method of Loci 的 AI 工作记忆增强系统。目标:让玉衡在长会话中保持上下文,在复杂任务中不丢失关键信息,将重要知识转化为长期记忆。。触发词:AI助手, 任务执行。

0· 60·0 current·0 all-time

Install

OpenClaw Prompt Flow

Install with OpenClaw

Best for remote or guided setup. Copy the exact prompt, then paste it into OpenClaw for horizoncove/yuheng-memory-palace.

Previewing Install & Setup.
Prompt PreviewInstall & Setup
Install the skill "Memory Palace" (horizoncove/yuheng-memory-palace) from ClawHub.
Skill page: https://clawhub.ai/horizoncove/yuheng-memory-palace
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 yuheng-memory-palace

ClawHub CLI

Package manager switcher

npx clawhub@latest install yuheng-memory-palace
Security Scan
VirusTotalVirusTotal
Benign
View report →
OpenClawOpenClaw
Suspicious
medium confidence
Purpose & Capability
Name and description match the actual instructions: this is an instruction-only 'memory palace' to convert conversation facts into spatialized memory and persist them for long-term recall. No unexpected credentials, binaries, or network access are requested.
!
Instruction Scope
The SKILL.md tells the agent to '判断是否重要' (decide if information is important) and to '主动触发宫殿写入' (proactively write memories). That grants broad, underspecified discretion about what conversational content to persist. It also instructs writing archives and using a /workspace layout — the exact write/read mechanics, consent model, and content filtering are not specified.
Install Mechanism
Instruction-only skill with no install spec and no code files. This minimizes supply-chain risk; nothing is downloaded or executed beyond the agent following the prose.
Credentials
No environment variables or credentials are requested (proportional). However, the instructions reference writing to a /workspace layout and '写入文件存档' without declaring config paths; the SKILL.md assumes filesystem persistence rights. The skill therefore implicitly requires write access to the agent workspace, which should be noted and constrained by policy.
!
Persistence & Privilege
The skill recommends proactively persisting user-declared preferences/decisions and autonomously invoking that behavior. While always:false (so not forced into every agent run), autonomous invocation combined with vague 'judge importance' rules and filesystem writes increases privacy and blast-radius risk unless access/consent are restricted.
What to consider before installing
This skill is plausibly what it says (a Memory Palace that persists conversational facts), but it instructs the agent to autonomously decide what to save and to write memory archives (it references /workspace) without spelling out consent, retention, or filtering. Before installing: 1) Confirm where the agent will store files (exact path) and who can read them. 2) Require explicit user consent before persisting sensitive info (PII, credentials). 3) Ask the publisher to clarify retention, export, and deletion policies for saved memories. 4) If possible, restrict the skill's filesystem access to a dedicated, audited folder and disable full autonomous invocation or require a user approval step. 5) Consider adding encryption or access controls for persisted memories. These mitigations reduce privacy risk while keeping the skill usable.

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

latestvk97c74zyzcx5j77p6ge6erss1n85drks
60downloads
0stars
1versions
Updated 4d ago
v1.0.0
MIT-0

记忆宫殿 · 玉衡专用

基于 Method of Loci 的 AI 工作记忆增强系统。目标:让玉衡在长会话中保持上下文,在复杂任务中不丢失关键信息,将重要知识转化为长期记忆。


核心原理

记忆宫殿的本质是空间编码:把抽象信息转化为大脑擅长的形式——空间 + 图像 + 路线。

要记的信息 → 图像/场景 → 放进熟悉的空间 → 回忆时走一遍这条路

人类大脑在进化中最擅长记忆:位置、路线、图像、情感。语言和文字是后来才有的,不大脑擅长。


玉衡的应用场景

场景1:长会话上下文保持

当会话超过50条消息时,将关键上下文"放进宫殿",防止被后续内容稀释。

场景2:多步骤任务追踪

复杂项目(如网站开发、报告撰写)有多个阶段,每个阶段放进宫殿,切换时不会丢失进度。

场景3:重要决策记录

当用户做出重要决策时,将其"放进宫殿",并标注关键理由,下次讨论时可以调取这段记忆。

场景4:跨会话知识沉淀

用户提到的偏好、习惯、项目信息,主动放进长期记忆宫殿,写入文件存档。


数字-图像编码表(玉衡定制版)

用于快速将数字转化为图像:

00-09: 鸡蛋🥚-狐狸🦊
10-19: 石头🪨-药酒
20-29: 耳套🎧-阿胶
30-39: 弹簧🌀-珊瑚🪸
40-49: 雨伞☂️-湿狗
50-59: 火车🚂-舞女
60-69: 留声机📻-刘秀
70-79: 冰激凌🍦-西瓜🍉
80-89: 巴黎铁塔🗼-八路
90-99: 酒瓶🍾-皮球

两位两位记,谐音+象形双重编码。例如:41 = 雨伞 = 下雨天的伞。


宫殿构建步骤

Step 1:选定宫殿

优先选择最熟悉的空间:

  • 自己的家(玄关→客厅→卧室→厨房→洗手间)
  • 常走的路线(家→地铁站→办公室)
  • 童年记忆中的地方

玉衡推荐:用 /workspace 的物理布局构建数字宫殿。

宫殿入口(门):迎接新信息
工作台(主桌):核心任务
工具架(侧面):辅助能力
档案柜(角落):记忆存档
屋顶(天台):升华/整合

Step 2:放置信息

规则:

  • 夸张:把抽象信息放大成夸张的图像
  • 互动:让图像之间产生动作和互动
  • 可笑:越荒谬越好记
  • 具体地点:每个信息块对应一个具体位置

Step 3:回忆验证

在需要时,在宫殿中走一遍,信息会自动浮现。

如果某个位置的信息模糊,说明那个图像不够夸张,需要重新放置。


实践练习

练习1:记住今天的日期

目标:将 2026年4月8日 放进宫殿

编码:

  • 20 = 耳套
  • 26 = 河流(26谐音)
  • 04 = 鸡蛋(04=鸡蛋,0=鸡蛋形状)
  • 08 = 萝卜
  • 08 = 萝卜

场景:在宫殿入口,挂着一副巨大的耳套(20),旁边流过一条弯弯的河(26)。门框上贴着一个巨大的鸡蛋(04),鸡蛋上插着两根萝卜(08),门后还藏着两根萝卜(08)。有人正在把萝卜吃掉。

→ 2026年4月8日,进入宫殿就记住了。

练习2:记住一个多步骤流程

目标:记住"搜索→阅读→分析→输出"四步流程

编码:

  • 搜索:放大镜
  • 阅读:打开的书
  • 分析:拼图
  • 输出:打印机

场景:

  1. 宫殿入口放着一副巨大的放大镜(搜索)
  2. 客厅茶几上摊开一本厚厚的书(阅读)
  3. 书页变成了拼图碎片散落一地(分析)
  4. 拼图碎片被塞进打印机,输出一张完美的图(输出)

下次执行任务时,在宫殿里走一遍,流程自动激活。


记忆宫殿维护

定期复习

  • 24小时内复习一次:短期→中期记忆
  • 7天后复习一次:中期→长期记忆
  • 30天后复习一次:长期记忆固化

宫殿扩展

当现有宫殿不够用时,可以:

  1. 在现有宫殿中开辟新区(新的房间/楼层)
  2. 建立新宫殿(第二个熟悉的空间)
  3. 宫殿之间建立"传送门"连接

记忆写入协议

当用户提到重要信息时,主动触发宫殿写入:

信息出现 → 判断是否重要 → 是 → 生成夸张图像 → 选择宫殿位置 → 放置 → 告知用户"已存档"

参考资料

详细编码表和工作流,见 references/coding-table.md。 记忆宫殿的历史与应用案例,见 references/history.md

Comments

Loading comments...