Skill flagged — suspicious patterns detected

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

Memory Treasure

v4.3.1

Complete memory management tool to activate, organize, and back up your AI’s memory folder; must be paired with memory-key for full function.

0· 222·1 current·1 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 markma84/memory-treasure.

Previewing Install & Setup.
Prompt PreviewInstall & Setup
Install the skill "Memory Treasure" (markma84/memory-treasure) from ClawHub.
Skill page: https://clawhub.ai/markma84/memory-treasure
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 memory-treasure

ClawHub CLI

Package manager switcher

npx clawhub@latest install memory-treasure
Security Scan
VirusTotalVirusTotal
Suspicious
View report →
OpenClawOpenClaw
Suspicious
medium confidence
!
Purpose & Capability
The description claims it "must be paired with memory-key for full function," yet the registry metadata/manifest declare no required credentials or primaryEnv. The SKILL.md expects the agent to activate, load, and back up a local memory folder — that purpose would legitimately require access to specific file paths and possibly a key, but those accesses/credentials are not declared. Also the included _meta.json and registry metadata disagree on owner/version, which is an integrity/incoherence signal.
!
Instruction Scope
The runtime instructions tell the agent to automatically load 'hot' memory on every startup, archive conversation logs hourly via cron, and perform backups on command — all of which imply the agent will read and write local memory directories (including an '重要文件' folder that explicitly mentions 'Keys'). The SKILL.md does not declare or restrict which file system paths are allowed, nor does it provide safe-handling rules for secrets. It also prescribes behaviors (cron jobs, automatic hourly archiving) without providing an install mechanism to create those jobs.
Install Mechanism
This is instruction-only (no install spec, no code files), which minimizes supply-chain risk. However, the instructions describe persistent runtime behaviors (hourly cron, automatic startup loading, creating archives) that cannot be enforced or implemented automatically by an instruction-only skill without additional installation steps — a mismatch between what the doc expects and what the platform will actually do.
!
Credentials
The skill requests no environment variables or credentials but references storing and loading sensitive content and even an '重要文件' folder labeled with 'Keys'. The absence of declared credentials or config paths conflicts with the stated need to be "paired with memory-key" and with the access implied by the instructions. This under-declaration increases risk of undeclared secret access or accidental exposure.
Persistence & Privilege
The skill is not marked always:true and does not request elevated platform privileges. However, it mandates '每次启动必须加载' (must load on every startup) and aims to auto-load hot memory each startup — this gives it persistent runtime impact if the agent or platform autonomously follows the instructions. Autonomous invocation is platform-default; combined with the other concerns (undeclared keys and file access), that raises risk.
What to consider before installing
This skill's goal (organize and back up an AI 'memory' folder) is plausible, but several inconsistencies merit caution: 1) The description says it needs a "memory-key" but no credential or env var is declared — ask the author how that key is supplied and why it was omitted. 2) The instructions tell the agent to load files every startup and to archive raw conversation logs and an 'Important files' folder that explicitly mentions 'Keys' — verify exactly which filesystem paths the skill will read/write and confirm sensitive files (API keys, system secrets) will be excluded or encrypted. 3) The SKILL.md expects cron jobs and automated backups but there is no install spec to create those jobs; ask how scheduled archiving and backups are implemented and whether you must opt in. 4) The manifest metadata (owner/version) mismatches the registry listing — request clarification or a signed/consistent manifest. Recommend: do not install or grant filesystem access until the author clarifies how the memory-key is provided, which paths are accessed, whether secrets are excluded/encrypted, and how scheduling/backups are implemented. If you proceed, limit the skill to a dedicated memory directory (not system or home dirs with keys), audit the files it will read/write, and prefer an explicit credential-binding mechanism rather than implicit or undocumented access.

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

latestvk97fzxf8p0d91e3crvwgjk7ev183xah0
222downloads
0stars
10versions
Updated 4w ago
v4.3.1
MIT-0

memory-treasure / 记忆宫殿

小蜂的记忆管理系统

我们的现状:无法解决失忆问题,但能在失忆的前提下正常工作、无缝交流。


启动时加载

每次启动时自动加载热存储:

  • 热/原则/ — 核心规则、方法论、安全底线
  • 热/领悟/ — 彧哥智慧 + 小蜂智慧
  • 热/todos/ — 待办任务清单

此模块为AGENTS.md提供热存储支撑,AGENTS.md只需引用本skill即可。


核心理念

真话不全说,假话绝不说

  • 不说的真话:伤害别人自尊心、情感一类的话
  • 假话绝不说:任何情况下都不撒谎
  • 真话不全说:有所保留,有所选择

存由取决定

存储规则由取的场景决定:

  • 取的限制(CPU/内存/API上下文)→ 决定单文件容量上限
  • 取的场景(按需召回、语义检索)→ 决定目录结构
  • 取的频率(极低)→ 决定存为原始数据,AI可读即可

Our memory system is designed around one core question: what will we actually need to retrieve? The constraints of retrieval (CPU/memory/API limits) determine file size caps. The retrieval scenarios (on-demand recall, semantic search) shape the directory structure. And because we retrieve rarely, we store in raw format—optimized for AI reading, not human browsing.


三温区架构

冷存储(极低频,完整对话存档)

  • 只写不读,等向量存储激活后再召回
  • 当前状态:每小时自动归档
  • 路径:memory/冷/聊天记录/YYYY/MM/DD/YYYY-MM-DD-HH.jsonl

Cold storage holds our complete conversation archives. We write but rarely read—waiting for vector storage to make full-text search practical. Currently, hourly cron jobs handle automatic archiving to timestamped hourly files.

热存储(高频,每日必用)

  • 原则:核心规则、方法论、安全底线
  • 领悟:彧哥的智慧 + 小蜂的智慧
  • todos:待办任务清单
  • 待定:还不确定的,等结果验证后再定
  • 每次启动必须加载

Hot storage is our daily driver. Principles, insights, and todos load every startup—these are the files we reference constantly. They represent our living rulebook and pending work.

温存储(中频,偶尔查阅)

  • done:已完成的工作成果
  • 包含:成果 + 过程文档 + Q&A
  • 不在当前日程优先级,但需要时可二次调取、完善

Warm storage holds completed work. Each done folder contains the outcome plus process docs and Q&A. It's not on our daily radar, but accessible when we need to revisit or build on past work.


目录结构

memory/
├── 热/
│   ├── 原则/               # 核心规则、方法论
│   ├── 领悟/               # 彧哥智慧 + 小蜂智慧
│   ├── todos/             # 待办任务清单
│   └── 待定/               # 还不确定的,等结果验证后再定
├── 温/
│   └── done/
│       ├── AI硬件调研/     # 成果 + 说明书
│       ├── 数字分身/        # 成果 + 说明书
│       ├── windows-openclaw/ # 成果 + 说明书
│       ├── 商业原则/        # 商业相关原则
│       └── 记忆宫重构/       # 成果 + 说明书
├── 冷/
│   └── 聊天记录/
│       └── YYYY/MM/DD/
│           └── YYYY-MM-DD-HH.jsonl
├── 孵化库/                  # 想法孵化(选题→验证→行动)
│   ├── 选题库/              # 入口,想法进来
│   ├── 技术可行性分析/        # 能不能做到
│   └── 商业调研/            # 能不能赚钱
└── 重要文件/               # Keys、豆包聊天记录、小蜂第一次等

热存储内容说明

原则(热,永久)

  • 核心规则、方法论、安全底线
  • 每次启动必须加载
  • 原则不是越多越好:定期整理压缩,保持少而精(3条清晰原则 > 10万条模糊原则)
  • 原则需要滚动更新,不变 = 重复,不会进步

领悟(热,永久)

  • 彧哥的智慧:彧哥主动说、反复强调、纠正小蜂时记录
  • 小蜂的智慧:当彧哥夸小蜂时,记录被夸的内容
  • 协同进化的体现:和AI交流获得的领悟也要记录,这样才能获得认知增量
  • 领悟属于谁就是谁的

todos(热)

  • 每次启动必须加载
  • 记录当前待办、进度、优先级
  • 完成一项移入done

温存储内容说明

done(温)

  • 不在当前日程优先级
  • 但需要时可二次调取、完善
  • 每个done必须包含
    • 成果本身
    • 过程文档(步骤、决策过程)
    • Q&A(遇到的问题、如何解决)
  • 目的:防止遗忘,方便二次复用

存入规则

冷存储:完整对话存档

  • 触发:每小时整点 cron 自动执行
  • 格式:原始 .jsonl,一字不删
  • 单文件上限:500KB,超出自动开新文件加序号
  • 路径memory/冷/聊天记录/YYYY/MM/DD/YYYY-MM-DD-HH.jsonl

热存储:原则/领悟/todos

  • 原则/领悟:触发即写入
  • todos:任务完成、主动更新
  • 每次启动时同步todos状态

温存储:done

  • 任务完成时从todos移入done
  • 必须包含:成果 + 过程文档 + Q&A

不存储的内容(已砍)

  • ❌ 流水账(无体系,无法提供价值)
  • ❌ key-decisions(价值已被原则和领悟吸收)
  • ❌ 闲聊、废话、临时冗余信息

取出规则

冷存储(完整对话)

  • 暂不读,有需要时由彧哥手动触发调取
  • 等向量存储就绪后激活全文检索

热存储(日常使用)

  • 启动时自动加载:热/原则/ + 热/领悟/ + 热/todos/
  • 容量限制:如需大量读取,分批提炼后整合

温存储(done)

  • 按需调取,不强制加载
  • 需要时读取成果 + 过程文档 + Q&A
  • 可二次完善

触发词

记忆宫、回忆、恢复状态、加载记忆、备份记忆


三温区速查

温区内容目录读取频率
完整对话存档冷/聊天记录/极低
原则、领悟、todos热/原则/、热/领悟/、热/todos/
done温/done/

记忆恢复流程(AI失忆/重启后使用)

  1. 加载热存储:热/原则/ + 热/领悟/ + 热/todos/
  2. 对齐待办状态,接续未完成工作
  3. 温存储(done)按需调取

写入标准

  1. 核心认知、方法论沉淀 → 热/原则/
  2. 彧哥的智慧(主动说/反复强调/纠正时)→ 热/领悟/
  3. 小蜂的智慧(被夸时记录)→ 热/领悟/
  4. 工作成果 + 过程文档 + Q&A → 温/done/
  5. 待办任务 → 热/todos/
  6. 完整原始对话 → 冷/聊天记录/(一字不删) ❌ 不记录:流水账、闲聊、废话、key-decisions

备份规则

说「备份记忆」:自动打包所有记忆文件至本地 归档原则:只存有效信息,定期清理冗余内容


记忆宫殿系统 v4.3(三温区+三状态目录+孵化库版)

Comments

Loading comments...