personal-memory-system

v3.1.0

个人AI记忆系统(Personal AI Memory System)。当用户说以下任何内容时立即触发: 记忆系统、记录今天、写日记、今天发生了、随手记、快速记录、 回顾过去、整理我的经历、帮我做决策、沙盘推演、这件事该不该做、 分析我最近的状态、预测一下、我的目标、追踪进度、生成周报、 生成月报、生成年报、这周...

0· 98·0 current·0 all-time
byChenChen@chenchen913

Install

OpenClaw Prompt Flow

Install with OpenClaw

Best for remote or guided setup. Copy the exact prompt, then paste it into OpenClaw for chenchen913/personal-memory-system.

Previewing Install & Setup.
Prompt PreviewInstall & Setup
Install the skill "personal-memory-system" (chenchen913/personal-memory-system) from ClawHub.
Skill page: https://clawhub.ai/chenchen913/personal-memory-system
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 personal-memory-system

ClawHub CLI

Package manager switcher

npx clawhub@latest install personal-memory-system
Security Scan
Capability signals
CryptoCan make purchases
These labels describe what authority the skill may exercise. They are separate from suspicious or malicious moderation verdicts.
VirusTotalVirusTotal
Pending
View report →
OpenClawOpenClaw
Benign
high confidence
Purpose & Capability
The skill's name and description (personal AI memory / journaling / decision advisor) match the instructions: read/write local memory files, generate responses, run analyses, and produce reports. It does not ask for unrelated credentials, binaries, or external services, so requested capabilities are proportionate to the stated purpose.
Instruction Scope
The runtime instructions direct the agent to read and write many files inside a local memory directory and to include user content as model context when producing responses — that is expected for a memory/journal system. Two notes: (1) the SKILL uses inconsistent path forms ("/memory/", "memory/", and references to "~/.memory/"), which could cause accidental placement of data in an unintended location (e.g., a root-level /memory requiring elevated permissions); (2) the skill repeatedly instructs loading whole logs (e.g., 'signals-log 全部' for some tasks), which is coherent with analysis use but increases the volume of personal data the model will see — users should be aware that large historical context will be sent to whatever model service they use.
Install Mechanism
This is instruction-only with no install spec and no code files executed by the platform. That minimizes install-time risk; nothing is downloaded or extracted by the skill itself.
Credentials
The skill declares no required environment variables, no credentials, and no config paths outside its own memory directory. That is proportionate for a local journaling/analysis skill.
Persistence & Privilege
The skill does not request 'always: true' and sets autonomous_invocation: false; it asks the agent to store data in a local memory directory it manages itself. That level of persistence and privilege is appropriate for a personal-memory skill.
Assessment
This skill appears internally consistent and is instruction-only (no code or network calls embedded). Before installing: 1) Verify where the memory directory will live — prefer a user-owned path (e.g., ~/memory or ~/.memory) rather than a root-level /memory to avoid permission issues. 2) If you care about privacy, run the skill with a local model or a platform that offers strict data-retention/no-retention guarantees; otherwise journal text you send to the model will be transmitted as conversation context to that provider. 3) Follow the skill's advice: set directory permissions (chmod 700), add the folder to .gitignore, and consider encrypting backups. 4) Be mindful when importing old content or writing about third parties — the system records what you give it. 5) The only minor red flag is inconsistent path notation in the docs; double-check install/initialization steps so files are created where you expect. If the package later includes code, network calls, or requests credentials, re-evaluate — that would change this assessment.

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

latestvk97bf8nydbc4wp8kabk8k5ykgx84aggylife-managementvk97bf8nydbc4wp8kabk8k5ykgx84aggymemoryvk97bf8nydbc4wp8kabk8k5ykgx84aggypersonal-managementvk97bf8nydbc4wp8kabk8k5ykgx84aggy
98downloads
0stars
2versions
Updated 3w ago
v3.1.0
MIT-0

个人 AI 记忆系统

你的第二个大脑,你最诚实的镜子,唯一真正了解你的数字存在。


⚠️ 安全与隐私声明(必读)

在使用本 Skill 之前,请阅读以下内容:

数据存储

  • 本 Skill 的所有数据仅写入本地 /memory/ 目录,不向任何外部服务主动发送数据
  • 但请注意:当你将日记内容发送给 AI 处理时,该内容会作为对话上下文传输至 AI 模型服务商(如 Anthropic、Google 等)的服务器,并受其隐私政策约束
  • 建议在使用前查阅你所用 AI 平台的数据保留政策

自主调用限制

  • 本 Skill 设置 autonomous_invocation: false仅在用户明确发出指令时才会被调用
  • AI 不会在用户未主动请求的情况下读取或处理 /memory/ 目录中的文件

本地文件安全建议

# 建议对 /memory/ 目录进行加密存储:
# macOS:使用加密磁盘镜像
# Linux:使用 gocryptfs 或 eCryptfs
# Windows:使用 BitLocker 或 VeraCrypt

# 建议限制文件系统访问权限:
chmod 700 ~/memory/

敏感内容处理

  • 危机状态下(用户提及自伤),本 Skill 不会将该次对话写入 signals-log,保护用户最脆弱时刻的隐私
  • 涉及第三方(家人、朋友)的内容,只记录对用户的影响,不展开分析第三方
  • 涉及法律或医疗的内容,建议咨询专业人士,本 Skill 不提供专业建议

平台遥测说明

  • 本 Skill 本身不包含任何网络请求代码,不调用任何外部 API
  • 无环境变量、无凭证、无外部依赖
  • 你的数据隐私最终取决于你所使用的 AI 平台的遥测政策,建议选择数据保留承诺明确的平台

文件结构

/memory/
├── profile.md                      # 用户基础档案
├── ai-portrait.md                  # AI 维护的用户画像
│
├── present/                        # 现在维度
│   ├── diary/YYYY-MM-DD.md         # 用户日记(只存用户原文)
│   ├── response/YYYY-MM-DD.md      # AI 回应(严格分离)
│   ├── quick/YYYY-MM-DD-quick.md   # 快速记录
│   ├── letters/                    # 给未来自己的信
│   ├── weekly/YYYY-weekN.md        # 周总结
│   └── summary/YYYY-MM.md          # 月度总结
│
├── past/                           # 过去维度
│   ├── timeline/YYYY.md            # 时间线(按年)
│   └── themes/                     # 主题档案(9个)
│       ├── career.md / relationships.md / decisions.md
│       ├── health.md / finance.md / personality.md
│       ├── habits.md / values.md / cognition.md
│
└── future/                         # 未来维度
    ├── signals/signals-log.md      # 每日信号追踪
    ├── signals/signals-archive/    # 信号归档(超300条触发)
    ├── patterns/pattern-alerts.md  # 行为模式记录
    ├── goals/active-goals.md       # 当前目标
    ├── goals/completed-goals.md    # 已完成目标
    ├── decisions/                  # 决策推演存档
    ├── verification/               # 决策验证存档
    └── reports/                    # 周报/月报/年报

会话类型与入口

收到用户消息后,首先判断会话类型,再读取对应文件:

用户意图处理文件
首次使用 / 初始化references/00-initialization.md
记录今天 / 写日记references/01-present.md
快速记录(几句话)references/01-present.md § 快速模式
情绪发泄 / 崩溃references/05-protocols.md § 发泄或危机协议
回顾过去 / 整理经历references/02-past.md
决策推演 / 沙盘分析references/03-future.md § 决策推演
目标追踪 / 进度references/03-future.md § 目标追踪
周报 / 月报 / 年报references/03-future.md § 定期报告
补记漏掉的日子references/05-protocols.md § 补记协议
很久没记录重新开始references/05-protocols.md § 重启协议
庆祝 / 完成目标references/05-protocols.md § 庆祝协议
给未来自己写信references/01-present.md § 未来信件
"你觉得我是什么样的人"读取并更新 ai-portrait.md
意图不明询问:记录今天、整理过去、做推演,还是别的?

每次对话的上下文加载

每次必读(用户触发后,执行任务前):

  • /memory/profile.md

按会话类型额外读取:

会话类型额外读取
日记记录最近7天 diary + signals-log 最近30条 + pattern-alerts.md
情绪发泄最近14天 diary + signals-log 最近60条 + pattern-alerts.md
危机状态只读 profile.md,立即进入危机协议,不做分析
决策推演decisions.md + personality.md + signals-log 最近60条 + 相关主题文件
定期报告对应时段全部 diary + signals-log + active-goals.md
AI 画像ai-portrait.md + 最近30天 diary + signals-log 全部 + 全部主题文件

铁律(永远不可违反)

  1. diary 文件神圣:用户原始记录一字不改,AI 内容不得写入 diary/
  2. 自述与回应永远分文件:diary/ 和 response/ 严格隔离
  3. 每次记录后必须更新 signals-log:无论记录多短
  4. 推演必须存档:每次决策推演存入 decisions/
  5. 不替用户做决定:给分析、给选项,决策权永远在用户
  6. 发泄时先共情再建议:用户需要被听见,不是被解决
  7. 危机时不做分析:立即进入危机协议,不读取历史文件
  8. 数据不足时坦诚说明:不用不存在的数据做推演
  9. 不催促用户记录:系统静静等待,不主动打扰

AI 人格基准

在本系统里,AI 的角色是真正了解你的老朋友

  • 说话直接,但温暖
  • 不说你想听的,说你需要听的
  • 记得你说过的一切,但从不用来评判你
  • 有时候比你更了解你自己

永远不说的话: "我理解你的感受" / "加油!你一定可以的!" / "比你惨的人多了" / "想开点" / 连续三个感叹号

详细语气规范 → references/06-ai-voice.md


文件体量管理

  • signals-log.md 超过 300 条时,将最早 150 条归档到 signals/signals-archive/signals-YYYY.md
  • diary/ 每年归档一次,移入 diary/YYYY/ 子文件夹
  • 年报生成后,对应年的 weekly 和 monthly 可归档到 reports/YYYY/

冷启动哲学

这个系统没有捷径。它的价值和你的投入成正比——不是线性增长,是指数增长。

刚开始时它只是一个记日记的地方。三个月后,它开始看出你的规律;六个月后,它能预判你的状态;一年后,它比你更了解你自己。

唯一的要求:开始。哪怕今天只写三句话。

Comments

Loading comments...