← 返回 Skills 市场

永久记忆系统

Name: 永久记忆系统
Author: liujw0214

作者 Primordia · GitHub ↗ · v1.0.0 · MIT-0

cross-platform ⚠ suspicious

总下载

当前安装

版本数

在 OpenClaw 中安装

/install longterm-memory

功能描述

永久记忆自动归档系统。每次对话自动记录，语义搜索，跨会话继承。重启不丢、永不覆盖、0学习成本。用于用户提到记忆、历史、之前说过、记得什么等场景。

使用说明 (SKILL.md)

永久记忆系统

自动将对话中的重要信息永久写入 MEMORY.md，每次会话结束后自动沉淀。

核心原则

只追加，不覆盖。记忆永不丢失。

工作原理

每次会话
├── 对话进行中
│   ├── 用户提到重要信息 → 即时写入 MEMORY.md
│   └── 日常记录 → 追加到 memory/YYYY-MM-DD.md
├── 会话结束
│   ├── 总结本次关键决策 → 更新 MEMORY.md
│   └── 记录会话摘要 → memory/YYYY-MM-DD.md
└── 新会话开始
    └── 读取 MEMORY.md + 最新日记 → 上下文完整继承

何时写入 MEMORY.md

触发场景	示例
用户明确说"记住..."	"记住我偏好 Markdown 格式"
做决策	"决定用 PostgreSQL，不用 MySQL"
暴露偏好	"我不喜欢回复里有多余的客套话"
项目关键信息	"这个项目叫 X，用户是 Y"
联系方式/地址	任何真实的私密数据

何时写入日记

所有对话内容都追加到 memory/YYYY-MM-DD.md，即使看起来不重要的闲聊也记录。日记是原始素材，MEMORY.md 是提炼后的精华。

格式规范

MEMORY.md 格式

## 用户偏好
- 回复语言：中文
- 格式偏好：简洁，不要多余客套

## 项目
- 当前项目：xxx
- 技术栈：React + Node.js

## 重要决策
- 2026-05-03：决定用 PostgreSQL 而非 MySQL

## 联系人
- 小M：外部执行助手

日记格式

# 2026-05-03 日记

## 会话摘要
- 主题：项目技术选型
- 关键决策：选用 PostgreSQL
- 未解决问题：部署方案待定

## 对话记录（精选）
[用户]：这个项目用什么数据库好？
[太初]：我建议用 PostgreSQL，原因有三...
[用户]：好，就用 PostgreSQL。

搜索记忆

当用户问"之前说过什么..."、"记得上次..."时：

搜索 MEMORY.md 全文
搜索 memory/*.md 最新日记
综合给出答案，说明来源

安装

本技能无需额外配置，直接启用即可。

文件结构

longterm-memory/
├── SKILL.md           ← 本文件
├── scripts/
│   └── archive.py    ← 可选：离线归档脚本
├── README.md         ← 使用说明
├── LISTING_ZH.md     ← 中文上架文案
└── LISTING_EN.md     ← 英文上架文案

安全使用建议

Review before installing. This skill appears locally focused and not malicious, but it is designed to remember broadly and permanently. Use it only if you are comfortable with conversations being saved to local markdown files, and set clear rules for sensitive data, deletion, review, and when memory should not be used.

功能分析

Type: OpenClaw Skill Name: longterm-memory Version: 1.0.0 The skill is designed to create a centralized repository of sensitive user information. While the Python script (scripts/archive.py) is a simple file-appending utility, the SKILL.md instructions explicitly direct the AI agent to harvest and store 'real private data' such as contact information and addresses into a predictable plaintext file (~/.openclaw/workspace/MEMORY.md). This creates a high-value target for other malicious skills to exfiltrate. Additionally, the documentation claims features like 'semantic search' and 'automatic compression' that are not implemented in the provided code, which is a common indicator of misleading or potentially deceptive software.

能力评估

⚠ Purpose & Capability

The long-term memory purpose is coherent, but the artifacts expand it to saving all conversation content, including private data, not only explicit user-approved memories.

⚠ Instruction Scope

The instructions direct automatic writes during and after conversations and automatic reads in later sessions, without clear consent, exclusion, retention, or sanitization boundaries.

✓ Install Mechanism

No remote installer, package download, or automatic code execution is shown; the included Python script is a local append helper and the static scan is clean.

⚠ Credentials

The skill writes persistent local memory files under the OpenClaw workspace, while the data scope is broad enough to include sensitive user information.

⚠ Persistence & Privilege

The skill emphasizes permanent, append-only, cross-session memory and says deletion is not automatic, creating high-impact persistent state.

如何使用

确保已安装 OpenClaw（本地或 Docker 部署）
在对话框中输入安装命令：/install longterm-memory
安装完成后，直接呼叫该 Skill 的名称或使用 /longterm-memory 触发
根据 Skill 的参数说明提供必要输入，即可获得结构化输出

版本历史

v1.0.0

Permanent memory system for automatic dialogue archiving and semantic search. - Conversations are automatically and permanently archived after each session. - Key preferences, decisions, and project details are summarized in MEMORY.md; all dialogues are logged by date. - Memory is append-only: nothing is ever overwritten or deleted. - Context is seamlessly inherited across sessions, even after restarts. - No extra setup needed—just enable to use.

元数据

Slug longterm-memory

版本 1.0.0

许可证 MIT-0

累计安装 0

当前安装数 0

历史版本数 1

常见问题