摇摇记忆系统

支持多层次记忆管理和标签分类，实现跨会话检索、持久化存储、定期整理及与IMA知识库双向同步。

Security Scan

The skill's functionality (local memory files + syncing to IMA) is coherent, but it will send workspace documents to an external IMA endpoint and the registry metadata fails to declare the required IMA credentials—this mismatch and the default-enabled sync behavior could lead to unintended data exfiltration.

Purpose & Capability

The name/description match the included scripts (local memory management + IMA sync). However, the registry metadata declares no required environment variables or credentials while both SKILL.md and scripts clearly expect IMA client_id/api_key (via config.json or env vars). That omission in declared requirements is an inconsistency users should know about.

Instruction Scope

Runtime instructions and scripts read and write many workspace files (MEMORY.md, memory/*.md, .learnings/*, knowledge/*) and may upload their contents to the remote IMA service. The SKILL.md indicates IMA sync is '默认启用' and the default config shows knowledge_sync.enabled: true, which means content could be sent externally if the user provides credentials or leaves defaults—this is scope expansion beyond purely local memory management and presents a potential data-exfiltration vector if users put sensitive data in the workspace.

Install Mechanism

No install spec (instruction-only + scripts) and no downloads are present. The skill does not perform archive downloads or install arbitrary third-party packages at install time. It relies on simple Python scripts and existing tooling (urllib/requests mention).

Credentials

The code expects IMA credentials (client_id / api_key) via config.json or environment variables (IMA_CLIENT_ID, IMA_API_KEY), but the registry metadata lists no required env or primary credential. Requesting an API key for the external IMA service is reasonable for sync functionality, but the missing declaration is a red flag. Also the skill accesses arbitrary workspace files which could contain secrets; syncing them to an external service is disproportionate unless the user explicitly consents and configures which files to sync.

Persistence & Privilege

The skill is not marked always:true and does not modify other skills or system-wide settings. Autonomous invocation is allowed by default (platform default) — combined with default-enabled sync this increases the blast radius but by itself is not an unusual privilege.

What to consider before installing

This skill implements local memory files plus an IMA sync script that will upload workspace documents to https://ima.qq.com/ using an IMA client_id and api_key. However, the registry metadata did not declare those required credentials. Before installing: 1) Treat the IMA sync as a potential data-exfiltration path — don't enable syncing if your workspace may contain sensitive data. 2) Inspect or create config.json and set knowledge_sync.enabled to false unless you intend to sync; verify the notes_mapping IDs. 3) If you will use sync, supply credentials only if you trust the IMA endpoint and understand what files will be uploaded. 4) Run the skill in a sandbox or with non-sensitive example files first. 5) Preferably ask the publisher for provenance (homepage, source repo) or request that required env vars be declared in the registry metadata.

Skill: yaoyaoya-memory

🦞 Universal Memory System for AI Assistants — with IMA by default

универсальная система памяти | 让 AI 助手拥有持久化记忆能力

---

与其他记忆方案的区别

特性	其他方案	yaoyaoya-memory
记忆分层	单一文件	MEMORY.md + 每日日记 + learnings 分层
标签系统	无/简单	完整标签体系（#项目 #教训 #决策 #待办）
知识同步	需手动	IMA 默认集成，可配置 Notion/Obsidian
可靠性标注	无	★★★★★ 到 ★☆☆☆☆ 分级
搜索能力	全文搜索	memory_search + 语义标签检索
配置方式	硬编码	config.json 完全可配置

---

核心能力

功能	触发词	说明
主动记忆	"我记得"/"之前说过"/"搜索记忆"	跨会话记忆检索
记忆写入	"记住"/"记录"/"存档"	持久化存储重要信息
记忆整理	"整理记忆"/"总结"	定期归纳提炼
IMA同步	"同步IMA"/"上传知识库"	与IMA笔记双向同步

---

触发规则

1. 主动记忆触发

触发条件：

• 用户说「我记得」「之前说过」「好像有过」「搜索记忆」
• 当用户询问涉及之前讨论过的内容时

执行流程：

1. 调用 memory_search 搜索记忆库
2. 分析查询意图，匹配相关记忆
3. 返回记忆片段，标注来源、时间和可靠性
4. 必要时用 memory_get 获取完整上下文

搜索范围：

• MEMORY.md — 长期记忆（精选沉淀）
• memory/*.md — 每日日记（按日期归档）
• .learnings/*.md — 错误教训和经验

---

2. 记忆写入触发

触发条件：

• 用户说「记住」「记录」「存档」「这个很重要」
• AI 判断某信息值得长期保存

执行流程：

1. 判断记忆类型和存储位置
   ├─ 重大决策 → MEMORY.md
   ├─ 每日事件 → memory/YYYY-MM-DD.md
   ├─ 错误教训 → .learnings/LEARNINGS.md
   └─ 项目详情 → knowledge/[项目名].md

2. 写入文件，带时间戳
3. 确认写入成功

写入规则：

• 只记录已验证事实，不记未证实推测
• 标注信息来源和可靠性等级
• 关键决策要记录上下文和理由

---

3. 记忆整理触发

触发条件：

• 用户说「整理记忆」「总结最近」「回顾」

执行流程：

1. 读取最近30天日记
2. 识别重复模式和重要事件
3. 将新洞察提炼到 MEMORY.md
4. 清理过时信息

---

4. IMA 知识库同步（默认启用）

触发条件：

• 用户配置了 IMA
• 用户说「同步IMA」「上传知识库」「更新IMA」

默认笔记矩阵：

笔记	note_id	用途
AI记忆库	7445750904545580	总目录
日记	7445750912911573	每日事件记录
决策库	7445750929715125	技术/产品重要决策
用户档案	7445750942269901	用户偏好和习惯
经验总结	7445750959047844	最佳实践和错误教训
项目	7445750967460566	各项目详情

IMA 配置：

{
  "knowledge_sync": {
    "enabled": true,
    "platform": "ima",
    "client_id": "your-ima-client-id",
    "api_key": "your-ima-api-key",
    "notes_mapping": {
      "AI记忆库": "7445750904545580",
      "日记": "7445750912911573",
      "决策库": "7445750929715125",
      "用户档案": "7445750942269901",
      "经验总结": "7445750959047844",
      "项目": "7445750967460566"
    }
  }
}

同步方向：

• → IMA：重要结论、决策、经验
• ← IMA：用户偏好、项目更新

---

文件结构

workspace/
├── MEMORY.md              # 长期记忆（精选沉淀）
├── memory/
│   └── YYYY-MM-DD.md     # 每日日记（原始流水）
├── .learnings/
│   ├── LEARNINGS.md      # 错误教训和最佳实践
│   ├── ERRORS.md         # 系统错误记录
│   └── FEATURE_REQUESTS.md  # 功能需求
├── knowledge/             # 项目知识库
│   └── [project].md      # 各项目详情
└── scripts/
    ├── memory_search.py   # 记忆搜索脚本
    └── sync_ima.py       # IMA同步脚本

---

标签体系

类型	示例
项目	#HarmonyOS #ArkTS #一日记账
教训	#教训 #踩坑 #错误
决策	#决策 #方案选型 #优先级
待办	#待办 #TODO #后续
时间	#今天 #本周 #2026-04

---

可靠性标注

验证方式	可靠性	记法
实际测试验证	★★★★★	直接记结论
用户明确告知	★★★★★	直接记结论
第三方确认	★★★☆☆	结论 + 注明来源
推测/推断	★★☆☆☆	结论 + 注明「推测」

---

快速指令

我记得之前...        # 主动记忆检索
记住这个...          # 写入长期记忆
搜索记忆...          # 按关键词检索
整理记忆            # 定期归纳总结
同步IMA             # IMA知识库同步

---

配置示例

在 skill 目录创建 config.json：

{
  "memory": {
    "long_term_file": "MEMORY.md",
    "daily_dir": "memory",
    "learnings_dir": ".learnings",
    "knowledge_dir": "knowledge"
  },
  "knowledge_sync": {
    "enabled": true,
    "platform": "ima",
    "client_id": "your-ima-client-id",
    "api_key": "your-ima-api-key",
    "notes_mapping": {
      "AI记忆库": "7445750904545580",
      "日记": "7445750912911573",
      "决策库": "7445750929715125",
      "用户档案": "7445750942269901",
      "经验总结": "7445750959047844",
      "项目": "7445750967460566"
    }
  }
}

---

依赖工具

• memory_search — 记忆语义搜索（OpenClaw 内置）
• memory_get — 读取指定记忆片段（OpenClaw 内置）
• scripts/sync_ima.py — IMA 同步脚本

---

安装

npx clawhub install yaoyaoya-memory

---

License

MIT — 可直接使用、修改、分发