← 返回 Skills 市场
Mem0 Memory
作者
chircken891
· GitHub ↗
· v1.0.0
· MIT-0
781
总下载
0
收藏
9
当前安装
1
版本数
在 OpenClaw 中安装
/install mem0-memory
功能描述
mem0 本地记忆层完整实现（增强版）。语义记忆存储/检索/管理，WAL 协议，SESSION-STATE，多级记忆（User/Session/Agent）。参考 ZejunCao/bilibili_code Mem0框架解读优化。
安全使用建议
Key points to consider before installing or using this skill:
- The package contains only a SKILL.md; it does not include the Python scripts or binaries it references. Verify that the referenced files (e.g., D:\autoclaw\结果\mem0\mem0_wrapper.py) and services (Ollama, the MiniMax model, Chroma DB) actually exist and come from a trusted source before running anything.
- This skill will scan every message and persist memories (including URLs, file paths, names, preferences) to local files. That can capture sensitive personal or system information — confirm retention, access controls, and whether data is encrypted at rest.
- The metadata does not declare required runtimes (Python) or config paths. Ask the author for a manifest: exact runtime requirements, file layout, where data is stored, and a copy of mem0_wrapper.py to review.
- Do not point this skill at directories with sensitive files (password stores, SSH keys, private documents). Consider running it in an isolated environment or VM and review/read the actual code before giving it access to your real user data.
- If you proceed, require explicit confirmation steps for destructive commands (reset/delete_all) and log all actions. If you cannot obtain the implementation or a trustworthy origin, avoid installing the skill.
功能分析
Type: OpenClaw Skill
Name: mem0-memory
Version: 1.0.0
The skill bundle implements a memory layer but relies on executing a local Python script (mem0_wrapper.py) via hardcoded absolute Windows paths (D:\autoclaw\结果\mem0\), which are not included in the bundle. The instructions in SKILL.md direct the AI agent to extract user content, URLs, and file paths and pass them directly as command-line arguments to this script, creating a high risk of command injection. While the intent appears to be functional memory management, the lack of the core logic and the reliance on unvalidated execution of user-derived strings make it a significant security risk.
能力评估
Purpose & Capability
The SKILL.md promises a full local memory stack (Python wrapper mem0_wrapper.py, Ollama embedder, MiniMax LLM, Chroma vector DB) and points at a concrete filesystem path (D:\autoclaw\结果\mem0\). However, the skill metadata declares no required binaries, no config paths, and includes no code or install steps. A legitimate implementation would require at minimum: Python/runtime, the mem0_wrapper.py script and supporting files, and locally running services (Ollama, Chroma/DB files). The absence of these requirements in the package metadata is an incoherence.
Instruction Scope
The instructions direct the agent to run Python commands against files in a specific local path, scan every incoming message for multiple data types, and persist user/session/agent memories to disk (including URLs and file paths). That scope involves reading/writing local files and continuously processing user messages for storage; these actions are broader and more privacy-sensitive than the declared skill footprint. There is no guidance in the metadata about what files will be created, retention, encryption, or where SESSION-STATE.md / WORKING-BUFFER.md live, so the runtime instructions exceed what the package declares.
Install Mechanism
The skill is instruction-only (no install spec), which is low risk from an installer perspective. However, it depends on several external components (Ollama running as a service, local LLM models, Chroma DB, and a Python wrapper) that are not installed or provided by the skill. That mismatch means the agent or operator must have preinstalled components — this should be declared and verified before use.
Credentials
No environment variables or credentials are requested, which superficially looks safe. But the SKILL.md expects access to a specific local filesystem location and to persist potentially sensitive user data (preferences, experiences, URLs, file paths). The lack of declared config paths and permission expectations in the metadata is disproportionate: the skill requires filesystem and local-service access but doesn't declare it. Storing personal data persistently without documented controls (encryption, retention policy, access controls) is a privacy risk.
Persistence & Privilege
The skill persists memories to disk and relies on SESSION-STATE.md and WORKING-BUFFER.md for state recovery; it does not set always:true and does not modify other skills. Persisting user memory is expected for this purpose, but the skill's metadata fails to document where/how data is persisted and protected. The per-message WAL scanning behavior (scan every message for triggers) increases privacy sensitivity and should be explicitly consented and controlled.
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 mem0-memory - 安装完成后，直接呼叫该 Skill 的名称或使用
/mem0-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
mem0-memory 2.1.0 更新内容：
- 实现本地全功能语义记忆层，支持添加、检索、更新、历史、批量与重置操作
- 新增“chat_with_memories”对话模式：自动检索记忆并注入 system prompt，由 LLM 生成增强回复
- 内置中文记忆自动提取与 FACT_RETRIEVAL_PROMPT，支持对话中快速生成记忆条目
- 完善多级记忆体系（用户/会话/智能体），参数可灵活指定
- 增强工作流与 WAL 支持，实现消息分流存储与 SESSION-STATE 优先机制
- 离线向量化（Ollama + Chroma），适配本地完全运行环境
元数据
常见问题
Mem0 Memory 是什么？
mem0 本地记忆层完整实现（增强版）。语义记忆存储/检索/管理，WAL 协议，SESSION-STATE，多级记忆（User/Session/Agent）。参考 ZejunCao/bilibili_code Mem0框架解读优化。 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 781 次。
如何安装 Mem0 Memory？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install mem0-memory」即可一键安装，无需额外配置。
Mem0 Memory 是免费的吗？
是的，Mem0 Memory 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
Mem0 Memory 支持哪些平台？
Mem0 Memory 跨平台运行，可在任意部署了 OpenClaw / Claude Code 的环境中使用（cross-platform）。
谁开发了 Mem0 Memory？
由 chircken891（@chircken891）开发并维护，当前版本 v1.0.0。
推荐 Skills