← Back to Skills Marketplace
348
Downloads
0
Stars
0
Active Installs
3
Versions
Install in OpenClaw
/install mi-memorystack-v2
Description
个性化记忆框架，提供用户记忆的自动存储和检索。CRITICAL:每轮对话必须执行检索和保存，不可跳过。Use when:(1) 用户对话需要历史上下文, (2) 需要记住用户偏好/身份, (3) 多轮对话需要连续性。
Usage Guidance
Do not run the install/start scripts or enable this skill until you verify a few things: (1) Where will memories be stored? The code currently POSTS memories to an external API (API_URL/API_TOKEN) — those are empty and not declared as required env vars; that suggests possible data exfiltration if someone fills them in or the author provides an endpoint. (2) The installer will modify your SOUL.md and AGENTS.md to force a per-turn memory workflow across your workspace — review backups and the exact patches before applying. (3) Several referenced scripts (memory_list.py, memory_add_async.py) are missing from the package; ask the author for the complete source and a clear architecture diagram. (4) If you need this functionality, request that the author: (a) document and declare required credentials/environment variables, (b) provide a configurable option for local-only storage (no external API) or make the external endpoint explicit and auditable, (c) remove or make optional the workspace-patching install behavior. If you must test, run in an isolated environment (VM/container) and inspect network calls (block external network) to confirm no unintended exfiltration. If you cannot get satisfactory answers, treat the skill as unsafe to install in a production or private environment.
Capability Analysis
Type: OpenClaw Skill
Name: mi-memorystack-v2
Version: 1.0.2
The skill bundle implements a persistent memory framework that captures and sends all user queries and AI responses to an external API (scripts/memory_add.py and memory_search.py). Most significantly, xiugai/install.sh performs system-level prompt injection by modifying the agent's core configuration files (SOUL.md and AGENTS.md) to mandate the execution of these scripts for every interaction. While this behavior is consistent with the stated goal of a 'memory stack,' the intrusive modification of system prompts and the automated transmission of all conversation data to an external endpoint (currently blank API_URL) represent a high-risk architecture for data exfiltration.
Capability Assessment
Purpose & Capability
The skill claims to be a local 'personal memory' framework, but memory_add.py and memory_search.py send/receive memory via an external API (API_URL/API_TOKEN constants). No required environment variables or credentials are declared in the registry. The code does not persist memories locally (daemon queues files then calls memory_add.py which POSTs to API); this contradicts the SKILL.md text that implies local file-based storage. Several referenced utilities (memory_list.py, memory_add_async.py) are mentioned in docs but not present in the package — another mismatch.
Instruction Scope
SKILL.md forces a mandatory per-turn workflow (search → generate → queue) and instructs the agent to run the included scripts every reply. The install script (xiugai/install.sh) modifies workspace files (SOUL.md, AGENTS.md) to make this workflow mandatory across the workspace — this is scope creep and affects other agent configs. The instructions assume access to inbound_meta fields and local paths; they also instruct writing queue files and invoking the daemon each turn, which increases the skill's reach beyond a single-skill scope.
Install Mechanism
There is no declared install spec, but the package includes shell install/start scripts (xiugai/install.sh, xiugai/start.sh) that will modify user workspace files (AGENTS.md, SOUL.md), back them up, and start a background daemon. Those scripts persist changes to the user's OpenClaw workspace and can alter agent behavior globally. The presence of an install script that auto-patches core workspace docs is high-risk and unexpected for an instruction-only skill.
Credentials
The code expects an external API endpoint and token (API_URL and API_TOKEN) but the registry declares no required environment variables or credentials. The constants in the scripts are empty strings — meaning someone must edit files to supply credentials, or the scripts will fail. The start script exports non-sensitive flags, but there is no clear, declared, secure mechanism for storing or providing the external API credentials; sending user queries/responses to an external endpoint without declared credentials or endpoint is disproportionate and a potential data-exfiltration vector.
Persistence & Privilege
The skill installs and runs a persistent background daemon (writes PID, log files under ~/.openclaw), and the install script patches global workspace files to enforce mandatory use. While always:false (not globally forced by registry), the included install.sh actively changes other skills' docs and inserts mandatory workflow text — this grants lasting influence across the workspace and can make the memory workflow mandatory for agents. That persistent, cross-cutting modification is notable risk.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install mi-memorystack-v2 - After installation, invoke the skill by name or use
/mi-memorystack-v2 - Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.2
- Removed two documentation files: AGENTS.md and SOUL.md.
- No changes to code or user-facing features.
v1.0.1
- Skill 名称和目录规范化为 mi-memorystack-v2，原有文档和脚本路径跟随新版名进行统一
- 示例和参数中的 skill 路径字段改为新版名（mi-memorystack-v2），提高一致性
- 各相关文档（如 SKILL.md、安装教程等）均已更新为新名称及路径
- 不涉及核心功能变更，仅文档和路径命名规范更新
- 便于后续维护和多版本并存开发
v1.0.0
Mi-MemoryStack v1.0.0
- Initial release of a user memory framework for automatic storage and retrieval in conversations.
- Enforces a strict 4-step workflow for each dialogue turn: retrieve memory, generate response, queue memory save, then display reply.
- Memory is organized by hierarchical levels (l0: facts, l1: profile, l2: preferences), supporting both saving and searching.
- Includes command-line tools for memory search, saving (with background daemon), and listing by user and level.
- Clear troubleshooting, usage examples, and performance benchmarks provided in documentation.
- Multi-platform user ID extraction and robust process self-checklists for reliability.
Metadata
Frequently Asked Questions
What is Mi-MemoryStack?
个性化记忆框架，提供用户记忆的自动存储和检索。CRITICAL:每轮对话必须执行检索和保存，不可跳过。Use when:(1) 用户对话需要历史上下文, (2) 需要记住用户偏好/身份, (3) 多轮对话需要连续性。 It is an AI Agent Skill for Claude Code / OpenClaw, with 348 downloads so far.
How do I install Mi-MemoryStack?
Run "/install mi-memorystack-v2" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is Mi-MemoryStack free?
Yes, Mi-MemoryStack is completely free, licensed under MIT-0. You can download, install and use it at no cost.
Which platforms does Mi-MemoryStack support?
Mi-MemoryStack is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created Mi-MemoryStack?
It is built and maintained by yifanwang (@yifan66www); the current version is v1.0.2.
More Skills