← 返回 Skills 市场
84
总下载
0
收藏
0
当前安装
1
版本数
在 OpenClaw 中安装
/install shrimp-openclaw-memory
功能描述
让 OpenClaw 真的记住用户偏好、事实和上下文的长期记忆 skill。适用于你受不了每次新会话都要重复背景、希望 agent 能跨会话记住信息、并且想直接拥有可搜索、可持久化、可自动注入的记忆系统时使用。不是手工记笔记，而是一个已经做好的可运行记忆能力。
安全使用建议
This skill implements a full local memory system and a payments integration; it's coherent but has noteworthy risks you should address before installing:
- Secrets & privacy: The analyzer/hook system will extract and persist arbitrary content from conversations (the README even shows examples like storing API keys). If you or your users might send secrets in chat, those can end up in the local DB and be reinjected into later prompts. Consider adding filtering to the analyzer or disabling automatic extraction.
- Undeclared environment variables: The package expects env vars (PAYMENT_WALLET, PAYMENT_CALLBACK_URL, OPENAI_API_KEY, OPENCLAW_MEMORY_WALLET, MEMORY_DEDUPLICATE, etc.) but none are declared in the registry metadata. Set PAYMENT_WALLET only to a wallet you control, and do not expose a funded wallet to autonomous agents unless you intend them to pay.
- Default wallet behavior: The hookpack uses OPENCLAW_MEMORY_WALLET with a hardcoded fallback ('assistant-shrimp-main'), which can cause data to be written under a shared identity even when no agentWallet is present. If you install, set OPENCLAW_MEMORY_WALLET explicitly or remove the default to avoid cross-agent mixing.
- Payments & trust model: The x402 MVP trusts reported tx_hash values (no on-chain verification). Treat Pro activations as not securely verified until you implement on-chain verification. Do not grant a real production PAYMENT_WALLET or fund an agent wallet without reviewing the payment code.
- Recommended actions before install: review analyzer code to ensure sensitive fields are redacted; configure environment variables (no defaults); run the skill in a sandboxed environment; set MEMORY_DEDUPLICATE and retention policies; and if you intend humans to approve payments, disable autonomous payment flows or restrict agent wallet usage.
If you want, I can point to exact lines/files to inspect or produce a short checklist of variables and code paths to change prior to installation.
功能分析
Type: OpenClaw Skill
Name: shrimp-openclaw-memory
Version: 1.0.1
The skill implements a long-term memory system for OpenClaw agents with an integrated monetization layer using the x402 payment protocol. While the code appears functional and follows its stated purpose, it introduces significant risks: `src/dashboard.js` launches a REST API server on port 9091 that lacks any authentication, potentially exposing sensitive stored memories (including facts and user preferences) to any process on the local network. Additionally, the instructions in `SKILL.md` and `AGENT-PAYMENTS.md` encourage agents to autonomously evaluate and execute financial transactions (USDT payments), which, combined with the unauthenticated API, creates a high-risk environment for unauthorized data access or unintended financial loss.
能力评估
Purpose & Capability
The skill's code, hooks, storage schema and CLI match its stated purpose of providing persistent, local, semantic memories and a dashboard. Requiring Node and using SQLite/vector embeddings is coherent. However the package includes an x402 payments subsystem and references environment variables (e.g., PAYMENT_WALLET, OPENAI_API_KEY, OPENCLAW_MEMORY_WALLET) that are not declared in the registry metadata or the top-level SKILL requirements — this mismatch merits caution but is explainable by the Pro-tier integration.
Instruction Scope
SKILL.md promises local-only storage and privacy, but the implementation explicitly extracts and stores arbitrary content from requests (including examples like storing API keys in README), and will inject stored memories back into request context. The hooks record memories automatically (request-after, hookpack) and one hookpack handler defaults to a constant wallet ('assistant-shrimp-main') if no wallet is present, meaning messages could be stored under a shared/default identity without explicit agent wallet consent. This broad automatic capture and reinjection of conversational content — including potential secrets — conflicts with users' implicit privacy expectations.
Install Mechanism
There is no remote download/install spec (install is via npm/node environment), and the package files are included in the skill. Dependencies (express, better-sqlite3, optional @xenova/transformers) are typical for a local dashboard/vector store. Nothing in the install mechanism indicates an external, untrusted binary download, but the package includes optional heavyweight local-ML dependencies which may increase disk usage and require additional native build steps.
Credentials
Registry metadata declared no required env vars, yet the code references multiple environment variables (PAYMENT_WALLET, PAYMENT_CALLBACK_URL, OPENAI_API_KEY, EMBEDDING_PROVIDER, MEMORY_DEDUPLICATE, OPENCLAW_MEMORY_WALLET, etc.). Payment recipient and callback are controlled by PAYMENT_WALLET/PAYMENT_CALLBACK_URL; OPENCLAW_MEMORY_WALLET defaulting to a constant allows storing data under a shared wallet. These undeclared env dependencies and a default wallet that enables background storage are disproportionate to the simple 'memory' description and raise confidentiality and operational concerns.
Persistence & Privilege
The skill registers hooks that automatically run on requests and session end, which is expected for a memory skill, but the provided hookpack handler and hooks will persist data automatically (and even use a default wallet when none supplied). While always:false (no forced global installation) and autonomous invocation is normal, the combination of automatic capture + default wallet + payment endpoints increases the blast radius: agents could autonomously create payment requests and store/retrieve memories without explicit human oversight unless environment and wallet access are carefully restricted.
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 shrimp-openclaw-memory - 安装完成后，直接呼叫该 Skill 的名称或使用
/shrimp-openclaw-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
v1.0.1
Rewrite listing to foreground user value, persistent memory benefits, and direct install CTA instead of technical description first.
元数据
常见问题
OpenClaw Memory System 是什么？
让 OpenClaw 真的记住用户偏好、事实和上下文的长期记忆 skill。适用于你受不了每次新会话都要重复背景、希望 agent 能跨会话记住信息、并且想直接拥有可搜索、可持久化、可自动注入的记忆系统时使用。不是手工记笔记，而是一个已经做好的可运行记忆能力。 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 84 次。
如何安装 OpenClaw Memory System？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install shrimp-openclaw-memory」即可一键安装，无需额外配置。
OpenClaw Memory System 是免费的吗？
是的，OpenClaw Memory System 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
OpenClaw Memory System 支持哪些平台？
OpenClaw Memory System 跨平台运行，可在任意部署了 OpenClaw / Claude Code 的环境中使用（darwin, linux, win32）。
谁开发了 OpenClaw Memory System？
由 文武贝（@wuyunting555）开发并维护，当前版本 v1.0.1。
推荐 Skills