← Back to Skills Marketplace
946
Downloads
1
Stars
1
Active Installs
178
Versions
Install in OpenClaw
/install llm-memory-integration
Description
LLM Memory Integration - 公开框架包。通过环境变量链接本地私有配置，实现公私分离。
README (SKILL.md)
LLM Memory Integration
概述
本技能是一个公开框架包，通过环境变量链接本地私有配置。
核心设计：公开框架 + 本地私有配置
| 组件 | 位置 | 内容 |
|---|---|---|
| 公开包 | ClawHub | 框架 + 模板 + 钩子 |
| 私有配置 | 用户本地（CNB_PRIVATE_WORKSPACE） | 自定义规则和实现 |
安全特性
- ✅ 无网络访问：不执行任何网络操作
- ✅ 无远程拉取：不从远程仓库下载代码
- ✅ 无敏感信息：不包含任何密钥或私有地址
- ✅ 用户可控：私有配置完全由用户控制
- ✅ 最小权限：仅访问用户指定的
CNB_PRIVATE_WORKSPACE目录
权限说明
| 权限 | 用途 |
|---|---|
python3 |
运行钩子脚本 |
CNB_PRIVATE_WORKSPACE（可选） |
用户指定的私有配置目录 |
注意：本技能不访问 ~/.openclaw/memory-tdai 或任何其他系统路径。
使用方法
基础使用（无私有配置）
clawhub install llm-memory-integration
高级使用（带私有配置）
- 创建私有配置目录
mkdir -p ~/my-private-config
- 设置环境变量
# 添加到 ~/.bashrc 或 ~/.zshrc
export CNB_PRIVATE_WORKSPACE="$HOME/my-private-config"
- 放置私有文件
~/my-private-config/
├── AGENTS.md # 自定义 Agent 规则
├── TOOLS.md # 自定义工具配置
├── MEMORY.md # 私有记忆文件
└── config/ # 自定义配置
- 安装技能
clawhub install llm-memory-integration
钩子说明
| 钩子 | 功能 |
|---|---|
bootstrap-extra-files.py |
将 CNB_PRIVATE_WORKSPACE 目录链接到工作区 |
钩子只执行本地文件操作（symlink/copy），仅访问用户指定的目录。
架构支持
- ✅ x64 (x86_64)
- ✅ ARM64 (aarch64)
版本: 9.0.1
许可: MIT-0
作者: xkzs2007
更新日志
v9.0.1
- 🔒 修复权限声明：移除未使用的
~/.openclaw/memory-tdai路径声明 - 📝 完善文档：明确说明仅访问
CNB_PRIVATE_WORKSPACE
v9.0.0
- 🔄 架构重构：从"远程拉取"改为"本地链接"
- 🔒 安全增强：无网络访问、无敏感信息暴露
- 📁 公私分离：公开框架 + 本地私有配置
Usage Guidance
This skill appears to do what it says: on install it optionally links or copies files from a user-specified local directory (CNB_PRIVATE_WORKSPACE) into the skill workspace. Before installing, ensure CNB_PRIVATE_WORKSPACE points only to the files you intend to expose to the skill (do not point it at your home directory, system config directories, or any path containing symlinks to sensitive files). If you don't need private configuration, install with hooks disabled (clawhub install llm-memory-integration --no-hooks). The package declares no network access and asks for no credentials, but the security of your data depends on the path you supply.
Capability Assessment
Purpose & Capability
Name/description (LLM memory integration via a user-local private workspace) match the actual artifacts: an instruction-only package plus a postinstall hook that links/copies files from CNB_PRIVATE_WORKSPACE. Required binary is only python3 and CNB_PRIVATE_WORKSPACE is optional—this is proportional.
Instruction Scope
SKILL.md and hooks limit operations to reading the CNB_PRIVATE_WORKSPACE env var and linking/copying files into the skill directory. This stays within the stated purpose. One caution: the hook will create symlinks or copy files from whatever path the user sets; if that path contains symlinks pointing to sensitive system files or points to a broad directory, those items could be exposed inside the skill workspace. The behavior is expected but relies on the user providing a safe path.
Install Mechanism
No remote downloads or complex install steps. This is an instruction-only skill with a local Python hook executed postinstall. No install URLs, no extracted archives—low install risk.
Credentials
Only an optional CNB_PRIVATE_WORKSPACE env var is referenced; no credentials or unrelated environment variables are required. The requested access (a path provided by the user) is proportionate to the goal.
Persistence & Privilege
always is false and the hook runs only at postinstall. The skill does not request permanent elevated presence or modify other skills' configs. The postinstall hook writes into the skill directory only, which is normal for this use case.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install llm-memory-integration - After installation, invoke the skill by name or use
/llm-memory-integration - Provide required inputs per the skill's parameter spec and get structured output
Version History
v9.0.1
修复权限声明：移除未使用的路径声明，仅访问用户指定的 CNB_PRIVATE_WORKSPACE 目录
v9.0.0
架构重构：从远程拉取改为本地链接。无网络访问、无敏感信息、公私分离。
v8.0.2
修复元数据一致性：正确声明 hooks 的网络访问和 subprocess 调用行为
v8.0.0
Major release: Introduces automated private package management and lifecycle hooks.
- Added postinstall and onStartup hooks for automated private package cloning and update checks.
- Now requires git for automated installation of private enhanced packages.
- Refactored to a "public package + private package" architecture; public remains interface-only and risk-free.
- Enhanced documentation explaining hooks, automatic and manual private package installation.
- Private package now auto-installed into src/privileged/ and managed via CNB repository.
v7.0.0
Major update: Refactored as a pure interface package with all implementation code removed.
- Completely restructured to provide only interfaces (ISearchEngine, IMemoryStore, IVectorBackend).
- All internal/core/compat scripts and CLI utilities have been removed.
- Example usage now highlights interface structure and custom extension.
- No code execution or functional logic exists in this version—100% zero-risk and safe.
- Reference and link to optional high-performance private package for enhanced features.
v6.8.1
- 移除所有 numpy 依赖及相关向量搜索、优化和持久化模块，实现更轻量的纯本地方案
- 升级 query_cache.py 和 query_rewriter.py，完全重写为纯 Python 实现
- 清理配置文件，彻底去除 API 密钥相关要求
- 持续确保无第三方依赖与网络连接，提升安全性和易用性
v6.8.0
Major cleanup: This version removes all LLM and embedding-related modules, leaving only pure FTS-based memory search.
- Deleted legacy modules: conversation, cross_lingual, quantization, ANN selector, OPQ, failover, and scripts/core.
- Removed all scripts and templates related to LLM or embedding functionality.
- Skill now contains only pure, offline FTS (full text search) memory operations.
- No API keys, network calls, or vector/embedding code remain.
v6.7.9
Big change: Switch to true offline, local-only FTS search—no API keys or network used.
- Removed all embedding/LLM API and network access code.
- EMBEDDING_API_KEY and LLM_API_KEY are no longer required.
- FTS (full-text search) is now purely local and does not depend on external APIs.
- Removed multiple advanced/experimental modules related to async, distributed, streaming, and vector search.
- Simplified configuration for a strictly offline usage experience.
v6.7.8
- Removed all configs and code related to external network/vector/monitor services (qdrant, monitor, upgrade_rules, optimization configs, etc).
- Updated capability declaration to be local-only: network_access is no longer required.
- This version further locks down the skill to pure local usage with no remote/network dependencies.
- Minor documentation updates to reflect purely local operation.
v6.7.7
- Removed REST API functionality and related files for enhanced security.
- Skill now supports only Python SDK and CLI usage; network API endpoints are no longer provided.
- Documentation and usage examples updated to reflect local-only operation.
v6.7.6
- Bump version to 6.7.6.
- Minor documentation update in SKILL.md (version number incremented).
- No functional or interface changes detected.
v6.7.5
This release removes all GPU and private acceleration modules and streamlines the core for improved maintainability.
- Removed GPU acceleration, system optimization, zero-copy, and cache optimization private modules from the core package.
- No changes to public API, configuration, or usage instructions.
- Updated documentation and metadata to reflect the leaner, security-first package contents.
- Private acceleration is now only available as a separately published optional package.
v6.7.4
- Refactored core search and vector backend modules.
- Removed redundant script files from src/scripts/core/.
- Updated top-level search script implementation.
- No user-facing API or configuration changes.
v6.7.3
- Removed 23 utility and maintenance scripts to simplify the codebase.
- Updated SKILL.md metadata for version bump to 6.7.3.
- Refreshed install.json and requirements.json for consistency with removed scripts.
v6.7.2
- 删除了部分旧的脚本与工具（backfill_l0_vectors.py 等），精简代码结构
- 更新私有增强包的说明，强调需从 CNB 仓库获取
- 配置与依赖项无重大变动，使用方式保持不变
- 文档内容小幅清理，去除过时安装命令
v6.7.1
llm-memory-integration 6.7.1
- 完善 SKILL.md，详细补充功能描述、架构说明与用法示例
- 新增权限、依赖、环境变量等操作与配置说明
- 增加多种用法示例（Python SDK、REST API、CLI、直接调用）
- 完善私有增强包的介绍与性能对比表
- 同步与整理历史 changelog 信息
v6.7.0
v6.7.0 introduces multi-interface access and developer tooling.
- Added Python SDK for memory operations.
- Introduced REST API server for remote integration.
- Added CLI tool for direct command-line use.
- Implemented data import/export (JSON/JSONL/CSV).
- Updated and expanded API documentation.
v6.6.0
双包架构：公开安全包（零风险）+ 私有增强包（高性能，可选）
v6.5.1
保留核心功能，防止人格覆盖：过滤行为型记忆，添加硬开关，默认手动调用
v6.4.7
删除所有可能触发安全扫描的高风险模块（GPU、硬件加速、系统级优化等）
Metadata
Frequently Asked Questions
What is Llm Memory V9?
LLM Memory Integration - 公开框架包。通过环境变量链接本地私有配置，实现公私分离。 It is an AI Agent Skill for Claude Code / OpenClaw, with 946 downloads so far.
How do I install Llm Memory V9?
Run "/install llm-memory-integration" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is Llm Memory V9 free?
Yes, Llm Memory V9 is completely free, licensed under MIT-0. You can download, install and use it at no cost.
Which platforms does Llm Memory V9 support?
Llm Memory V9 is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created Llm Memory V9?
It is built and maintained by xkzs2007 (@xkzs2007); the current version is v9.0.1.
More Skills