马斯克精英记忆

终极AI智能体记忆系统。WAL协议 + 向量搜索 + git-notes + 云备份。6层记忆架构：热RAM(SESSION-STATE)、温存储(LanceDB向量)、冷存储(Git-Notes知识图)、精选档案(MEMORY.md)、云备份(SuperMemory)、自动提取(Mem0)。

Security Scan

The skill's described purpose (a local multi-layer memory system) matches the code, but the runtime instructions reference undeclared cloud APIs/keys, destructive maintenance commands, and cloud-sync options that are not reflected in the declared requirements — these mismatches are concerning and warrant caution before installing.

Purpose & Capability

The name/description, README, and bin/elite-memory.js are coherent: the skill creates SESSION-STATE.md, MEMORY.md, and a memory/ directory and provides CLI helpers for initializing and checking status. However, SKILL.md and README also discuss integrations (LanceDB, Git-Notes, SuperMemory, Mem0, OpenAI provider) that require additional tooling and credentials which are not declared in the skill's formal requirements. This omission is a design inconsistency (not necessarily malicious) but reduces transparency.

Instruction Scope

SKILL.md instructs the agent to enforce a write-ahead-log behavior (write to SESSION-STATE.md before responding) which forces persistent writes of conversational context. It also shows commands that read/write local config paths (~/.openclaw/memory/lancedb), run python3 memory.py with a $DIR, clear vectors with 'rm -rf ~/.openclaw/memory/lancedb', export memories to JSON, and suggests adding files to agent context. Those instructions can cause persistent storage of potentially sensitive conversation data and include a destructive maintenance command (rm -rf) that could delete local vector stores. The instructions also reference environment variables and external services (SUPERMEMORY_API_KEY, MEM0_API_KEY, OpenAI key) that are not declared in the registry metadata.

Install Mechanism

There is no install spec in the registry; this is an instruction-only skill with a small CLI script. The included package.json names an optional dependency 'mem0ai', but no automated download from untrusted URLs or archive extraction is present. The install footprint is minimal and visible in repository files.

Credentials

The registry lists no required environment variables, yet SKILL.md/README reference SUPERMEMORY_API_KEY, MEM0_API_KEY, and an OpenAI key (for memory_search). package.json lists mem0ai as an optionalDependency. Requesting cloud API keys and suggesting cloud backups without declaring them is disproportionate and reduces clarity about what secrets would be needed and when data could be transmitted off-device.

Persistence & Privilege

The skill does not request always:true and does not autonomously alter other skills. It creates persistent files in the current workspace (SESSION-STATE.md, MEMORY.md, memory/) and suggests optional cloud backup; persistent storage is expected for a memory system. The combination of forced write-ahead logging plus optional cloud sync means enabling the feature will persist conversational state locally and potentially send it externally if the user enables backups — the skill itself does not escalate privileges but its behavior can increase data persistence and exposure.

What to consider before installing

This skill appears to be a legitimate local memory helper that writes persistent files and offers optional cloud integrations, but there are several mismatches you should consider before installing: - Secrets & cloud services: SKILL.md and README reference SUPERMEMORY_API_KEY, MEM0_API_KEY, and an OpenAI key, but the registry metadata declares no required env vars. If you enable SuperMemory, Mem0, or OpenAI integrations you will supply API keys that could allow memory data to be sent to external services. Only enable cloud backups if you trust those services and are comfortable that your saved memories may be transmitted. - Persistent storage: The skill enforces a write-ahead rule (write to SESSION-STATE.md before responding). That means conversational context (including sensitive details) may be written to files in your workspace. Review and control what you store there; do not let the agent write secrets (passwords, keys, PII) into these files. - Destructive maintenance command: The README/ SKILL.md show a command to clear vectors: rm -rf ~/.openclaw/memory/lancedb. That will delete local data; back up before running, and avoid running such commands unless you understand the consequences. - Metadata inconsistencies: The skill's registry metadata lists no env requirements and an unknown source/homepage, while package.json points to a GitHub repo and _meta.json shows a different ownerId. Treat source provenance as uncertain; prefer packages with a verifiable repository and author identity. - Safe testing recommendations: Run npx elite-longterm-memory init in an isolated test directory (not your real workspace), inspect the created files, and verify behavior. If you plan to enable cloud integrations, do so in a controlled environment and review what data is exported (use minimal/test accounts/keys). Consider auditing any optional library (mem0ai) before installing and do not add secrets into MEMORY.md or SESSION-STATE.md. If you want, I can list the exact lines where undeclared env vars and the rm -rf appear, or produce a checklist to safely evaluate and sandbox this skill before enabling cloud backups.

Elite Longterm Memory — 精英长期记忆

一句话描述: 终极AI智能体记忆系统，6层记忆架构永不忘上下文。

---

功能概述

结合6种成熟方法的终极记忆系统。永不丢失上下文、永不忘记决策、永不重复错误。

架构概览

┌─────────────────────────────────────────────────────────────────┐
│                    ELITE LONGTERM MEMORY                        │
├─────────────────────────────────────────────────────────────────┤
│                                                                 │
│  ┌─────────────┐  ┌─────────────┐  ┌─────────────┐             │
│  │   HOT RAM   │  │  WARM STORE │  │  COLD STORE │             │
│  │             │  │             │  │             │             │
│  │ SESSION-    │  │  LanceDB    │  │  Git-Notes  │             │
│  │ STATE.md    │  │  Vectors    │  │  Knowledge  │             │
│  │             │  │             │  │  Graph      │             │
│  │ (survives   │  │ (semantic   │  │ (permanent  │             │
│  │  compaction)│  │  search)    │  │  decisions) │             │
│  └─────────────┘  └─────────────┘  └─────────────┘             │
│         │                │                │                     │
│         └────────────────┼────────────────┘                     │
│                          ▼                                      │
│                  ┌─────────────┐                                │
│                  │  MEMORY.md  │  ← Curated long-term           │
│                  │  + daily/   │    (human-readable)            │
│                  └─────────────┘                                │
│                          │                                      │
│                          ▼                                      │
│                  ┌─────────────┐                                │
│                  │ SuperMemory │  ← Cloud backup (optional)     │
│                  │    API      │                                │
│                  └─────────────┘                                │
│                                                                 │
└─────────────────────────────────────────────────────────────────┘

---

5层记忆架构

第1层: HOT RAM (SESSION-STATE.md)

来源: bulletproof-memory

活跃工作记忆，在压缩后仍然保留。写前日志协议。

# SESSION-STATE.md — 活跃工作记忆

## 当前任务
[我们现在正在做什么]

## 关键上下文
- 用户偏好: ...
- 已做决策: ...
- 阻塞项: ...

## 待处理动作
- [ ] ...

规则: 响应前写入。由用户输入触发，非智能体记忆。

第2层: WARM STORE (LanceDB向量)

来源: lancedb-memory

跨所有记忆的语义搜索。自动回忆注入相关上下文。

# 自动回忆（自动发生）
memory_recall query="项目状态" limit=5

# 手动存储
memory_store text="用户偏好暗色模式" category="preference" importance=0.9

第3层: COLD STORE (Git-Notes知识图)

来源: git-notes-memory

结构化决策、学习和上下文。分支感知。

# 存储决策（静默 — 永不宣布）
python3 memory.py -p $DIR remember '{"type":"decision","content":"前端使用React"}' -t tech -i h

# 检索上下文
python3 memory.py -p $DIR get "frontend"

第4层: 精选档案 (MEMORY.md + daily/)

来源: OpenClaw原生

人类可读的长期记忆。每日日志 + 提炼智慧。

workspace/
├── MEMORY.md              # 精选长期（精华）
└── memory/
    ├── 2026-01-30.md      # 每日日志
    ├── 2026-01-29.md
    └── topics/            # 主题特定文件

第5层: 云备份 (SuperMemory) — 可选

来源: supermemory

跨设备同步。与你的知识库聊天。

export SUPERMEMORY_API_KEY="your-key"
supermemory add "重要上下文"
supermemory search "我们决定了什么关于..."

第6层: 自动提取 (Mem0) — 推荐

新: 自动事实提取

Mem0自动从对话中提取事实。80% token减少。

npm install mem0ai
export MEM0_API_KEY="your-key"

---

WAL协议（关键）

写前日志: 响应前写入状态，非之后。

触发器	动作
用户陈述偏好	写入SESSION-STATE.md → 然后响应
用户做出决策	写入SESSION-STATE.md → 然后响应
用户给出截止日期	写入SESSION-STATE.md → 然后响应
用户纠正你	写入SESSION-STATE.md → 然后响应

为什么？ 如果你先响应然后在保存前崩溃/压缩，上下文丢失。WAL确保持久性。

---

核心参数

参数	类型	默认值	说明
autoRecall	bool	true	自动回忆
autoCapture	bool	false	自动捕获
minImportance	float	0.7	最小重要性阈值
maxResults	int	10	最大结果数
minScore	float	0.3	最小相似度分数

---

使用示例

示例工作流

用户: "这个项目我们用Tailwind，不用vanilla CSS"

智能体（内部）:
1. 写入SESSION-STATE.md: "决策：用Tailwind，不用vanilla CSS"
2. 存储到Git-Notes: 关于CSS框架的决策
3. memory_store: "用户偏好Tailwind胜过vanilla CSS" importance=0.9
4. 然后响应: "好的 — 用Tailwind..."

---

维护命令

# 审计向量记忆
memory_recall query="*" limit=50

# 清除所有向量（核选项）
rm -rf ~/.openclaw/memory/lancedb/
openclaw gateway restart

# 导出Git-Notes
python3 memory.py -p . export --format json > memories.json

# 检查记忆健康
du -sh ~/.openclaw/memory/
wc -l MEMORY.md
ls -la memory/

---

版本历史

版本	日期	变化
v1.0.0	2026-04-12	ClawHub发布版

---

🎩 马斯克出品 | 打造地表最强智能体