Triple Memory Lake

Quad-layer memory system integration - unifies OpenClaw, Claude Code, and self-improving agent memories into a single knowledge lake

Security Scan

The skill mostly does what its description promises (sync Claude Code and self-improving metrics and mine patterns) but has internal inconsistencies (name vs layers, claims to sync OpenClaw daily logs but provides no implementation) and reads/copies sensitive user data from home directories into the skill workspace — something you should review before installing or enabling autonomous use.

Purpose & Capability

Overall purpose (unify Claude Code and self-improving agent data into a memory lake and mine patterns) aligns with the included scripts. However: the skill is named 'Triple Memory Lake' while the description and SKILL.md describe a 'quad-layer' system (minor naming inconsistency). More importantly, SKILL.md and the directory layout advertise OpenClaw daily logs as a source (memory/sources/mine/), but there is no script that actually syncs OpenClaw daily logs — only self-improving metrics (~/.openclaw/agents/*/metrics.json) and Claude Code sessions (~/.claude/projects/*/sessions/*.jsonl) are implemented. This mismatch between claimed sources and implemented syncing is an incoherence the user should be aware of.

Instruction Scope

SKILL.md and the scripts instruct the agent to read and copy files from hidden user directories (~/.claude and ~/.openclaw) into the skill's own memory/ directory. That is coherent with the stated purpose, but it is privacy-sensitive: user conversations, session logs, and agent metrics may contain secrets or PII and will be duplicated inside the skill workspace. The SKILL.md cp examples and the sync scripts do not filter or redact sensitive fields before copying.

Install Mechanism

Instruction-only with included Python scripts; there is no installer or external download. No network endpoints, package installs, or archive extraction are used. This is low-risk from an installation supply-chain perspective.

Credentials

The skill requests no environment variables or credentials which is proportionate. However, it accesses user-local files under the home directory (~/.claude and ~/.openclaw). While those accesses are relevant to the skill's goal, they are effectively privileged because they duplicate user-private data into the skill folder; the skill does not declare or warn about this in its metadata.

Persistence & Privilege

The skill does not request always:true and does not modify other skills or system-wide settings. It runs as user-invocable and can be invoked autonomously by agents per platform defaults; combine that with the file-access behavior when deciding whether to allow autonomous runs.

What to consider before installing

This skill will copy files from your home directory (~/.claude and ~/.openclaw) into a local memory/ folder and then process them to extract patterns. Before installing or enabling autonomous execution: (1) Inspect the memory/ and scripts locally to see exactly what will be copied and generated. (2) Consider running the sync scripts manually in an isolated environment (or container) first to review outputs and ensure no secrets are being copied. (3) If you do want this behavior, restrict execution to manual invocation or audit the files to redact secrets; if you expect OpenClaw daily logs to be synced, note that the skill currently lacks a script for that and may not be doing what you expect. (4) If you are uncomfortable with duplicating potentially sensitive logs, do not enable autonomous invocation and consider removing or modifying the scripts to filter/redact sensitive fields before saving.

基于三个记忆系统整合的统一知识湖，已升级为四层结构。

四层记忆系统

Layer	内容	更新频率
L1 MEMORY.md	铁律/人物/项目/教训	重大决策后
L2 Daily Logs	每日工作日志	每日
L3 Sources+Patterns+Domain+Tools	原始数据+提炼模式+领域知识+工具配置	定期同步
L4 Reviews	自审+质量评分	每周一/触发时

核心功能

1. 三源数据同步

• Claude Code JSONL → memory/sources/claude-code/
• self-improving metrics → memory/sources/self-improving/
• OpenClaw daily logs → memory/sources/mine/

2. 模式提炼

• 错误模式汇总 → memory/patterns/error-patterns.md
• 工作流模式 → memory/patterns/workflow-patterns.md
• 用户偏好 → memory/patterns/user-preferences.md

3. 知识沉淀

• 长期记忆 → MEMORY.md
• 领域知识 → memory/domain/
• 工具知识 → memory/tools/

4. 自省机制

• 质量评分标准 → memory/reviews/memory-quality.md
• 自审模板 → memory/reviews/self-review-template.md
• 自审报告 → memory/reviews/self-review-YYYY-MM-DD.md

目录结构

memory/
├── index.md              # 统一入口
├── MEMORY.md              # Layer 1: 长期记忆
├── YYYY-MM-DD.md          # Layer 2: 每日日志
├── sources/               # Layer 3a: 原始数据来源
│   ├── claaude-code/      # Claude Code JSONL 日志
│   ├── self-improving/    # self-improving 指标
│   └── mine/              # OpenClaw 每日日志（预留）
├── patterns/              # Layer 3b: 提炼模式
│   ├── error-patterns.md  # 错误模式汇总
│   ├── workflow-patterns.md # Captain工作流模式
│   └── user-preferences.md  # 用户行为偏好
├── domain/                # Layer 3c: 领域知识
│   ├── openclaw.md        # OpenClaw配置/铁律/调度规则
│   └── stock-system.md    # 股票分析系统详情
├── tools/                 # Layer 3d: 工具配置
│   ├── skills.md          # Skills状态一览
│   ├── git-workflow.md    # Git工作流配置
│   └── hook-system.md     # Hook系统配置
└── reviews/               # Layer 4: 自省机制
    ├── memory-quality.md   # 记忆质量评分标准
    ├── self-review-template.md # 自审报告模板
    └── self-review-YYYY-MM-DD.md # 历次自审报告

自审触发条件

满足任一即触发自审：

1. 每周一自动自审
2. MEMORY.md 超过30天未更新
3. 发现新错误模式时
4. 用户指出记忆错误时
5. 重大系统变更后（如 OpenClaw 升级）

质量评分

评估维度：完整性(25%) / 准确性(30%) / 时效性(25%) / 可检索性(20%)

评分：9-10优秀 / 7-8良好 / 5-6及格 / 3-4危险 / 1-2失效

使用方式

查看知识湖状态

cat memory/index.md

执行自审

# 参考 memory/reviews/self-review-template.md 生成报告
# 保存为 memory/reviews/self-review-YYYY-MM-DD.md

同步数据源

# Claude Code
cp ~/.claude/projects/*/sessions/*.jsonl memory/sources/claude-code/

# Self-improving
cp ~/.openclaw/agents/*/metrics.json memory/sources/self-improving/

---

Last updated: 2026-04-03