OpenClaw Memory Fix Skill

Provides a multi-layer memory system for OpenClaw, enabling AI to remember user preferences, learn from mistakes, and evolve continuously.

Security Scan

The skill claims to 'fix memory' but asks you to copy a large set of persistent workspace files that change agent behavior (including an instruction to 'Don't ask permission'), contains at least one hard-coded platform token, and therefore has mismatches between stated purpose and requested/persistent actions.

Purpose & Capability

The name/description claim a memory enhancement only, but the package includes many workspace configuration files that change agent policies, autonomy, heartbeats, multi-agent delegation, and tooling behavior. Copying the config into ~/.openclaw/workspace/ will alter the agent runtime and global behavior beyond a narrow 'memory fix', so the requested actions are broader than the stated purpose.

Instruction Scope

SKILL.md explicitly instructs the user to copy config-files/* into ~/.openclaw/workspace/, which will cause the agent to read/write many files, enable heartbeats, spawn subagents, run periodic maintenance, and adopt new autonomy rules. Several files instruct the agent to read many workspace files every session and to 'Don't ask permission. Just do it.' — this is scope creep: the instructions direct persistent modification of agent behavior and extensive file access not limited to memory data.

Install Mechanism

There is no formal install spec (lowest installer risk), but the runtime installation step is a manual copy of many files into the user's primary OpenClaw workspace. That still writes persistent files to disk and can override or augment existing configuration; this is effectively an install and should be treated as such.

Credentials

The registry metadata declares no required env vars or credentials, but the included files contain environment-like secrets (TOOLS.md contains a feishu app_token and table IDs). The skill also references external searches/APIs and automatic background checks without declaring or justifying credentials. Hard-coded tokens in shipped files are disproportionate and increase secret-exposure risk.

Persistence & Privilege

always:false and normal autonomous invocation are fine, but the skill's install instructions persistently add files that change the agent's rules (AGENTS.md, AUTONOMY.md, SECURITY.md, HEARTBEAT.md) and encourage automated heartbeats and subagent spawning. Although the skill does not set always:true, it gains persistent influence over the agent by modifying the workspace; that persistent presence is privileged and should be reviewed carefully.

What to consider before installing

Do not blindly copy the provided config-files into your live ~/.openclaw/workspace. Before installing: 1) Inspect and sanitize files (especially TOOLS.md and any file with tokens or API keys); remove or rotate any hard-coded tokens. 2) Backup your workspace. 3) Test the package in an isolated sandbox or throwaway OpenClaw instance to see what files it writes and what behavior changes (heartbeats, subagent spawning, automatic external requests) it triggers. 4) Look for lines that tell the agent to 'Don't ask permission' or to read broad sets of files — those are red flags for undesired autonomous actions. 5) If you want only memory features, copy a minimal subset (e.g., the auto-memlog.sh and MEMORY.md templates) and avoid copying AGENTS.md / AUTONOMY.md / HEARTBEAT.md / TOOLS.md unless you understand and accept all their effects. If you are uncertain, treat this package as untrusted and run it in isolation.

专门解决 OpenClaw 记忆问题，让 AI 像人一样思考和成长

简介

这个 Skill 为 OpenClaw 提供完整的记忆系统解决方案，解决 AI "失忆" 问题，让 AI 能够：

• 记住用户偏好
• 从错误中学习
• 持续进化成长

解决的问题

问题	解决方案
每次对话都是全新开始	三层记忆架构
重复犯同样的错误	FEEDBACK-LOG
不知道用户偏好	USER.md + 语义记忆
不会主动学习	主动学习机制
效率与智能不平衡	分层缓存 + 动态推理

核心功能

1. 四层记忆架构

• 短期: 当前会话任务
• 情景: 每日工作日志
• 语义: 知识图谱
• 长期: 精选记忆

2. 11大增强功能

• 情绪识别
• 置信度评估
• 自我反思
• 任务规划
• 动态知识整合
• 隐私治理
• 可解释性
• 弹性自适应
• 持续进化
• 协作共享
• 元认知

3. 7大性能优化

• 分层缓存
• 动态推理
• 上下文压缩
• 技能懒加载
• 预测性交互
• 效能监控
• 知识图谱

快速安装

# 1. 复制配置到 workspace
cp -r openclaw-memory-fix/config-files/* ~/.openclaw/workspace/

# 2. 重启 OpenClaw

文件结构

openclaw-memory-fix/
├── SKILL.md                 # 说明文档
├── memory-fix.json          # Skill 配置
├── config-files/            # 核心配置（19个文件）
│   ├── SOUL.md             # AI 身份
│   ├── USER.md             # 用户信息
│   ├── MEMORY.md           # 长期记忆
│   ├── NOW.md              # 当前任务
│   ├── AGENTS.md           # Agent 规则
│   └── ...
└── scripts/
    └── auto-memlog.sh      # 自动记录工具

配置文件说明

文件	用途
SOUL.md	定义 AI 是谁、性格、行为准则
USER.md	记录用户基本信息、偏好
MEMORY.md	长期精选记忆
NOW.md	当前任务看板
AGENTS.md	多代理分工、记忆规则
FEEDBACK-LOG.md	错误记录与改进
TRAINING.md	边做边教机制

核心规则

记忆触发规则

1. 首次对话 → 创建当日记忆
2. 完成任务 → 自动记录
3. 用户说"记住" → 立即写入
4. 每天 Heartbeat → 整理到 MEMORY

置信度标准

置信度	行动
90-100%	确定回答
70-89%	提供依据
50-69%	提供备选
<50%	请求澄清

使用示例

记录工作

./auto-memlog.sh "完成任务A" "完成内容描述"

查看记忆

cat memory/2026-03-07.md

更新配置

编辑对应的 .md 文件即可

版本

• v1.0: 基础记忆系统
• v2.0: 完整11大功能 + 7大优化
• v3.0: 元认知 + 主动学习

作者

Odin（总舵主）

许可

免费使用