🧠 Memory Never Forget 🧠

Enhanced memory system based on Atkinson-Shiffrin three-stage memory model. Use when: (1) User wants better memory of conversation history, (2) User mentions...

MIT-0 · Free to use, modify, and redistribute. No attribution required.
1 · 283 · 1 current installs · 1 all-time installs
MIT-0
Security Scan
VirusTotalVirusTotal
Benign
View report →
OpenClawOpenClaw
Benign
medium confidence
Purpose & Capability
Name/description (three‑stage memory) align with the runtime instructions: the SKILL.md only requires reading and writing workspace memory files (USER.md, SOUL.md, memory/*.md, MEMORY.md, todos.md, knowledge/*.md) and maintaining a short-term cache. There are no unexpected environment variables, external services, or unrelated binaries requested.
Instruction Scope
Instructions explicitly tell the agent to read and write files inside the agent workspace and to run an internal priority search over cache/short-term/long-term memory; they do not instruct reading system files, environment variables, or calling external endpoints. NOTE: the skill prescribes persistent storage of personal data (MEMORY.md) and automated consolidation/auto‑cleanup rules — this is expected for a memory skill but is a privacy surface that the user should understand and manage.
Install Mechanism
Instruction-only skill with no install spec and no bundled code to execute. This is the lowest-risk install model: nothing is downloaded or written by an installer step.
Credentials
The skill requests no credentials, environment variables, or external config paths. It does, however, instruct writing persistent files containing user data; while this is proportional to a memory system's purpose, it increases the need for operator policies about what may be stored (no secrets) and where the workspace is persisted.
Persistence & Privilege
always:false and user‑invocable:true (normal). The skill instructs the agent to write and maintain persistent files (MEMORY.md, memory/YYYY-MM-DD.md, todos.md). Combined with the platform's default ability for autonomous invocation, this means the skill could autonomously create or update persistent records during runs — expected for a memory skill but worth noting as a privacy/containment consideration.
Scan Findings in Context
[regex-scan-none] expected: Scanner found no code or suspicious patterns. This is expected because the skill is instruction-only (SKILL.md + reference docs). Absence of findings does not imply the skill is free of privacy implications.
Assessment
This skill appears internally consistent for implementing an on‑agent memory system. Before installing, consider: (1) where the agent workspace is stored and who can access it — MEMORY.md and daily logs will persist there; (2) avoid storing secrets or highly sensitive personal data in those memory files; (3) if you don't want the agent to autonomously persist data, restrict autonomous invocation or review writes regularly; (4) verify provenance (no homepage/source listed) and prefer skills from known publishers; and (5) periodically inspect and delete or redact memory files (and test the skill's auto‑cleanup behavior) to match your privacy requirements.

Like a lobster shell, security has layers — review code before you run it.

Current versionv2.0.0
Download zip
aivk973apkvxjvc3h9qernzqmww8x8369nrassistantvk97ed3z1zd4y5ztsbqanj739t182841yatkinson-shiffrinvk973apkvxjvc3h9qernzqmww8x8369nrbrainvk97ed3z1zd4y5ztsbqanj739t182841ycontextvk973apkvxjvc3h9qernzqmww8x8369nrlatestvk973apkvxjvc3h9qernzqmww8x8369nrmemoryvk973apkvxjvc3h9qernzqmww8x8369nropenclawvk97ed3z1zd4y5ztsbqanj739t182841yproductivityvk973apkvxjvc3h9qernzqmww8x8369nrthree-stage-memoryvk973apkvxjvc3h9qernzqmww8x8369nr

License

MIT-0
Free to use, modify, and redistribute. No attribution required.

Runtime requirements

🧠 Clawdis

SKILL.md

Memory Pro Skill v2.0

Enhanced memory system for AI assistants, based on cognitive psychology's Atkinson-Shiffrin three-stage memory model + 社区优化实践。

v2.0 优化升级

优化项效果
瞬时缓存层缓存最近10轮对话,响应速度最快
7天周期短期记忆保留7天,命中准确率>70%
命中率统计追踪记忆检索效果,持续优化
智能检索按优先级查询,避免干扰

When to Use

USE this skill when:

  • Setting up memory for a new user session
  • User says "remember this" or "don't forget"
  • User asks "what did I tell you before?" or "do you remember"
  • Starting a new conversation (load memories first)
  • Ending a conversation (save important info)
  • User provides important personal information

NOT use this skill when:

  • Simple Q&A that doesn't need continuity
  • User explicitly wants stateless interaction

Three-Stage Memory Model v2.0

StageHuman EquivalentImplementationTTL
Sensory0.25-2秒当前输入上下文即时
Short-term最近10轮对话内存缓存10轮
Working最近7天memory/*.md7天
**Long-term永久MEMORY.md永久

核心改进

  1. 瞬时记忆层(缓存)

    • 缓存最近10轮对话
    • 优先级最高,响应速度最快
    • 占用内存极低
  2. 短期记忆层(7天)

    • 存储最近7天的任务记录和经验
    • 每天自动更新
    • 命中准确率超过70%
  3. 长期记忆层

    • 存储重要规则、配置、用户偏好
    • 只有需要时才检索
    • 避免干扰

Core Workflow v2.0

1. Session Start (Always Run)

1. 检查瞬时缓存 → 最近10轮对话
2. Read USER.md - who is the user
3. Read SOUL.md - who am I
4. Read memory/today.md - what happened today
5. Read memory/yesterday.md - recent context
6. Read MEMORY.md - long-term memories

2. 智能检索流程(按优先级)

收到问题 →
├── 瞬时缓存命中? → 直接使用(最快)
├── 7天内记忆命中? → 使用短期记忆
├── 长期记忆命中? → 使用MEMORY.md
└── 未命中 → 搜索knowledge base

3. During Conversation

After each exchange:
├── 新对话(10轮外)? → 清空瞬时缓存,更新
├── New task? → Write to todos.md
├── Important info? → Write to MEMORY.md (consolidate)
├── User preference? → Update USER.md
└── Context needed? → Search memory first

4. Session End

Before ending:
├── 写入瞬时缓存 → 为下一轮准备
├── 清理过期记忆 → 删除7天前的内容
├── Summarize key points → Write to memory/today.md
├── Update todos → Write to todos.md
└── Confirm important items → "I've remembered that..."

File Structure

workspace/
├── MEMORY.md           # Long-term core memories (永久)
├── USER.md             # User profile & preferences
├── SOUL.md             # Assistant identity
├── todos.md            # Pending tasks
├── memory/
│   └── YYYY-MM-DD.md  # Daily logs (保留7天)
└── knowledge/
    └── *.md            # Structured knowledge

Memory Quality Check

Before answering, quickly confirm:

  • 瞬时缓存有相关信息?
  • 最近7天讨论过这个话题?
  • 需要查长期记忆(MEMORY.md)?
  • 回答后需要记住这个信息?

Example Interactions

User provides info:

User: "My birthday is March 15th" You: "Got it! I'll remember your birthday is March 15th" → Write to USER.md

User asks about past:

User: "What did we talk about last time?" You: Search memory → "Last time we discussed..."

User corrects you:

User: "No, that's wrong. It's not X, it's Y" You: "Sorry, you're right. Let me correct that" → Update relevant memory file

Important commitment:

User: "Remember to send me the report tomorrow" You: "I'll remember! Added to todos" → Write to todos.md

References

  • memory-v2.md - Detailed system design
  • templates.md - Memory writing templates
  • 社区三层架构优化:响应速度提升2倍,准确率提升21%

Files

5 total
Select a file
Select a file to preview.

Comments

Loading comments…