memory-system

What This Skill Does

The memory-system skill provides a standardized, robust framework for managing an AI agent's persistent memory. It bridges the gap between ephemeral chat sessions and long-term knowledge retention by implementing a multi-layered storage architecture. The skill covers the creation and maintenance of a central 'MEMORY.md' file for long-term facts, daily notes under the 'memory/' directory for raw data, and volatile workspace files like 'SESSION-STATE.md' and 'working-buffer.md' for task management and recovery. It is designed to work seamlessly with Obsidian, allowing for a hybrid workflow where the AI manages technical documentation and the user maintains a personal knowledge base.

Installation

To install this skill, run the following command in your terminal within your OpenClaw environment: clawhub install openclaw/skills/skills/cjke84/agent-memory-system-guide

Ensure your project repository has the required templates/ directory structure for 'SESSION-STATE.md', 'working-buffer.md', and 'OBSIDIAN-NOTE.md' to take full advantage of the automated setup scripts.

Use Cases

• Project Continuity: Ideal for developers or writers who need to pick up complex tasks over multiple days without repeating context.
• Agent Training: Perfect for refining an agent's understanding of user preferences, specific project constraints, and historical decision-making logs.
• Knowledge Management: Best for users who want their AI assistant to act as an extension of their Obsidian-based second brain.
• Error Prevention: Ensures that previous technical 'pitfalls' and their corresponding solutions are indexed for future reference.

Example Prompts

1. "记忆系统，请帮我初始化当前的记忆架构并根据我最近的三个工作日笔记更新 MEMORY.md。"
2. "memory-setup: 请检查我的 SESSION-STATE.md 是否符合模板契约，并提取今天已完成的任务更新到归档笔记中。"
3. "搭建记忆：我需要根据我的项目进展，创建一个新的工作区状态，确保我的卡点和决策记录被正确格式化。"

Tips & Limitations

• Maintenance: The system works best when you adhere to the distillation cadence. Regularly review your daily notes to prune noise, keeping MEMORY.md concise and high-value.
• Hierarchy: Always respect the retrieval order: Start with SESSION-STATE for task awareness, then recent daily notes, and finally use memory_search for deep, historical queries.
• Constraints: Do not create duplicate WAL (Write-Ahead-Log) files. Stick to the defined 'working-buffer.md' to avoid data fragmentation across different agent skills. This system is designed for local text-file management and does not natively sync across devices unless your workspace is synced via tools like Git or Obsidian Sync.