A disciplined, five-layer file-based memory system that gives AI agents permanent recall and structured context management without external APIs.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install afrexai-agent-memory
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install afrexai-agent-memory using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Agent Memory Architecture?
The Agent Memory Architecture is a comprehensive framework designed to provide AI agents with a robust, zero-dependency memory system. Instead of relying on volatile session history or complex external databases, this architecture uses a structured hierarchy of Markdown files to store everything from immediate tasks to long-term curated knowledge. By incorporating these Openclaw Skills, developers can ensure their agents maintain continuity across sessions, learn from past mistakes, and manage large amounts of context efficiently.
This system operates on the principle that text is permanent while mental notes are fleeting. It organizes information into five distinct layers—Hot, Warm, Daily, Topic, and Cold—allowing the agent to prioritize what it needs to know right now while archiving historical data for future retrieval. This approach makes Openclaw Skills more reliable, human-aware, and capable of handling complex, long-running projects.
Agent Memory Architecture Use Cases
- Maintaining task priorities and session state across multiple agent restarts.
- Building a permanent knowledge base of user preferences, communication styles, and project history.
- Managing deep technical context for specialized projects without bloating the primary context window.
- Creating a searchable audit trail of agent decisions and lessons learned over time.
- Facilitating seamless session handoffs when context limits are reached in complex Openclaw Skills workflows.
How Agent Memory Architecture Works
- The agent begins every session by executing a startup protocol, reading identity files and the ACTIVE-CONTEXT.md file to establish immediate priorities.
- Throughout the interaction, the agent logs every significant event, decision, and outcome to a daily note file (memory/YYYY-MM-DD.md).
- Critical insights or permanent facts are promoted to the long-term MEMORY.md file to ensure they are always available in main sessions.
- For specialized subjects, the agent creates and updates topic-specific files in the memory/topics directory to keep the main memory file lean.
- Periodic hygiene tasks or heartbeats trigger the agent to prune the active context, extract lessons to long-term storage, and archive old daily notes into quarterly summaries.
Agent Memory Architecture Setup
To implement this architecture within your Openclaw Skills environment, initialize the required file structure using the following commands:
mkdir -p memory/topics memory/archive
touch MEMORY.md ACTIVE-CONTEXT.md AGENTS.md SOUL.md USER.md
Once the files are created, populate the SOUL.md and USER.md files with identity context, and configure your agent's system instructions to follow the session startup and write protocols defined in the architecture.
Agent Memory Architecture Data Schema & Taxonomy
The architecture utilizes a tiered Markdown schema to organize data by volatility and relevance:
| Layer | Primary File | Purpose |
|---|---|---|
| Hot | ACTIVE-CONTEXT.md | Current priorities, active tasks, and blockers (Max 50 lines). |
| Warm | MEMORY.md | Curated long-term knowledge, preferences, and project status (Max 500 lines). |
| Daily | memory/YYYY-MM-DD.md | Raw event logs, timestamps, and daily takeaways. |
| Topic | memory/topics/*.md | Deep-dive context for specific projects or technical domains. |
| Cold | memory/archive/*.md | Historical quarterly summaries and project post-mortems. |
Agent Memory Architecture Advanced Features
- Heartbeat Integration: Leverages periodic agent wake-ups to perform automated memory audits and hygiene without human intervention.
- Context Overflow Protocol: Uses a HANDOFF.md pattern to transfer state between sessions when token limits are exceeded.
- Progressive Disclosure: A tiered loading strategy that reduces token costs by only loading essential files for specific session types (e.g., group vs. private chats).
- Write-Ahead Protocol: Ensures data persistence by recording instructions to daily notes before executing them to prevent loss during crashes.
- Semantic Search Support: Enables Openclaw Skills to perform fuzzy queries across the file-based knowledge web for precise information retrieval.
SKILL.md
Loading
Related Openclaw Skills
AI Agent Manager Playbook
A strategic operational framework for managing, monitoring, and scaling autonomous AI agent workforces within an organization.
Agent Engineering
A comprehensive framework for designing, deploying, and operating production-ready AI agent systems and autonomous swarms.
Accounts Receivable Manager
Automate accounts receivable workflows including aging analysis, collection prioritization, and bad debt forecasting to optimize business cash flow.
Accounts Payable Automation Framework
A sophisticated automation framework for managing the end-to-end accounts payable lifecycle, from invoice OCR to fraud prevention.
Agent Observability & Monitoring
A professional framework to score, monitor, and troubleshoot production AI agent fleets across six critical performance dimensions.
Agent Ops Runbook
A comprehensive framework for generating production-ready operations runbooks to deploy, monitor, and scale AI agents safely.
Featured*
