An intelligent multi-store memory system for AI agents featuring human-like encoding, reflection cycles, and decay-based forgetting.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install cognitive-memory-temp
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 cognitive-memory-temp 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 Cognitive Memory System?
The Cognitive Memory System is a sophisticated architectural framework designed to move beyond basic flat-file memory for autonomous agents. It implements a layered cognitive stack including episodic, semantic, procedural, and core memory stores, allowing AI agents to retain context, learn from historical interactions, and evolve their internal identity over time. By integrating this system into your Openclaw Skills workflow, you empower agents to handle complex, long-term tasks with human-like recall and philosophical self-reflection.
This architecture ensures that the most relevant information remains in the context window while less important data is archived through a mathematical decay model. It provides a robust foundation for building sophisticated agents that don't just process commands, but grow through experience.
Cognitive Memory System Use Cases
- Persistent personalization across multiple agent sessions based on user preferences.
- Knowledge graph construction and entity relationship tracking from ongoing conversations.
- Automated reflection and self-improvement cycles for long-running autonomous agents.
- Multi-agent coordination requiring shared memory access with controlled write permissions.
- Establishing a permanent audit trail for agent decisions and knowledge mutations.
How Cognitive Memory System Works
- The Trigger Engine monitors natural language inputs for keywords like remember or forget to route information to the appropriate memory store.
- Data is categorized into specialized stores: Core (always loaded), Episodic (event logs), Semantic (knowledge graph), or Procedural (learned workflows).
- A Decay Model automatically calculates relevance scores based on access frequency and time, deprioritizing stale information.
- Reflection Cycles periodically consolidate episodic logs into semantic knowledge, updating the agent self-image and soul files.
- The Multi-Agent Layer manages shared read access and a gated write model where sub-agents propose memories for the primary agent to review.
- The Audit System tracks every file mutation using Git commits and structured logs for full transparency.
Cognitive Memory System Setup
To get started with this system for Openclaw Skills, run the initialization script to prepare your workspace:
bash scripts/init_memory.sh /path/to/workspace
Next, update your global configuration file (e.g., ~/.clawdbot/clawdbot.json) to enable the memory provider:
{
"memorySearch": {
"enabled": true,
"provider": "voyage",
"sources": ["memory", "sessions"],
"indexMode": "hot",
"minScore": 0.3,
"maxResults": 20
}
}
Finally, append the agent instruction block located in assets/templates/agents-memory-block.md to your project's AGENTS.md file.
Cognitive Memory System Data Schema & Taxonomy
The system organizes data using a hierarchical file structure to balance context window limits with deep retrieval:
| Component | File Path | Description |
|---|---|---|
| Core Memory | MEMORY.md |
Critical facts (~3K tokens) always present in the context window. |
| Identity | IDENTITY.md |
Defines facts, self-image, and logs self-awareness entries. |
| Soul | SOUL.md |
Encapsulates core values, principles, and behavioral boundaries. |
| Episodes | memory/episodes/ |
Chronological logs of interactions organized by date. |
| Knowledge Graph | memory/graph/ |
Stores entities and relationships as distinct Markdown files. |
| Meta Tracking | memory/meta/ |
Contains decay scores, reflection archives, and reward logs. |
Cognitive Memory System Advanced Features
- Philosophical Internal Monologue: Agents engage in third-person self-examination to deepen understanding of the user and themselves.
- Token-Based Evolution: A reward system where users grant token budgets for agents to perform deeper reflection cycles.
- Decay-Based Forgetting: Mathematical relevance scoring that naturally archives dormant memories to save context space.
- Gated Multi-Agent Memory: A robust proposal system allowing sub-agents to contribute to the long-term memory without corrupting core data.
- Self-Image Consolidation: Periodic analysis of self-awareness logs to rewrite and compact the agent's internal identity.
SKILL.md
Loading
Related Openclaw Skills
Fund Investment Advisory Assistant (fin-advisor)
A specialized AI investment assistant that delivers professional fund analysis, market comparisons, and data-driven decision support.
Brave API Search
A powerful suite of tools for real-time web search, intelligent query suggestions, and AI-driven answers grounded in live web data.
Excel Generator Skill
An intelligent data assistant for AI agents to generate professionally formatted Excel spreadsheets from structured data or chat inputs.
GitLab MR Review
An automated code review utility that analyzes GitLab merge requests and posts standardized feedback via CLI integration.
IMAP IDLE Email Notifier
A real-time email monitoring skill that uses IMAP IDLE to push instant notifications to Feishu.
QC Deep Feature Forensics
A 12-dimensional technical feature attribution engine that identifies the market microstructure conditions separating winning trades from losing ones.
Featured*
