Claude Memory for Openclaw

A structured, layered memory architecture designed to give AI agents long-term persistence and organized context management.

kaiji-z
v1.2.0
Apr 5, 2026
0
1.4k
0

Install & Download

1. ClawHub CLI

The fastest way to install a skill directly from the registry.

npx clawhub@latest install claude-memory

2. Manual Installation

Copy the skill folder to one of these locations

Global
~/.openclaw/skills/
Workspace
<project>/skills/

Priority: Workspace > Local > Bundled

3. Prompt Installation

Copy this prompt to OpenClaw to install it automatically.

Help me install claude-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 Claude Memory?

Claude Memory is a sophisticated framework for managing an agent's long-term intelligence and context. By utilizing Openclaw Skills, this system implements a 4-type classification—user, feedback, project, and reference—to ensure that information is not just stored, but effectively prioritized. It solves the problem of token limits by using a hybrid approach: high-frequency data stays in a lean MEMORY.md file, while deeper, low-frequency details are offloaded to specialized topic files.

This skill transforms how agents interact by allowing them to remember human corrections and confirmations instantly. It moves beyond simple chat history into a professional memory management system that maintains a heartbeat routine to consolidate notes, verify technical states, and tidy up data structures for maximum efficiency.

Claude Memory Use Cases

  • Establishing a permanent memory system for new Openclaw Skills agents.
  • Improving agent reliability by recording specific human feedback and confirmed approaches.
  • Managing complex technical project states that persist across multiple conversation sessions.
  • Organizing large volumes of reference material without exhausting LLM token windows.

How Claude Memory Works

  1. The system initializes a standardized directory structure including a primary MEMORY.md file and a memory/ directory for feedback and topics.
  2. Information is classified into four distinct types to determine its location and retrieval priority.
  3. The agent follows strict write rules to record human feedback immediately, ensuring that mistakes are not repeated and successes are reinforced.
  4. A heartbeat maintenance routine is triggered to consolidate daily notes into long-term topic files and verify that technical information remains accurate.
  5. When responding to queries, the agent uses progressive disclosure, reading detailed topic files only when specifically referenced in the high-level memory index.

Claude Memory Setup

To deploy this memory system within your Openclaw Skills environment, execute the initialization script:

python scripts/init_memory.py <workspace-path>

After running the script, you must manually update your AGENTS.md to incorporate the memory management rules and configure the HEARTBEAT.md for autonomous maintenance routines.

Claude Memory Data Schema & Taxonomy

The architecture uses a specific hierarchy to balance speed and depth:

Component Max Size Function
MEMORY.md 10KB High-frequency info and pointers to topics
feedback.md Unlimited Critical human corrections and confirmed approaches
topics/ Unlimited Deep-dive technical resources and domain-specific knowledge
YYYY-MM-DD.md N/A Raw daily notes awaiting consolidation

All entries are tagged with absolute dates to ensure they remain interpretable over long periods of time.

Claude Memory Advanced Features

  • Confirmation Recording: Unlike basic systems that only track errors, this skill records positive validations to prevent agents from becoming overly cautious.
  • Heartbeat Grooming: A 4-phase routine (Catch-up, Consolidate, Verify, Tidy) that manages memory health during background polling.
  • Selective Verification: Automatically identifies which memories need real-time validation (like service configs) versus stable facts (like user history).
  • Hybrid Retrieval: Combines inline readability for speed with external pointers for technical depth.

SKILL.md


Loading

Related Openclaw Skills

METADATA

Github Stars: 0
forks: 0

Featured*