Why is there a 200-line budget for MEMORY.md?

The budget acts as implicit decay. It forces the agent to prioritize the most durable and important facts, ensuring the memory file stays within the LLM's efficient processing range.

Can multiple agents share the same memory in OpenClaw?

Yes, the skill supports three shared patterns: a team-memory directory for domain files, GROUP_MEMORY.md for shared channels, and SHARED_KNOWLEDGE.json for system-wide machine-readable facts.

What is the SCRIBE compression pattern?

SCRIBE is a weekly maintenance cycle that extracts durable patterns and facts from daily logs to update a curated MEMORY.md file, ensuring essential knowledge survives while noise is discarded.

What are the four layers of the memory model?

The architecture includes Layer 1 (Permanent bootstrap files), Layer 1.5 (Semi-permanent daily logs), Layer 2 (Temporary session transcripts), and Layer 3 (Permanent vector retrieval index).

How does the Pre-compaction Flush work?

When context usage hits 80%, the skill identifies unsaved facts in the current session and writes them to persistent storage before the LLM's context window is summarized or cleared.

OpenClaw Memory Management

Name: OpenClaw Memory Management
Author: TheophilusChinomona

byTheophilusChinomona

0•

データサイエンスとML

Manages persistent AI agent memory using a multi-layer architecture to prevent information loss and optimize context usage.

This skill provides a comprehensive framework for managing AI agent memory within the OpenClaw gateway. It implements a robust four-layer memory model—comprising bootstrap files, daily logs, session transcripts, and a vector-based retrieval index—to ensure critical data survives context compaction and session resets. It includes specialized patterns like SCRIBE compression for long-term knowledge curation and shared memory structures for cross-agent collaboration, making it essential for developers building complex, multi-agent autonomous systems where data persistence and factual accuracy are paramount.

主な機能

010 GitHub stars

02Four-layer memory architecture (Bootstrap, Daily Logs, Session, Vector DB)

03Pre-compaction flush mechanism to prevent data loss during context resets

04SCRIBE compression for weekly long-term memory curation and pruning

05Shared memory patterns for seamless team-based agent collaboration

06LanceDB vector store integration for semantic memory retrieval

ユースケース

01Building persistent autonomous agents that retain user preferences over months

02Optimizing LLM context windows by compressing raw logs into durable facts

03Coordinating multi-agent teams using shared memory directories and knowledge bases

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add theophiluschinomona/claude-openclaw-plugin openclaw-memory

For use in Claude.ai and ChatGPT

主な機能

010 GitHub stars

02Four-layer memory architecture (Bootstrap, Daily Logs, Session, Vector DB)

03Pre-compaction flush mechanism to prevent data loss during context resets

04SCRIBE compression for weekly long-term memory curation and pruning

05Shared memory patterns for seamless team-based agent collaboration

06LanceDB vector store integration for semantic memory retrieval

ユースケース

01Building persistent autonomous agents that retain user preferences over months

02Optimizing LLM context windows by compressing raw logs into durable facts

03Coordinating multi-agent teams using shared memory directories and knowledge bases

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add theophiluschinomona/claude-openclaw-plugin openclaw-memory

For use in Claude.ai and ChatGPT