A structured two-tier memory system for AI agents that categorizes learnings into GOOD, BAD, and NEUTRAL entries to ensure a lean and searchable knowledge base.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install minimal-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 minimal-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 Minimal Memory Management?
Minimal Memory Management is a robust framework designed to keep Openclaw Skills running efficiently by organizing agent memory into two distinct tiers. It differentiates between long-term essential learnings stored in MEMORY.md and daily operational logs stored in dated files. By enforcing a strict categorization system, the skill ensures that only high-value insights are retained in the main memory file, preventing information bloat.
This skill is essential for developers using Openclaw Skills who want to maintain a clean, actionable, and semantic search-ready history of their agent's performance. It utilizes a tagging system—[GOOD], [BAD], and [NEUTRAL]—to help agents quickly identify successful patterns, avoid past mistakes, and manage transient context without cluttering the long-term knowledge base.
Minimal Memory Management Use Cases
- Managing agent memory when MEMORY.md grows too large or becomes disorganized.
- Deciding what technical context should be stored long-term vs. temporarily.
- Searching past agent experiences to avoid repeating errors.
- Automated memory cleanup and archiving of expired daily logs.
- Organizing project-specific knowledge for Openclaw Skills during long-running tasks.
How Minimal Memory Management Works
- Every new memory entry is tagged as [GOOD], [BAD], or [NEUTRAL] within a daily markdown file.
- The system uses specific scripts to search through categorized entries or create new daily templates.
- A weekly cleanup process migrates high-value [GOOD] and [BAD] entries to the main MEMORY.md file.
- [NEUTRAL] facts and temporary context are automatically archived after 30 days to prevent bloat.
- The agent references MEMORY.md first for established patterns and falls back to daily files for recent context.
Minimal Memory Management Setup
To integrate this into your environment for Openclaw Skills, use the following bash commands to initialize and manage your memory files:
# Create a daily memory file with the standard template
~/.openclaw/skills/minimal-memory/scripts/daily.sh
# Add a specific learning entry with auto-tagging
~/.openclaw/skills/minimal-memory/scripts/add.sh GOOD "Optimized database query performance"
# Run cleanup to migrate learnings to long-term storage
~/.openclaw/skills/minimal-memory/scripts/cleanup.sh
Minimal Memory Management Data Schema & Taxonomy
The skill organizes data into a tiered structure to ensure high-performance retrieval for Openclaw Skills:
| File Type | Purpose | Retention | Max Size |
|---|---|---|---|
MEMORY.md |
Long-term [GOOD]/[BAD] learnings | Permanent | < 150 lines |
memory/YYYY-MM-DD.md |
Daily operation logs with all tags | 30 days | Unlimited |
scripts/ |
Search, add, and cleanup utilities | Permanent | N/A |
All entries must use prefixes like [GOOD], [BAD], or [NEUTRAL] to remain compatible with the search scripts and migration logic.
Minimal Memory Management Advanced Features
- Semantic search capabilities across all historical memory files using the specialized search.sh script.
- Automated migration of successful patterns and failure avoidance strategies to long-term memory.
- Configurable expiration policies for [NEUTRAL] data to keep the Openclaw Skills workspace lean.
- Support for multi-agent coordination by standardizing memory formats across different sessions.
- Quick-tagging system that ensures every piece of information has a clear value proposition.
SKILL.md
Loading
Related Openclaw Skills
Workspace Anchor
A multi-agent workspace awareness and safety system that prevents agent drift by anchoring contexts through environment-based naming.
Doc - Sysadmin Ubuntu 24.04
An automated IT specialist skill for Ubuntu 24.04 designed to monitor system health, optimize performance, and perform safe disk maintenance.
claudemem
claudemem provides a persistent, local memory layer for AI agents to remember technical decisions, API specifications, and session context across multiple conversations.
AIEOS (AI Entity Object Specification)
AIEOS is a standardization framework that provides a consistent data structure for defining how AI agents speak, react, and remember across different ecosystems.
Remotion Video Generator
An AI-driven video production workflow that generates professional motion graphics programmatically using React and Remotion.
Skill Creator
A comprehensive development pipeline for designing, testing, and publishing high-performance AI agent skills with built-in best practices.
Featured*
