SkillAgentSearch skills...

Openclaw Memory

No description available

Install / Use

/learn @aquiladev/Openclaw Memory
About this skill

Quality Score

0/100

Supported Platforms

Universal

README

OpenClaw Memory Management Guide

A comprehensive guide to understanding, configuring, and optimizing memory in OpenClaw agents — so your agent stops forgetting everything.

Based on insights from an OpenClaw codebase maintainer with 2+ months of daily production use. Source: Why your OpenClaw agent forgets everything (and how to fix it)

What's Inside

The Guide

A complete reference covering:

  • The four memory layers — bootstrap files, session transcript, LLM context window, and retrieval index
  • Three failure modes — why your agent forgets and how to diagnose each one
  • Compaction vs. pruning — the difference most guides get wrong
  • Configuration — recommended openclaw.json settings with reserveTokensFloor, memory flush, and hybrid search
  • File architecture — what goes in SOUL.md, AGENTS.md, USER.md, MEMORY.md, and daily logs
  • The memory protocol — a rule for AGENTS.md that makes the agent search before acting
  • Retrieval tracks — Track A (built-in search), Track A+ (extra paths), Track B (QMD for large vaults)
  • Diagnostics — using /context list, /status, /compact, and /verbose
  • Memory hygiene — daily, weekly, and git backup practices

Claude Code Skill

A ready-to-use skill for Claude Code that helps you manage OpenClaw memory:

skills/
├── SKILL.md                        # Main skill instructions
└── references/
    └── config-examples.md          # Copy-paste config templates

To install: Copy the skills/ directory into your Claude Code skills location (e.g., ~/.claude/skills/openclaw-memory/ or your project's .claude/skills/ directory).

The skill triggers whenever you mention memory issues, compaction, context loss, bootstrap files, or any OpenClaw memory-related topic. It guides Claude through diagnostics, configuration, and best practices.

Quick Start

If you read nothing else, do these three things:

  1. Put durable rules in files, not chat. Write important instructions to MEMORY.md and AGENTS.md. Chat instructions don't survive compaction.

  2. Enable and tune the memory flush. Add this to ~/.openclaw/openclaw.json:

    {
      "agents": {
        "defaults": {
          "compaction": {
            "reserveTokensFloor": 40000,
            "memoryFlush": {
              "enabled": true,
              "softThresholdTokens": 4000,
              "systemPrompt": "Session nearing compaction. Store durable memories now.",
              "prompt": "Write any lasting notes to memory/YYYY-MM-DD.md; reply with NO_REPLY if nothing to store."
            }
          }
        }
      }
    }
    
  3. Make retrieval mandatory. Add to your AGENTS.md:

    ## Memory Protocol
    Before doing anything non-trivial, search memory first.
    

License

MIT

View on GitHub
GitHub Stars0
CategoryDevelopment
Updated17d ago
Forks0

Security Score

80/100

Audited on Mar 10, 2026

No findings