jzOcb Memory Management Skill
Manage agent memory efficiently using a three-layer architecture with automatic archival.
Quick Setup
- Format MEMORY.md with priority tags:
- [P0] Core identity item (never expires)
- [P1][2026-02-10] Active project (90-day TTL)
- [P2][2026-02-10] Temporary item (30-day TTL)
- Set up auto-archive cron:
0 4 * * * python3 ~/.openclaw/workspace/scripts/memory-janitor.py
- Store lessons in
memory/lessons/*.jsonl:
{"id": "lesson-001", "date": "2026-02-10", "category": "infra", "title": "Problem title", "problem": "What happened", "solution": "How to fix", "tags": ["tag1"]}
Three-Layer Architecture
Layer 1: Hot Memory (MEMORY.md)
- Always loaded, ≤200 lines
- P0: Core identity, never expires
- P1: Active projects, 90-day TTL
- P2: Temporary, 30-day TTL
Layer 2: Cold Memory (searchable)
memory/lessons/*.jsonl— structured lessonsmemory/archive/— expired content- Use
memory_searchto recall
Layer 3: Raw Logs
memory/YYYY-MM-DD.md— daily logs- Not loaded automatically
Priority Guidelines
| Priority | Use For | TTL |
|---|---|---|
| P0 | User identity, preferences, safety rules | Never |
| P1 | Active projects, current strategies | 90 days |
| P2 | Debug notes, one-time events | 30 days |
Core Principles (max 5 in AGENTS.md)
Keep only essential rules in AGENTS.md. Other lessons go to lessons/*.jsonl.
Example 5 rules:
- Real money = correctness > speed
- External actions require confirmation
- Check both cron systems (system + OpenClaw)
- Long-running processes need setsid isolation
- Read platform rules before trading
When Memory Gets Too Large
Run janitor with stats:
python3 scripts/memory-janitor.py --stats
If >200 lines:
- Review P0 entries — are they all truly permanent?
- Add dates to P1/P2 entries missing them
- Move detailed content to
lessons/*.jsonl - Run
--dry-runthen archive