memory-management
Manage tri-tier agent memory (hot/warm/cold) with event consolidation and weekly cleanup. Use for memory updates after completing tasks.
When & Why to Use This Skill
This Claude skill provides a sophisticated tri-tier memory architecture—comprising hot, warm, and cold storage—to optimize agent persistence and contextual intelligence. It enables agents to maintain a high-performance workspace by automatically consolidating recent events, preserving successful strategies in mid-term storage, and archiving historical patterns for long-term reference, ensuring that the most relevant information is always at the agent's fingertips without overwhelming the context window.
Use Cases
- Continuous Learning: Automatically capturing 'lessons learned' and outcomes after every task to build a persistent knowledge base of what works and what doesn't.
- Context Window Optimization: Managing token usage and focus by migrating older events from 'hot' memory to 'warm' or 'cold' storage based on frequency and recency.
- Long-term Project Tracking: Maintaining continuity across multi-week projects by retrieving historical patterns and known issues from archived memory tiers.
- Strategy Refinement: Storing successful execution patterns in warm memory to allow the agent to reuse proven workflows for recurring technical tasks.
| name | memory-management |
|---|---|
| description | Manage tri-tier agent memory (hot/warm/cold) with event consolidation and weekly cleanup. Use for memory updates after completing tasks. |
Memory Management - Tri-Tier System
Memory Structure
{
"hot_memory": {
"_description": "Last 20 events - always loaded",
"recent_events": [],
"active_tasks": [],
"recent_learnings": []
},
"warm_memory": {
"_description": "Events 21-100 - patterns",
"patterns": [],
"known_issues": [],
"successful_strategies": []
},
"cold_memory": {
"_description": "Events 101+ - archived",
"historical_patterns": []
}
}
When to Update
After Task Completion
Update hot_memory.recent_events:
- What was done
- Outcome
- Lessons learned
Weekly Consolidation
hot → warm (after 20 events)
warm → cold (after 100 events)
Update Pattern
{
"recent_events": [
{
"date": "2025-11-24",
"task": "Built FastAPI endpoint",
"outcome": "Success",
"lessons": ["Async patterns work well", "Remember CORS"]
}
]
}