triple-memory-skill（三重记忆能力）

Triple Memory System

A comprehensive memory architecture combining three complementary systems for maximum context retention across sessions.

Architecture Overview

User Message
     ↓
[LanceDB auto-recall] → injects relevant conversation memories
     ↓
Agent responds (using all 3 systems)
     ↓
[LanceDB auto-capture] → stores preferences/decisions automatically
     ↓
[Git-Notes] → structured decisions with entity extraction
     ↓
[File updates] → persistent workspace docs

The Three Systems

1. LanceDB (Conversation Memory)

• Auto-recall: Relevant memories injected before each response
• Auto-capture: Preferences/decisions/facts stored automatically
• Tools: memory_recall, memory_store, memory_forget
• Triggers: "remember", "prefer", "my X is", "I like/hate/want"

2. Git-Notes Memory (Structured, Local)

• Branch-aware: Memories isolated per git branch
• Entity extraction: Auto-extracts topics, names, concepts
• Importance levels: critical, high, normal, low
• No external API calls

3. File Search (Workspace)

• Searches: MEMORY.md, memory/*.md, any workspace file
• Script: scripts/file-search.sh

Setup

Enable LanceDB Plugin

{
  "plugins": {
    "slots": { "memory": "memory-lancedb" },
    "entries": {
      "memory-lancedb": {
        "enabled": true,
        "config": {
          "embedding": { "apiKey": "${OPENAI_API_KEY}", "model": "text-embedding-3-small" },
          "autoRecall": true,
          "autoCapture": true
        }
      }
    }
  }
}

Enable Auto Memory Flush (Pre-Compaction)

Add to your Clawdbot config to automatically preserve context before compaction:

{
  "agents": {
    "defaults": {
      "compaction": {
        "mode": "safeguard",
        "memoryFlush": {
          "enabled": true,
          "softThresholdTokens": 8000,
          "systemPrompt": "Session nearing compaction. Preserve context using triple memory system (git-notes, LanceDB, files).",
          "prompt": "Context is at ~80%. Write session summary to memory/YYYY-MM-DD.md, update MEMORY.md if needed, store key facts to git-notes. Reply NO_REPLY when done."
        }
      }
    }
  }
}

This triggers automatic memory dump when context reaches ~80%, before any information is lost to compaction.

Install Git-Notes Memory

clawdhub install git-notes-memory

Create File Search Script

Copy scripts/file-search.sh to your workspace.

Usage

Session Start (Always)

python3 skills/git-notes-memory/memory.py -p $WORKSPACE sync --start

Store Important Decisions

python3 skills/git-notes-memory/memory.py -p $WORKSPACE remember \
  '{"decision": "Use PostgreSQL", "reason": "Team expertise"}' \
  -t architecture,database -i h

Search Workspace Files

./scripts/file-search.sh "database config" 5

Conversation Memory (Automatic)

LanceDB handles this automatically. Manual tools:

• memory_recall "query" - search conversation memory
• memory_store "text" - manually store something
• memory_forget - delete memories (GDPR)

Importance Levels

When to Use Each System

File Structure

workspace/
├── MEMORY.md              # Long-term curated memory
├── memory/
│   ├── active-context.md  # Current session state
│   └── YYYY-MM-DD.md      # Daily logs
├── scripts/
│   └── file-search.sh     # Workspace search
└── skills/
    └── git-notes-memory/  # Structured memory

Silent Operation

Never announce memory operations to users. Just do it:

• ❌ "I'll remember this"
• ❌ "Saving to memory"
• ✅ (silently store and continue)