MetaClaw Skill - Memory Management & Skill Evolution#

Use when: Memory-Suche mit Hybrid Retrieval (LanceDB + FTS), automatische Memory-Extraktion, PluginEval Quality Checks, oder Skill Evolution. Dieser Skill bündelt die besten MetaClaw-Konzepte für generische Memory-Systeme.

---

Overview#

MetaClaw ist ein generischer Skill für fortgeschrittene Memory-Management-Lösungen mit:

• Hybrid Retrieval - Kombiniert semantische Suche (LanceDB) mit FTS/Keyword-Suche
• Memory Extraction - Automatische Extraktion aus Sessions in preferences.md, project-state.md
• Session Counter - Tracking für konsolidierte Memory-Konsolidation (Trigger: N Sessions)
• PluginEval Integration - Auto-Fix für Quality Badge ≥ Gold (≥85)
• Auto-Dream Consolidation - Automatische Memory-Konsolidierung basierend auf Session-Zähler

---

Installation#

bash

# VEnv für Dependencies erstellen (wird automatisch verwendet)
cd ~/.openclaw/workspace/skills/metaclaw
python3 -m venv .venv-metaclaw
source .venv-metaclaw/bin/activate
pip install lancedb sentence-transformers

---

Usage#

CLI Interface#

bash

# Hybrid Retrieval (default: 0.6 semantic + 0.4 keyword)
python3 scripts/memory-vector-wrapper.py "query" --mode hybrid

# Semantic Search
python3 scripts/memory-vector-wrapper.py "query" --mode semantic

# Keyword Search (FTS)
python3 scripts/memory-vector-wrapper.py "query" --mode keyword

# Memory Extraction
python3 scripts/memory-extract.py [--date YYYY-MM-DD] [--dry-run]

# Session Counter
./scripts/session-counter.sh --increment
./scripts/session-counter.sh --check  # exit 0 = consolidation ready

# PluginEval Quality Check
python3 scripts/skill-eval.py --layer1 skills/metaclaw/

# Auto-Fix (mit --allow-write zum Schreiben)
python3 scripts/skill-eval.py --auto-fix --allow-write skills/metaclaw/

Output#

Memory Vector Search (JSON)

json

{
  "id": "memory/2026-04-01.md",
  "path": "/Users/donmeusi/.openclaw/workspace/memory/2026-04-01.md",
  "snippet": "Extrahierter Inhalt...",
  "date": "2026-04-01",
  "type": "daily",
  "tags": [],
  "score": 0.0,
  "relevance": 0.0,
  "method": "hybrid",
  "combined_score": 0.0
}

Memory Extraction (JSON)

json

{
  "extracted": "2026-04-01",
  "type": "project_state",
  "confidence": 0.85,
  "items": [
    {
      "category": "Project",
      "content": "MetaClaw Integration Phase 1",
      "type": "update",
      "source": "session"
    }
  ],
  "preferences": [],
  "project_state": []
}

Session Counter

bash

$ ./scripts/session-counter.sh --check
# exit 0: consolidation ready
# exit 1: not ready yet

---

Config#

memory/config.yaml#

yaml

# Hybrid Retrieval Settings
retrieval:
  default_mode: "hybrid"  # semantic, keyword, fts, hybrid
  semantic_weight: 0.6    # 0.6 semantic + 0.4 keyword/fts
  default_limit: 5

# Consolidation Settings
consolidation:
  interval_sessions: 10   # alle N Sessions
  min_interval_hours: 6   # min 6h zwischen Consolidations
  max_memory_lines: 200   # Maximum Zeilen in MEMORY.md

# Logging
logging:
  level: "INFO"  # DEBUG, INFO, WARNING, ERROR
  file: "~/.openclaw/workspace/memory/.log"

---

Auto-Dream Consolidation#

Trigger-Logik#

• Session Counter: Zählt Sessions (default: 10)
• Mindestintervall: 6 Stunden zwischen Consolidations
• Auto-Detect: In memory-digest-enhanced.sh

Cron-Job (Optional)#

bash

# crontab -e
# Tägliche Auto-Dream Consolidation (03:00)
0 3 * * * "$HOME/.openclaw/workspace/scripts/memory-digest-enhanced.sh" --auto-dream

---

PluginEval Integration#

Quality Check (Layer 1)#

bash

# Read-Only Check
python3 scripts/skill-eval.py --layer1 skills/metaclaw/

# Anti-Pattern Detection
python3 scripts/skill-eval.py --anti-patterns skills/metaclaw/

# Auto-Fix Preview (Read-Only)
python3 scripts/skill-eval.py --auto-fix skills/metaclaw/

# Auto-Fix (mit Schreiben)
python3 scripts/skill-eval.py --auto-fix --allow-write skills/metaclaw/

Quality Targets#

• Gold Badge: ≥85 Punkte (Ziel)
• Silver Badge: ≥70 Punkte
• Bronze Badge: ≥60 Punkte

---

Implementation Details#

Hybrid Retrieval Algorithm#

python

# Combined Score = 0.6 * Semantic + 0.4 * FTS
combined_score = semantic_weight * semantic_relevance + (1 - semantic_weight) * fts_score

Mode Selection#

ModeAlgorithmusUse CasesemanticVector-SucheSemantische ÄhnlichkeitkeywordFile-Inhalts-SucheExakte BegriffeftsLanceDB FTSSchnelle Keyword-SuchehybridKombiniertStandard (empfohlen)

---

Session Counter#

scripts/session-counter.sh#

Trackt Sessions für konsolidierte Consolidation:

bash

# Increment on Heartbeat
scripts/session-counter.sh --increment

# Get current count
scripts/session-counter.sh --get

# Reset after Consolidation
scripts/session-counter.sh --reset

# Check if consolidation should run
scripts/session-counter.sh --check  # exit 0 = yes, exit 1 = no

---

Memory Digest#

scripts/memory-digest-enhanced.sh#

Konsolidiert Memory-Files mit Session-basierter Logik:

bash

# Normal run (check if consolidation needed)
scripts/memory-digest-enhanced.sh

# Force run (override check)
scripts/memory-digest-enhanced.sh --force

# Dry run (show what would be done)
scripts/memory-digest-enhanced.sh --dry-run

# Auto-dream mode (with consolidation)
scripts/memory-digest-enhanced.sh --auto-dream

---

Integration in OpenClaw#

Heartbeat Integration#

Füge folgenden Task zur Heartbeat-Logik hinzu:

bash

# scripts/heartbeat.sh
"$HOME/.openclaw/workspace/scripts/session-counter.sh" --increment

Cron Integration#

bash

# crontab -e
# Auto-Dream tägliche Consolidation (Optional)
0 3 * * * "$HOME/.openclaw/workspace/scripts/memory-digest-enhanced.sh" --auto-dream

---

Troubleshooting#

Problem: "Dependencies nicht verfügbar"#

Lösung:

bash

source .venv-metaclaw/bin/activate
pip install lancedb sentence-transformers numpy

Problem: "Tabelle nicht gefunden"#

Lösung:

bash

python3 skills/metaclaw/scripts/memory_vector_init.py

Problem: Session-Counter füllt sich nicht auf#

Lösung:

bash

./scripts/session-counter.sh --status
# Prüfe ob --increment in Heartbeat aufgerufen wird

---

Changelog#

Version 1.0.0 (2026-04-01)#

Added:

• Generischer MetaClaw Skill für ClawHub
• Hybrid Retrieval (semantic + FTS)
• Memory Extraction Pipeline
• PluginEval Integration
• Session Counter für konsolidierte Consolidation
• Memory Digest mit Session-basierter Logik

Changed:

• Default Mode: hybrid (statt semantic)
• Score-Kombination: 0.6 * semantic + 0.4 * keyword

Removed:

• Keine Breaking Changes (Backwards compatible)

---

Referenzen#

• ~/.openclaw/workspace/skills/memory-vector/ - Original Hybrid Retrieval
• ~/.openclaw/workspace/scripts/memory-extract.py - Memory Extraction
• ~/.openclaw/workspace/scripts/skill-eval.py - PluginEval
• ~/.openclaw/workspace/notes/projects/metaclaw-integration.md - Integration Plan

---

Last updated: 2026-04-01 Category: automation, memory Tags: #hybrid-retrieval #memory #skill-evolution #plugineval