2026-06-16 11:59:02 +02:00
|
|
|
---
|
|
|
|
|
created: 2026-06-16
|
|
|
|
|
updated: 2026-06-16
|
|
|
|
|
sources: []
|
|
|
|
|
tags: [architecture, agents, orchestration, openclaw]
|
|
|
|
|
---
|
|
|
|
|
|
2026-06-07 11:47:41 +02:00
|
|
|
# Agent Orchestration Patterns
|
|
|
|
|
|
|
|
|
|
> Migriert aus ByteRover context-tree (April 2026)
|
|
|
|
|
|
|
|
|
|
## Kernprinzipien
|
|
|
|
|
|
|
|
|
|
1. **Starte mit 1 Agent, nicht 7** — Orchestrator-Pattern gewinnt
|
|
|
|
|
2. **STATE.yaml statt Content-Forwarding** — spart ~80% Tokens
|
|
|
|
|
3. **Telegram Multi-Bot:** Disable Sub-Bot-Commands, um 1006 WebSocket-Fehler zu vermeiden
|
|
|
|
|
|
|
|
|
|
## Swarm-Adoptionsplan (3-Stufen)
|
|
|
|
|
|
|
|
|
|
| Stufe | Tool | Aufwand | Status (2026-04) |
|
|
|
|
|
|-------|------|---------|------------------|
|
|
|
|
|
| 1 | ClawRecipes (@jiggai/recipes, 13.3K Installs/Monat) | Sofort, 0 Kosten | Empfohlen |
|
|
|
|
|
| 2 | ClawTeam (HKUDS, 4.8K Stars, Leader+Worker, Git Worktree) | Woche 1-2 | Zu evaluieren |
|
|
|
|
|
| 3 | Multi-Agent: 2-3 Specialists, Hector als Orchestrator | Nach Stufen 1-2 | Native OpenClaw verfügbar |
|
|
|
|
|
|
|
|
|
|
## Framework-Vergleich
|
|
|
|
|
|
|
|
|
|
| Framework | Bewertung | Kosten |
|
|
|
|
|
|-----------|-----------|--------|
|
|
|
|
|
| OpenClaw Native | ⭐⭐⭐⭐⭐ | 0 |
|
|
|
|
|
| ClawRecipes | ⭐⭐⭐⭐ | Niedrig |
|
|
|
|
|
| ClawTeam | ⭐⭐⭐⭐ | Mittel |
|
|
|
|
|
| CrewAI | ⭐⭐⭐ | Hoch |
|
|
|
|
|
| LangGraph | ⭐⭐⭐ | Hoch |
|
|
|
|
|
| AutoGen | ⭐⭐ | Sehr hoch |
|
|
|
|
|
| Google ADK | ⭐ | — |
|
|
|
|
|
|
|
|
|
|
## Community Patterns
|
|
|
|
|
|
|
|
|
|
- Die meisten Teams brauchen 2-4 Agenten, nicht 7
|
|
|
|
|
- Task Board + Git Worktree Isolation (ClawTeam) für parallele Entwicklung
|
|
|
|
|
- Sub-Agent + agentToAgent als native OpenClaw-Alternative
|
