# Multi Agent Patterns

> Master orchestrator, peer-to-peer, and hierarchical multi-agent architectures.

- Skill: `xiangteng007/multi-agent-patterns` (Agent Skill)
- Install (CLI): `npx skillmds add xiangteng007/multi-agent-patterns`
- Raw SKILL.md: https://api.skillmd.com/api/skills/xiangteng007/multi-agent-patterns/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: xiangteng007 (https://skillmd.com/u/xiangteng007)
- Updated: 2026-09-10
- Page: https://skillmd.com/skills/xiangteng007/multi-agent-patterns

---


# Multi-Agent Architectural Patterns

## Orchestrator Pattern (Supervisor)
- Use a central "Supervisor" agent to route tasks to sub-agents (e.g., Argus, Titan, Rusty) based on defined schemas.
- **State Management**: Maintain a central Graph State (like LangGraph) where each agent modifies specific state keys instead of passing massive text context strings.
- **Interruption Mechanisms**: Equip the Supervisor with explicit `<INTERRUPT>` tools to stop endless loops or arguments between agents.

## Peer-to-Peer Consensus (War Room)
- Allow agents to communicate over a pub/sub event bus (like OpenClaw).
- Enforce strict "Vote" formatting so that text generation naturally converges to a decision, lowering token costs.