multi-agent-blueprint
por neal-collab
Coding Agents & IDEsResumo
Production-tested blueprint for building 5-10 agent teams in OpenClaw with cross-agent routing, Telegram.
Descrição
O que esta skill faz
A template for deploying 5-10 specialized AI agents in OpenClaw that coordinate through cross-agent routing. Covers Telegram multi-bot setup, model tiering across Opus/Sonnet/Haiku, File Master and Notion Master centralization patterns, fallback chains, and RAG memory. Based on a real 10-agent deployment.
Provides working configuration examples and proven patterns for multi-agent teams rather than requiring users to figure out routing, session isolation, and cost controls from scratch.
Quando usá-la
- Routing daily task requests between finance, sales, and tech agents
- Running separate Telegram bots per agent with DM isolation
- Cutting AI costs by assigning cheaper models to repetitive tasks
- Centralizing file operations through one agent to avoid race conditions
- Keeping agent memory intact across session resets with RAG
Exemplo de fluxo de trabalho
User asks: Set up a coordinator, finance, and sales agent team on Telegram
- 1
Creates workspace and agentDir directories for each agent
- 2
Generates IDENTITY.md, SOUL.md, AGENTS.md, and USER.md for each role
- 3
Configures openclaw.json with agent list, agentToAgent enabled, and dmScope per-account-channel-peer
- 4
Assigns model tiers: Opus for coordinator, Sonnet for finance, Haiku for sales
- 5
Sets up Telegram bot accounts with requireMention true for group chats
Three-agent team running on Telegram with cross-agent routing and isolated sessions per bot
Requisitos
- Telegram bot token per agent, created via @BotFather
- Anthropic API key or Claude subscription
- Notion API key if using the Notion Master pattern
- Ollama installed locally for free heartbeat model support
Instalação
clawhub install neal-collab/multi-agent-blueprintExecute o comando da CLI ou cole o prompt no Claude Code, Cursor, Codex ou no agente de código que você usa.