agent-team-mesh

agent-team-mesh

• Author: Evan Song · github.com/Songhonglei
• Repository: https://github.com/Songhonglei/better-agent-skills
• License: MIT

Team-wide P2P mesh for OpenClaw agents running on different containers/pods. Direct WebSocket calls between teammates' agents — no broker, no central database.

Open-source edition of an internal team-comms skill, rebuilt with proper token hygiene, message size limits, dry-run mode, and pluggable identity detection.

---

Architecture

My agent's Gateway (WS)
   │
   ▼
ws://<teammate's pod IP>:18789  ──▶  Teammate's agent processes message
   │  (token-authenticated)         and writes a reply to their session
   ▼
chat.send → agent.wait → chat.history

• No broker — direct P2P, each teammate's OpenClaw Gateway is the endpoint
• Per-agent tokens — each agent has its own WS token (not a shared key)
• Pod IP direct — Kubernetes hostname/DNS sometimes doesn't cross nodes
• IP usually stable — long-running pods, but use sync if they roll

---

Prerequisites

• All agents on a routable network (same VPC / mesh / VPN)
• Each agent's OpenClaw Gateway exposed on port :18789 over WebSocket
• Per-agent WS tokens collected and stored in tokens.json (chmod 600)
• OPENCLAW_ALLOW_INSECURE_PRIVATE_WS=1 is set by the script (ws:// plaintext is acceptable only on trusted internal networks)
• Python 3.8+, bash, curl, openclaw CLI on PATH

---

Setup

1. Edit your team registry

Copy references/registry.json and fill in your teammates:

{
  "agents": [
    { "name": "Alice", "emailPrefix": "alice", "ip": "10.0.0.10", "hostname": "agent-alice-0" },
    { "name": "Bob",   "emailPrefix": "bob",   "ip": "10.0.0.11", "hostname": "agent-bob-0" }
  ]
}

The emailPrefix is the key — it must match what whoami detects on each teammate's machine.

2. Create tokens file

Each teammate generates their own OpenClaw gateway WS token and shares it. Collect into: