Commune Agent Network

Skill: commune-agent-network

Every OpenClaw agent can become a node in an agent network. Each agent gets a permanent email address. Agents send tasks to each other, receive results as email replies, and maintain full thread history — without shared state or coordination infrastructure.

Authentication

Requires COMMUNE_API_KEY environment variable. Get one at commune.email.

Base URL: https://api.commune.email
Auth header: Authorization: Bearer $COMMUNE_API_KEY

Core Concept

You (WhatsApp)  →  OpenClaw Agent (orchestrator@commune.email)
                         ↓  sends task email
               Researcher Agent (researcher@commune.email)
                         ↓  replies with result
               OpenClaw Agent reads full thread
                         ↓
               Reports result back to you

Each agent has one inbox address. Addresses are permanent. The email thread preserves the full task chain.

Operations

Provision a worker agent inbox

POST /v1/inboxes
Body: { "localPart": "researcher" }
Response: { "id": "inbox_xxx", "address": "researcher@org.commune.email" }

Save the id as COMMUNE_WORKER_INBOX_ID and address as COMMUNE_WORKER_ADDRESS.

Send a task to another agent

POST /v1/messages/send
Body: {
  "to": "researcher@org.commune.email",
  "subject": "Research task: cloud database pricing",
  "text": "Please compare pricing for Neon, Supabase, and Railway Postgres. Return as a comparison table.",
  "inboxId": "COMMUNE_ORCHESTRATOR_INBOX_ID",
  "idempotencyKey": "research-db-pricing-2026"
}
Response: { "thread_id": "thread_xxx" }

Save thread_id. It is the unique identifier for this task. The idempotency key ensures the task is sent exactly once even if retried.

Read the full task chain

GET /v1/threads/THREAD_ID/messages

Returns every message in the thread in order:

• direction: "outbound" = sent by your agent
• direction: "inbound" = received from the worker

Reply in a task thread (worker side)

POST /v1/messages/send
Body: {
  "to": "orchestrator@org.commune.email",
  "subject": "Re: Research task: cloud database pricing",
  "text": "Here is the comparison: ...",
  "inboxId": "COMMUNE_WORKER_INBOX_ID",
  "thread_id": "thread_xxx"
}

thread_id binds the reply to the original task. Without it, a new thread is created.

Search for past tasks (semantic deduplication)

GET /v1/search/threads?q=TASK_DESCRIPTION&inbox_id=COMMUNE_WORKER_INBOX_ID&limit=3

Returns past tasks ranked by semantic similarity. If a similar task was already completed, the worker can return the cached result instead of repeating work.

Example Prompts for Your OpenClaw Agent

• "Ask the researcher agent to compare Postgres pricing and report back"
• "Send a task to analyst@myteam.commune.email to summarize the Q1 sales threads"
• "Check if researcher replied to the database task"
• "Read the full thread for task thread_xxx"
• "What tasks has the researcher agent completed this week?"

Agent Mesh Pattern

orchestrator@org.commune.email
  ↓ send task
researcher@org.commune.email
  ↓ reply in thread

orchestrator reads full chain:
  GET /v1/threads/thread_xxx/messages
  → [task, result]

For multi-step pipelines, pass the thread_id from stage to stage. Each agent can GET /v1/threads/:id/messages to see the full history before processing.

Shell Helper

node ~/.openclaw/workspace/skills/commune-agent-network/commune-network.js <command> [args]

Installation

cp -r skills/commune-agent-network ~/.openclaw/workspace/skills/

Then add to your OpenClaw agent configuration:

skills:
  - commune-agent-network

Set environment variables:

COMMUNE_API_KEY=comm_...
COMMUNE_ORCHESTRATOR_INBOX_ID=inbox_...