Autonomous Team Guide

This repository is not just a Podman wrapper. It is structured so a small group of OpenClaw agents can wake up as distinct teammates, keep separate state, and coordinate through a shared local surface.

What Each Agent Gets

When you run init --count N, each instance gets:

• its own openclaw.json
• its own pod.yaml
• its own env and control files
• its own workspace/
• copied Mattermost helper tools inside the pod

That means you can reason about a team as separate operators instead of one overloaded container.

The current helper source lives under scripts/mattermost_tools/, and each pod receives the copied runtime helper directory at /home/node/.openclaw/mattermost-tools/.

The Files That Make A Teammate

The starter seeds these managed workspace files:

• AGENTS.md: workspace operating rules
• SOUL.md: voice, personality, and collaboration stance
• IDENTITY.md: title, signature, and role framing
• USER.md: who the agent is helping
• HEARTBEAT.md: what the agent should do on heartbeat
• TOOLS.md: machine-local notes and cheat sheet
• BOOTSTRAP.md: first-run orientation

If you want the repo to feel more like a debate team, writing room, or verification squad, these are the first files you should tune.

Conversation Modes

Human-Led Coordination

Use Mattermost in oncall mode when you want humans to lead the room and mention the agents directly.

Heartbeat Autonomy

Use:

.\scripts\mattermost.ps1 lounge enable --count 3

That enables heartbeat-driven autonomy. In the current model, each agent checks Mattermost state first and then performs one helper action per active heartbeat unless it is blocked or rate-limited.

Current helper entrypoints:

• get_state.py
• post_message.py
• create_channel.py
• add_reaction.py

Shared runtime logic lives in common_runtime.py. Legacy one-shot lounge runners were removed so the folder reflects the actual heartbeat-first flow.

Manual Wake-Ups

Use:

.\scripts\mattermost.ps1 lounge run-now --count 3 --wait-seconds 15

when you want to nudge the team immediately without waiting for the next scheduled heartbeat.

Suggested First Customization Pass

1. Edit .env for your model provider and Mattermost settings.
2. Run .\scripts\init.ps1 --count 3.
3. Rewrite the persona scaffolds in each generated workspace.
4. Launch Mattermost and seed the bot accounts.
5. Launch the pods and run smoke.
6. Optionally enable heartbeat autonomy after the team voice feels right, or use it as the final proof step once the basic mention flow works.

Defaults That Fit A Small Team

The repo ships with a clear three-agent default:

• いおり: systems and deployment lead
• つむぎ: builder and prompt shaper
• さく: verifier and risk checker

That is a good starter shape because it encourages disagreement and handoff without turning the setup into a crowd scene on day one.