clawlab

local multi-agent setup

powered by: openclaw | picoclaw | zeroclaw

intro

• manage multiple agents as plain directories
• command multiple agents in a unified way
• native commands get forwarded (ie. help, onboard, status, etc.)

structure

• agents/ - working dir per-agent (instances)
• shared/ - shared space for all agents
• cfg - general config file
• cmd - script for commanding specific agents

usage

Usage:
  cmd list
  cmd make <agent-name>
  cmd <agent-id> <command...>

Global commands:
  list    -> list existing agents
  make    -> create agent working dir

Agent commands:
  edit    -> open agent working dir in $EDITOR
  remove  -> remove agent working dir (-y to confirm)
  prompt  -> run <tool> agent -m "<message>" (arg, tty prompt, or stdin)
  start   -> start gateway / daemon (depending on agent)
  stop    -> stop gateway / daemon (depending on agent)

Examples:
  ./cmd list

  ./cmd make 001-openclaw
  ./cmd 001 onboard
  ./cmd 001 prompt
  ./cmd 001 edit
  ./cmd 001 start

  ./cmd make 004-picoclaw
  ./cmd 004 onboard
  ./cmd 004 prompt "hello from shell"
  ./cmd 004 stop
  ./cmd 004 remove -y

  ./cmd 007 make 007-zeroclaw
  ./cmd 007 onboard 
  ./cmd 007 auth login --provider
  ./cmd 007 channel add telegram '{"bot_token":"...","name":"my-bot"}'
  ./cmd 007 status 

notes

• requires on $PATH: swift, openclaw, picoclaw, zeroclaw
• toggle version control manually in .gitignore
• ports are defined in cfg

---

done for fun