claw-ecosystem

name	claw-ecosystem
description	This skill should be used when the user asks to "start all claws", "start claw gateways", "run all gateways", "start openclaw nanobot picoclaw zeroclaw nanoclaw", "launch claw ecosystem", "stop all claws", "stop gateways", "claw status", "check gateway status", or "open claws". Manages all five Claw ecosystem gateways (OpenClaw, ZeroClaw, NanoBot, PicoClaw, NanoClaw) as a unified fleet.
version	2.0.0

Claw Ecosystem Gateway Manager

Manage all five Claw/NanoBot gateways as one fleet. Supports start, stop, status, and restart across OpenClaw (TypeScript), ZeroClaw (Rust), NanoBot (Python), PicoClaw (Go), and NanoClaw (Node.js).

Project Overview

Project	Language	Binary/Entrypoint	Runtime Workspace	Port
OpenClaw	TypeScript	openclaw (npm global)	~/.openclaw	18789
ZeroClaw	Rust	zeroclaw (brew)	~/.zeroclaw	42617
NanoBot	Python	nanobot (pip)	~/.nanobot	18790
PicoClaw	Go	picoclaw (binary)	~/.picoclaw	dynamic
NanoClaw	Node.js	node dist/index.js (source)	~/Open-Universe/NanoClaw	—

Quick Commands

# Start all (skips already-running ones)
~/.claude/skills/claw-ecosystem/scripts/claw-gateway.sh start

# Status of all
~/.claude/skills/claw-ecosystem/scripts/claw-gateway.sh status

# Stop all
~/.claude/skills/claw-ecosystem/scripts/claw-gateway.sh stop

# Full restart
~/.claude/skills/claw-ecosystem/scripts/claw-gateway.sh restart

# Single project operations
~/.claude/skills/claw-ecosystem/scripts/claw-gateway.sh start zeroclaw
~/.claude/skills/claw-ecosystem/scripts/claw-gateway.sh restart nanoclaw
~/.claude/skills/claw-ecosystem/scripts/claw-gateway.sh stop nanobot

Logs: /tmp/claw-gateways/<project>.log

Project Details

OpenClaw (TypeScript)

• Start: openclaw gateway start — managed via launchd (ai.openclaw.gateway)
• Stop: openclaw gateway stop
• Status: openclaw gateway status
• Port: 18789
• Config: ~/.openclaw/openclaw.json
• Detection: launchctl list ai.openclaw.gateway (NOT pgrep — it's a LaunchAgent)
• Important: Do NOT pkill -f openclaw — always use openclaw gateway stop

ZeroClaw (Rust)

• Start: zeroclaw daemon ← must use daemon, not gateway
  • daemon = gateway + channels + heartbeat + scheduler (all components)
  • gateway = HTTP only — no channels, no scheduler, no heartbeat
• Stop: pkill -f "zeroclaw daemon"
• Status: zeroclaw status
• Config: ~/.zeroclaw/config.toml
• Port: 42617
• Default provider: custom:http://localhost:20128/v1 (model: glm/glm-4.7)
• Provider note: Only recognises 37 built-in IDs + custom:<URL>. For local proxy use:

  default_provider = "custom:http://localhost:20128/v1"
  [[model_routes]]
  hint = "glm-4.7"
  provider = "custom:http://localhost:20128/v1"
  model = "glm/glm-4.7"
  api_key = "sk-6e4dbe05f673220e-m4u65c-def63b28"
  

• TOML pitfall: Never mix model_routes = [] and [[model_routes]] — causes duplicate key parse error

NanoBot (Python)

• Start: nanobot gateway
• Stop: pkill -f "nanobot gateway"
• Status: nanobot status
• Config: ~/.nanobot/config.json
• Port: 18790
• Channels: Telegram (@MetaSocialDataBot), Discord, Email (IMAP polling)
• Default provider: zai (model: glm-4.7)
• Critical: Provider MUST be zai, NOT local or custom. GLM models return 422 errors for tool-calling (rejects tool role + tool_choice: auto). The key "custom" in providers is NanoBot's built-in name for custom OpenAI-compatible endpoints.
• Fallbacks: kimi/kimi-k2-thinking-turbo, google/gemini-2.5-flash, openrouter/meta-llama/llama-3.3-70b-instruct:free

PicoClaw (Go)

• Start: picoclaw gateway
• Stop: pkill -f "picoclaw gateway"
• Status: picoclaw status
• Config: ~/.picoclaw/config.json
• Health endpoint: http://127.0.0.1:<port>/health
• Memory files: Auto-created at ~/.picoclaw/workspace/memory/YYYYMM/YYYYMMDD.md

NanoClaw (Node.js — run from source)

• Start: cd ~/Open-Universe/NanoClaw && node dist/index.js
• Stop: pkill -f "Open-Universe/NanoClaw"
• Source dir: ~/Open-Universe/NanoClaw/
• Config: ~/Open-Universe/NanoClaw/.env
• Channels: Telegram (@NanoClaw97_bot), Discord (NanoClaw#2256)
• Trigger phrase: @Andy
• Database: ~/Open-Universe/NanoClaw/store/messages.db (SQLite)
• Two prerequisites (auto-handled by script):
  1. Apple container system running: container system status → apiserver is running
     • Fix: container system start (downloads kernel on first run, ~2 min)
  2. nanoclaw-agent:latest image built: container image list → shows nanoclaw-agent
     • Build: cd ~/Open-Universe/NanoClaw/container && bash build.sh (~3 min, first time only)
• Architecture: Each conversation runs in an isolated Linux VM via Apple's native container CLI. The image contains claude-code, Chromium (agent-browser), and the TypeScript agent-runner.
• Rebuild dist: cd ~/Open-Universe/NanoClaw && npm run build
• Rebuild image: cd ~/Open-Universe/NanoClaw/container && bash build.sh

Local LLM Proxy

Shared proxy at http://localhost:20128/v1, key: sk-6e4dbe05f673220e-m4u65c-def63b28

Model ID	Used by
glm/glm-4.7	ZeroClaw
qwen/qwen3-coder-plus	ZeroClaw
gh/claude-sonnet-4.5	ZeroClaw

NanoBot uses zai (not the local proxy) because GLM lacks tool-calling support.

Troubleshooting

OpenClaw port conflict:

openclaw gateway stop && openclaw gateway start
# or: openclaw gateway run --force

ZeroClaw "API key not set" for all providers: Named providers need zeroclaw onboard. Use custom:http://... instead — no onboarding needed.

ZeroClaw TOML "duplicate key: model_routes": Remove model_routes = [] line; keep only [[model_routes]] sections.

NanoBot "No API key configured" on start: Provider local is not recognised. Fix:

python3 -c "
import json; f='$HOME/.nanobot/config.json'
d=json.load(open(f))
d['agents']['defaults']['provider']='zai'
d['agents']['defaults']['model']='glm-4.7'
json.dump(d,open(f,'w'),indent=2)"
nanobot gateway &

NanoBot "422 tool role" error: GLM doesn't support tool-calling via the local proxy. Switch provider to zai or use kimi/kimi-k2-thinking-turbo.

NanoClaw "XPC connection error: Connection invalid":

container system start   # first run downloads kata kernel (~2 min)

NanoClaw container exits immediately (code 1):

cd ~/Open-Universe/NanoClaw/container && bash build.sh

NanoClaw flooding health-check alerts (runaway scheduled tasks):

sqlite3 ~/Open-Universe/NanoClaw/store/messages.db "
UPDATE scheduled_tasks SET status='paused'
WHERE prompt LIKE '%health check%' OR prompt LIKE '%monitoring/%'
   OR prompt LIKE '%code-reviewer agent%' OR prompt LIKE '%health summary%';"

ZeroClaw stale heartbeat/scheduler/telegram errors: You're running zeroclaw gateway instead of zeroclaw daemon. Restart with daemon:

pkill -f "zeroclaw gateway" && zeroclaw daemon &

Scripts

• scripts/claw-gateway.sh — Fleet management script (start/stop/status/restart [project])