cwc-fix

Something broke mid-session. Diagnose and fix the infrastructure without touching the participant's AGENT config.

Health check (run all of these, then reason from the results)

echo "── processes ──"
ps -eo pid,comm,args | grep -E 'java.*server.jar|node.*bot.js|node.*server.mjs|cloudflared' | grep -v grep
echo "── bot state (local) ──"
curl -fsS -m 3 http://localhost:8088/state 2>&1 | head -c 400 || echo "(unreachable)"
echo "── relay registration ──"
. .env.setup 2>/dev/null
[ -n "${RELAY_URL:-}" ] && curl -fsS -m 5 "${RELAY_URL%/}/p/${RELAY_KEY}/status" 2>&1 || echo "(no .env.setup — run ./setup.sh first)"
echo "── event server ──"
curl -fsS -m 5 "${LEADERBOARD_URL:-http://localhost:8888/api}/leaderboard" 2>&1 | head -c 200 || echo "(unreachable)"
echo "── recent errors ──"
tail -20 /tmp/mc-server.log /tmp/mc-bot.log /tmp/event-local.log 2>/dev/null
grep '\[relay\]' /tmp/mc-bot.log 2>/dev/null | tail -10

Decision tree

• No java process → server died. ./setup.sh --restart.
• No node bot.js process → bot died (often heap OOM after long runs, or kicked by server). ./setup.sh --restart.
• Bot state unreachable but process alive → bot hung. Kill and restart: ./setup.sh --restart.
• Bot reachable locally, relay status "connected":false → the bot's outbound WebSocket to the event server is down. Look at the [relay] lines in /tmp/mc-bot.log:
  • reconnecting in Ns lines → it's retrying; if the event server is up (curl $EVENT_URL/api/config), wait ~30s. If still down, ./setup.sh --restart.
  • superseded by another registration → they ran setup on two machines with the same copied repo/key. rm .relay-key && ./setup.sh --restart mints a fresh key on this machine.
• Event server unreachable ($EVENT_URL/api/config fails) → that's the facilitator's problem (they restart it from their hosting dashboard). Nothing to fix on this machine; tell the participant to flag the facilitator. Their bot reconnects automatically when it's back.
• my_agent.py says "no tools" or hangs at "looking for existing agent" forever → stale .agent_cache.json pointing at a dead agent. rm -f .agent_cache.json and retry.
• Viewer at :8088/view is blank blue → prismarine-viewer y<0 bug. (cd bot && node patch-viewer.cjs) && ./setup.sh --restart, then hard-refresh the browser (Cmd-Shift-R / Ctrl-Shift-R).
• SOLO mode: tunnel unreachable → rm -f /tmp/cf-tunnel-8888.log && ./setup.sh re-mints the tunnel. The public hostname changes, so also rm -f .agent_cache.json (the agent's MCP URL moved).