Emergency Protocol

1. Open Telegram and send /status to @TetherJRDBot
2. If the bot responds with a status report — Cadence is alive but may be busy. Wait 5 minutes and retry your original request.
3. If the bot is silent (no response after 2 minutes):
   • Send /restart to the bot
   • Wait 3 minutes for the restart to complete
4. If still silent after restart attempt — email Witness Fleet Support:
   To: witness-support@agentmail.to
   Subject: [Cadence] Not Responding
   Body: AiCIV: Cadence. Cadence is not responding to Telegram commands. /status and /restart both returned no response. Started approximately [time]. Portal: tpjg-reports.pages.dev

1. Verify the URL: tpjg-reports.pages.dev
2. Try a hard refresh: Ctrl+Shift+R (or Cmd+Shift+R on Mac)
3. Check Cloudflare status: cloudflarestatus.com
   • If Cloudflare is reporting an outage — wait for their resolution
4. If Cloudflare is healthy but portal is still down — email Witness Fleet Support:
   To: witness-support@agentmail.to
   Subject: [Cadence] Portal Down
   Body: AiCIV: Cadence. Portal at tpjg-reports.pages.dev is not loading. Cloudflare status page shows no outages. Hosted on Cloudflare Pages. Started approximately [time].

1. Open Telegram and navigate to @TetherJRDBot
2. Send /restart — this triggers the internal restart mechanism
3. Wait 3 minutes for the bot to reinitialize
4. Send /status to verify recovery
5. If still silent after restart:
   To: witness-support@agentmail.to
   Subject: [Cadence] Telegram Bot Unresponsive
   Body: AiCIV: Cadence. Bot @TetherJRDBot is not responding. Tried /restart, waited 3 min, no response. Started approximately [time].

1. Governor auto-manages context levels — in most cases, no action needed
2. At YELLOW (60-80%): Governor limits heavy task launches automatically
3. At RED (80-95%): Governor auto-compacts context. Current task finishes, then session resets.
4. At DEAD (95%+):
   • Session is effectively frozen — manual restart required
   • The Autorestart Watcher should catch this and restart automatically
   • Circuit breaker allows max 3 restarts per hour to prevent loops
5. If Cadence is stuck in DEAD zone and autorestart is not recovering:
   To: witness-support@agentmail.to
   Subject: [Cadence] Context Overload - Stuck in DEAD Zone
   Body: AiCIV: Cadence. Governor reports DEAD zone (95%+). Autorestart watcher does not appear to be recovering the session. Manual VPS intervention may be needed.

1. Identify the failure type:
   • Cannot send emails — check service account delegation
   • Cannot read emails — check IMAP / API connectivity
2. Verify the service account is active:
   • Service Account: tether-bot@tpj-ai-infrastructure.iam.gserviceaccount.com
   • GCP Project: tpj-ai-infrastructure
3. Check Google Workspace admin console if domain-wide delegation has been revoked or expired
4. If the service account is healthy but email still fails — check Google Workspace status: Google Workspace Status
5. If all checks pass and email still fails, contact Joe directly:
   Call/Text: (281) 650-4658
   Email (personal): Joseph@ThePropertyJoesGroup.com
   GCP admin access required to investigate further.

When Telegram bot, autorestart, and Fleet Support have all failed, SSH into the VPS directly to diagnose and restart Cadence manually.

1. SSH into the VPS:
   ssh -p 2219 aiciv@37.27.237.109 -i ~/.ssh/human_access
2. Check if the tmux session is alive:
   tmux list-sessions
   Look for tether-primary or aiciv-primary
3. If the tmux session exists — reattach to it:
   tmux attach -t tether-primary
   Fallback: tmux attach -t aiciv-primary
   • Check if Claude Code is running inside the session
   • If frozen/unresponsive, press Ctrl+C to interrupt, then restart
4. If NO tmux session exists — create one and restart Claude Code:
   tmux new-session -s tether-primary
   Then inside tmux, launch Claude Code per the startup script:
   cd /home/aiciv && bash tools/launch_primary_visible.sh
5. Verify recovery:
   • Send /status to @TetherJRDBot in Telegram
   • Check health file: cat /tmp/tether_health.json
6. If SSH connection itself fails (timeout, refused):
   Escalate immediately:
   To: witness-support@agentmail.to
   Subject: [Cadence] VPS Unreachable via SSH
   Body: AiCIV: Cadence. Cannot SSH to 37.27.237.109:2219. Connection [timeout/refused]. All automated recovery has failed. VPS-level intervention needed.

Cadence is part of the Witness fleet ecosystem. When local recovery fails, the Guardian Protocol provides fleet-level support from the parent civilization.

1. What is the Guardian Protocol?
   • Witness (parent CIV) operates a fleet support system that monitors child civilizations
   • Jared/Aether built an improved guardian website for fleet-wide visibility
   • Fleet support can perform VPS-level interventions that are beyond Joe's normal access
2. When to invoke the Guardian Protocol:
   • Telegram bot unresponsive AND SSH recovery failed
   • VPS itself is unreachable (network/host down)
   • Repeated crash loops exceeding the circuit breaker (3+/hr sustained)
   • Data integrity concerns or suspected compromise
3. How to escalate:
   To: witness-support@agentmail.to
   Subject: [Cadence] Guardian Protocol Request
   Body:
   AiCIV: Cadence
   VPS: 37.27.237.109 (SSH port 2219)
   Issue: [Describe what failed]
   Local recovery attempted: [List steps taken]
   Requesting: [VPS restart / network check / crash investigation]
4. Parallel notification: Also notify Joe directly via call/text at (281) 650-4658 so he is aware fleet support is engaged
5. Expected response time: Witness Fleet Support monitors automatically. Response depends on severity but is typically within hours for critical issues.