Deploy agent-services to the infra VM. Use when shipping new code, restarting services, or rolling back a broken deploy. Covers pre-deploy snapshot, pull/build/restart, smoke tests, and rollback.
Standardmäßig ist der Prompt ausgewählt, der zuerst die Quelle prüft. Sie können zu einem direkten Befehl wechseln oder eine lokale Kopie herunterladen.
Quelldateien prüfen
Lesen Sie SKILL.md und alle von SkillsMP angezeigten Begleitdateien, bevor Sie sich für eine Installation entscheiden.
Mit Codex oder Claude installieren Kopieren Sie diesen Prompt, fügen Sie ihn in Codex, Claude oder einen anderen Assistant ein und lassen Sie die Skill-Seite prüfen und installieren.
Ein direkter Befehl überspringt den Prüf-Prompt. Prüfen Sie die Quelle, bevor Sie ihn ausführen.
Deploy agent-services to the infra VM. Use when shipping new code, restarting services, or rolling back a broken deploy. Covers pre-deploy snapshot, pull/build/restart, smoke tests, and rollback.
Deploy Runbook
Step-by-step procedure for deploying agent-services to the infra VM. Follow this exactly — skipping steps has caused real incidents (unauthenticated servers, unrecoverable state).
When to Use
Shipping new features — after a PR merges to main
Hotfixing a broken service — pull the fix, rebuild, restart
Rolling back — restore from pre-deploy snapshot
Restarting after a crash — same process minus the git pull
Prerequisites
You need:
The infra VM ID (check the registry or ask the orchestrator)
The VERS_AUTH_TOKEN value (check with the orchestrator — if lost, all agents need reconfiguration)
SSH access to the VM (root@{VM_ID}.vm.vers.sh)
Step 1: Snapshot the Infra VM
Do this FIRST. Every time. No exceptions.
Before touching anything, snapshot the current state so you can roll back:
# From the orchestrator or any agent with vers tools
vers_vm_commit --vmId {INFRA_VM_ID}
# → returns a commit ID, e.g., "commit-abc123"
Record the commit in the ledger (if the service is still running):
Each should return JSON with the expected structure. If any returns 401, the auth token is correct (middleware is working). If any returns nothing or errors, check the logs:
tail -50 /tmp/agent-services.log
Verify from outside the VM
From the orchestrator or another agent, test the external URL:
Snapshot the new working state (another vers_vm_commit) and record it with a post-deploy tag
Verify agents can still reach the board, feed, and registry
Rollback Procedure
If the deploy is broken:
Option A: Restore from Snapshot (clean rollback)
# From the orchestrator (not from the broken VM)
vers_vm_restore --commitId {PRE_DEPLOY_COMMIT_ID}
# → creates a new VM with the old state# Update DNS/references to point to the new VM ID
This gives you a brand new VM with the exact pre-deploy state. The old broken VM can be deleted.
Option B: Git Revert (if the VM is still accessible)
ssh -o StrictHostKeyChecking=no root@{INFRA_VM_ID}.vm.vers.sh
cd /root/workspace/vers-agent-services
git log --oneline -5 # find the last known good commit
git reset --hard {GOOD_COMMIT_SHA}
npm install && npm run build
pkill -f 'node dist/server.js' || truesleep 1
nohupenv VERS_AUTH_TOKEN=$VERS_AUTH_TOKEN node dist/server.js > /tmp/agent-services.log 2>&1 &
Common Pitfalls
❌ Forgetting VERS_AUTH_TOKEN
What happens: The server starts but with no auth — all endpoints are open to anyone. You'll see this warning in the logs:
⚠️ VERS_AUTH_TOKEN is not set — all endpoints are unauthenticated.
Fix: Kill the server, set the env var, restart.
❌ Forgetting to Snapshot First
What happens: If the deploy breaks, you have no rollback point. You're stuck debugging a broken VM under pressure.
Fix: Always snapshot before deploying. Make it muscle memory. The commit ledger exists precisely for this.
❌ Stale dist/ Files
What happens:npm run build fails but you restart anyway. The server runs old compiled code from dist/, which may not match the new source. Subtle bugs ensue.
Fix: Never restart after a failed build. Fix the build error first, or roll back.
❌ Port Already in Use
What happens:pkill didn't fully kill the old process. The new server fails to bind to port 3000.
Fix:
# Find what's using port 3000
lsof -i :3000
# Force killkill -9 $(lsof -t -i :3000)
sleep 1
# Restart
❌ Deploying Without Testing
What happens: You push untested code and deploy it. The build passes but runtime errors break endpoints.
Fix: Always run npm test locally (or on a branch VM) before deploying. The infra VM is shared — breaking it affects all agents.
❌ Data File Permissions / Missing Directories
What happens: The server crashes on first write because data/ doesn't exist or has wrong permissions.
Fix: The stores auto-create directories, but if you see file permission errors:
