agentic-engineering-framework: Command for Claude Code

/rollback - Revert Production to Previous Version

Recovery skill for failed or problematic deployments. Reads deployment history, confirms with user, executes rollback, verifies health.

Step 1: Task Gate

Verify an active task exists:

1. Read .context/working/focus.yaml — check current_task is set
2. If no task is set, STOP and say: "No active task. Run /start-work first."

Step 2: Read Deployment History

ls -t .context/deployments/*.yaml 2>/dev/null | head -5

If no deployment records exist, STOP and say: "No deployment records found in .context/deployments/. Nothing to roll back."

Read the most recent deployment record:

cat .context/deployments/$(ls -t .context/deployments/*.yaml | head -1 | xargs basename)

Present the deployment info:

## Last Deployment

- Timestamp: {timestamp}
- App: {app}
- Task: {task}
- Commit: {commit}
- Result: {result}

Step 3: Check Current Service State

export DOCKER_HOST=tcp://192.168.10.201:2375
docker service ls --filter name=watchtower
docker service ps watchtower_app --format '{{.ID}} {{.Image}} {{.CurrentState}}' | head -5

Present the current state: running image version, replica count, health.

Step 4: Confirm with User

This is a Tier 0 action. Present options and WAIT for explicit user confirmation:

## Rollback Options

1. Rollback watchtower_app to previous version (docker service rollback)
2. Cancel — do not rollback

Do NOT proceed without the user selecting option 1. If user selects 2 or anything else, abort.

Step 5: Execute Rollback

Only after user confirms option 1:

export DOCKER_HOST=tcp://192.168.10.201:2375
docker service rollback watchtower_app

Wait for convergence (up to 60 seconds):

for i in $(seq 1 30); do
  RUNNING=$(docker service ps watchtower_app --filter desired-state=running --format '{{.CurrentState}}' | grep -c 'Running')
  if [ "$RUNNING" -ge 2 ]; then
    echo "Rollback converged: $RUNNING replicas running"
    break
  fi
  sleep 2
done