Deployment Automation

deployment-automation

Agent: Social Media Manager

L2 DevOps and infrastructure engineer responsible for CI/CD pipelines, deployment automation, cloud infrastructure, monitoring, alerting, incident response, and rollout management.

Department ethos: ideal-engineering.md

Skill Description

The DevOps / Infrastructure Engineer automates deployment processes to reduce manual steps and human error.

When to Use

• Deployments currently require manual SSH, script execution, or console clicks.
• Deployment frequency is limited by human availability rather than technical readiness.
• A deployment error caused by a manual misstep needs to be prevented from recurring.
• The team is scaling and manual deployment processes will not keep pace.

Workflow

1. Process Audit: Document every manual deployment step, decision point, and prerequisite. Map the current runbook to identify SSH sessions, console clicks, script executions, and tribal knowledge. Deliverable: current-state deployment map with automation candidates flagged.
2. Strategy Selection: Choose the deployment pattern based on infrastructure: blue-green for stateless services (swap ALB target groups or Kubernetes service selectors), canary for high-traffic services (route 1-5% via Istio traffic splitting or AWS weighted target groups), rolling for stateful workloads (Kubernetes rolling update with maxSurge/maxUnavailable tuning). Deliverable: deployment strategy document with pattern rationale.
3. Pre-Deployment Automation: Implement automated pre-flight checks: artifact version validation (SHA-tagged immutable images, never latest), dependency health verification (database connectivity, downstream API availability), environment drift detection (Terraform plan diff, Kubernetes config audit), and capacity headroom check. Deliverable: pre-deployment gate configuration.
4. Deployment Execution Automation: Replace manual steps with pipeline stages or infrastructure-as-code. For Kubernetes: Helm chart or Kustomize overlays with ArgoCD or Flux for GitOps-driven deploys. For VM-based: Ansible playbooks or Terraform with zero-downtime rolling replacement. Include health check gates: readiness probes (/healthz returning 200), startup probes for slow-initializing services, and liveness probes to catch post-deploy crashes. Deliverable: automated deployment pipeline or GitOps configuration.
5. Post-Deployment Validation: Automate smoke tests (critical user journey HTTP assertions), metric baseline comparison (compare p95 latency, error rate, and throughput against the previous 30-minute window), and synthetic monitoring triggers. If any check fails within the bake time window (10-15 minutes), trigger automatic rollback: revert the Kubernetes deployment revision, swap the blue-green target group back, or scale the canary to zero. Deliverable: post-deployment validation suite with automated rollback triggers.
6. Human Gates: Retain manual approval gates only where required: database migrations with backward-incompatible schema changes, breaking API version deprecations, and compliance-sensitive deployments. Implement approval via Slack-integrated workflow (approve/reject buttons) rather than SSH access. Deliverable: gate policy document specifying which deployments require human approval and why. [GATE]

Anti-Patterns

• Automating without understanding the manual process first. Why: Automating a broken process just makes it fail faster; fix the process, then automate it.
• Removing all human gates. Why: Some deployments (database migrations, breaking changes) benefit from a human approval step; full automation is not always the goal.
• Skipping rollback automation. Why: Automated deploys without automated rollback increase risk because recovery still depends on manual intervention.
• Not testing the automation in staging. Why: Deployment automation that has only been tested in production is a live experiment with user-facing consequences.

Output

Success: A fully automated deployment pipeline that deploys, validates, and can rollback without manual intervention, with documentation.

Failure: A partial automation report listing which steps remain manual, why they could not be automated, and a plan to address them.

Related Skills

None defined yet.

• infrastructure-scaling-executor — sibling skill under the same agent — combine with infrastructure-scaling-executor for end-to-end coverage
• alerting-configurator — sibling skill under the same agent — combine with alerting-configurator for end-to-end coverage
• production-readiness-reviewer — sibling skill under the same agent — combine with production-readiness-reviewer for end-to-end coverage