Purpose

Select, configure, and execute cloud deployment strategies — blue-green swaps, canary percentage ramps, and rolling updates — with automated health checks, promotion gates, and instant rollback paths across any cloud provider.

When to use this skill

• Deploying a service to staging or production on any cloud provider.
• Designing or modifying a CI/CD pipeline that includes a deploy stage.
• Adding health-check endpoints, readiness probes, or liveness probes to a deploy target.
• Choosing between blue-green, canary, or rolling strategies for a specific service.
• Writing rollback automation or manual rollback runbooks.
• Configuring promotion gates (approval steps, smoke tests, metric thresholds).

Do not use this skill when

• The task requires provider-specific CLI commands or console workflows — prefer

  aws

  ,

  gcp

  , or

  vercel

  skills.
• The change is purely application logic with no deployment artifact or pipeline impact.
• The target is a local development environment (docker-compose up, localhost).
• Container image building is the focus — prefer

  docker-containers

  skill.

Operating procedure

1. Identify the deployment target. Determine the cloud provider, compute service (VM, container, serverless), region, and environment (staging, production).
2. Select the deployment strategy. Choose blue-green if zero-downtime cutover is required and two full environments are affordable. Choose canary if gradual traffic shifting with metric-based promotion is needed. Choose rolling if the service supports N-1 compatibility and incremental pod/instance replacement.
3. Define health-check criteria. Specify the health endpoint path (e.g.,

   /healthz

   ), expected HTTP status (200), response timeout (≤5s), and failure threshold (3 consecutive failures).
4. Configure promotion gates. List the checks that must pass before traffic shifts: smoke test suite, error-rate threshold (<0.5%), p99 latency threshold, and any manual approval steps.
5. Write the rollback trigger. Define automatic rollback conditions: error rate exceeds threshold for >2 minutes, health check fails on new version, or deployment times out past the configured window.
6. Implement the pipeline stage. Add the deploy step to the CI/CD config (GitHub Actions, Cloud Build, CodePipeline). Wire in the strategy, health checks, and promotion gates as pipeline conditions.
7. Execute a dry-run deployment. Run the pipeline against staging with

   --dry-run

   or equivalent flag. Verify the deployment plan output matches expectations before applying.
8. Deploy to staging and validate. Apply the deployment, monitor the health-check endpoint, run the smoke test suite, and confirm metrics stay within thresholds.
9. Promote to production. Trigger the promotion gate. For blue-green, switch the load balancer target group. For canary, ramp traffic from 5% → 25% → 100% with metric checks at each step. For rolling, confirm each batch reaches healthy state before proceeding.
10. Verify post-deployment. Confirm all instances/pods report healthy, check error rates and latency in the monitoring dashboard, and validate that rollback artifacts (previous image tag, previous task definition) are recorded.
11. Document the deployment. Record the deployed version, strategy used, promotion gate results, and rollback instructions in the deployment log or ticket.

Decision rules

• If the service cannot run two versions simultaneously, use blue-green over canary.
• If cost constraints prevent duplicate environments, use rolling updates.
• If the service has database migrations, deploy the migration separately before the application deploy.
• Never skip health checks — a deploy without health verification is incomplete.
• If any promotion gate fails, halt the rollout and trigger rollback automatically.
• Prefer immutable deployment artifacts (container images, versioned bundles) over in-place mutations.
• Record the previous known-good version before every deployment.

Output requirements

1. Deployment plan — target, strategy, health-check config, promotion gates, rollback triggers.
2. Pipeline configuration — the CI/CD stage definition with strategy and gates wired in.
3. Health-check specification — endpoint, thresholds, timeout values.
4. Rollback runbook — exact commands or automation to revert to the previous version.
5. Post-deploy verification — checklist of metrics and endpoints confirmed healthy.

References

• Blue-green deployment pattern: https://martinfowler.com/bliki/BlueGreenDeployment.html
• Canary releases: https://martinfowler.com/bliki/CanaryRelease.html
• Kubernetes rolling update strategy: https://kubernetes.io/docs/concepts/workloads/controllers/deployment/#rolling-update-deployment

• references/preflight-checklist.md

• references/verification-checks.md

• references/rollback-notes.md

Related skills

• aws

  — AWS-specific deploy tooling (CodeDeploy, ECS, Lambda).

• gcp

  — GCP-specific deploy tooling (Cloud Run, Cloud Deploy).

• vercel

  — Vercel deployment and preview environments.

• docker-containers

  — container image building and optimization.

Anti-patterns

• Deploying directly to production without a staging validation step.
• Using mutable tags (e.g.,

  latest

  ) as the deployment artifact — breaks rollback.
• Configuring health checks that always return 200 regardless of application state.
• Skipping database migration ordering — deploying app code that assumes a schema not yet applied.
• Setting canary traffic to 50% on the first step — defeats the purpose of gradual rollout.

Failure handling

• If health checks fail on the new version within the configured window, execute automatic rollback to the previous known-good version.
• If the CI/CD pipeline times out, mark the deployment as failed, do not promote, and alert the on-call channel.
• If rollback itself fails, escalate immediately — page the on-call engineer and freeze further deployments.
• If the deployment strategy is unclear from the repo context, ask for clarification rather than defaulting to a potentially destructive strategy.
• If provider-specific details are needed, redirect to the appropriate provider skill (

  aws

  ,

  gcp

  ,

  vercel

  ).