ML-MLOps-Portfolio: Skill for Claude Code

Release Checklist

Pre-Release Verification

Code Quality

• All CI checks passing (lint, test, build)
• Coverage >= 90% lines, >= 80% branches
• No TODO/FIXME items in critical paths
• Type hints on all public functions

Model Quality

• Quality gates passing for all services
• Fairness checks passed (DIR >= 0.80)
• SHAP consistency verified
• No data leakage detected

Infrastructure

• Terraform plan shows no unexpected changes
• tfsec/checkov clean (no HIGH/CRITICAL findings)
• Secrets rotated if approaching expiry

Documentation

• ADRs up to date for any new decisions
• Service READMEs updated with new metrics
• AGENTS.md updated if new invariants
• CHANGELOG.md updated

Version Tagging

# Update version
export VERSION=v{MAJOR}.{MINOR}.{PATCH}

# Tag and push
git tag -a ${VERSION} -m "Release ${VERSION}: {summary}"
git push origin ${VERSION}

Build and Push Images

# For each service
for SERVICE in service-a service-b service-c; do
  # GCP Artifact Registry
  docker build -f ${SERVICE}/Dockerfile -t ${GCP_REGISTRY}/${SERVICE}:${VERSION} .   # repo root: uv.lock
  docker push ${GCP_REGISTRY}/${SERVICE}:${VERSION}

  # AWS ECR
  docker build -f ${SERVICE}/Dockerfile -t ${AWS_REGISTRY}/${SERVICE}:${VERSION} .
  docker push ${AWS_REGISTRY}/${SERVICE}:${VERSION}
done

Deploy to GCP

# Update overlays (production environment for release)
# k8s/overlays/gcp-prod/kustomization.yaml → newTag: ${VERSION} (CI replaces with @sha256 digest)
kubectl apply -k k8s/overlays/gcp-prod/
kubectl rollout status deployment --all -n {namespace} --timeout=300s

Deploy to AWS

# Update overlays (production environment for release)
# k8s/overlays/aws-prod/kustomization.yaml → newTag: ${VERSION} (CI replaces with @sha256 digest)
kubectl apply -k k8s/overlays/aws-prod/
kubectl rollout status deployment --all -n {namespace} --timeout=300s

Post-Deploy Verification