rad-modules: Command for Claude Code

Guided deployment and test campaign for one or more rad-modules modules: $ARGUMENTS

$ARGUMENTS is a comma/space-separated list of module names (e.g. "Istio_GKE", "Bank_GKE EKS_GKE"). If empty, include every deployable module under modules/.

Deployments go through rad-launcher (rad-launcher/radlab.py), which stores Terraform state in a GCS bucket — state is never kept in the repo.

---

PHASE 1 — CONFIGURE

Resolve campaign parameters from the environment, prompting for anything missing:

export PROJECT_ID="${RAD_PROJECT_ID:-$(gcloud config get-value project 2>/dev/null)}"
export STATE_BUCKET="${RAD_STATE_BUCKET:-${PROJECT_ID}-radlab-tfstate}"
export RESOURCE_CREATOR_IDENTITY="${RESOURCE_CREATOR_IDENTITY:-}"

Display the resolved values and confirm before continuing:

Campaign parameters
  Project:       $PROJECT_ID
  State bucket:  $STATE_BUCKET   (must exist; create with: gcloud storage buckets create gs://$STATE_BUCKET)
  Creator SA:    $RESOURCE_CREATOR_IDENTITY (blank = caller ADC)

Confirm prerequisites are installed once: cd rad-launcher && python3 installer_prereq.py. It must run from inside rad-launcher/ — it invokes requirements.txt, opentofu_installer.py and cloudsdk_kubectl_installer.py by CWD-relative path.

---

PHASE 2 — RESOLVE MODULE LIST

Scan modules/ for deployable directories (has main.tf/variables.tf). Apply the $ARGUMENTS filter; report "Unknown module: " for any token without a directory.

For each selected module, read variables.tf for REQUIRED inputs (declared type, no default) beyond the standard set, and for any cloud credentials it needs:

• AKS_GKE: Azure client_id, client_secret, azure_tenant_id, subscription_id (via tfvars or ARM_* env vars — never hardcode).
• EKS_GKE: AWS aws_access_key / aws_secret_key; Migration_Center: aws_access_key_id / aws_secret_access_key (via tfvars or AWS_* env vars). For each module, prepare a minimal tfvars file capturing project_id, deployment_id (optional), and any required inputs. Report any input you cannot fill so the user can supply it.

---

PHASE 3 — DEPLOY (one module at a time)

Deploy modules SERIALLY — each GKE-based module provisions a cluster and post-provisioning installers; running several at once contends on cluster operations. For each module:

python3 rad-launcher/radlab.py \
  -m <Module> -a create \
  -p "$PROJECT_ID" -b "$STATE_BUCKET" \
  -f /path/to/<module>.tfvars

Stream the output. On failure, capture the error, classify it (auth/quota/API/timeout vs a real module bug), and report — do not silently retry destructive operations.

---

PHASE 4 — VERIFY

After a successful create, sanity-check the deployment:

• Read the module outputs (deployment_id, project_id, and any endpoint output).
• For GKE modules, fetch credentials and confirm core workloads are Ready (kubectl get pods -A), e.g. Bank of Anthos frontend, Istio ingressgateway.
• For an exposed endpoint, curl the ingress IP for an HTTP 200 where applicable. Record PASS/FAIL per module with the evidence (the failing pod, the HTTP status).

---

PHASE 5 — DESTROY (if this is a create→destroy test)

python3 rad-launcher/radlab.py \
  -m <Module> -a delete \
  -p "$PROJECT_ID" -b "$STATE_BUCKET" \
  -f /path/to/<module>.tfvars

Confirm destroy completes cleanly (best-effort destroy provisioners may print warnings — that is expected). Verify no orphaned clusters/networks remain.

---

REPORT

Per module: deploy result, verify result (with evidence), destroy result, and any bug found with a concrete fix suggestion. Summarise as a table at the end.