diff --git a/.github/workflows/pr-checks.yml b/.github/workflows/pr-checks.yml
index 19ebfde2..4fad504a 100644
--- a/.github/workflows/pr-checks.yml
+++ b/.github/workflows/pr-checks.yml
@@ -23,5 +23,12 @@ jobs:
       - name: Install dependencies
         run: sudo apt-get update && sudo apt-get install -y bats jq
 
+      - name: Install gomplate
+        run: |
+          sudo curl -fsSL -o /usr/local/bin/gomplate \
+            https://github.com/hairyhenderson/gomplate/releases/download/v4.3.3/gomplate_linux-amd64
+          sudo chmod +x /usr/local/bin/gomplate
+          gomplate --version
+
       - name: Run unit tests
         run: make test-unit
diff --git a/k8s/deployment/tests/build_deployment.bats b/k8s/deployment/tests/build_deployment.bats
index 41adfe2a..be88f49d 100644
--- a/k8s/deployment/tests/build_deployment.bats
+++ b/k8s/deployment/tests/build_deployment.bats
@@ -213,6 +213,8 @@ _render_context() {
     "capabilities": {
       "cpu_millicores": 100,
       "ram_memory": 128,
+      "cpu_millicores_limit": 200,
+      "ram_memory_limit": 256,
       "additional_ports": [],
       "scaling_type": "fixed",
       "autoscaling": {
diff --git a/k8s/diagnose/tests/scope/health_probe_endpoints.bats b/k8s/diagnose/tests/scope/health_probe_endpoints.bats
index 3621480d..92471cb4 100644
--- a/k8s/diagnose/tests/scope/health_probe_endpoints.bats
+++ b/k8s/diagnose/tests/scope/health_probe_endpoints.bats
@@ -180,7 +180,10 @@ EOF
   stripped=$(strip_ansi "$output")
   assert_contains "$stripped" "Readiness Probe on HTTP://8080/health:"
   assert_contains "$stripped" "HTTP 404 - Health check endpoint not found"
-  assert_contains "$stripped" "Update probe path or implement the endpoint in application"
+  # Remediation guidance lives in the structured evidence (suggested_actions),
+  # not in stdout - the check publishes it for the AI summarizer, not operators.
+  actions=$(jq -r '.evidence.suggested_actions[]' "$SCRIPT_OUTPUT_FILE")
+  assert_contains "$actions" "Update probe path or implement health endpoint in application"
 }
 
 @test "scope/health_probe_endpoints: updates status to failed on 404" {
@@ -261,7 +264,10 @@ EOF
   stripped=$(strip_ansi "$output")
   assert_contains "$stripped" "Readiness Probe on HTTP://8080/health:"
   assert_contains "$stripped" "HTTP 500 - Application error"
-  assert_contains "$stripped" "Check application logs and fix internal errors or dependencies"
+  # Remediation guidance lives in the structured evidence (suggested_actions),
+  # not in stdout - the check publishes it for the AI summarizer, not operators.
+  actions=$(jq -r '.evidence.suggested_actions[]' "$SCRIPT_OUTPUT_FILE")
+  assert_contains "$actions" "Check application logs for internal errors"
 }
 
 @test "scope/health_probe_endpoints: updates status to warning on 500" {
diff --git a/k8s/scope/tests/build_context.bats b/k8s/scope/tests/build_context.bats
index bd86e56b..61b8563c 100644
--- a/k8s/scope/tests/build_context.bats
+++ b/k8s/scope/tests/build_context.bats
@@ -154,6 +154,7 @@ teardown() {
     "gateway_name": "co-gateway-public",
     "alb_name": "co-balancer-public",
     "component": "test-namespace-test-app",
+    "base_domain": "cloud-domain.io",
     "k8s_modifiers": {}
   }'
 
@@ -213,6 +214,7 @@ teardown() {
     "gateway_name": "co-gateway-private",
     "alb_name": "co-balancer-private",
     "component": "test-namespace-test-app",
+    "base_domain": "cloud-domain.io",
     "k8s_modifiers": {}
   }'
 
diff --git a/k8s/scope/tests/wait_on_balancer.bats b/k8s/scope/tests/wait_on_balancer.bats
index 83d384d4..96f4406c 100644
--- a/k8s/scope/tests/wait_on_balancer.bats
+++ b/k8s/scope/tests/wait_on_balancer.bats
@@ -72,9 +72,15 @@ teardown() {
 # external_dns: Success after retries
 # =============================================================================
 @test "wait_on_balancer: external_dns success after retries" {
-  local call_count=0
+  # The script reads kubectl output via command substitution ($(kubectl ...)),
+  # which runs the mock in a subshell — an in-memory counter would reset every
+  # call. Persist the attempt count in a file so it survives across subshells.
+  export CALL_COUNT_FILE="$BATS_TEST_TMPDIR/dnsendpoint_calls"
+  echo 0 > "$CALL_COUNT_FILE"
   kubectl() {
-    call_count=$((call_count + 1))
+    local call_count
+    call_count=$(($(cat "$CALL_COUNT_FILE") + 1))
+    echo "$call_count" > "$CALL_COUNT_FILE"
     case "$*" in
       "get dnsendpoint k8s-my-app-my-scope-scope-123-dns -n default-namespace -o jsonpath={.status.observedGeneration}")
         if [ "$call_count" -ge 2 ]; then
diff --git a/k8s/utils/assume_role_lib b/k8s/utils/assume_role_lib
index 58c08352..ca008e13 100644
--- a/k8s/utils/assume_role_lib
+++ b/k8s/utils/assume_role_lib
@@ -17,11 +17,14 @@ arn_for_selector() {
   local json="$1" selector="$2"
   [ -n "$json" ] || return 0
   [ -n "$selector" ] || return 0
+  # jq exits non-zero (status 5) on malformed JSON; swallow it so the helper
+  # honors its "never crashes, returns empty" contract instead of propagating
+  # jq's failure as the function's return code.
   printf '%s' "$json" | jq -r --arg sel "$selector" '
     [ .iam_role_arns.arns[]?
       | select(.selector == $sel)
       | .arn ]
-    | first // ""' 2>/dev/null
+    | first // ""' 2>/dev/null || true
 }
 
 # resolve_assume_role_arn <iam_provider_json> <selector>
diff --git a/testing/run_bats_tests.sh b/testing/run_bats_tests.sh
index d17384e6..91c7edc7 100755
--- a/testing/run_bats_tests.sh
+++ b/testing/run_bats_tests.sh
@@ -82,11 +82,24 @@ run_tests_in_dir() {
   local exit_code=0
   (
     cd "$test_dir"
-    # Use script to force TTY for colored output
-    # Exclude integration directory - those tests are run by run_integration_tests.sh
-    # --print-output-on-failure: only show test output when a test fails
-    script -q /dev/null bats --formatter pretty --print-output-on-failure $(find . -name "*.bats" -not -path "*/integration/*" | sort)
-  ) 2>&1 | tee "$temp_output" || exit_code=$?
+    # Force a TTY so the bats "pretty" formatter renders; --print-output-on-failure
+    # only dumps test output when a test fails. Exclude integration tests (run
+    # separately). `script`'s command syntax differs by platform: BSD/macOS takes a
+    # positional command (script <file> <cmd...>), while util-linux (Linux CI) needs
+    # -c "<cmd>" <file>. Using the wrong one silently runs zero tests.
+    local test_files
+    test_files=$(find . -name "*.bats" -not -path "*/integration/*" | sort | tr '\n' ' ')
+    local bats_cmd="bats --formatter pretty --print-output-on-failure $test_files"
+    # The pretty formatter shells out to `tput`, which needs $TERM. CI runners and
+    # containers often leave it unset - default it so tput can find a terminfo entry.
+    export TERM="${TERM:-xterm}"
+    if [ "$(uname)" = "Darwin" ]; then
+      script -q /dev/null $bats_cmd
+    else
+      script -qec "$bats_cmd" /dev/null
+    fi
+  ) 2>&1 | tee "$temp_output"
+  exit_code=${PIPESTATUS[0]}
 
   # Extract failed tests from output
   # Strip all ANSI escape codes (colors, cursor movements, etc.)
@@ -191,4 +204,12 @@ if [ ${#FAILED_TESTS[@]} -gt 0 ]; then
   exit 1
 fi
 
+# A non-zero runner exit with no parsed "✗" lines means bats itself failed to run
+# (e.g. a malformed .bats file or a broken `script` invocation) - never report success.
+if [ "$HAS_FAILURES" -ne 0 ]; then
+  echo -e "${RED}BATS runner exited with errors but no individual test failures were parsed.${NC}"
+  echo -e "${RED}This usually means the test harness failed to execute - check the output above.${NC}"
+  exit 1
+fi
+
 echo -e "${GREEN}All BATS tests passed!${NC}"