diff --git a/.claude/skills/skill-linter/scripts/validate-skill.sh b/.claude/skills/skill-linter/scripts/validate-skill.sh
index 34f014a4..1d3fe00e 100755
--- a/.claude/skills/skill-linter/scripts/validate-skill.sh
+++ b/.claude/skills/skill-linter/scripts/validate-skill.sh
@@ -158,9 +158,17 @@ if [ -n "$COMPAT" ]; then
   pass "Compatibility valid ($COMPAT_LEN chars)"
 fi
 
-# Check allowed-tools (if present)
+# Check allowed-tools (optional)
 TOOLS=$(echo "$FRONTMATTER" | grep '^allowed-tools:' | sed 's/^allowed-tools:[[:space:]]*//' || true)
-if [ -n "$TOOLS" ]; then
+if [ -z "$TOOLS" ]; then
+  # Check for YAML multi-line array (allowed-tools:\n  - item)
+  if echo "$FRONTMATTER" | grep -q '^allowed-tools:$'; then
+    NEXT_LINE=$(echo "$FRONTMATTER" | grep -A1 '^allowed-tools:$' | sed -n '2p')
+    if echo "$NEXT_LINE" | grep -qE '^[[:space:]]*-[[:space:]]'; then
+      fail "allowed-tools must be space-delimited, not YAML array"
+    fi
+  fi
+else
   # FAIL if commas found (must be space-delimited)
   if echo "$TOOLS" | grep -qE ','; then
     fail "allowed-tools must be space-delimited, not comma-separated: $TOOLS"
@@ -171,13 +179,6 @@ if [ -n "$TOOLS" ]; then
     pass "Allowed-tools field valid"
   fi
 fi
-# FAIL if YAML multi-line array detected (allowed-tools:\n  - item)
-if echo "$FRONTMATTER" | grep -q '^allowed-tools:$'; then
-  NEXT_LINE=$(echo "$FRONTMATTER" | grep -A1 '^allowed-tools:$' | sed -n '2p')
-  if echo "$NEXT_LINE" | grep -qE '^[[:space:]]*-[[:space:]]'; then
-    fail "allowed-tools must be space-delimited, not YAML array"
-  fi
-fi
 
 # 8. Check line count (max 500)
 LINE_COUNT=$(wc -l < "$SKILL_FILE" | tr -d ' ')
diff --git a/.github/workflows/skill-code-review.yml b/.github/workflows/skill-code-review.yml
index 88f8d228..cc2a1367 100644
--- a/.github/workflows/skill-code-review.yml
+++ b/.github/workflows/skill-code-review.yml
@@ -297,9 +297,28 @@ jobs:
           set -euo pipefail
           MARKER="<!-- gemini-code-review -->"
           WORKFLOW_URL="${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}"
+          TIMESTAMP=$(date -u '+%Y-%m-%d %H:%M UTC')
+
+          EXISTING_COMMENT_ID=$(gh api --paginate \
+            "/repos/${{ github.repository }}/issues/${PR_NUMBER}/comments" \
+            --jq ".[] | select(.body | startswith(\"$MARKER\")) | .id" \
+            | head -1)
+
+          EXISTING_BODY=""
+          RUN_NUMBER=1
+          if [ -n "$EXISTING_COMMENT_ID" ]; then
+            EXISTING_BODY=$(gh api \
+              "/repos/${{ github.repository }}/issues/comments/$EXISTING_COMMENT_ID" \
+              --jq '.body')
+            RUN_NUMBER=$(( $(echo "$EXISTING_BODY" | grep -c '### 🔄 Run #' || true) + 1 ))
+          fi
 
           {
-            echo "$MARKER"
+            if [ "$RUN_NUMBER" -eq 1 ]; then
+              echo "$MARKER"
+            fi
+            echo "### 🔄 Run #${RUN_NUMBER} — ${TIMESTAMP}"
+            echo ""
             echo "## Gemini Code Review"
             echo ""
 
@@ -317,21 +336,38 @@ jobs:
             fi
 
             echo ""
-            echo "---"
             echo "> [Workflow run]($WORKFLOW_URL)"
-          } > /tmp/comment.md
-
-          EXISTING_COMMENT_ID=$(gh api --paginate \
-            "/repos/${{ github.repository }}/issues/${PR_NUMBER}/comments" \
-            --jq ".[] | select(.body | startswith(\"$MARKER\")) | .id" \
-            | head -1)
+          } > /tmp/new_report.md
 
           if [ -n "$EXISTING_COMMENT_ID" ]; then
-            gh api --method PATCH \
+            {
+              echo "$EXISTING_BODY"
+              echo ""
+              echo "---"
+              echo ""
+              cat /tmp/new_report.md
+            } > /tmp/comment.md
+            gh api \
+              --method PATCH \
               "/repos/${{ github.repository }}/issues/comments/$EXISTING_COMMENT_ID" \
               -f body="$(cat /tmp/comment.md)"
           else
             gh pr comment "$PR_NUMBER" \
               --repo "${{ github.repository }}" \
-              --body-file /tmp/comment.md
+              --body-file /tmp/new_report.md
           fi
+
+      - name: React to PR
+        if: always()
+        uses: actions/github-script@v7
+        with:
+          script: |
+            const result = '${{ job.status }}';
+            const prNumber = Number('${{ steps.inputs.outputs.pr_number }}');
+            await github.rest.reactions.createForIssue({
+              owner: context.repo.owner,
+              repo: context.repo.repo,
+              issue_number: prNumber,
+              content: result === 'success' ? '+1' : '-1'
+            });
+
diff --git a/.github/workflows/skill-security-scan.yml b/.github/workflows/skill-security-scan.yml
index 81333e3d..64d2c265 100644
--- a/.github/workflows/skill-security-scan.yml
+++ b/.github/workflows/skill-security-scan.yml
@@ -220,6 +220,7 @@ jobs:
               --recursive \
               --use-behavioral \
               --use-llm \
+              --llm-consensus-runs 3 \
               --check-overlap \
               --enable-meta \
               --fail-on-severity medium \
@@ -237,6 +238,7 @@ jobs:
           fi
 
       - name: Upload security reports
+        id: upload
         if: steps.detect.outputs.pack_count != '0' && steps.creds.outputs.available == 'true' && always()
         uses: actions/upload-artifact@v4
         with:
@@ -257,6 +259,8 @@ jobs:
           set -euo pipefail
           MARKER="<!-- skill-security-scan -->"
           SCAN_RESULT="${{ steps.scan.outputs.scan_result }}"
+          ARTIFACT_URL="${{ steps.upload.outputs.artifact-url }}"
+          TIMESTAMP=$(date -u '+%Y-%m-%d %H:%M UTC')
 
           if [ "$SCAN_RESULT" = "passed" ]; then
             ICON="✅"
@@ -264,8 +268,26 @@ jobs:
             ICON="❌"
           fi
 
+          EXISTING_COMMENT_ID=$(gh api --paginate \
+            "/repos/${{ github.repository }}/issues/${PR_NUMBER}/comments" \
+            --jq ".[] | select(.body | startswith(\"$MARKER\")) | .id" \
+            | head -1)
+
+          EXISTING_BODY=""
+          RUN_NUMBER=1
+          if [ -n "$EXISTING_COMMENT_ID" ]; then
+            EXISTING_BODY=$(gh api \
+              "/repos/${{ github.repository }}/issues/comments/$EXISTING_COMMENT_ID" \
+              --jq '.body')
+            RUN_NUMBER=$(( $(echo "$EXISTING_BODY" | grep -c '### 🔄 Run #' || true) + 1 ))
+          fi
+
           {
-            echo "$MARKER"
+            if [ "$RUN_NUMBER" -eq 1 ]; then
+              echo "$MARKER"
+            fi
+            echo "### 🔄 Run #${RUN_NUMBER} — ${TIMESTAMP}"
+            echo ""
             echo "## $ICON Skill Security Scan"
             echo ""
 
@@ -286,20 +308,51 @@ jobs:
             fi
 
             echo ""
-            echo "> [Workflow run](${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }})"
-          } > /tmp/comment.md
-
-          EXISTING_COMMENT_ID=$(gh api --paginate \
-            "/repos/${{ github.repository }}/issues/${PR_NUMBER}/comments" \
-            --jq ".[] | select(.body | startswith(\"$MARKER\")) | .id" \
-            | head -1)
+            if [ -n "$ARTIFACT_URL" ]; then
+              echo "> 📦 [Download security reports]($ARTIFACT_URL) | [Workflow run](${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }})"
+            else
+              echo "> [Workflow run](${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }})"
+            fi
+          } > /tmp/new_report.md
 
           if [ -n "$EXISTING_COMMENT_ID" ]; then
-            gh api --method PATCH \
+            {
+              echo "$EXISTING_BODY"
+              echo ""
+              echo "---"
+              echo ""
+              cat /tmp/new_report.md
+            } > /tmp/comment.md
+            gh api \
+              --method PATCH \
               "/repos/${{ github.repository }}/issues/comments/$EXISTING_COMMENT_ID" \
               -f body="$(cat /tmp/comment.md)"
           else
             gh pr comment "$PR_NUMBER" \
               --repo "${{ github.repository }}" \
-              --body-file /tmp/comment.md
+              --body-file /tmp/new_report.md
+          fi
+
+      - name: Check scan results
+        if: steps.detect.outputs.pack_count != '0' && steps.creds.outputs.available == 'true'
+        run: |
+          if [ "${{ steps.scan.outputs.scan_result }}" = "failed" ]; then
+            echo "❌ Security scan found MEDIUM or higher severity issues — blocking merge"
+            exit 1
+          else
+            echo "✅ Security scan passed — no MEDIUM or higher severity issues found"
           fi
+
+      - name: React to PR
+        if: always()
+        uses: actions/github-script@v7
+        with:
+          script: |
+            const result = '${{ job.status }}';
+            const prNumber = Number('${{ steps.inputs.outputs.pr_number }}');
+            await github.rest.reactions.createForIssue({
+              owner: context.repo.owner,
+              repo: context.repo.repo,
+              issue_number: prNumber,
+              content: result === 'success' ? '+1' : '-1'
+            });
diff --git a/rh-virt/mcps.json b/rh-virt/mcps.json
index 22f6d970..74a95e99 100644
--- a/rh-virt/mcps.json
+++ b/rh-virt/mcps.json
@@ -9,8 +9,8 @@
         "--network=host",
         "--userns=keep-id:uid=65532,gid=65532",
         "-v", "${KUBECONFIG}:/kubeconfig:ro,Z",
-        "--entrypoint", "/openshift-mcp-server",
-        "quay.io/redhat-user-workloads/crt-nshift-lightspeed-tenant/openshift-mcp-server@sha256:2f52c860f91ab3c8a5129b727bdef0d620e733013f073b10355866c45eafd053",
+        "--entrypoint", "/app/kubernetes-mcp-server",
+        "quay.io/ecosystem-appeng/openshift-mcp-server@sha256:3531cb78f51f8c7ebcdb21adc21358ab8924116994848a3ce1ff542b3fc23742",
         "--kubeconfig", "/kubeconfig",
         "--toolsets", "core,kubevirt"
       ],
diff --git a/rh-virt/skills/vm-clone/SKILL.md b/rh-virt/skills/vm-clone/SKILL.md
index 5bc3a990..f19f3719 100644
--- a/rh-virt/skills/vm-clone/SKILL.md
+++ b/rh-virt/skills/vm-clone/SKILL.md
@@ -151,7 +151,11 @@ Parse: Extract storage names, calculate size, determine DataSources vs container
 
 **Option 2 warning**: `⚠️ Shared Storage Dangerous - Both VMs share disk, data corruption risk. Only safe if source stopped. Use Option 1 instead. Proceed anyway? (yes/cancel)` Wait for explicit "yes".
 
-### Step 3: Present Clone Configuration for Confirmation
+### Step 3: Check Namespace Quota and Present Clone Configuration
+
+**3.0: Check ResourceQuota** (before presenting confirmation)
+
+Use `resources_list` (apiVersion="v1", kind="ResourceQuota", namespace=`<target-namespace>`) to check if quotas exist. If a quota is found, compare current usage against limits for CPU, memory, and storage. If the clone would push usage above 80% of any limit, include a warning in the confirmation summary.
 
 **Present configuration summary:**
 
@@ -275,8 +279,8 @@ Would you like help troubleshooting this error?
 
 ### Batch Cloning
 **User request:** "Create 3 copies of template-vm named web-01, web-02, web-03"
-**Workflow**: Validate source once, generate/check target names, present combined scope, ask storage strategy, confirm, execute sequentially
-**Batch confirmation**: Show source, targets list, strategy, total impact (VMs, storage, vCPU, memory), estimated time
+**Limit**: Maximum 5 clones per batch request. If user requests more, refuse and explain the limit exists to prevent resource exhaustion.
+**Workflow**: Validate source once, generate/check target names, ask storage strategy once, then process each clone individually through Steps 3-5 with separate confirmation per VM. Stop on first failure.
 
 ### Cross-Namespace Cloning
 **User request:** "Clone production-vm from production to staging namespace"
@@ -417,17 +421,22 @@ Agent: "✓ VM Cloned Successfully"
 ```
 User: "Create 3 copies of template-vm named web-01, web-02, web-03 in production"
 
-Agent: [Validates source, checks all names, presents batch review]
-       "Source: template-vm, Targets: 3 VMs (web-01, web-02, web-03)"
-       "Total Impact: 90Gi, 6 vCPU, 12Gi, ~20-30 min"
-       "Proceed? (yes/no)"
+Agent: [Validates source, checks all names, asks storage strategy once]
+       "Storage strategy for all clones: Clone Storage (1)"
+
+       ## Clone 1 of 3: web-01
+       "Source: template-vm → Target: web-01 (production), 30Gi clone"
+       "Proceed with VM cloning? (yes/no)"
+User: "yes"
+Agent: "✓ web-01 cloned"
 
+       ## Clone 2 of 3: web-02
+       "Source: template-vm → Target: web-02 (production), 30Gi clone"
+       "Proceed with VM cloning? (yes/no)"
 User: "yes"
+Agent: "✓ web-02 cloned"
 
-Agent: "📦 Cloning VM 1 of 3: web-01... ✓"
-       "📦 Cloning VM 2 of 3: web-02... ✓"
-       "📦 Cloning VM 3 of 3: web-03... ✓"
-       "✓ Batch Cloning Completed - 3 VMs, 90Gi storage, all Stopped"
+       [... repeats for each VM with individual confirmation ...]
 ```
 
 ### Example 4: Shared Storage Warning
diff --git a/rh-virt/skills/vm-create/SKILL.md b/rh-virt/skills/vm-create/SKILL.md
index 2468ea52..8afe1430 100644
--- a/rh-virt/skills/vm-create/SKILL.md
+++ b/rh-virt/skills/vm-create/SKILL.md
@@ -83,8 +83,7 @@ Create virtual machines in OpenShift Virtualization using the `vm_create` tool f
 **Optional (use defaults):** OS (fedora), Size (medium), Storage (30Gi), Performance (u1), Autostart (false)
 
 **Gather cluster info:**
-- Detect current namespace: `kubectl config view --minify -o jsonpath='{..namespace}' || echo "default"`
-- List namespaces: `namespaces_list` (from openshift-virtualization)
+- List namespaces: `namespaces_list` (from openshift-virtualization) — ask user to select if not provided
 - List StorageClasses: `resources_list` with apiVersion="storage.k8s.io/v1", kind="StorageClass"
 - Identify default SC: annotation `storageclass.kubernetes.io/is-default-class`="true"
 - Analyze SC: `.volumeBindingMode` (Immediate/WaitForFirstConsumer), provisioner (rbd/cephfs=RWX hint)
@@ -166,15 +165,13 @@ Confirm: yes/no/modify
 **Status interpretation:**
 - Stopped/Halted → Success (VM created, not started)
 - Running → Success (if autostart=true)
-- Provisioning → Wait 5s, check again
+- Provisioning → Wait 5s, check again (max 3 retries, then report status as "still provisioning" and suggest the user check back later)
 - ErrorUnschedulable → Execute diagnostic workflow (Step 5a)
 - ErrorDataVolumeNotReady → Storage issue (see Common Issues)
 
 #### 5a. Diagnostic Workflow (ErrorUnschedulable)
 
-**CRITICAL: Document Consultation FIRST:**
-1. Read [scheduling-errors.md](../../docs/troubleshooting/scheduling-errors.md) using Read tool
-2. Output: "I detected ErrorUnschedulable. I consulted [scheduling-errors.md] to understand diagnosis strategies."
+**Reference**: See [scheduling-errors.md](../../docs/troubleshooting/scheduling-errors.md) for diagnosis strategies.
 
 **Gather diagnostics:**
 - List events: `events_list` (namespace=`<namespace>`) → Filter for VM/VMI
@@ -194,7 +191,7 @@ Confirm: yes/no/modify
 
 ### Recommended Solution
 <workaround-description>
-**Command**: `oc patch vm <name> -n <namespace> ...`
+**Action**: Update VM via `resources_create_or_update`
 **Impact**: <what-changes>
 **Options**: 1) Apply workaround, 2) Manual, 3) Cancel, 4) Ignore
 ⚠️ MCP limitation: vm_create doesn't support tolerations
@@ -203,7 +200,7 @@ Confirm: yes/no/modify
 **Wait for user decision.**
 
 **If user confirms:**
-1. Apply patch: `resources_create_or_update` (fetch, add tolerations, update) OR `oc patch`
+1. Apply patch: `resources_create_or_update` (fetch, add tolerations, update)
 2. Verify: `resources_get` → Check `.spec.template.spec.tolerations`
 3. **Restart VM**: `vm_lifecycle` (action="restart") to apply new spec
 4. Wait 15-20s, check status → Stopped → Provisioning → Running
@@ -241,7 +238,7 @@ Start: "Start VM <name>" | View: "Show VM <name>"
 **Error**: <error-message>
 
 **Common Causes**:
-- Namespace not exists → `oc create namespace <name>`
+- Namespace not exists → Create via `resources_create_or_update`
 - RBAC denied → Check ServiceAccount permissions
 - Resource constraints → Try smaller size
 - Invalid parameters → Verify OS, size, storage format
@@ -254,7 +251,7 @@ Troubleshooting: See Common Issues
 
 ### Issue 1: Namespace Not Found
 **Error**: "Namespace 'xyz' not found"
-**Solution**: List with `namespaces_list`, create with `resources_create_or_update` or `oc create namespace <name>`
+**Solution**: List with `namespaces_list`, create with `resources_create_or_update`
 
 ### Issue 2: Insufficient Permissions
 **Error**: "Forbidden: User cannot create VirtualMachines"
@@ -331,7 +328,7 @@ Troubleshooting: See Common Issues
 - **RBAC**: Requires create VirtualMachines (kubevirt.io/v1) in namespace
 - **Namespace Isolation**: VMs in specified namespace only
 - **Storage Quotas**: Respects ResourceQuotas
-- **Image Security**: Uses official images from trusted registries
+- **Image Security**: Default OS options (fedora, ubuntu, rhel, centos-stream, debian, opensuse) use upstream container disk images. Custom images via the `workload` parameter are not validated — warn the user when a custom image URL is provided and confirm the source is trusted before proceeding
 - **KUBECONFIG**: Never exposed (presence only)
 - **Audit**: All ops logged via K8s audit
 
@@ -392,6 +389,6 @@ Agent: [Creates]
 
 ## Advanced Features
 
-**Custom Images**: `vm_create({"workload": "quay.io/containerdisks/fedora:latest", ...})`
+**Custom Images**: `vm_create({"workload": "quay.io/containerdisks/fedora:latest", ...})` — ⚠️ When a custom image URL is provided, warn the user that image provenance is not verified and confirm the source is trusted before creating the VM
 **Secondary Networks**: `vm_create({"networks": ["vlan-network"], ...})` or `{"networks": [{"name": "eth1", "networkName": "vlan"}], ...}`
 **Explicit Instance Type**: `vm_create({"instancetype": "u1.large", ...})`
diff --git a/rh-virt/skills/vm-delete/SKILL.md b/rh-virt/skills/vm-delete/SKILL.md
index ac079445..e6d2d5f2 100644
--- a/rh-virt/skills/vm-delete/SKILL.md
+++ b/rh-virt/skills/vm-delete/SKILL.md
@@ -394,7 +394,7 @@ This is preview only. No resources deleted.
 ## Advanced Features
 
 ### Batch Deletion
-Delete multiple VMs with confirmation for each: `"Delete VMs test-01, test-02, test-03 in dev"` → Process each individually with full workflow. Use typed confirmation: `DELETE-3-VMS` for batch.
+Delete multiple VMs: `"Delete VMs test-01, test-02, test-03 in dev"` → Process each VM individually through the full workflow (Steps 1-5), requiring typed confirmation of each VM name separately. Never use a generic batch confirmation — each VM must be confirmed by its exact name.
 
 ### Dry-Run Mode
 Show deletion scope without executing: Execute Step 1-2, skip Steps 3-4. User request: "Show what would be deleted if I delete VM xyz"
diff --git a/rh-virt/skills/vm-inventory/SKILL.md b/rh-virt/skills/vm-inventory/SKILL.md
index 3b9c78f9..8a292540 100644
--- a/rh-virt/skills/vm-inventory/SKILL.md
+++ b/rh-virt/skills/vm-inventory/SKILL.md
@@ -30,10 +30,6 @@ List and inspect virtual machines in OpenShift Virtualization clusters. This ski
 - `resources_list` (from openshift-virtualization) - List Kubernetes resources including VirtualMachines
 - `resources_get` (from openshift-virtualization) - Get specific Kubernetes resource details
 
-**Fallback CLI Commands** (if MCP tools unavailable):
-- `oc get virtualmachines` / `oc get vm` - List VirtualMachines
-- `oc get vm <name> -n <namespace> -o yaml` - Get VM details
-
 **Required Environment Variables**:
 - `KUBECONFIG` - Path to Kubernetes configuration file with cluster access
 
@@ -51,12 +47,7 @@ List and inspect virtual machines in OpenShift Virtualization clusters. This ski
 
 **Human Notification Protocol:** `❌ Cannot execute vm-inventory: MCP server not available. Setup: Add to mcps.json, set KUBECONFIG, restart Claude Code. Docs: https://github.com/openshift/openshift-mcp-server`
 
-⚠️ **SECURITY**: Never display KUBECONFIG path or credential values.
-
-**Note on Fallback Behavior**:
-- If MCP server is unavailable but KUBECONFIG is set, the skill CAN proceed with CLI commands
-- Always offer the user the choice between setup (MCP) or CLI fallback
-- CLI fallback requires explicit user confirmation before executing any commands
+⚠️ **SECURITY**: Never display KUBECONFIG path or credential values. Never fall back to CLI commands (`oc`, `kubectl`) — all operations must go through MCP tools exclusively.
 
 ## When to Use This Skill
 
@@ -83,11 +74,9 @@ List and inspect virtual machines in OpenShift Virtualization clusters. This ski
 ## Workflow
 
 **CRITICAL EXECUTION PATTERN**:
-1. **ALWAYS attempt MCP server tools FIRST** - Try `resources_list` or `resources_get`
-2. **If MCP tools fail** - Propose CLI commands (`oc get vm`) with user confirmation
-3. **Never skip MCP attempt** - Always try them first
-
-**Tool Execution Priority**: MCP tools (primary) → CLI commands (fallback with confirmation)
+1. **Use MCP server tools exclusively** - `resources_list` or `resources_get`
+2. **If MCP tools fail** - Report the error and guide the user to fix MCP setup
+3. **Never use CLI commands** - No `oc` or `kubectl` execution under any circumstances
 
 ### Workflow A: List All VMs (Across All Namespaces)
 
@@ -95,11 +84,13 @@ List and inspect virtual machines in OpenShift Virtualization clusters. This ski
 
 **MCP Tool**: `resources_list` (apiVersion="kubevirt.io/v1", kind="VirtualMachine", allNamespaces=true)
 
-**Errors:** Tool not found/connection error → Report, offer CLI fallback: `oc get virtualmachines -A -o json`
+**Errors:** Tool not found/connection error → Report error, guide user to fix MCP setup
 
 **Step 2: Get Resource Details for Running VMs**
 
-**CRITICAL**: To display complete VM information, query VirtualMachineInstance (VMI) resources:
+**Large cluster safeguard**: If Step 1 returned more than 20 VMs, skip VMI queries entirely. Display a summary table using only VM resource data (Name, Namespace, Status) and suggest the user narrow down by namespace: `⚠️ Found <N> VMs across all namespaces. Showing summary view. Use "List VMs in namespace <ns>" for full details.`
+
+To display complete VM information (when 20 or fewer VMs), query VirtualMachineInstance (VMI) resources:
 
 **MCP Tool**: `resources_list` (apiVersion="kubevirt.io/v1", kind="VirtualMachineInstance")
 
@@ -158,7 +149,7 @@ Ask user for namespace if not provided.
 
 **MCP Tool**: `resources_list` (apiVersion="kubevirt.io/v1", kind="VirtualMachine", namespace=`<namespace>`)
 
-**Errors:** Tool fails → Report, offer CLI fallback: `oc get virtualmachines -n <namespace> -o json`
+**Errors:** Tool fails → Report error, guide user to fix MCP setup
 
 **Step 3: Get Resource Details and Display**
 
@@ -186,14 +177,11 @@ Required: VM name, Namespace (ask if not provided)
 
 **MCP Tool**: `resources_get` (apiVersion="kubevirt.io/v1", kind="VirtualMachine", namespace=`<namespace>`, name=`<vm-name>`)
 
-**Errors:** Tool fails → Report, offer CLI fallback: `oc get vm <vm-name> -n <namespace> -o yaml`
+**Errors:** Tool fails → Report error, guide user to fix MCP setup
 
-**Step 3: Interpret Status and Conditions (Optional)**
+**Step 3: Interpret Status and Conditions**
 
-**OPTIONAL**: If VM has error status (ErrorUnschedulable, ErrorDataVolumeNotReady, CrashLoopBackOff), consult [troubleshooting/INDEX.md](../../docs/troubleshooting/INDEX.md) using Read tool. Output: "Consulted INDEX.md to interpret status."
-
-**When to consult**: VM status is Error/Warning or stuck state (CrashLoopBackOff, Terminating)
-**When NOT to consult**: VM status is normal (Running, Stopped, Provisioning)
+Report the VM status as-is from the API response. Do NOT read external documentation files to interpret status — use the status indicators defined in the Output Formatting Guidelines section below. If the user needs troubleshooting guidance, suggest they use a dedicated troubleshooting skill instead.
 
 **Step 4: Display Detailed Information**
 
@@ -253,7 +241,7 @@ Required: VM name, Namespace (ask if not provided)
 - By Status (post-processing): Filter results by `status.printableStatus` field
 - By Resource Size (post-processing): Parse instance type or VMI resource specs
 
-**Errors:** Tool fails → Report, offer CLI fallback: `oc get virtualmachines -A -l <labelSelector> -o json`
+**Errors:** Tool fails → Report error, guide user to fix MCP setup
 
 **Step 2: Display Filtered Results**
 
@@ -304,14 +292,7 @@ Display with explanation: `## 📋 VMs with label 'app=web'` + list/table using
 - `resources_list` - List resources (apiVersion, kind, namespace optional, allNamespaces optional, labelSelector optional)
 - `resources_get` - Get resource details (apiVersion, kind, namespace, name)
 
-### CLI Fallback Commands (Use only if MCP tools fail)
-- `oc get virtualmachines` / `oc get vm` - List VirtualMachines
-- `oc get vm <name> -n <namespace>` - Get specific VM
-- `oc get vm -A` - List VMs across all namespaces
-- `oc get vm -n <namespace>` - List VMs in specific namespace
-- `oc get vm -l <selector>` - Filter VMs by label selector
-
-**Important**: Always attempt MCP tools first. Only use CLI commands after MCP tool failure and with user confirmation.
+**Important**: All operations must use MCP tools exclusively. CLI commands (`oc`, `kubectl`) are prohibited to prevent command injection risks.
 
 ### Related Skills
 - `vm-create` - Create VMs after checking inventory
@@ -319,28 +300,14 @@ Display with explanation: `## 📋 VMs with label 'app=web'` + list/table using
 - `vm-troubleshooter` (planned) - Diagnose problematic VMs from inventory
 
 ### Reference Documentation
-- [Troubleshooting INDEX](../../docs/troubleshooting/INDEX.md) - VM status interpretation (optionally consulted when displaying VM details with error states)
 - [OpenShift Virtualization Documentation](https://docs.redhat.com/en/documentation/openshift_container_platform/4.21/html-single/virtualization/index#virt/about_virt/about-virt.html)
 - [KubeVirt VirtualMachine API](https://kubevirt.io/api-reference/)
 - [Accessing VMs](https://docs.redhat.com/en/documentation/openshift_container_platform/4.21/html-single/virtualization/index#virt/virtual_machines/virt-accessing-vm-consoles.html)
 - [VM Status Conditions](https://kubevirt.io/user-guide/virtual_machines/vm_status_conditions/)
 
-## Critical: Human-in-the-Loop Requirements
-
-**Not applicable** - This skill performs read-only operations and does not modify any cluster resources. No user confirmation required.
-
-**Read-only operations:**
-- Listing VirtualMachines across namespaces or in specific namespaces
-- Retrieving VM details, status, and resource configurations
-- Displaying VM health conditions and resource usage
-- Filtering VMs by labels or field selectors
-- Viewing VM network, storage, and node placement information
+## Human-in-the-Loop Requirements
 
-**No modifications performed:**
-- ✓ Does not change VM state (start/stop/restart)
-- ✓ Does not modify VM configuration
-- ✓ Does not delete VMs or resources
-- ✓ Does not consume cluster resources
+This skill is **read-only** — no user confirmation is required. It does not change VM state, modify configuration, delete resources, or consume cluster capacity.
 
 ## Security Considerations
 
@@ -362,15 +329,14 @@ Agent: [MCP: resources_list(apiVersion="kubevirt.io/v1", kind="VirtualMachine",
        [Displays table format from Workflow A Step 3]
 ```
 
-### Example 2: CLI fallback when MCP unavailable
+### Example 2: MCP unavailable
 
 ```
 User: "List all VMs"
 Agent: [MCP tool fails]
-       ⚠️ MCP tool 'resources_list' not available. Use CLI: `oc get virtualmachines -A`?
-User: "yes"
-Agent: [Executes: oc get virtualmachines -A -o json]
-       [Displays table format]
+       ❌ Cannot execute vm-inventory: MCP server not available.
+       Setup: Add openshift-virtualization to mcps.json, set KUBECONFIG, restart Claude Code.
+       Docs: https://github.com/openshift/openshift-mcp-server
 ```
 
 ### Example 3: Get specific VM details
diff --git a/rh-virt/skills/vm-lifecycle-manager/SKILL.md b/rh-virt/skills/vm-lifecycle-manager/SKILL.md
index be1903f6..b7f8ed71 100644
--- a/rh-virt/skills/vm-lifecycle-manager/SKILL.md
+++ b/rh-virt/skills/vm-lifecycle-manager/SKILL.md
@@ -220,6 +220,8 @@ Confirm: yes/no
 
 If user says "no" or wants to reconsider, do not proceed.
 
+**Batch operations**: When the user requests actions on multiple VMs, process each VM individually through the full workflow (Step 1-3) with separate confirmation for each. Never use a single bulk confirmation for multiple VMs.
+
 **Why**: start (consumes resources), stop (interrupts services), restart (brief downtime). User should verify correct VM and understand impact.
 
 ## Security Considerations
@@ -282,7 +284,7 @@ Impact: Running after stop+start. Brief interruption. Monitor app logs.
 ```
 User: "Start web-server in namespace vms"
 Agent: [vm-lifecycle-manager skill]
-       [vm_lifecycle(action="start")]
+       [Checks VM status via resources_get — already Running]
 ## ℹ️ VM Already Running
 VM: `web-server` | Namespace: `vms` | Status: Running
 Result: No action taken - VM already in desired state.
@@ -293,16 +295,18 @@ To restart: "Restart VM web-server in namespace vms"
 
 ```
 User: "Stop VMs web-01, web-02, web-03 in namespace production"
-Agent: [vm-lifecycle-manager skill - batch mode]
-## Batch Lifecycle Operation
-Stopping 3 VMs in 'production': web-01, web-02, web-03
-Impact: All 3 VMs will shut down, services interrupted.
+Agent: [vm-lifecycle-manager skill — processes each VM individually with full workflow]
+
+## VM Lifecycle Operation
+| VM Name | `web-01` | Namespace | `production` | Action | `stop` | graceful shutdown |
+Confirm: yes/no
+User: "yes"
+Agent: [vm_lifecycle(namespace="production", name="web-01", action="stop")]
+✓ web-01: Stopped
+
+## VM Lifecycle Operation
+| VM Name | `web-02` | Namespace | `production` | Action | `stop` | graceful shutdown |
 Confirm: yes/no
 User: "yes"
-Agent: [Executes vm_lifecycle for each VM sequentially]
-## ✓ Batch Stop Successful
-- web-01: Stopped
-- web-02: Stopped
-- web-03: Stopped
-All VMs stopped. Resources freed.
+[...repeats for each VM with individual confirmation...]
 ```
diff --git a/rh-virt/skills/vm-rebalance/SKILL.md b/rh-virt/skills/vm-rebalance/SKILL.md
index 8b77b5c0..af1e04ee 100644
--- a/rh-virt/skills/vm-rebalance/SKILL.md
+++ b/rh-virt/skills/vm-rebalance/SKILL.md
@@ -90,8 +90,8 @@ Orchestrate VM migrations across OpenShift cluster nodes for load balancing, mai
 **For Manual Mode:**
 
 **Document Consultation** (REQUIRED - Execute FIRST):
-1. Read [REBALANCE_MANUAL.md](./REBALANCE_MANUAL.md) using Read tool
-2. Output: "I consulted [REBALANCE_MANUAL.md](rh-virt/skills/vm-rebalance/REBALANCE_MANUAL.md) to understand the manual migration workflow."
+1. **Action**: Read [REBALANCE_MANUAL.md](./REBALANCE_MANUAL.md) using Read tool
+2. **Output to user**: "I consulted [REBALANCE_MANUAL.md](./REBALANCE_MANUAL.md) to understand the manual migration workflow."
 3. **Then execute**: Follow workflow in REBALANCE_MANUAL.md
 
 ---
@@ -99,8 +99,8 @@ Orchestrate VM migrations across OpenShift cluster nodes for load balancing, mai
 **For Automatic Mode:**
 
 **Document Consultation** (REQUIRED - Execute FIRST):
-1. Read [REBALANCE_AUTOMATIC.md](./REBALANCE_AUTOMATIC.md) using Read tool
-2. Output: "I consulted [REBALANCE_AUTOMATIC.md](rh-virt/skills/vm-rebalance/REBALANCE_AUTOMATIC.md) to understand the automatic rebalancing workflow."
+1. **Action**: Read [REBALANCE_AUTOMATIC.md](./REBALANCE_AUTOMATIC.md) using Read tool
+2. **Output to user**: "I consulted [REBALANCE_AUTOMATIC.md](./REBALANCE_AUTOMATIC.md) to understand the automatic rebalancing workflow."
 3. **Then execute**: Follow workflow in REBALANCE_AUTOMATIC.md
 
 ## Common Validation Logic
@@ -143,7 +143,7 @@ Orchestrate VM migrations across OpenShift cluster nodes for load balancing, mai
 
 **Errors**: Not found → "Node doesn't exist" | Not Ready → "Choose different target" | Cordoned → "Uncordon or choose different target"
 
-**Reference**: [../../docs/troubleshooting/scheduling-errors.md](../../docs/troubleshooting/scheduling-errors.md)
+**Reference**: [scheduling-errors.md](../../docs/troubleshooting/scheduling-errors.md)
 
 ## Node Selection for Automatic Rebalancing
 
@@ -272,12 +272,12 @@ If any node's CPU or Memory percentage **exceeds 100%** after rebalancing:
 ### Error 1: Live Migration Fails - Storage Not RWX
 **Symptom**: "Cannot live migrate: PVC access mode is ReadWriteOnce"
 **Solution**: Use cold migration OR convert PVC to RWX
-**Reference**: [../../docs/troubleshooting/storage-errors.md](../../docs/troubleshooting/storage-errors.md)
+**Reference**: [storage-errors.md](../../docs/troubleshooting/storage-errors.md)
 
 ### Error 2: VM Stuck ErrorUnschedulable After Cold Migration
 **Symptom**: "VM cannot be scheduled: ErrorUnschedulable"
 **Solution**: Check node capacity (`nodes_top`), verify no blocking taints (`resources_get` Node), add tolerations, choose different target, remove nodeSelector
-**Reference**: [../../docs/troubleshooting/scheduling-errors.md](../../docs/troubleshooting/scheduling-errors.md)
+**Reference**: [scheduling-errors.md](../../docs/troubleshooting/scheduling-errors.md)
 
 ### Error 3: Live Migration Times Out
 **Symptom**: "Migration exceeded timeout: 150s per GiB"
@@ -331,10 +331,10 @@ If any node's CPU or Memory percentage **exceeds 100%** after rebalancing:
 - [references/production-considerations.md](./references/production-considerations.md) - HA, capacity, security
 
 **Troubleshooting**:
-- [../../docs/troubleshooting/INDEX.md](../../docs/troubleshooting/INDEX.md) - Master index
-- [../../docs/troubleshooting/scheduling-errors.md](../../docs/troubleshooting/scheduling-errors.md) - ErrorUnschedulable, taints
-- [../../docs/troubleshooting/storage-errors.md](../../docs/troubleshooting/storage-errors.md) - PVC access modes
-- [../../docs/troubleshooting/lifecycle-errors.md](../../docs/troubleshooting/lifecycle-errors.md) - VM start/stop
+- [Troubleshooting INDEX](../../docs/troubleshooting/INDEX.md) - Master index
+- [scheduling-errors.md](../../docs/troubleshooting/scheduling-errors.md) - ErrorUnschedulable, taints
+- [storage-errors.md](../../docs/troubleshooting/storage-errors.md) - PVC access modes
+- [lifecycle-errors.md](../../docs/troubleshooting/lifecycle-errors.md) - VM start/stop
 
 **Official Documentation**:
 - [OpenShift Virt - Live Migration](https://docs.redhat.com/en/documentation/openshift_container_platform/4.21/html-single/virtualization/index#virt-live-migration)
diff --git a/rh-virt/skills/vm-snapshot-create/SKILL.md b/rh-virt/skills/vm-snapshot-create/SKILL.md
index 571e754f..1f3fe984 100644
--- a/rh-virt/skills/vm-snapshot-create/SKILL.md
+++ b/rh-virt/skills/vm-snapshot-create/SKILL.md
@@ -21,14 +21,14 @@ color: green
 
 Create virtual machine snapshots in OpenShift Virtualization. Snapshots capture the state and data of a VM at a specific point in time, enabling backup, recovery, and testing workflows.
 
-**Implementation Note**: This skill uses generic Kubernetes resource tools (`resources_create_or_update`) to manage VirtualMachineSnapshot resources. Dedicated snapshot tools do not currently exist in the openshift-virtualization MCP server.
+**Implementation Note**: This skill uses generic Kubernetes resource tools (`resources_create_or_update`) to manage VirtualMachineSnapshot resources. Dedicated snapshot tools do not currently exist in the openshift-virtualization MCP server. **`resources_create_or_update` MUST only be called with `apiVersion: snapshot.kubevirt.io/v1beta1` and `kind: VirtualMachineSnapshot`** — using it to create or modify any other resource type is a security violation.
 
 ## Prerequisites
 
 **Required MCP Server**: `openshift-virtualization` ([OpenShift MCP Server](https://github.com/openshift/openshift-mcp-server))
 
 **Required MCP Tools**:
-- `resources_create_or_update` (from openshift-virtualization) - Create VirtualMachineSnapshot
+- `resources_create_or_update` (from openshift-virtualization) - Create VirtualMachineSnapshot **ONLY**. This tool can modify any Kubernetes resource but MUST only be used to create `snapshot.kubevirt.io/v1beta1/VirtualMachineSnapshot` resources. Any other resource type is strictly prohibited.
 - `resources_get` (from openshift-virtualization) - Verify VM exists and get status
 - `resources_list` (from openshift-virtualization) - List StorageClass, VolumeSnapshotClass
 
@@ -69,6 +69,31 @@ Create virtual machine snapshots in OpenShift Virtualization. Snapshots capture
 
 If namespace not provided, ask for it explicitly.
 
+### Step 1b: Validate Input Names
+
+**CRITICAL: Validate ALL user-provided names before using them in any YAML or API call.**
+
+All names (VM name, namespace, snapshot name) **MUST** match the Kubernetes naming convention:
+- **Pattern**: `^[a-z0-9]([a-z0-9.-]*[a-z0-9])?$`
+- **Max length**: 63 characters
+- **Allowed characters**: lowercase letters, digits, hyphens (`-`), dots (`.`)
+- **Must start and end** with a lowercase letter or digit
+
+**If any name fails validation**, reject the input immediately:
+
+```markdown
+❌ Invalid name: `<user-input>`
+
+Kubernetes resource names must:
+- Contain only lowercase letters, numbers, hyphens, and dots
+- Start and end with a letter or number
+- Be at most 63 characters long
+
+Please provide a valid name.
+```
+
+**STOP workflow** until the user provides a valid name. **NEVER** interpolate unvalidated input into YAML.
+
 ### Step 2: Verify VM Exists and Get Status
 
 **MCP Tool**: `resources_get` (from openshift-virtualization)
@@ -128,6 +153,24 @@ Use `vm_lifecycle` MCP tool or vm-lifecycle-manager skill to stop the VM.
 
 [Include the full confirmation template with storage backend analysis, guest agent status, volumes to snapshot, etc.]
 
+**Display the exact YAML that will be sent to the API** so the user can verify it before confirming:
+
+```markdown
+**YAML to be applied:**
+\`\`\`yaml
+apiVersion: snapshot.kubevirt.io/v1beta1
+kind: VirtualMachineSnapshot
+metadata:
+  name: <snapshot-name>
+  namespace: <namespace>
+spec:
+  source:
+    apiGroup: kubevirt.io
+    kind: VirtualMachine
+    name: <vm-name>
+\`\`\`
+```
+
 **Wait for user confirmation.**
 
 **Handle response:**
@@ -140,7 +183,11 @@ Use `vm_lifecycle` MCP tool or vm-lifecycle-manager skill to stop the VM.
 
 **MCP Tool**: `resources_create_or_update` (from openshift-virtualization)
 
-**Construct VirtualMachineSnapshot YAML:**
+**If snapshot name not provided by user**, generate one:
+- Format: `<vm-name>-snapshot-<timestamp>`
+- Example: `database-01-snapshot-20260218-143022`
+
+**Fixed YAML Template — use this EXACT structure. Do NOT add, remove, or modify any fields beyond the three placeholder values (`<snapshot-name>`, `<namespace>`, `<vm-name>`):**
 
 ```yaml
 apiVersion: snapshot.kubevirt.io/v1beta1
@@ -155,10 +202,6 @@ spec:
     name: <vm-name>
 ```
 
-**If snapshot name not provided by user**, generate one:
-- Format: `<vm-name>-snapshot-<timestamp>`
-- Example: `database-01-snapshot-20260218-143022`
-
 **Parameters**:
 ```json
 {
@@ -166,6 +209,18 @@ spec:
 }
 ```
 
+### Step 8b: Post-Construction Verification
+
+**CRITICAL: Before calling `resources_create_or_update`, verify the constructed YAML. This is a MANDATORY gate — never skip it.**
+
+1. `apiVersion` is exactly `snapshot.kubevirt.io/v1beta1` — any other apiVersion is a **security violation**, STOP immediately
+2. `kind` is exactly `VirtualMachineSnapshot` — any other kind is a **security violation**, STOP immediately
+3. Only these fields exist: `metadata.name`, `metadata.namespace`, `spec.source.apiGroup`, `spec.source.kind`, `spec.source.name`
+4. No additional fields, labels, annotations, or nested objects are present
+5. All placeholder values (`<snapshot-name>`, `<namespace>`, `<vm-name>`) have been replaced with validated user input from Step 1b
+
+**If ANY check fails, STOP and report a validation error. Do NOT send it to the API. Do NOT attempt to fix the YAML — cancel the operation entirely.**
+
 **Report progress:**
 ```markdown
 📸 Creating VM snapshot...
@@ -233,7 +288,7 @@ Check `status.phase`:
 - `openshift-virtualization` - OpenShift MCP server with kubevirt toolset
 
 ### Required MCP Tools
-- `resources_create_or_update` (from openshift-virtualization) - Create VirtualMachineSnapshot
+- `resources_create_or_update` (from openshift-virtualization) - Create VirtualMachineSnapshot **ONLY** (never for other resource types)
 - `resources_get` (from openshift-virtualization) - Verify VM and snapshot status
 - `resources_list` (from openshift-virtualization) - List StorageClass, VolumeSnapshotClass
 
@@ -327,13 +382,22 @@ Check `status.phase`:
 
 ## Security Considerations
 
-- **RBAC Enforcement**: Requires permissions for VirtualMachineSnapshot resources
+- **Input Validation**: All user-provided names (VM, namespace, snapshot) are validated against Kubernetes naming rules (`^[a-z0-9]([a-z0-9.-]*[a-z0-9])?$`, max 63 chars) before use
+- **Fixed YAML Template**: Resource YAML uses a fixed structure with only three replaceable values — no additional fields may be added
+- **Post-Construction Verification**: The constructed YAML is verified against expected structure before being sent to the API
+- **Human Review of YAML**: The exact YAML to be applied is shown to the user in the confirmation step
+- **RBAC Enforcement**: Requires permissions for VirtualMachineSnapshot resources — the ServiceAccount should be scoped to only VirtualMachineSnapshot create/get in target namespaces
 - **Storage Quotas**: Respects namespace storage quotas
 - **Hot-Plugged Volume Detection**: Prevents snapshots when hot-plugged volumes present
 - **KUBECONFIG Security**: Credentials never exposed in output
 - **Namespace Isolation**: Snapshots scoped to namespace boundaries
 - **Audit Trail**: All snapshot operations logged in Kubernetes API audit logs
 
+**Defense-in-depth note**: This skill uses `resources_create_or_update`, a generic MCP tool capable of creating or modifying any Kubernetes resource. The controls in this skill (input validation, fixed YAML template, post-construction verification, allowed-tools declaration) are instruction-level — they guide the LLM but are not hard security boundaries. The real enforcement layers are:
+- **Kubernetes RBAC**: The KUBECONFIG ServiceAccount should follow least-privilege — grant only `create` and `get` for `VirtualMachineSnapshot` in target namespaces, not broad resource permissions
+- **Admission webhooks**: Can enforce server-side policies on resource creation
+- **Future migration**: If the openshift-virtualization MCP server adds dedicated snapshot tools, this skill should migrate to them and drop `resources_create_or_update`
+
 ## Example Usage
 
 ### Example 1: Create Snapshot Before Upgrade
diff --git a/rh-virt/skills/vm-snapshot-delete/SKILL.md b/rh-virt/skills/vm-snapshot-delete/SKILL.md
index 561cabf2..f7c292ad 100644
--- a/rh-virt/skills/vm-snapshot-delete/SKILL.md
+++ b/rh-virt/skills/vm-snapshot-delete/SKILL.md
@@ -53,15 +53,11 @@ Permanently delete virtual machine snapshots in OpenShift Virtualization. Deleti
 ## When to Use This Skill
 
 **Trigger this skill when:**
-- User wants to free storage by removing old snapshots
-- User wants to delete a specific snapshot
-- User wants to implement snapshot retention policies
+- User wants to delete a specific snapshot by name
 
 **User phrases that trigger this skill:**
 - "Delete snapshot pre-upgrade-backup"
-- "Remove old snapshots for VM database-01"
-- "Delete all snapshots older than 7 days"
-- "Free up snapshot storage"
+- "Remove snapshot pre-upgrade-backup in production"
 
 **Do NOT use this skill when:**
 - User wants to create snapshots → Use `vm-snapshot-create` skill
@@ -171,14 +167,14 @@ Delete operation cancelled.
 
 ---
 
-**Proceed with snapshot deletion? This action cannot be undone. (yes/no)**
+**Type the snapshot name `<snapshot-name>` to confirm deletion (cannot be undone):** _____
 ```
 
-**Wait for user confirmation.**
+**Wait for user typed confirmation.**
 
 **Handle response:**
-- If "yes" → Proceed to Step 5 (execute deletion)
-- If "no", "cancel", or anything else → Cancel operation
+- If input matches snapshot name exactly (case-sensitive) → Proceed to Step 5
+- If mismatch → Report: `❌ Confirmation failed. You typed: <input>. Expected: <snapshot-name>. Cancelled.` **STOP workflow.**
 
 **On cancellation:**
 ```markdown
@@ -368,11 +364,12 @@ Would you like help troubleshooting this error?
    - Show snapshot details (VM, age, size)
    - Confirm snapshot won't be needed for recovery
    - List other available snapshots for the VM
-   - Ask: "Proceed with snapshot deletion? (yes/no)"
-   - Wait for explicit "yes"
+   - **Require typed confirmation**: user must type the exact snapshot name to confirm
+   - Accept only exact match (case-sensitive) — mismatch cancels the operation
 
 2. **Never Auto-Execute**
-   - **NEVER delete without user confirmation**
+   - **NEVER delete without typed confirmation**
+   - **NEVER accept a simple "yes" — require the snapshot name**
    - **ALWAYS show what will be lost before deletion**
 
 **Why This Matters:**
@@ -383,7 +380,7 @@ Would you like help troubleshooting this error?
 ## Security Considerations
 
 - **RBAC Enforcement**: Requires delete permissions for VirtualMachineSnapshot resources
-- **User Confirmation**: Always requires explicit "yes" before deletion
+- **Typed Confirmation**: Requires exact snapshot name to confirm deletion — prevents accidental "yes"
 - **Last Snapshot Warning**: Warns users when deleting the only snapshot for a VM
 - **Namespace Isolation**: Snapshots scoped to namespace boundaries
 - **Audit Trail**: Deletions logged in Kubernetes API audit logs
@@ -421,9 +418,9 @@ Impact of Deletion:
 Available snapshots for VM `database-01`:
 - `database-01-pre-upgrade` (created 2024-01-15 10:30)
 
-Proceed with snapshot deletion? This action cannot be undone. (yes/no)
+Type the snapshot name `database-01-daily-backup` to confirm deletion (cannot be undone): _____
 
-User: "yes"
+User: "database-01-daily-backup"
 
 Agent: [Step 5: Deletes snapshot]
 
diff --git a/rh-virt/skills/vm-snapshot-restore/SKILL.md b/rh-virt/skills/vm-snapshot-restore/SKILL.md
index 05f70803..cd3f395f 100644
--- a/rh-virt/skills/vm-snapshot-restore/SKILL.md
+++ b/rh-virt/skills/vm-snapshot-restore/SKILL.md
@@ -43,14 +43,11 @@ Restore virtual machines from snapshots in OpenShift Virtualization. **CRITICAL*
 ## When to Use This Skill
 
 **Trigger this skill when:**
-- User wants to restore a VM to a previous state
-- User wants to recover from failed changes/upgrades
-- User explicitly requests snapshot restore
+- User explicitly requests restoring a VM from a named snapshot
 
 **User phrases that trigger this skill:**
 - "Restore VM api-server from snapshot snapshot-20240115"
-- "Roll back database-01 to pre-upgrade snapshot"
-- "Recover VM web-server from backup"
+- "Roll back database-01 to snapshot pre-upgrade"
 
 **Do NOT use this skill when:**
 - User wants to create snapshots → Use `vm-snapshot-create` skill