fix(two-node): support proxy env file in bare metal diagnostic mode

.claude-plugin/marketplace.json

@@ -106,7 +106,7 @@
       "name": "two-node",
       "source": "./plugins/two-node",
       "description": "Two-node topology (TNA/TNF) workflow automation — RHEL verification, Jira integration, and more",
-      "version": "1.1.0"
+      "version": "1.1.1"
     }
   ]
 }

plugins/two-node/.claude-plugin/plugin.json

@@ -1,7 +1,7 @@
 {
   "name": "two-node",
   "description": "Two-node topology (TNA/TNF) workflow automation — RHEL verification, Jira integration, and more",
-  "version": "1.1.0",
+  "version": "1.1.1",
   "author": {
     "name": "lucaconsalvi"
   },

plugins/two-node/scripts/diagnose-cluster.sh

@@ -11,6 +11,7 @@
 #   BMC_USER, BMC_PASS — BMC credentials (required if BMC_0/BMC_1 set)
 #   SSH_KEY         — Path to SSH private key (default: ~/.ssh/id_rsa if it exists)
 #   KUBECONFIG      — Path to kubeconfig (bare metal mode; hypervisor mode auto-detects)
+#   PROXY_ENV       — Path to a proxy env file sourced before oc commands (bare metal mode)
 #   DEBUG           — Set to "1" to show SSH/command stderr (default: suppressed)
 #
 # Collects:
@@ -83,7 +84,10 @@ run_on_node() {
 
 run_oc() {
     if [ "$ACCESS_MODE" = "direct" ]; then
-        if [ -n "${KUBECONFIG:-}" ]; then
+        if [ -n "${PROXY_ENV:-}" ] && [ -f "$PROXY_ENV" ]; then
+            # shellcheck source=/dev/null
+            (source "$PROXY_ENV" > /dev/null 2>&1; oc "$@") 2>"$STDERR_REDIRECT"
+        elif [ -n "${KUBECONFIG:-}" ]; then
             KUBECONFIG="$KUBECONFIG" oc "$@" 2>"$STDERR_REDIRECT"
         else
             oc "$@" 2>"$STDERR_REDIRECT"

plugins/two-node/skills/cluster-diagnostic/README.md

@@ -0,0 +1,143 @@
+# TNF Cluster Diagnostic — Getting Started
+
+A Claude Code skill that diagnoses shutdown/recovery issues on Two-Node with
+Fencing (TNF) clusters. It gathers live cluster state via SSH, matches it
+against tested failure scenarios, and reports findings with severity
+classification. Read-only — it never modifies the cluster.
+
+## Install
+
+1. Install [Claude Code](https://docs.anthropic.com/en/docs/claude-code/overview)
+   if you haven't already.
+
+2. Add the edge-tooling plugin marketplace:
+
+   ```text
+   /plugin marketplace add openshift-eng/edge-tooling
+   ```
+
+3. Install the `two-node` plugin:
+
+   ```text
+   /plugin install two-node
+   ```
+
+## Set Up Cluster Access
+
+The skill connects to your TNF cluster over SSH. Pick whichever access pattern
+matches your setup:
+
+### Dev-scripts (hypervisor hop)
+
+```bash
+export HYPERVISOR=10.0.0.5   # your EC2 / hypervisor IP
+```
+
+Or, if you have the `two-node-toolbox/` submodule checked out, it auto-detects.
+
+### Bare metal (direct SSH)
+
+```bash
+export NODE_0=10.1.155.141
+export NODE_1=10.1.155.142
+```
+
+### Tip: use an env file
+
+Create `~/.tnf-cluster.env` with all your settings to avoid repeating exports:
+
+```bash
+export NODE_0=10.1.155.141
+export NODE_1=10.1.155.142
+export SSH_KEY=~/.ssh/my-key         # defaults to ~/.ssh/id_rsa
+export KUBECONFIG=/path/to/kubeconfig
+
+# If your cluster requires a proxy for oc commands, point PROXY_ENV
+# at the file instead of sourcing it directly. The diagnostic script
+# sources it internally — only for oc calls — so proxy vars don't
+# leak into Claude Code's shell and break its API/MCP connections.
+export PROXY_ENV=/path/to/auth/proxy.env
+
+# BMC vars — see "Debugging Fencing / BMC Issues" below
+export BMC_0=bmc-node0.example.com
+export BMC_1=bmc-node1.example.com
+export BMC_USER=admin
+export BMC_PASS="add BMC password here"
+```
+
+Source it before launching Claude Code: `source ~/.tnf-cluster.env`
+
+## Debugging Fencing / BMC Issues
+
+If you're debugging fencing failures, add the BMC variables to your env file
+(see above). Without them the skill still runs — it just skips BMC checks.
+With them, the
+diagnostic tests each layer of the fencing path from each node to its peer's
+BMC:
+
+1. **DNS resolution** — can the node resolve the peer BMC hostname?
+2. **Network reachability** — can it ping the peer BMC?
+3. **Redfish API** — is the BMC's HTTPS endpoint responding?
+4. **`fence_redfish --action=status`** — does the actual fence agent work
+   end-to-end with the configured credentials?
+5. **Power state** — what does the BMC report (On / Off / unknown)?
+
+The skill also checks STONITH configuration, fencing history, and failcounts
+from Pacemaker (these don't need BMC vars). Common issues it catches:
+
+- **BMC unreachable from the peer node** — fencing will fail even if the BMC
+  responds from your workstation. The test must run from the surviving node.
+- **Stale failcount after a transient BMC timeout** — the BMC recovered but
+  Pacemaker won't retry fencing until you run `pcs resource cleanup`.
+- **STONITH timeout too short** — must exceed `pcmk_delay_base` plus worst-case
+  Redfish response time.
+- **DNS or VLAN/routing failures** — management network can fail independently
+  of the cluster (Corosync) network.
+
+To debug a BMC issue, source your env file and ask in Claude Code:
+
+```text
+Can you help debug the BMC not being connected
+```
+
+The skill runs the full cluster diagnostic and the BMC sections will show
+which layer is failing.
+
+## Usage
+
+### Diagnose a live cluster
+
+```text
+/two-node:cluster-diagnostic
+```
+
+Connects via SSH, collects pacemaker/etcd/corosync/OCP state, and produces a
+diagnostic report with severity-classified findings and a scenario match.
+
+### Validate a proposed procedure
+
+```text
+/two-node:cluster-diagnostic validate "cordon all nodes, drain, then shutdown -h 1 on each"
+```
+
+Checks your procedure against known failure modes. No SSH required.
+
+### Get a recovery guide
+
+```text
+/two-node:cluster-diagnostic recovery-guide full-shutdown
+/two-node:cluster-diagnostic recovery-guide standby
+/two-node:cluster-diagnostic recovery-guide split-brain
+```
+
+Returns the correct step-by-step recovery procedure. No SSH required. Run
+without a scenario name to see the full list.
+
+### Interactive training
+
+```text
+/two-node:cluster-diagnostic game
+```
+
+Practice diagnosing TNF issues in quiz, scenario, or rapid-fire mode. No SSH
+required.