Merge pull request #175 from anurag008w/fix/whatsapp-runtime-ok-false-positive

start.sh

@@ -1630,9 +1630,11 @@ _hc_allow_openclaw_plugins() {
 
   local plugins_json
   plugins_json=$(printf '%s\n' "${plugins[@]}" | jq -R 'select(length > 0)' | jq -s 'unique') || return 0
-  jq --argjson plugins "$plugins_json" \
+  local patched
+  patched=$(jq --argjson plugins "$plugins_json" \
     '.plugins.allow = (((.plugins.allow // []) + $plugins) | unique)' \
-    "$config" > "$config.tmp" && mv "$config.tmp" "$config"
+    "$config" 2>/dev/null) || { echo "Warning: could not update plugins.allow for $*" >&2; return 0; }
+  write_json_atomic "$config" "$patched" || echo "Warning: could not write plugins.allow update to config." >&2
 }
 _hc_has_arg() {
   local needle="$1"
@@ -1986,7 +1988,7 @@ sync_installed_plugins_into_allow() {
     return 0
   }
 
-  echo "$patched" > "$config.tmp" && mv "$config.tmp" "$config"
+  write_json_atomic "$config" "$patched" || echo "Warning: could not write synced plugins.allow to config." >&2
 }
 
 hc_finish_startup_commands() {
@@ -2106,17 +2108,42 @@ if [ -s "$STARTUP_FILE" ]; then
   echo "Workspace startup script complete."
 fi
 whatsapp_plugin_runtime_ok() {
+  # Check both the bare and scoped install paths that OpenClaw uses across
+  # stable/beta releases.  The scoped path (@openclaw/whatsapp) is used when
+  # the plugin is installed via `openclaw plugins install @openclaw/whatsapp`.
   local ext_dir="/home/node/.openclaw/extensions/whatsapp"
-  if [ -f "$ext_dir/dist/setup-entry.js" ] && [ -f "$ext_dir/dist/index.js" ]; then
-    return 0
-  fi
+  local ext_dir_scoped="/home/node/.openclaw/extensions/@openclaw/whatsapp"
 
-  # Prefer OpenClaw's own runtime inspector over hard-coded paths; plugin
-  # layout can differ across stable/beta releases. If inspect can load the
-  # WhatsApp plugin, do not attempt any install.
-  if command -v openclaw >/dev/null 2>&1; then
-    openclaw plugins inspect whatsapp --runtime --json \
-      >/tmp/openclaw-whatsapp-plugin-inspect.log 2>&1 && return 0
+  for dir in "$ext_dir" "$ext_dir_scoped"; do
+    if [ -f "$dir/dist/setup-entry.js" ] && [ -f "$dir/dist/index.js" ]; then
+      return 0
+    fi
+  done
+
+  # Use openclaw's own inspector to discover a non-standard install directory,
+  # but ALWAYS verify the actual dist files exist afterwards.
+  #
+  # IMPORTANT: `openclaw plugins inspect whatsapp --runtime --json` exits with
+  # code 0 based purely on the plugin registry record — even when the dist
+  # files (dist/setup-entry.js, dist/index.js) are absent.  That is exactly
+  # the condition the gateway.startup_failed error reports.  Trusting the exit
+  # code alone (the previous behaviour) caused `install_whatsapp_plugin_runtime`
+  # and `repair_broken_whatsapp_plugin_entry` to both return early thinking the
+  # runtime was healthy, letting the gateway start with `enabled=true` and
+  # missing dist files → crash loop.
+  if command -v openclaw >/dev/null 2>&1 && command -v jq >/dev/null 2>&1; then
+    local inspect_json
+    inspect_json=$(openclaw plugins inspect whatsapp --runtime --json 2>/dev/null) || true
+    if [ -n "$inspect_json" ]; then
+      local root_dir
+      root_dir=$(printf '%s' "$inspect_json" \
+        | jq -r '(.rootDir // .root // .dir // .path // empty)' 2>/dev/null || true)
+      if [ -n "$root_dir" ] && \
+         [ -f "$root_dir/dist/setup-entry.js" ] && \
+         [ -f "$root_dir/dist/index.js" ]; then
+        return 0
+      fi
+    fi
   fi
 
   return 1
@@ -2223,7 +2250,7 @@ repair_broken_whatsapp_plugin_entry() {
     echo "Warning: could not patch broken WhatsApp plugin entry; gateway may still reject config." >&2
     return 0
   }
-  echo "$patched" > "$config.tmp" && mv "$config.tmp" "$config"
+  write_json_atomic "$config" "$patched" || echo "Warning: could not write patched WhatsApp plugin config; gateway may still reject config." >&2
 }
 
 hc_finish_startup_commands

wa-guardian.js

@@ -293,6 +293,13 @@ process.on("unhandledRejection", (reason) => {
 writeStatus({ configured: true, connected: false, pairing: false });
 console.log("[guardian] WhatsApp Guardian active. Monitoring pairing status...");
 _checkInterval = setInterval(checkStatus, CHECK_INTERVAL);
-// Allow the process to exit even if this interval is the only active handle.
-if (_checkInterval.unref) _checkInterval.unref();
+// NOTE: Do NOT call _checkInterval.unref() here.
+// With unref(), Node.js exits between interval ticks the moment the
+// short-lived WebSocket from checkStatus() closes in its finally block
+// (typically ~25 s after the first run). The interval never fires again,
+// so the guardian stops monitoring entirely after a single check.
+// start_guardian_once() in start.sh has no monitoring loop that would
+// revive it — it is only called at gateway startup. The comment above
+// ("A previous one-shot exit caused 'works once then stops' behavior")
+// documents exactly this failure; removing unref() is the correct fix.
 setTimeout(checkStatus, 15000);