fix: include Claude cache tokens in usage

Add dir="auto" to chat message content and composer textarea so
Persian and Arabic text automatically renders right-to-left
while English and other LTR text remains unaffected.

Co-authored-by: Haile <118998054+blackmammoth@users.noreply.github.com>

package-lock.json

@@ -39,6 +39,7 @@
         "cmdk": "^1.1.1",
         "cors": "^2.8.5",
         "cross-spawn": "^7.0.3",
+        "dompurify": "^3.4.7",
         "express": "^4.18.2",
         "fuse.js": "^7.0.0",
         "gray-matter": "^4.0.3",
@@ -4580,6 +4581,13 @@
         "@types/node": "*"
       }
     },
+    "node_modules/@types/trusted-types": {
+      "version": "2.0.7",
+      "resolved": "https://registry.npmjs.org/@types/trusted-types/-/trusted-types-2.0.7.tgz",
+      "integrity": "sha512-ScaPdn1dQczgbl0QFTeTOmVHFULt394XJgOQNoyVhZ6r2vLnMLJfBPd53SB52T/3G36VI1/g2MZaX0cwDuXsfw==",
+      "license": "MIT",
+      "optional": true
+    },
     "node_modules/@types/unist": {
       "version": "3.0.3",
       "resolved": "https://registry.npmjs.org/@types/unist/-/unist-3.0.3.tgz",
@@ -7485,6 +7493,15 @@
         "node": ">=0.10.0"
       }
     },
+    "node_modules/dompurify": {
+      "version": "3.4.7",
+      "resolved": "https://registry.npmjs.org/dompurify/-/dompurify-3.4.7.tgz",
+      "integrity": "sha512-2jBxDJY4RR06tQNy4w5FlFH7kfxsQZlufd0sbv+chfHCxeJwrFw2baUDsSwvBISD4K4RDbd0PTfy3uNXsR6siA==",
+      "license": "(MPL-2.0 OR Apache-2.0)",
+      "optionalDependencies": {
+        "@types/trusted-types": "^2.0.7"
+      }
+    },
     "node_modules/dot-prop": {
       "version": "5.3.0",
       "resolved": "https://registry.npmjs.org/dot-prop/-/dot-prop-5.3.0.tgz",

package.json

@@ -96,6 +96,7 @@
     "cmdk": "^1.1.1",
     "cors": "^2.8.5",
     "cross-spawn": "^7.0.3",
+    "dompurify": "^3.4.7",
     "express": "^4.18.2",
     "fuse.js": "^7.0.0",
     "gray-matter": "^4.0.3",

public/modelConstants.js

@@ -11,7 +11,7 @@ export const CLAUDE_MODELS = {
     {
       value: "default",
       label: "Default (recommended)",
-      description: "Use the default model (currently Opus 4.7 (1M context)) · $5/$25 per Mtok",
+      description: "Use the default model (currently Opus 4.8 (1M context)) · $5/$25 per Mtok",
     },
     {
       value: "sonnet",

server/claude-sdk.js

@@ -304,7 +304,11 @@ function extractTokenBudget(sdkMessage) {
 
   const messageUsage = sdkMessage.message?.usage || sdkMessage.usage;
   if (messageUsage && typeof messageUsage === 'object') {
-    const inputTokens = readNumber(messageUsage.input_tokens ?? messageUsage.inputTokens);
+    const directInputTokens = readNumber(messageUsage.input_tokens ?? messageUsage.inputTokens);
+    const cacheCreationTokens = readNumber(messageUsage.cache_creation_input_tokens ?? messageUsage.cacheCreationInputTokens ?? messageUsage.cacheCreationTokens);
+    const cacheReadTokens = readNumber(messageUsage.cache_read_input_tokens ?? messageUsage.cacheReadInputTokens ?? messageUsage.cacheReadTokens);
+    const cacheTokens = cacheCreationTokens + cacheReadTokens;
+    const inputTokens = directInputTokens + cacheTokens;
     const outputTokens = readNumber(messageUsage.output_tokens ?? messageUsage.outputTokens);
     const totalUsed = inputTokens + outputTokens;
     const contextWindow = parseInt(process.env.CONTEXT_WINDOW, 10) || 160000;
@@ -314,6 +318,9 @@ function extractTokenBudget(sdkMessage) {
       total: contextWindow,
       inputTokens,
       outputTokens,
+      cacheReadTokens,
+      cacheCreationTokens,
+      cacheTokens,
       breakdown: {
         input: inputTokens,
         output: outputTokens,

server/index.js

@@ -87,6 +87,11 @@ const installMode = fs.existsSync(path.join(APP_ROOT, '.git')) ? 'git' : 'npm';
 
 console.log('SERVER_PORT from env:', process.env.SERVER_PORT);
 
+function readUsageNumber(value) {
+    const parsed = Number(value);
+    return Number.isFinite(parsed) ? parsed : 0;
+}
+
 const app = express();
 const server = http.createServer(app);
 
@@ -1386,6 +1391,8 @@ app.get('/api/projects/:projectId/sessions/:sessionId/token-usage', authenticate
         const contextWindow = Number.isFinite(parsedContextWindow) ? parsedContextWindow : 160000;
         let inputTokens = 0;
         let outputTokens = 0;
+        let cacheReadTokens = 0;
+        let cacheCreationTokens = 0;
 
         // Find the latest assistant message with usage data (scan from end)
         for (let i = lines.length - 1; i >= 0; i--) {
@@ -1397,8 +1404,11 @@ app.get('/api/projects/:projectId/sessions/:sessionId/token-usage', authenticate
                     const usage = entry.message.usage;
 
                     // Use token counts from latest assistant message only
-                    inputTokens = usage.input_tokens || 0;
+                    const directInputTokens = readUsageNumber(usage.input_tokens ?? usage.inputTokens);
-                    outputTokens = usage.output_tokens || 0;
+                    cacheReadTokens = readUsageNumber(usage.cache_read_input_tokens ?? usage.cacheReadInputTokens ?? usage.cacheReadTokens);
+                    cacheCreationTokens = readUsageNumber(usage.cache_creation_input_tokens ?? usage.cacheCreationInputTokens ?? usage.cacheCreationTokens);
+                    inputTokens = directInputTokens + cacheReadTokens + cacheCreationTokens;
+                    outputTokens = readUsageNumber(usage.output_tokens ?? usage.outputTokens);
 
                     break; // Stop after finding the latest assistant message
                 }
@@ -1409,12 +1419,16 @@ app.get('/api/projects/:projectId/sessions/:sessionId/token-usage', authenticate
         }
 
         const totalUsed = inputTokens + outputTokens;
+        const cacheTokens = cacheReadTokens + cacheCreationTokens;
 
         res.json({
             used: totalUsed,
             total: contextWindow,
             inputTokens,
             outputTokens,
+            cacheReadTokens,
+            cacheCreationTokens,
+            cacheTokens,
             breakdown: {
                 input: inputTokens,
                 output: outputTokens

server/modules/providers/list/claude/claude-auth.provider.ts

@@ -83,6 +83,10 @@ export class ClaudeProviderAuth implements IProviderAuth {
   private async checkCredentials(): Promise<ClaudeCredentialsStatus> {
     const missingCredentialsError = 'Claude CLI is not authenticated. Run claude /login or configure ANTHROPIC_API_KEY.';
 
+    if (process.env.ANTHROPIC_AUTH_TOKEN?.trim()) {
+      return { authenticated: true, email: 'Auth Token', method: 'api_key' };
+    }
+
     if (process.env.ANTHROPIC_API_KEY?.trim()) {
       return { authenticated: true, email: 'API Key Auth', method: 'api_key' };
     }

server/modules/websocket/services/websocket-server.service.ts

@@ -31,6 +31,24 @@ export function createWebSocketServer(
   });
 
   wss.on('connection', (ws, request) => {
+    // Keep WebSocket alive across reverse-proxy idle timeouts (Cloudflare ~100s,
+    // AWS ALB 60s, nginx 60s, etc.). Without app-level pings these connections
+    // are silently torn down even when the UI is active, causing repeated
+    // reconnect cycles. ws library heartbeat is opt-in.
+    const HEARTBEAT_INTERVAL_MS = 30_000;
+    const heartbeat = setInterval(() => {
+      if (ws.readyState === ws.OPEN) {
+        try {
+          ws.ping();
+        } catch {
+          // socket may have been closed concurrently — interval will be cleared below
+        }
+      }
+    }, HEARTBEAT_INTERVAL_MS);
+    const stopHeartbeat = () => clearInterval(heartbeat);
+    ws.on('close', stopHeartbeat);
+    ws.on('error', stopHeartbeat);
+
     const incomingRequest = request as AuthenticatedWebSocketRequest;
     const url = incomingRequest.url ?? '/';
     const pathname = new URL(url, 'http://localhost').pathname;

server/routes/agent.js

@@ -592,12 +592,14 @@ class ResponseCollector {
       }
     }
 
+    const inputTokens = totalInput + totalCacheRead + totalCacheCreation;
+
     return {
-      inputTokens: totalInput,
+      inputTokens,
       outputTokens: totalOutput,
       cacheReadTokens: totalCacheRead,
       cacheCreationTokens: totalCacheCreation,
-      totalTokens: totalInput + totalOutput + totalCacheRead + totalCacheCreation
+      totalTokens: inputTokens + totalOutput
     };
   }
 }

server/routes/commands.js

@@ -268,16 +268,35 @@ Custom commands can be created in:
           tokenUsage.contextWindow ??
           0,
       ) || 0;
-    const inputTokensRaw =
+    const normalizedInputValue =
+      tokenUsage.inputTokens ??
+      tokenUsage.input ??
+      tokenUsage.cumulativeInputTokens ??
+      tokenUsage.breakdown?.input ??
+      tokenUsage.promptTokens;
+    const directInputTokens =
       Number(
-        tokenUsage.inputTokens ??
+        normalizedInputValue ??
-          tokenUsage.input ??
           tokenUsage.input_tokens ??
-          tokenUsage.cumulativeInputTokens ??
+          0
-          tokenUsage.breakdown?.input ??
+      ) || 0;
-          tokenUsage.promptTokens ??
+    const cacheReadTokens =
+      Number(
+        tokenUsage.cacheReadTokens ??
+          tokenUsage.cache_read_input_tokens ??
+          tokenUsage.cacheReadInputTokens ??
           0,
       ) || 0;
+    const cacheCreationTokens =
+      Number(
+        tokenUsage.cacheCreationTokens ??
+          tokenUsage.cache_creation_input_tokens ??
+          tokenUsage.cacheCreationInputTokens ??
+          0,
+      ) || 0;
+    const inputTokens = normalizedInputValue == null
+      ? directInputTokens + cacheReadTokens + cacheCreationTokens
+      : directInputTokens;
     const outputTokens =
       Number(
         tokenUsage.outputTokens ??
@@ -288,8 +307,9 @@ Custom commands can be created in:
           tokenUsage.completionTokens ??
           0,
       ) || 0;
-    const hasTokenBreakdown = inputTokensRaw > 0 || outputTokens > 0;
+    const computedUsed = inputTokens + outputTokens;
-    const used = reportedUsed || inputTokensRaw + outputTokens;
+    const hasTokenBreakdown = computedUsed > 0;
+    const used = Math.max(reportedUsed, computedUsed);
 
     return {
       type: "builtin",
@@ -302,7 +322,7 @@ Custom commands can be created in:
         ...(hasTokenBreakdown
           ? {
               tokenBreakdown: {
-                input: inputTokensRaw,
+                input: inputTokens,
                 output: outputTokens,
               },
             }

src/components/chat/view/subcomponents/ChatComposer.tsx

@@ -295,6 +295,7 @@ export default function ChatComposer({
 
             <PromptInputTextarea
               ref={textareaRef}
+              dir="auto"
               value={input}
               onChange={onInputChange}
               onClick={onTextareaClick}

src/components/chat/view/subcomponents/MessageComponent.tsx

@@ -120,7 +120,7 @@ const MessageComponent = memo(({ message, prevMessage, createDiff, onFileOpen, o
         /* User message bubble on the right */
         <div className="flex w-full items-end space-x-0 sm:w-auto sm:max-w-[85%] sm:space-x-3 md:max-w-md lg:max-w-lg xl:max-w-xl">
           <div className="group flex-1 rounded-2xl rounded-br-md bg-blue-600 px-3 py-2 text-white shadow-sm sm:flex-initial sm:px-4">
-            <div className="whitespace-pre-wrap break-words text-sm">
+            <div dir="auto" className="whitespace-pre-wrap break-words text-sm">
               {message.content}
             </div>
             {message.images && message.images.length > 0 && (
@@ -405,7 +405,7 @@ const MessageComponent = memo(({ message, prevMessage, createDiff, onFileOpen, o
                 </ReasoningContent>
               </Reasoning>
             ) : (
-              <div className="text-sm text-gray-700 dark:text-gray-300">
+              <div dir="auto" className="text-sm text-gray-700 dark:text-gray-300">
                 {/* Reasoning accordion */}
                 {showThinking && message.reasoning && (
                   <Reasoning className="mb-3" defaultOpen={false}>

src/components/plugins/view/PluginIcon.tsx

@@ -1,4 +1,6 @@
 import { useState, useEffect } from 'react';
+import DOMPurify from 'dompurify';
+
 import { authenticatedFetch } from '../../../utils/api';
 
 type Props = {
@@ -10,6 +12,48 @@ type Props = {
 // Module-level cache so repeated renders don't re-fetch
 const svgCache = new Map<string, string>();
 
+const FORBIDDEN_SVG_TAGS = [
+  'script',
+  'foreignObject',
+  'iframe',
+  'object',
+  'embed',
+  'link',
+  'meta',
+  'style',
+  'animate',
+  'set',
+  'animateTransform',
+  'animateMotion',
+];
+
+const FORBIDDEN_SVG_ATTRS = [
+  'href',
+  'xlink:href',
+  'src',
+  'style',
+];
+
+function sanitizeSvg(svgText: string): string | null {
+  const sanitized = DOMPurify.sanitize(svgText, {
+    USE_PROFILES: { svg: true, svgFilters: true },
+    FORBID_TAGS: FORBIDDEN_SVG_TAGS,
+    FORBID_ATTR: FORBIDDEN_SVG_ATTRS,
+  });
+
+  if (!sanitized) return null;
+
+  try {
+    const doc = new DOMParser().parseFromString(sanitized, 'image/svg+xml');
+    const root = doc.documentElement;
+    if (!root || root.nodeName.toLowerCase() !== 'svg') return null;
+    if (doc.querySelector('parsererror')) return null;
+    return sanitized;
+  } catch {
+    return null;
+  }
+}
+
 export default function PluginIcon({ pluginName, iconFile, className }: Props) {
   const url = iconFile
     ? `/api/plugins/${encodeURIComponent(pluginName)}/assets/${encodeURIComponent(iconFile)}`
@@ -24,9 +68,11 @@ export default function PluginIcon({ pluginName, iconFile, className }: Props) {
         return r.text();
       })
       .then((text) => {
-        if (text && text.trimStart().startsWith('<svg')) {
+        if (!text) return;
-          svgCache.set(url, text);
+        const sanitized = sanitizeSvg(text);
-          setSvg(text);
+        if (sanitized) {
+          svgCache.set(url, sanitized);
+          setSvg(sanitized);
         }
       })
       .catch(() => {});
@@ -35,10 +81,6 @@ export default function PluginIcon({ pluginName, iconFile, className }: Props) {
   if (!svg) return <span className={className} />;
 
   return (
-    <span
+    <span className={className} dangerouslySetInnerHTML={{ __html: svg }} />
-      className={className}
-      // SVG is fetched from the user's own installed plugin — same trust level as the plugin code itself
-      dangerouslySetInnerHTML={{ __html: svg }}
-    />
   );
 }

src/contexts/WebSocketContext.tsx

@@ -36,8 +36,12 @@ const useWebSocketProviderState = (): WebSocketContextType => {
   const { token } = useAuth();
 
   useEffect(() => {
+    // The cleanup below sets unmountedRef = true. Without this reset, every
+    // re-run of the effect (e.g. on token refresh) would short-circuit connect()
+    // at its unmounted guard and leave the socket permanently disconnected.
+    unmountedRef.current = false;
     connect();
-    
+
     return () => {
       unmountedRef.current = true;
       if (reconnectTimeoutRef.current) {

vite.config.js

@@ -37,6 +37,10 @@ export default defineConfig(({ mode }) => {
         '/shell': {
           target: `ws://${proxyHost}:${serverPort}`,
           ws: true
+        },
+        '/plugin-ws': {
+          target: `ws://${proxyHost}:${serverPort}`,
+          ws: true
         }
       }
     },