Account for full context tokens in compression (#908)

* Account for full context tokens in compression

* Fix group chat final context updates

---------

Co-authored-by: Codex <codex@openai.com>

packages/server/src/services/hermes/group-chat/index.ts

@@ -925,9 +925,7 @@ export class GroupChatServer {
         ack?.({ id: savedMsg.id })
 
         const mentionDepth = normalizeMentionDepth(data.mentionDepth)
-        const shouldRouteMentions =
-            savedMsg.role === 'user' ||
-            (savedMsg.role === 'assistant' && mentionDepth < 2)
+        const shouldRouteMentions = savedMsg.role === 'user'
 
         if (shouldRouteMentions) {
             // Server-side @mention routing — parse user mentions and invoke agents directly.
@@ -1046,7 +1044,7 @@ export class GroupChatServer {
         })
     }
 
-    private handleContextStatus(socket: Socket, data: { roomId?: string; agentName?: string; status?: string }): void {
+    private handleContextStatus(socket: Socket, data: { roomId?: string; agentName?: string; status?: string; totalTokens?: number }): void {
         const roomId = data.roomId || 'general'
         const agentName = data.agentName || ''
         const status = data.status || ''
@@ -1072,6 +1070,11 @@ export class GroupChatServer {
             agentName,
             status,
         })
+
+        if (typeof data.totalTokens === 'number' && Number.isFinite(data.totalTokens) && data.totalTokens >= 0) {
+            this.storage.updateRoomTotalTokens(roomId, Math.floor(data.totalTokens))
+            this.nsp.to(roomId).emit('room_updated', { roomId, totalTokens: Math.floor(data.totalTokens) })
+        }
     }
 
     private async handleInterruptAgent(socket: Socket, data: { roomId?: string; agentName?: string }, ack?: (response?: unknown) => void): Promise<void> {