diff --git a/scarf/Packages/ScarfCore/Sources/ScarfCore/ViewModels/RichChatViewModel.swift b/scarf/Packages/ScarfCore/Sources/ScarfCore/ViewModels/RichChatViewModel.swift
index 1a3e4db..f78f191 100644
--- a/scarf/Packages/ScarfCore/Sources/ScarfCore/ViewModels/RichChatViewModel.swift
+++ b/scarf/Packages/ScarfCore/Sources/ScarfCore/ViewModels/RichChatViewModel.swift
@@ -44,11 +44,41 @@ public final class RichChatViewModel {
     public var messages: [HermesMessage] = []
     public var currentSession: HermesSession?
     public var messageGroups: [MessageGroup] = []
+    /// True from the moment the user sends a prompt until the ACP
+    /// `promptComplete` event arrives. Covers the whole round-trip
+    /// including auxiliary post-processing (title generation, usage
+    /// accounting, etc.). UIs should prefer the `isGenerating` /
+    /// `isPostProcessing` pair below — they distinguish "agent is
+    /// thinking about your message" from "agent is closing out" and
+    /// avoid the misleading "spinner after the reply has landed" UX
+    /// we saw in pass-1 (M7 #4).
     public var isAgentWorking = false
     public var pendingPermission: PendingPermission?
     /// Mutated to trigger a scroll-to-bottom in the message list.
     public var scrollTrigger = UUID()
 
+    /// True while the assistant hasn't yet emitted a complete reply
+    /// for the latest user prompt. Renders the prominent "Agent is
+    /// thinking…" indicator in the chat. Flips false as soon as we've
+    /// finalized an assistant message with content — even if the ACP
+    /// `promptComplete` event hasn't arrived yet (Hermes auxiliary
+    /// work like title generation delays that event).
+    public var isGenerating: Bool {
+        isAgentWorking && !isPostProcessing
+    }
+
+    /// True while ACP hasn't closed out the prompt but the assistant
+    /// has already finalized a reply the user can see. Renders a
+    /// subtle "Finishing up…" pill instead of the prominent spinner.
+    /// Avoids the pass-1 M7 #4 UX where users stared at "Agent is
+    /// working…" forever because `promptComplete` was held up by
+    /// auxiliary server-side work.
+    public var isPostProcessing: Bool {
+        guard isAgentWorking else { return false }
+        guard let last = messages.last else { return false }
+        return last.isAssistant && !last.content.isEmpty
+    }
+
     // Cumulative ACP token tracking (ACP returns tokens per prompt but DB has none)
     public private(set) var acpInputTokens = 0
     public private(set) var acpOutputTokens = 0
diff --git a/scarf/Scarf iOS/Chat/ChatView.swift b/scarf/Scarf iOS/Chat/ChatView.swift
index 9a81bc0..f42932c 100644
--- a/scarf/Scarf iOS/Chat/ChatView.swift	
+++ b/scarf/Scarf iOS/Chat/ChatView.swift	
@@ -94,15 +94,26 @@ struct ChatView: View {
                         MessageBubble(message: msg)
                             .id(msg.id)
                     }
-                    if controller.vm.isAgentWorking {
+                    if controller.vm.isGenerating {
                         HStack {
                             ProgressView()
-                            Text("Agent is working…")
+                            Text("Agent is thinking…")
                                 .font(.caption)
                                 .foregroundStyle(.secondary)
                         }
                         .frame(maxWidth: .infinity, alignment: .leading)
                         .padding(.horizontal)
+                    } else if controller.vm.isPostProcessing {
+                        HStack(spacing: 6) {
+                            Image(systemName: "ellipsis")
+                                .font(.caption2)
+                                .foregroundStyle(.tertiary)
+                            Text("Finishing up…")
+                                .font(.caption2)
+                                .foregroundStyle(.tertiary)
+                        }
+                        .frame(maxWidth: .infinity, alignment: .leading)
+                        .padding(.horizontal)
                     }
                     Color.clear
                         .frame(height: 1)
diff --git a/scarf/scarf/Features/Chat/Views/RichChatView.swift b/scarf/scarf/Features/Chat/Views/RichChatView.swift
index 1347c3e..863fd85 100644
--- a/scarf/scarf/Features/Chat/Views/RichChatView.swift
+++ b/scarf/scarf/Features/Chat/Views/RichChatView.swift
@@ -15,7 +15,13 @@ struct RichChatView: View {
         VStack(spacing: 0) {
             SessionInfoBar(
                 session: richChat.currentSession,
-                isWorking: richChat.isAgentWorking,
+                // Prefer `isGenerating` over the raw `isAgentWorking`
+                // so the info bar drops the spinner as soon as the
+                // assistant's reply is visible, even while ACP
+                // auxiliary work (title gen, usage accounting) is
+                // still in flight. See RichChatViewModel docs — same
+                // fix as ScarfGo for pass-1 M7 #4.
+                isWorking: richChat.isGenerating,
                 acpInputTokens: richChat.acpInputTokens,
                 acpOutputTokens: richChat.acpOutputTokens,
                 acpThoughtTokens: richChat.acpThoughtTokens,
@@ -34,7 +40,7 @@ struct RichChatView: View {
             // which manifests as a white flash.
             RichChatMessageList(
                 groups: richChat.messageGroups,
-                isWorking: richChat.isAgentWorking,
+                isWorking: richChat.isGenerating,
                 isLoadingSession: chatViewModel.isPreparingSession,
                 scrollTrigger: richChat.scrollTrigger
             )