🤖 Change interrupt sentinel from [INTERRUPTED] to [CONTINUE] (#170)

## Problem

When users interrupt a stream and then resume, we inject a synthetic
`[INTERRUPTED]` message to give the model context. However, this wording
can confuse models into thinking there was an error or that they should
stop.

## Solution

Change the sentinel text from `[INTERRUPTED]` to `[CONTINUE]` to make it
clearer that the model should continue its previous work.

This is a quick UX improvement while we work on more comprehensive
handling of interrupted messages with thinking (see
#investigate-thinking-error branch).

## Changes

- Updated sentinel text: `[INTERRUPTED]` → `[CONTINUE]`
- Updated all comments and tests to reflect the new wording
- All 297 unit tests pass ✅

## Testing

- ✅ Unit tests updated and passing
- Verified with `make test`

_Generated with `cmux`_

Author: ammar-agent

src/services/aiService.ts

@@ -440,7 +440,7 @@ export class AIService extends EventEmitter {
         log.debug("Keeping reasoning parts for OpenAI (fetch wrapper handles item_references)");
       }
 
-      // Add [INTERRUPTED] sentinel to partial messages (for model context)
+      // Add [CONTINUE] sentinel to partial messages (for model context)
       const messagesWithSentinel = addInterruptedSentinel(filteredMessages);
 
       // Convert CmuxMessage to ModelMessage format using Vercel AI SDK utility

src/services/ipcMain.ts

@@ -1055,7 +1055,7 @@ export class IpcMain {
             // so that stream-end can properly clean up the streaming indicator
             this.aiService.replayStream(workspaceId);
           } else if (partial) {
-            // No active stream but there's a partial - send as regular message (shows INTERRUPTED)
+            // No active stream but there's a partial - send as regular message (shows CONTINUE)
             this.mainWindow?.webContents.send(chatChannel, partial);
           }
         }

src/types/message.ts

@@ -13,7 +13,7 @@ export interface CmuxMetadata {
   providerMetadata?: Record<string, unknown>; // Raw AI SDK provider data
   systemMessageTokens?: number; // Token count for system message sent with this request (calculated by AIService)
   partial?: boolean; // Whether this message was interrupted and is incomplete
-  synthetic?: boolean; // Whether this message was synthetically generated (e.g., [INTERRUPTED] sentinel)
+  synthetic?: boolean; // Whether this message was synthetically generated (e.g., [CONTINUE] sentinel)
   error?: string; // Error message if stream failed
   errorType?: StreamErrorType; // Error type/category if stream failed
   compacted?: boolean; // Whether this message is a compacted summary of previous history

src/utils/messages/modelMessageTransform.test.ts

@@ -420,13 +420,13 @@ describe("modelMessageTransform", () => {
 
       const result = addInterruptedSentinel(messages);
 
-      // Should have 3 messages: user, assistant, [INTERRUPTED] user
+      // Should have 3 messages: user, assistant, [CONTINUE] user
       expect(result).toHaveLength(3);
       expect(result[0].id).toBe("user-1");
       expect(result[1].id).toBe("assistant-1");
       expect(result[2].id).toBe("interrupted-assistant-1");
       expect(result[2].role).toBe("user");
-      expect(result[2].parts).toEqual([{ type: "text", text: "[INTERRUPTED]" }]);
+      expect(result[2].parts).toEqual([{ type: "text", text: "[CONTINUE]" }]);
       expect(result[2].metadata?.synthetic).toBe(true);
       expect(result[2].metadata?.timestamp).toBe(2000);
     });
@@ -472,10 +472,10 @@ describe("modelMessageTransform", () => {
 
       const result = addInterruptedSentinel(messages);
 
-      // Should have 3 messages: user, assistant (reasoning only), [INTERRUPTED] user
+      // Should have 3 messages: user, assistant (reasoning only), [CONTINUE] user
       expect(result).toHaveLength(3);
       expect(result[2].role).toBe("user");
-      expect(result[2].parts).toEqual([{ type: "text", text: "[INTERRUPTED]" }]);
+      expect(result[2].parts).toEqual([{ type: "text", text: "[CONTINUE]" }]);
     });
 
     it("should handle multiple partial messages", () => {

src/utils/messages/modelMessageTransform.ts

@@ -62,8 +62,8 @@ export function stripReasoningForOpenAI(messages: CmuxMessage[]): CmuxMessage[]
 }
 
 /**
- * Add [INTERRUPTED] sentinel to partial messages by inserting a user message.
- * This helps the model understand that a message was interrupted and incomplete.
+ * Add [CONTINUE] sentinel to partial messages by inserting a user message.
+ * This helps the model understand that a message was interrupted and to continue.
  * The sentinel is ONLY for model context, not shown in UI.
  *
  * We insert a separate user message instead of modifying the assistant message
@@ -77,12 +77,12 @@ export function addInterruptedSentinel(messages: CmuxMessage[]): CmuxMessage[] {
   for (const msg of messages) {
     result.push(msg);
 
-    // If this is a partial assistant message, insert [INTERRUPTED] user message after it
+    // If this is a partial assistant message, insert [CONTINUE] user message after it
     if (msg.role === "assistant" && msg.metadata?.partial) {
       result.push({
         id: `interrupted-${msg.id}`,
         role: "user",
-        parts: [{ type: "text", text: "[INTERRUPTED]" }],
+        parts: [{ type: "text", text: "[CONTINUE]" }],
         metadata: {
           timestamp: msg.metadata.timestamp,
           // Mark as synthetic so it can be identified if needed