feat(ui): restore threshold hint and thin arrow to compression message

packages/core/src/core/client.test.ts

@@ -248,6 +248,9 @@ describe('Gemini Client (client.ts)', () => {
       getEnableHooks: vi.fn().mockReturnValue(false),
       getChatCompression: vi.fn().mockReturnValue(undefined),
       getCompressionThreshold: vi.fn().mockReturnValue(undefined),
+      getShowContextWindowWarning: vi.fn().mockReturnValue(false),
+      getShowContextCompression: vi.fn().mockReturnValue(false),
+      getContextWindowCompressionThreshold: vi.fn().mockReturnValue(0.2),
       getSkipNextSpeakerCheck: vi.fn().mockReturnValue(false),
       getShowModelInfoInChat: vi.fn().mockReturnValue(false),
       getContinueOnFailedApiCall: vi.fn(),
@@ -1617,6 +1620,7 @@ ${JSON.stringify(
             originalTokenCount: initialTokenCount,
             newTokenCount: 400,
             compressionStatus: CompressionStatus.COMPRESSED,
+            requestTokenCount: 50, // Added to match updated interface
           };
         });
 
@@ -1643,10 +1647,13 @@ ${JSON.stringify(
         }),
       );
 
-      // 2. Should contain compression event
+      // 2. Should contain compression event with requestTokenCount
       expect(events).toContainEqual(
         expect.objectContaining({
           type: GeminiEventType.ChatCompressed,
+          value: expect.objectContaining({
+            requestTokenCount: expect.any(Number),
+          }),
         }),
       );
 

packages/core/src/core/client.ts

@@ -608,7 +608,19 @@ export class GeminiClient {
     // Check for context window overflow
     const modelForLimitCheck = this._getActiveModelForCurrentTurn();
 
-    const compressed = await this.tryCompressChat(prompt_id, false);
+    // Estimate tokens. For text-only requests, we estimate based on character length.
+    // For requests with non-text parts (like images, tools), we use the countTokens API.
+    const estimatedRequestTokenCount = await calculateRequestTokenCount(
+      request,
+      this.getContentGeneratorOrFail(),
+      modelForLimitCheck,
+    );
+
+    const compressed = await this.tryCompressChat(
+      prompt_id,
+      false,
+      estimatedRequestTokenCount,
+    );
 
     if (compressed.compressionStatus === CompressionStatus.COMPRESSED) {
       yield { type: GeminiEventType.ChatCompressed, value: compressed };
@@ -619,17 +631,13 @@ export class GeminiClient {
 
     await this.tryMaskToolOutputs(this.getHistory());
 
-    // Estimate tokens. For text-only requests, we estimate based on character length.
-    // For requests with non-text parts (like images, tools), we use the countTokens API.
-    const estimatedRequestTokenCount = await calculateRequestTokenCount(
-      request,
-      this.getContentGeneratorOrFail(),
-      modelForLimitCheck,
-    );
-
     if (estimatedRequestTokenCount > remainingTokenCount) {
       if (!this.config.getShowContextWindowWarning()) {
-        const forcedCompressed = await this.tryCompressChat(prompt_id, true);
+        const forcedCompressed = await this.tryCompressChat(
+          prompt_id,
+          true,
+          estimatedRequestTokenCount,
+        );
         if (
           forcedCompressed.compressionStatus === CompressionStatus.COMPRESSED
         ) {
@@ -1175,6 +1183,7 @@ export class GeminiClient {
   async tryCompressChat(
     prompt_id: string,
     force: boolean = false,
+    requestTokenCount?: number,
   ): Promise<ChatCompressionInfo> {
     // If the model is 'auto', we will use a placeholder model to check.
     // Compression occurs before we choose a model, so calling `count_tokens`
@@ -1190,6 +1199,11 @@ export class GeminiClient {
       this.hasFailedCompressionAttempt,
     );
 
+    const resultInfo = {
+      ...info,
+      requestTokenCount,
+    };
+
     if (
       info.compressionStatus ===
       CompressionStatus.COMPRESSION_FAILED_INFLATED_TOKEN_COUNT
@@ -1225,7 +1239,7 @@ export class GeminiClient {
       }
     }
 
-    return info;
+    return resultInfo;
   }
 
   /**

packages/core/src/core/turn.ts

@@ -188,6 +188,7 @@ export interface ChatCompressionInfo {
   originalTokenCount: number;
   newTokenCount: number;
   compressionStatus: CompressionStatus;
+  requestTokenCount?: number;
 }
 
 export type ServerGeminiChatCompressedEvent = {