feat(cli): lower compression threshold to 0.2 and update UX

- Change default compression threshold to 200k

- Introduce 'ChatCompressing' event to handle UI loading

- Subtly display context optimization in message footer

- Fix associated snapshot tests

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

@@ -1077,6 +1077,7 @@ ${JSON.stringify(
         new AbortController().signal,
         'test-prompt-id',
       );
+      await stream.next(); // Trigger ChatCompressing
       await stream.next(); // Trigger the generator
 
       expect(countTokensSpy).toHaveBeenCalledWith(
@@ -1924,8 +1925,10 @@ ${JSON.stringify(
 
       // Assert
       expect(events).toEqual([
+        { type: GeminiEventType.ChatCompressing },
         { type: GeminiEventType.ModelInfo, value: 'default-routed-model' },
         { type: GeminiEventType.InvalidStream },
+        { type: GeminiEventType.ChatCompressing },
         { type: GeminiEventType.Content, value: 'Continued content' },
       ]);
 
@@ -1980,6 +1983,7 @@ ${JSON.stringify(
 
       // Assert
       expect(events).toEqual([
+        { type: GeminiEventType.ChatCompressing },
         { type: GeminiEventType.ModelInfo, value: 'default-routed-model' },
         { type: GeminiEventType.InvalidStream },
       ]);
@@ -2017,8 +2021,8 @@ ${JSON.stringify(
       const events = await fromAsync(stream);
 
       // Assert
-      // We expect 3 events (model_info + original + 1 retry)
-      expect(events.length).toBe(3);
+      // We expect 5 events (chat_compressing + model_info + original + 1 retry chat_compressing + 1 retry model_info)
+      expect(events.length).toBe(5);
       expect(
         events
           .filter((e) => e.type === GeminiEventType.ModelInfo)

packages/core/src/core/client.ts

@@ -571,6 +571,7 @@ export class GeminiClient {
     // Check for context window overflow
     const modelForLimitCheck = this._getActiveModelForCurrentTurn();
 
+    yield { type: GeminiEventType.ChatCompressing };
     const compressed = await this.tryCompressChat(prompt_id, false);
 
     if (compressed.compressionStatus === CompressionStatus.COMPRESSED) {

packages/core/src/core/turn.ts

@@ -68,6 +68,7 @@ export enum GeminiEventType {
   ModelInfo = 'model_info',
   AgentExecutionStopped = 'agent_execution_stopped',
   AgentExecutionBlocked = 'agent_execution_blocked',
+  ChatCompressing = 'chat_compressing',
 }
 
 export type ServerGeminiRetryEvent = {
@@ -192,6 +193,10 @@ export type ServerGeminiChatCompressedEvent = {
   value: ChatCompressionInfo | null;
 };
 
+export type ServerGeminiChatCompressingEvent = {
+  type: GeminiEventType.ChatCompressing;
+};
+
 export type ServerGeminiMaxSessionTurnsEvent = {
   type: GeminiEventType.MaxSessionTurns;
 };
@@ -213,6 +218,7 @@ export type ServerGeminiCitationEvent = {
 // The original union type, now composed of the individual types
 export type ServerGeminiStreamEvent =
   | ServerGeminiChatCompressedEvent
+  | ServerGeminiChatCompressingEvent
   | ServerGeminiCitationEvent
   | ServerGeminiContentEvent
   | ServerGeminiErrorEvent

packages/core/src/services/chatCompressionService.ts

@@ -36,7 +36,7 @@ import { PreCompressTrigger } from '../hooks/types.js';
  * Default threshold for compression token count as a fraction of the model's
  * token limit. If the chat history exceeds this threshold, it will be compressed.
  */
-export const DEFAULT_COMPRESSION_TOKEN_THRESHOLD = 0.5;
+export const DEFAULT_COMPRESSION_TOKEN_THRESHOLD = 0.2;
 
 /**
  * The fraction of the latest chat history to keep. A value of 0.3