feat(core): thinking block cross-turn retention with idle cleanup

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

@@ -427,6 +427,119 @@ describe('Gemini Client (client.ts)', () => {
     });
   });
 
+  describe('thinking block idle cleanup and latch', () => {
+    let mockChat: Partial<GeminiChat>;
+
+    beforeEach(() => {
+      const mockStream = (async function* () {
+        yield {
+          type: GeminiEventType.Content,
+          value: 'response',
+        };
+      })();
+      mockTurnRunFn.mockReturnValue(mockStream);
+
+      mockChat = {
+        addHistory: vi.fn(),
+        getHistory: vi.fn().mockReturnValue([]),
+        stripThoughtsFromHistory: vi.fn(),
+        stripThoughtsFromHistoryKeepRecent: vi.fn(),
+      };
+      client['chat'] = mockChat as GeminiChat;
+    });
+
+    it('should not strip thoughts on active session (< 1h idle)', async () => {
+      // Simulate a recent API completion (5 minutes ago)
+      client['lastApiCompletionTimestamp'] = Date.now() - 5 * 60 * 1000;
+      client['thinkingClearLatched'] = false;
+
+      const gen = client.sendMessageStream(
+        [{ text: 'Hello' }],
+        new AbortController().signal,
+        'prompt-1',
+        { type: SendMessageType.UserQuery },
+      );
+      for await (const _ of gen) {
+        /* drain */
+      }
+
+      expect(
+        mockChat.stripThoughtsFromHistoryKeepRecent,
+      ).not.toHaveBeenCalled();
+    });
+
+    it('should latch and strip thoughts after > 1h idle', async () => {
+      // Simulate an old API completion (2 hours ago)
+      client['lastApiCompletionTimestamp'] = Date.now() - 2 * 60 * 60 * 1000;
+      client['thinkingClearLatched'] = false;
+
+      const gen = client.sendMessageStream(
+        [{ text: 'Hello' }],
+        new AbortController().signal,
+        'prompt-2',
+        { type: SendMessageType.UserQuery },
+      );
+      for await (const _ of gen) {
+        /* drain */
+      }
+
+      expect(client['thinkingClearLatched']).toBe(true);
+      expect(mockChat.stripThoughtsFromHistoryKeepRecent).toHaveBeenCalledWith(
+        1,
+      );
+    });
+
+    it('should keep stripping once latched even if idle < 1h', async () => {
+      // Pre-set latch with a recent timestamp
+      client['lastApiCompletionTimestamp'] = Date.now() - 5 * 60 * 1000;
+      client['thinkingClearLatched'] = true;
+
+      const gen = client.sendMessageStream(
+        [{ text: 'Hello' }],
+        new AbortController().signal,
+        'prompt-3',
+        { type: SendMessageType.UserQuery },
+      );
+      for await (const _ of gen) {
+        /* drain */
+      }
+
+      expect(client['thinkingClearLatched']).toBe(true);
+      expect(mockChat.stripThoughtsFromHistoryKeepRecent).toHaveBeenCalledWith(
+        1,
+      );
+    });
+
+    it('should update lastApiCompletionTimestamp after API call', async () => {
+      client['lastApiCompletionTimestamp'] = null;
+
+      const before = Date.now();
+      const gen = client.sendMessageStream(
+        [{ text: 'Hello' }],
+        new AbortController().signal,
+        'prompt-4',
+        { type: SendMessageType.UserQuery },
+      );
+      for await (const _ of gen) {
+        /* drain */
+      }
+
+      expect(client['lastApiCompletionTimestamp']).toBeGreaterThanOrEqual(
+        before,
+      );
+    });
+
+    it('should reset latch and timestamp on resetChat', async () => {
+      client['lastApiCompletionTimestamp'] = Date.now();
+      client['thinkingClearLatched'] = true;
+
+      await client.resetChat();
+
+      expect(client['thinkingClearLatched']).toBe(false);
+      expect(client['lastApiCompletionTimestamp']).toBeNull();
+    });
+  });
+
   describe('tryCompressChat', () => {
     const mockGetHistory = vi.fn();
 
@@ -436,6 +549,7 @@ describe('Gemini Client (client.ts)', () => {
         addHistory: vi.fn(),
         setHistory: vi.fn(),
         stripThoughtsFromHistory: vi.fn(),
+        stripThoughtsFromHistoryKeepRecent: vi.fn(),
       } as unknown as GeminiChat;
     });
 
@@ -457,6 +571,7 @@ describe('Gemini Client (client.ts)', () => {
         getHistory: vi.fn((_curated?: boolean) => chatHistory),
         setHistory: vi.fn(),
         stripThoughtsFromHistory: vi.fn(),
+        stripThoughtsFromHistoryKeepRecent: vi.fn(),
       };
       client['chat'] = mockOriginalChat as GeminiChat;
 
@@ -1149,6 +1264,7 @@ describe('Gemini Client (client.ts)', () => {
         addHistory: vi.fn(),
         getHistory: vi.fn().mockReturnValue([]),
         stripThoughtsFromHistory: vi.fn(),
+        stripThoughtsFromHistoryKeepRecent: vi.fn(),
       } as unknown as GeminiChat;
       client['chat'] = mockChat;
 
@@ -1204,6 +1320,7 @@ Other open files:
         addHistory: vi.fn(),
         getHistory: vi.fn().mockReturnValue([]),
         stripThoughtsFromHistory: vi.fn(),
+        stripThoughtsFromHistoryKeepRecent: vi.fn(),
       };
       client['chat'] = mockChat as GeminiChat;
 
@@ -1260,6 +1377,7 @@ Other open files:
         addHistory: vi.fn(),
         getHistory: vi.fn().mockReturnValue([]),
         stripThoughtsFromHistory: vi.fn(),
+        stripThoughtsFromHistoryKeepRecent: vi.fn(),
       };
       client['chat'] = mockChat as GeminiChat;
 
@@ -1326,6 +1444,7 @@ hello
         addHistory: vi.fn(),
         getHistory: vi.fn().mockReturnValue([]),
         stripThoughtsFromHistory: vi.fn(),
+        stripThoughtsFromHistoryKeepRecent: vi.fn(),
       };
       client['chat'] = mockChat as GeminiChat;
 
@@ -1365,6 +1484,7 @@ Other open files:
         addHistory: vi.fn(),
         getHistory: vi.fn().mockReturnValue([]),
         stripThoughtsFromHistory: vi.fn(),
+        stripThoughtsFromHistoryKeepRecent: vi.fn(),
       };
       client['chat'] = mockChat as GeminiChat;
 
@@ -1410,6 +1530,7 @@ Other open files:
         addHistory: vi.fn(),
         getHistory: vi.fn().mockReturnValue([]),
         stripThoughtsFromHistory: vi.fn(),
+        stripThoughtsFromHistoryKeepRecent: vi.fn(),
       };
       client['chat'] = mockChat as GeminiChat;
 
@@ -1498,6 +1619,7 @@ Other open files:
         addHistory: vi.fn(),
         getHistory: vi.fn().mockReturnValue([]),
         stripThoughtsFromHistory: vi.fn(),
+        stripThoughtsFromHistoryKeepRecent: vi.fn(),
       };
       client['chat'] = mockChat as GeminiChat;
 
@@ -1555,6 +1677,7 @@ Other open files:
         addHistory: vi.fn(),
         getHistory: vi.fn().mockReturnValue([]),
         stripThoughtsFromHistory: vi.fn(),
+        stripThoughtsFromHistoryKeepRecent: vi.fn(),
       };
       client['chat'] = mockChat as GeminiChat;
 
@@ -1636,6 +1759,7 @@ Other open files:
               { role: 'user', parts: [{ text: 'previous message' }] },
             ]),
           stripThoughtsFromHistory: vi.fn(),
+          stripThoughtsFromHistoryKeepRecent: vi.fn(),
         };
         client['chat'] = mockChat as GeminiChat;
       });
@@ -1889,6 +2013,7 @@ Other open files:
           getHistory: vi.fn().mockReturnValue([]), // Default empty history
           setHistory: vi.fn(),
           stripThoughtsFromHistory: vi.fn(),
+          stripThoughtsFromHistoryKeepRecent: vi.fn(),
         };
         client['chat'] = mockChat as GeminiChat;
 
@@ -2228,6 +2353,7 @@ Other open files:
         addHistory: vi.fn(),
         getHistory: vi.fn().mockReturnValue([]),
         stripThoughtsFromHistory: vi.fn(),
+        stripThoughtsFromHistoryKeepRecent: vi.fn(),
       };
       client['chat'] = mockChat as GeminiChat;
 
@@ -2265,6 +2391,7 @@ Other open files:
         addHistory: vi.fn(),
         getHistory: vi.fn().mockReturnValue([]),
         stripThoughtsFromHistory: vi.fn(),
+        stripThoughtsFromHistoryKeepRecent: vi.fn(),
       };
       client['chat'] = mockChat as GeminiChat;
 
@@ -2305,6 +2432,7 @@ Other open files:
         addHistory: vi.fn(),
         getHistory: vi.fn().mockReturnValue([]),
         stripThoughtsFromHistory: vi.fn(),
+        stripThoughtsFromHistoryKeepRecent: vi.fn(),
       };
       client['chat'] = mockChat as GeminiChat;
 
@@ -2329,6 +2457,7 @@ Other open files:
           getHistory: vi.fn().mockReturnValue([]),
           setHistory: vi.fn(),
           stripThoughtsFromHistory: vi.fn(),
+          stripThoughtsFromHistoryKeepRecent: vi.fn(),
           stripOrphanedUserEntriesFromHistory: vi.fn(),
         };
         client['chat'] = mockChat as GeminiChat;
@@ -2361,6 +2490,7 @@ Other open files:
           getHistory: vi.fn().mockReturnValue([]),
           setHistory: vi.fn(),
           stripThoughtsFromHistory: vi.fn(),
+          stripThoughtsFromHistoryKeepRecent: vi.fn(),
           stripOrphanedUserEntriesFromHistory: vi.fn(),
         };
         client['chat'] = mockChat as GeminiChat;
@@ -2405,6 +2535,7 @@ Other open files:
           addHistory: vi.fn(),
           getHistory: vi.fn().mockReturnValue([]),
           stripThoughtsFromHistory: vi.fn(),
+          stripThoughtsFromHistoryKeepRecent: vi.fn(),
         };
         client['chat'] = mockChat as GeminiChat;
       });

packages/core/src/core/client.ts

@@ -111,6 +111,13 @@ export interface SendMessageOptions {
   };
 }
 
+/**
+ * Idle threshold for thinking block cleanup. After this period without any
+ * API call the old thinking blocks are unlikely to aid reasoning coherence
+ * and only waste context tokens.
+ */
+const THINKING_IDLE_THRESHOLD_MS = 60 * 60 * 1000; // 1 hour
+
 export class GeminiClient {
   private chat?: GeminiChat;
   private sessionTurnCount = 0;
@@ -126,6 +133,25 @@ export class GeminiClient {
    */
   private hasFailedCompressionAttempt = false;
 
+  /**
+   * Timestamp (epoch ms) of the last completed API call.
+   * Used to detect idle periods for thinking block cleanup.
+   * Starts as null — on the first query there is no prior thinking to clean,
+   * so the idle check is skipped until the first API call completes.
+   */
+  private lastApiCompletionTimestamp: number | null = null;
+
+  /**
+   * Sticky-on latch for clearing thinking blocks from prior turns.
+   * Triggered when >1h since last API call — old thinking is no longer
+   * useful for reasoning coherence. Once latched, stays true to prevent
+   * oscillation: without it, thinking would accumulate → get stripped →
+   * accumulate again, causing the message prefix to change repeatedly
+   * (bad for any provider-side prompt caching and wastes context).
+   * Reset on /clear (resetChat).
+   */
+  private thinkingClearLatched = false;
+
   constructor(private readonly config: Config) {
     this.loopDetector = new LoopDetectionService(config);
   }
@@ -199,6 +225,9 @@ export class GeminiClient {
   }
 
   async resetChat(): Promise<void> {
+    // Reset thinking clear latch — fresh chat, no prior thinking to clean up
+    this.thinkingClearLatched = false;
+    this.lastApiCompletionTimestamp = null;
     await this.startChat();
   }
 
@@ -537,8 +566,24 @@ export class GeminiClient {
       // record user message for session management
       this.config.getChatRecordingService()?.recordUserMessage(request);
 
-      // strip thoughts from history before sending the message
-      this.stripThoughtsFromHistory();
+      // Thinking block cross-turn retention with idle cleanup:
+      // - Active session (< 1h idle): keep thinking blocks for reasoning coherence
+      // - Idle > 1h: clear old thinking, keep only last 1 turn to free context
+      // - Latch: once triggered, never revert — prevents oscillation
+      if (
+        !this.thinkingClearLatched &&
+        this.lastApiCompletionTimestamp !== null
+      ) {
+        if (
+          Date.now() - this.lastApiCompletionTimestamp >
+          THINKING_IDLE_THRESHOLD_MS
+        ) {
+          this.thinkingClearLatched = true;
+        }
+      }
+      if (this.thinkingClearLatched) {
+        this.getChat().stripThoughtsFromHistoryKeepRecent(1);
+      }
     }
     if (messageType !== SendMessageType.Retry) {
       this.sessionTurnCount++;
@@ -680,6 +725,7 @@ export class GeminiClient {
           if (arenaAgentClient) {
             await arenaAgentClient.reportError('Loop detected');
           }
+          this.lastApiCompletionTimestamp = Date.now();
           return turn;
         }
       }
@@ -698,9 +744,14 @@ export class GeminiClient {
               : 'Unknown error';
           await arenaAgentClient.reportError(errorMsg);
         }
+        this.lastApiCompletionTimestamp = Date.now();
         return turn;
       }
     }
+
+    // Track API completion time for thinking block idle cleanup
+    this.lastApiCompletionTimestamp = Date.now();
+
     // Fire Stop hook through MessageBus (only if hooks are enabled and registered)
     // This must be done before any early returns to ensure hooks are always triggered
     if (