🤖 feat: add 5% buffer between auto-compaction threshold and force-compaction

src/browser/components/AIView.tsx

@@ -594,6 +594,7 @@ const AIViewInner: React.FC<AIViewProps> = ({
           <CompactionWarning
             usagePercentage={autoCompactionResult.usagePercentage}
             thresholdPercentage={autoCompactionResult.thresholdPercentage}
+            isStreaming={canInterrupt}
             onCompactClick={handleCompactClick}
           />
         )}

src/browser/components/CompactionWarning.tsx

@@ -1,38 +1,58 @@
 import React from "react";
+import { FORCE_COMPACTION_BUFFER_PERCENT } from "@/common/constants/ui";
 
 /**
  * Warning indicator shown when context usage is approaching the compaction threshold.
  *
  * Displays as subtle right-aligned text:
  * - Below threshold: "Auto-Compact in X% usage" (where X = threshold - current)
- * - At/above threshold: Bold "Next message will Auto-Compact"
+ * - At/above threshold (not streaming): Bold "Next message will Auto-Compact"
+ * - At/above threshold (streaming): "Force-compacting in N%" (where N = force threshold - current usage)
  *
- * Both states are clickable to insert /compact command.
+ * All states are clickable to insert /compact command.
  *
- * @param usagePercentage - Current token usage as percentage (0-100)
+ * @param usagePercentage - Current token usage as percentage (0-100), reflects live usage when streaming
  * @param thresholdPercentage - Auto-compaction trigger threshold (0-100, default 70)
+ * @param isStreaming - Whether currently streaming a response
  * @param onCompactClick - Callback when user clicks to trigger manual compaction
  */
 export const CompactionWarning: React.FC<{
   usagePercentage: number;
   thresholdPercentage: number;
+  isStreaming: boolean;
   onCompactClick?: () => void;
 }> = (props) => {
   // At threshold or above, next message will trigger compaction
   const willCompactNext = props.usagePercentage >= props.thresholdPercentage;
   const remaining = props.thresholdPercentage - props.usagePercentage;
 
-  const text = willCompactNext
-    ? "Next message will Auto-Compact"
-    : `Auto-Compact in ${Math.round(remaining)}% usage`;
+  // When streaming and above threshold, show countdown to force-compaction
+  const forceCompactThreshold = props.thresholdPercentage + FORCE_COMPACTION_BUFFER_PERCENT;
+  const showForceCompactCountdown =
+    props.isStreaming && willCompactNext && props.usagePercentage < forceCompactThreshold;
+  const forceCompactRemaining = forceCompactThreshold - props.usagePercentage;
+
+  let text: string;
+  let isUrgent: boolean;
+
+  if (showForceCompactCountdown) {
+    text = `Force-compacting in ${Math.round(forceCompactRemaining)}%`;
+    isUrgent = false;
+  } else if (willCompactNext) {
+    text = "Next message will Auto-Compact";
+    isUrgent = true;
+  } else {
+    text = `Auto-Compact in ${Math.round(remaining)}% usage`;
+    isUrgent = false;
+  }
 
   return (
     <div className="mx-4 mt-2 mb-1 text-right text-[10px]">
       <button
         type="button"
         onClick={props.onCompactClick}
         className={`cursor-pointer hover:underline ${
-          willCompactNext ? "text-plan-mode font-semibold" : "text-muted"
+          isUrgent ? "text-plan-mode font-semibold" : "text-muted"
         }`}
         title="Click to insert /compact command"
       >

src/browser/utils/compaction/autoCompactionCheck.test.ts

@@ -3,7 +3,6 @@ import { checkAutoCompaction } from "./autoCompactionCheck";
 import type { WorkspaceUsageState } from "@/browser/stores/WorkspaceStore";
 import type { ChatUsageDisplay } from "@/common/utils/tokens/usageAggregator";
 import { KNOWN_MODELS } from "@/common/constants/knownModels";
-import { FORCE_COMPACTION_TOKEN_BUFFER } from "@/common/constants/ui";
 
 // Helper to create a mock usage entry
 const createUsageEntry = (
@@ -302,63 +301,63 @@ describe("checkAutoCompaction", () => {
     });
   });
 
-  describe("Force Compaction (Live Usage)", () => {
-    const SONNET_MAX_TOKENS = 200_000;
-    const BUFFER = FORCE_COMPACTION_TOKEN_BUFFER;
+  describe("Force Compaction (threshold + 5% buffer)", () => {
+    // Force-compact triggers at threshold + 5%
+    // With default 70% threshold, force-compact at 75%
 
-    test("shouldForceCompact is false when no liveUsage (falls back to lastUsage with room)", () => {
-      const usage = createMockUsage(100_000); // 100k remaining - plenty of room
+    test("shouldForceCompact is false when usage just below force threshold", () => {
+      // 74% usage, threshold 70%, force at 75% - should NOT trigger
+      const usage = createMockUsage(148_000); // 74%
       const result = checkAutoCompaction(usage, KNOWN_MODELS.SONNET.id, false);
 
       expect(result.shouldForceCompact).toBe(false);
     });
 
-    test("shouldForceCompact is false when currentUsage has plenty of room", () => {
-      const liveUsage = createUsageEntry(100_000); // 100k remaining
-      const usage = createMockUsage(50_000, undefined, KNOWN_MODELS.SONNET.id, liveUsage);
+    test("shouldForceCompact is true when usage at force threshold", () => {
+      // 75% usage, threshold 70%, force at 75% - should trigger
+      const usage = createMockUsage(150_000); // 75%
       const result = checkAutoCompaction(usage, KNOWN_MODELS.SONNET.id, false);
 
-      expect(result.shouldForceCompact).toBe(false);
+      expect(result.shouldForceCompact).toBe(true);
     });
 
-    test("shouldForceCompact is true when remaining <= buffer", () => {
-      // Exactly at buffer threshold
-      const liveUsage = createUsageEntry(SONNET_MAX_TOKENS - BUFFER);
-      const usage = createMockUsage(50_000, undefined, KNOWN_MODELS.SONNET.id, liveUsage);
+    test("shouldForceCompact is true when usage above force threshold", () => {
+      // 80% usage, threshold 70%, force at 75% - should trigger
+      const usage = createMockUsage(160_000); // 80%
       const result = checkAutoCompaction(usage, KNOWN_MODELS.SONNET.id, false);
 
       expect(result.shouldForceCompact).toBe(true);
     });
 
-    test("shouldForceCompact is true when over context limit", () => {
-      const liveUsage = createUsageEntry(SONNET_MAX_TOKENS + 5000);
-      const usage = createMockUsage(50_000, undefined, KNOWN_MODELS.SONNET.id, liveUsage);
+    test("shouldForceCompact uses liveUsage when available", () => {
+      // lastUsage at 50%, liveUsage at 75% - should trigger based on live
+      const liveUsage = createUsageEntry(150_000); // 75%
+      const usage = createMockUsage(100_000, undefined, KNOWN_MODELS.SONNET.id, liveUsage);
       const result = checkAutoCompaction(usage, KNOWN_MODELS.SONNET.id, false);
 
       expect(result.shouldForceCompact).toBe(true);
+      expect(result.usagePercentage).toBe(75); // usagePercentage reflects live when streaming
     });
 
-    test("shouldForceCompact is false when just above buffer", () => {
-      // 1 token above buffer threshold
-      const liveUsage = createUsageEntry(SONNET_MAX_TOKENS - BUFFER - 1);
-      const usage = createMockUsage(50_000, undefined, KNOWN_MODELS.SONNET.id, liveUsage);
-      const result = checkAutoCompaction(usage, KNOWN_MODELS.SONNET.id, false);
+    test("shouldForceCompact respects custom threshold", () => {
+      // 55% usage with 50% threshold - force at 55%, should trigger
+      const usage = createMockUsage(110_000); // 55%
+      const result = checkAutoCompaction(usage, KNOWN_MODELS.SONNET.id, false, 0.5);
 
-      expect(result.shouldForceCompact).toBe(false);
+      expect(result.shouldForceCompact).toBe(true);
     });
 
     test("shouldForceCompact respects 1M context mode", () => {
-      // With 1M context, exactly at buffer threshold
-      const liveUsage = createUsageEntry(1_000_000 - BUFFER);
+      // 75% of 1M = 750k tokens
+      const liveUsage = createUsageEntry(750_000);
       const usage = createMockUsage(50_000, undefined, KNOWN_MODELS.SONNET.id, liveUsage);
       const result = checkAutoCompaction(usage, KNOWN_MODELS.SONNET.id, true);
 
       expect(result.shouldForceCompact).toBe(true);
     });
 
-    test("shouldForceCompact triggers with empty history but liveUsage near limit", () => {
-      // Bug fix: empty history but liveUsage should still trigger
-      const liveUsage = createUsageEntry(SONNET_MAX_TOKENS - BUFFER);
+    test("shouldForceCompact triggers with empty history but liveUsage at force threshold", () => {
+      const liveUsage = createUsageEntry(150_000); // 75%
       const usage: WorkspaceUsageState = {
         usageHistory: [],
         totalTokens: 0,
@@ -367,12 +366,35 @@ describe("checkAutoCompaction", () => {
       const result = checkAutoCompaction(usage, KNOWN_MODELS.SONNET.id, false);
 
       expect(result.shouldForceCompact).toBe(true);
-      expect(result.usagePercentage).toBe(0); // No lastUsage for percentage
+      expect(result.usagePercentage).toBe(75); // usagePercentage reflects live even with empty history
+    });
+
+    test("shouldShowWarning uses live usage when no history exists", () => {
+      // No lastUsage, liveUsage at 65% - should show warning (65% >= 60%)
+      const liveUsage = createUsageEntry(130_000); // 65%
+      const usage: WorkspaceUsageState = {
+        usageHistory: [],
+        totalTokens: 0,
+        liveUsage,
+      };
+      const result = checkAutoCompaction(usage, KNOWN_MODELS.SONNET.id, false);
+
+      expect(result.shouldShowWarning).toBe(true);
+      expect(result.shouldForceCompact).toBe(false); // 65% < 75%
+    });
+
+    test("shouldShowWarning uses max of last and live usage", () => {
+      // lastUsage at 50% (below warning), liveUsage at 72% (above warning)
+      const liveUsage = createUsageEntry(144_000); // 72%
+      const usage = createMockUsage(100_000, undefined, KNOWN_MODELS.SONNET.id, liveUsage);
+      const result = checkAutoCompaction(usage, KNOWN_MODELS.SONNET.id, false);
+
+      expect(result.shouldShowWarning).toBe(true); // 72% >= 60%
+      expect(result.shouldForceCompact).toBe(false); // 72% < 75%
     });
 
     test("shouldForceCompact is false when auto-compaction disabled", () => {
-      const liveUsage = createUsageEntry(199_000); // Very close to limit
-      const usage = createMockUsage(50_000, undefined, KNOWN_MODELS.SONNET.id, liveUsage);
+      const usage = createMockUsage(190_000); // 95% - would trigger if enabled
       const result = checkAutoCompaction(usage, KNOWN_MODELS.SONNET.id, false, 1.0); // disabled
 
       expect(result.shouldForceCompact).toBe(false);

src/browser/utils/compaction/autoCompactionCheck.ts

@@ -21,7 +21,7 @@ import { getModelStats } from "@/common/utils/tokens/modelStats";
 import { supports1MContext } from "@/common/utils/ai/models";
 import {
   DEFAULT_AUTO_COMPACTION_THRESHOLD,
-  FORCE_COMPACTION_TOKEN_BUFFER,
+  FORCE_COMPACTION_BUFFER_PERCENT,
 } from "@/common/constants/ui";
 
 /** Sum all token components from a ChatUsageDisplay */
@@ -37,8 +37,9 @@ function getTotalTokens(usage: ChatUsageDisplay): number {
 
 export interface AutoCompactionCheckResult {
   shouldShowWarning: boolean;
-  /** True when live usage shows ≤FORCE_COMPACTION_TOKEN_BUFFER remaining in context */
+  /** True when usage exceeds threshold + buffer (gives user control before force-compact) */
   shouldForceCompact: boolean;
+  /** Current usage percentage - live when streaming, otherwise last completed */
   usagePercentage: number;
   thresholdPercentage: number;
 }
@@ -94,30 +95,22 @@ export function checkAutoCompaction(
     };
   }
 
-  // Current usage: live when streaming, else last historical
-  // Use lastContextUsage (last step) for accurate context window size
+  // Current usage: live when streaming, else last completed
   const lastUsage = usage.lastContextUsage;
   const currentUsage = usage.liveUsage ?? lastUsage;
 
-  // Force-compact when approaching context limit (can trigger even with empty history if streaming)
-  let shouldForceCompact = false;
-  if (currentUsage) {
-    const remainingTokens = maxTokens - getTotalTokens(currentUsage);
-    shouldForceCompact = remainingTokens <= FORCE_COMPACTION_TOKEN_BUFFER;
-  }
+  // Usage percentage from current context (live when streaming, otherwise last completed)
+  const usagePercentage = currentUsage ? (getTotalTokens(currentUsage) / maxTokens) * 100 : 0;
 
-  // Warning/percentage based on lastUsage (completed requests only)
-  if (!lastUsage) {
-    return {
-      shouldShowWarning: false,
-      shouldForceCompact,
-      usagePercentage: 0,
-      thresholdPercentage,
-    };
-  }
+  // Force-compact when usage exceeds threshold + buffer
+  const forceCompactThreshold = thresholdPercentage + FORCE_COMPACTION_BUFFER_PERCENT;
+  const shouldForceCompact = usagePercentage >= forceCompactThreshold;
 
-  const usagePercentage = (getTotalTokens(lastUsage) / maxTokens) * 100;
-  const shouldShowWarning = usagePercentage >= thresholdPercentage - warningAdvancePercent;
+  // Warning uses max of last completed and current (live when streaming)
+  // This ensures warning shows when live usage spikes above threshold mid-stream
+  const lastUsagePercentage = lastUsage ? (getTotalTokens(lastUsage) / maxTokens) * 100 : 0;
+  const shouldShowWarning =
+    Math.max(lastUsagePercentage, usagePercentage) >= thresholdPercentage - warningAdvancePercent;
 
   return {
     shouldShowWarning,

src/common/constants/ui.ts

@@ -41,14 +41,11 @@ export const DEFAULT_COMPACTION_WORD_TARGET = 2000;
 export const WORDS_TO_TOKENS_RATIO = 1.3;
 
 /**
- * Force-compaction token buffer.
- * When auto-compaction is enabled and live usage shows this many tokens or fewer
- * remaining in the context window, force a compaction immediately.
- * Set to 2x the expected compaction output size to ensure room for the summary.
- */
-export const FORCE_COMPACTION_TOKEN_BUFFER = Math.round(
-  2 * DEFAULT_COMPACTION_WORD_TARGET * WORDS_TO_TOKENS_RATIO
-); // = 5200 tokens
+ * Force-compact this many percentage points after threshold.
+ * Gives user a buffer zone between warning and force-compaction.
+ * E.g., with 70% threshold, force-compact triggers at 75%.
+ */
+export const FORCE_COMPACTION_BUFFER_PERCENT = 5;
 
 /**
  * Duration (ms) to show "copied" feedback after copying to clipboard