diff --git a/packages/core/src/agent/context-window/index.ts b/packages/core/src/agent/context-window/index.ts
index 8d04b64f..62ebdf4b 100644
--- a/packages/core/src/agent/context-window/index.ts
+++ b/packages/core/src/agent/context-window/index.ts
@@ -43,7 +43,6 @@ export {
   splitMessagesForSummary,
   detectSplitTurn,
   computeAdaptiveChunkRatio,
-  compactMessagesWithSummary,
   compactMessagesWithChunkedSummary,
 } from "./summarization.js";
 
diff --git a/packages/core/src/agent/context-window/summarization.ts b/packages/core/src/agent/context-window/summarization.ts
index 1ff99c11..859b8e11 100644
--- a/packages/core/src/agent/context-window/summarization.ts
+++ b/packages/core/src/agent/context-window/summarization.ts
@@ -238,116 +238,6 @@ function createSummaryMessage(summary: string, previousSummary?: string): AgentM
   };
 }
 
-/**
- * Execute summary-based compaction
- *
- * Uses LLM to generate summary of historical messages, then combines summary with recent messages
- */
-export async function compactMessagesWithSummary(
-  params: SummaryCompactionParams,
-): Promise<SummaryCompactionResult | null> {
-  const {
-    messages,
-    model,
-    apiKey,
-    availableTokens,
-    targetRatio,
-    minKeepMessages,
-    reserveTokens = 2048,
-    customInstructions,
-    previousSummary,
-    signal,
-  } = params;
-
-  // Split messages
-  const split = splitMessagesForSummary(messages, availableTokens, {
-    targetRatio,
-    minKeepMessages,
-  });
-
-  if (!split) {
-    return null;
-  }
-
-  let { toSummarize, toKeep } = split;
-
-  // Detect and handle split turn
-  const splitTurn = detectSplitTurn(toSummarize, toKeep);
-  let splitPrefixSummary = "";
-
-  if (splitTurn) {
-    toSummarize = splitTurn.adjustedToSummarize;
-    toKeep = splitTurn.adjustedToKeep;
-
-    // Summarize the split prefix separately
-    const instructions = customInstructions || DEFAULT_SUMMARY_INSTRUCTIONS;
-    const prefixResult = await summarizeWithFallback({
-      messages: splitTurn.splitPrefix,
-      model,
-      reserveTokens,
-      apiKey,
-      signal,
-      instructions,
-      previousSummary,
-      availableTokens,
-    });
-    splitPrefixSummary = prefixResult.summary;
-  }
-
-  // Generate summary with fallback (toSummarize no longer contains split prefix messages)
-  const instructions = customInstructions || DEFAULT_SUMMARY_INSTRUCTIONS;
-  let finalSummary = "";
-
-  if (toSummarize.length > 0) {
-    const { summary } = await summarizeWithFallback({
-      messages: toSummarize,
-      model,
-      reserveTokens,
-      apiKey,
-      signal,
-      instructions,
-      previousSummary,
-      availableTokens,
-    });
-    finalSummary = summary;
-  }
-
-  // Append split prefix summary if present
-  if (splitPrefixSummary) {
-    finalSummary += (finalSummary ? "\n\n" : "") + `## Split Turn Context\n${splitPrefixSummary}`;
-  }
-
-  // Append metadata sections (all compacted = adjusted toSummarize + splitPrefix)
-  const allCompactedMessages = splitTurn
-    ? [...toSummarize, ...splitTurn.splitPrefix]
-    : toSummarize;
-  const failures = collectToolFailures(allCompactedMessages);
-  const fileOps = collectFileOperations(allCompactedMessages);
-
-  finalSummary += formatToolFailuresSection(failures);
-  finalSummary += formatFileOperationsSection(fileOps);
-
-  // Create summary message
-  const summaryMessage = createSummaryMessage(finalSummary, previousSummary);
-
-  // Combine results
-  const kept = [summaryMessage, ...toKeep];
-
-  const tokensRemoved = estimateMessagesTokens(allCompactedMessages);
-  const tokensKept = estimateMessagesTokens(kept);
-
-  return {
-    kept,
-    removedCount: allCompactedMessages.length,
-    tokensRemoved,
-    tokensKept,
-    summary: finalSummary,
-    reason: "summary",
-    fileOperations: (fileOps.readFiles.length > 0 || fileOps.modifiedFiles.length > 0) ? fileOps : undefined,
-    toolFailures: failures.length > 0 ? failures : undefined,
-  };
-}
-
 /**
  * Generate summary in chunks (for very large history)
  *
diff --git a/packages/core/src/agent/session/compaction.test.ts b/packages/core/src/agent/session/compaction.test.ts
index e3087816..b5cf3d70 100644
--- a/packages/core/src/agent/session/compaction.test.ts
+++ b/packages/core/src/agent/session/compaction.test.ts
@@ -1,6 +1,5 @@
 import { describe, it, expect, vi } from "vitest";
 import {
-  compactMessagesByCount,
   compactMessagesByTokens,
   compactMessages,
   type CompactionResult,
@@ -27,96 +26,6 @@ describe("compaction", () => {
     })) as AgentMessage[];
   }
 
-  function createMessagesWithToolUse(): AgentMessage[] {
-    return [
-      { role: "user", content: "Start" },
-      {
-        role: "assistant",
-        content: [{ type: "tool_use", id: "tool-1", name: "test", input: {} }],
-      } as any,
-      {
-        role: "user",
-        content: [{ type: "tool_result", tool_use_id: "tool-1", content: "Result" }],
-      } as any,
-      { role: "assistant", content: "Done" },
-      { role: "user", content: "Next message" },
-    ];
-  }
-
-  describe("compactMessagesByCount", () => {
-    it("should return null when under max messages", () => {
-      const messages = createMessages(50);
-      const result = compactMessagesByCount(messages, 80, 60);
-      expect(result).toBeNull();
-    });
-
-    it("should compact when over max messages", () => {
-      const messages = createMessages(100);
-      const result = compactMessagesByCount(messages, 80, 60);
-
-      expect(result).not.toBeNull();
-      expect(result!.reason).toBe("count");
-      expect(result!.kept.length).toBeLessThanOrEqual(100);
-      expect(result!.removedCount).toBeGreaterThan(0);
-    });
-
-    it("should keep the specified number of last messages", () => {
-      const messages = createMessages(100);
-      const result = compactMessagesByCount(messages, 80, 50);
-
-      if (result) {
-        // Should keep approximately keepLast messages
-        expect(result.kept.length).toBeGreaterThanOrEqual(40);
-        expect(result.kept.length).toBeLessThanOrEqual(60);
-      }
-    });
-
-    it("should return null when exact at max messages", () => {
-      const messages = createMessages(80);
-      const result = compactMessagesByCount(messages, 80, 60);
-      expect(result).toBeNull();
-    });
-
-    it("should not break tool_use/tool_result pairs", () => {
-      // Create many messages followed by a tool pair
-      const regularMessages = createMessages(70);
-      const toolMessages = createMessagesWithToolUse();
-      const messages = [...regularMessages, ...toolMessages];
-
-      const result = compactMessagesByCount(messages, 80, 20);
-
-      if (result) {
-        // Check that we didn't end up with orphaned tool_result
-        let hasOrphanedToolResult = false;
-        for (let i = 0; i < result.kept.length; i++) {
-          const msg = result.kept[i] as any;
-          if (Array.isArray(msg.content)) {
-            const hasToolResult = msg.content.some((b: any) => b.type === "tool_result");
-            if (hasToolResult) {
-              // Check if previous message has corresponding tool_use
-              const prevMsg = result.kept[i - 1] as any;
-              if (!prevMsg || !Array.isArray(prevMsg.content)) {
-                hasOrphanedToolResult = true;
-              }
-            }
-          }
-        }
-        // This test verifies the safe compaction point logic
-        // The exact behavior depends on findSafeCompactionPoint implementation
-      }
-    });
-
-    it("should return null when would keep almost all messages", () => {
-      const messages = createMessages(85);
-      const result = compactMessagesByCount(messages, 80, 82);
-
-      // If we'd only remove 2-3 messages, should return null
-      if (result) {
-        expect(result.removedCount).toBeGreaterThan(2);
-      }
-    });
-  });
-
   describe("compactMessagesByTokens", () => {
     it("should return null when under token limit", () => {
       const messages = createMessages(5);
@@ -161,41 +70,16 @@ describe("compaction", () => {
   });
 
   describe("compactMessages (unified entry point)", () => {
-    describe("count mode", () => {
-      it("should use count-based compaction", () => {
-        const messages = createMessages(100);
-        const result = compactMessages(messages, {
-          mode: "count",
-          maxMessages: 80,
-          keepLast: 60,
-        });
-
-        expect(result).not.toBeNull();
-        expect(result!.reason).toBe("count");
-      });
-
-      it("should use default max and keep values", () => {
-        const messages = createMessages(100);
-        const result = compactMessages(messages, {
-          mode: "count",
-        });
-
-        // Default: maxMessages: 80, keepLast: 60
-        expect(result).not.toBeNull();
-        expect(result!.reason).toBe("count");
-      });
-    });
-
     describe("tokens mode", () => {
       it("should use token-based compaction when utilization is high", () => {
         const messages = createMessages(100);
         // ~300 message tokens (real estimator: ~3 tokens/msg)
-        // systemPromptTokens ≈ 7, reserveTokens = 0
-        // available = 500 - 7 = 493
-        // utilization = (300 * 1.5) / 493 ≈ 0.91 > 0.8 → should compact
+        // systemPromptTokens ≈ 4, reserveTokens = 0
+        // available = 400 - 4 = 396
+        // utilization = (300 * 1.2) / 396 ≈ 0.91 > 0.8 → should compact
         const result = compactMessages(messages, {
           mode: "tokens",
-          contextWindowTokens: 500,
+          contextWindowTokens: 400,
           systemPrompt: "System prompt",
           reserveTokens: 0,
         });
@@ -207,8 +91,8 @@ describe("compaction", () => {
       it("should return null when utilization is low", () => {
         const messages = createMessages(5);
         // ~15 message tokens
-        // available = 10000 - 7 - 1024 = 8969
-        // utilization = (15 * 1.5) / 8969 ≈ 0.003 < 0.8
+        // available = 10000 - 4 - 1024 = 8972
+        // utilization = (15 * 1.2) / 8972 ≈ 0.002 < 0.8
         const result = compactMessages(messages, {
           mode: "tokens",
           contextWindowTokens: 10000,
diff --git a/packages/core/src/agent/session/compaction.ts b/packages/core/src/agent/session/compaction.ts
index 3651f972..5543588b 100644
--- a/packages/core/src/agent/session/compaction.ts
+++ b/packages/core/src/agent/session/compaction.ts
@@ -1,11 +1,9 @@
 import type { AgentMessage } from "@mariozechner/pi-agent-core";
 import type { Model } from "@mariozechner/pi-ai";
 import {
-  estimateMessagesTokens,
   compactMessagesTokenAware,
   estimateTokenUsage,
   shouldCompact as shouldCompactTokens,
-  compactMessagesWithSummary,
   compactMessagesWithChunkedSummary,
   COMPACTION_TARGET_RATIO,
   MIN_KEEP_MESSAGES,
@@ -30,96 +28,12 @@ export type CompactionResult = {
   fileOperations?: { readFiles: string[]; modifiedFiles: string[] } | undefined;
   /** Tool failures extracted from compacted messages */
   toolFailures?: Array<{ toolName: string; summary: string }> | undefined;
-  /** Reason for compaction: count, tokens, summary, or pruning (tool result trimming only) */
-  reason: "count" | "tokens" | "summary" | "pruning";
+  /** Reason for compaction: tokens, summary, or pruning (tool result trimming only) */
+  reason: "tokens" | "summary" | "pruning";
   /** Tool result pruning statistics (when Phase 1 pruning was applied) */
   pruningStats?: PruningStats | undefined;
 };
 
-/**
- * Find a safe compaction point that doesn't break tool_use/tool_result pairs.
- * Returns the index to start keeping messages from.
- */
-function findSafeCompactionPoint(messages: AgentMessage[], targetStart: number): number {
-  let start = targetStart;
-
-  // Move forward until we find a safe starting point
-  while (start < messages.length) {
-    const msg = messages[start];
-    if (!msg) {
-      start++;
-      continue;
-    }
-
-    // Safe to start from a user message
-    if (msg.role === "user") {
-      // But make sure it's not a toolResult without corresponding tool_use
-      const msgAny = msg as any;
-      if (Array.isArray(msgAny.content)) {
-        const hasToolResult = msgAny.content.some((b: any) => b.type === "tool_result");
-        if (!hasToolResult) {
-          break; // Safe: user message without tool_result
-        }
-      } else {
-        break; // Safe: simple user message
-      }
-    }
-
-    // toolResult messages need their corresponding tool_use, skip them
-    // assistant messages are ok to start from if they don't reference missing tool calls
-    if (msg.role === "assistant") {
-      // Check if previous messages have the required tool_use for any following tool_result
-      const nextMsg = messages[start + 1];
-      if (nextMsg && nextMsg.role === "user") {
-        const nextAny = nextMsg as any;
-        if (Array.isArray(nextAny.content)) {
-          const hasToolResult = nextAny.content.some((b: any) => b.type === "tool_result");
-          if (hasToolResult) {
-            // This assistant message has tool_use that's needed by next message
-            break;
-          }
-        }
-      }
-    }
-
-    start++;
-  }
-
-  return start;
-}
-
-/**
- * Simple compression based on message count (legacy logic, maintains backward compatibility)
- */
-export function compactMessagesByCount(
-  messages: AgentMessage[],
-  maxMessages: number,
-  keepLast: number,
-): CompactionResult | null {
-  if (messages.length <= maxMessages) return null;
-
-  const targetStart = messages.length - keepLast;
-  const safeStart = findSafeCompactionPoint(messages, targetStart);
-
-  // If we can't find a safe point, don't compact
-  if (safeStart >= messages.length) {
-    return null;
-  }
-
-  const kept = messages.slice(safeStart);
-
-  // Don't compact if we'd keep almost everything anyway
-  if (kept.length >= messages.length - 2) {
-    return null;
-  }
-
-  return {
-    kept,
-    removedCount: messages.length - kept.length,
-    reason: "count",
-  };
-}
-
 /**
  * Token-based intelligent compression
  */
@@ -143,13 +57,9 @@ export function compactMessagesByTokens(
   };
 }
 
-/** Synchronous compaction options (count/tokens modes) */
-export type SyncCompactionOptions = {
-  mode: "count" | "tokens";
-  // count mode parameters
-  maxMessages?: number | undefined;
-  keepLast?: number | undefined;
-  // tokens mode parameters
+/** Token-based compaction options */
+export type TokenCompactionOptions = {
+  mode: "tokens";
   contextWindowTokens?: number | undefined;
   systemPrompt?: string | undefined;
   reserveTokens?: number | undefined;
@@ -163,39 +73,28 @@ export type SummaryCompactionOptions = {
   // Required parameters
   model: Model<any>;
   apiKey: string;
-  // tokens mode parameters (reused)
+  // Token parameters (reused)
   contextWindowTokens?: number | undefined;
   systemPrompt?: string | undefined;
   reserveTokens?: number | undefined;
   targetRatio?: number | undefined;
   minKeepMessages?: number | undefined;
-  // summary-specific parameters
+  // Summary-specific parameters
   customInstructions?: string | undefined;
   previousSummary?: string | undefined;
   signal?: AbortSignal | undefined;
   maxChunkTokens?: number | undefined;
 };
 
-export type CompactionOptions = SyncCompactionOptions | SummaryCompactionOptions;
+export type CompactionOptions = TokenCompactionOptions | SummaryCompactionOptions;
 
 /**
- * Unified compaction entry point (synchronous version, for count/tokens modes)
- *
- * Selects compaction strategy based on mode
+ * Synchronous token-based compaction
  */
 export function compactMessages(
   messages: AgentMessage[],
-  options: SyncCompactionOptions,
+  options: TokenCompactionOptions,
 ): CompactionResult | null {
-  if (options.mode === "count") {
-    return compactMessagesByCount(
-      messages,
-      options.maxMessages ?? 80,
-      options.keepLast ?? 60,
-    );
-  }
-
-  // Token mode
   const contextWindowTokens = options.contextWindowTokens ?? 200_000;
   const estimation = estimateTokenUsage({
     messages,
@@ -204,7 +103,6 @@ export function compactMessages(
     reserveTokens: options.reserveTokens,
   });
 
-  // 检查是否需要压缩
   if (!shouldCompactTokens(estimation)) {
     return null;
   }
diff --git a/packages/core/src/agent/session/session-manager.ts b/packages/core/src/agent/session/session-manager.ts
index d11773c6..3b901fdd 100644
--- a/packages/core/src/agent/session/session-manager.ts
+++ b/packages/core/src/agent/session/session-manager.ts
@@ -33,12 +33,8 @@ export type SessionManagerOptions = {
   baseDir?: string | undefined;
 
   // Compaction mode configuration
-  /** Compaction mode: "count" uses message count, "tokens" uses token awareness, "summary" uses LLM summary */
-  compactionMode?: "count" | "tokens" | "summary" | undefined;
-
-  // Count mode parameters
-  maxMessages?: number | undefined;
-  keepLast?: number | undefined;
+  /** Compaction mode: "tokens" uses token awareness, "summary" uses LLM summary (default) */
+  compactionMode?: "tokens" | "summary" | undefined;
 
   // Token mode parameters
   /** Context window token count */
@@ -61,7 +57,7 @@ export type SessionManagerOptions = {
   customInstructions?: string | undefined;
 
   // Tool result pruning
-  /** Whether to enable tool result pruning before compaction (default: true in tokens/summary mode) */
+  /** Whether to enable tool result pruning before compaction (default: true) */
   enableToolResultPruning?: boolean | undefined;
   /** Tool result pruning settings */
   toolResultPruning?: Partial<ToolResultPruningSettings> | undefined;
@@ -74,10 +70,7 @@ export type SessionManagerOptions = {
 export class SessionManager {
   private readonly sessionId: string;
   private readonly baseDir: string | undefined;
-  private readonly compactionMode: "count" | "tokens" | "summary";
-  // Count mode
-  private readonly maxMessages: number;
-  private readonly keepLast: number;
+  private readonly compactionMode: "tokens" | "summary";
   // Token mode
   private readonly contextWindowTokens: number;
   private systemPrompt: string | undefined;
@@ -105,10 +98,6 @@ export class SessionManager {
     // Compaction mode (default: summary with LLM-based summarization)
     this.compactionMode = options.compactionMode ?? "summary";
 
-    // Count mode parameters
-    this.maxMessages = options.maxMessages ?? 80;
-    this.keepLast = options.keepLast ?? 60;
-
     // Token mode parameters
     this.contextWindowTokens = options.contextWindowTokens ?? 200_000;
     this.systemPrompt = options.systemPrompt;
@@ -121,10 +110,8 @@ export class SessionManager {
     this.apiKey = options.apiKey;
     this.customInstructions = options.customInstructions;
 
-    // Tool result pruning (enabled by default in tokens/summary mode)
-    this.enableToolResultPruning =
-      options.enableToolResultPruning ??
-      (this.compactionMode === "tokens" || this.compactionMode === "summary");
+    // Tool result pruning (enabled by default)
+    this.enableToolResultPruning = options.enableToolResultPruning ?? true;
     this.toolResultPruning = options.toolResultPruning;
 
     // Observability
@@ -164,7 +151,7 @@ export class SessionManager {
   /**
    * Get current compaction mode
    */
-  getCompactionMode(): "count" | "tokens" | "summary" {
+  getCompactionMode(): "tokens" | "summary" {
     return this.compactionMode;
   }
 
@@ -264,10 +251,6 @@ export class SessionManager {
 
   /** Check whether compaction would trigger for the given messages (without executing it) */
   needsCompaction(messages: AgentMessage[]): boolean {
-    if (this.compactionMode === "count") {
-      return messages.length > this.maxMessages;
-    }
-    // Token and summary modes use the same token-based threshold
     const estimation = estimateTokenUsage({
       messages,
       systemPrompt: this.systemPrompt,
@@ -376,12 +359,9 @@ export class SessionManager {
         }
       }
     } else {
+      // tokens mode
       result = compactMessages(workingMessages, {
-        mode: this.compactionMode,
-        // Count mode parameters
-        maxMessages: this.maxMessages,
-        keepLast: this.keepLast,
-        // Token mode parameters
+        mode: "tokens",
         contextWindowTokens: this.contextWindowTokens,
         systemPrompt: this.systemPrompt,
         reserveTokens: this.reserveTokens,
diff --git a/packages/core/src/agent/types.ts b/packages/core/src/agent/types.ts
index 3db04fca..7663bba6 100644
--- a/packages/core/src/agent/types.ts
+++ b/packages/core/src/agent/types.ts
@@ -50,11 +50,10 @@ export type AgentOptions = {
   reserveTokens?: number | undefined;
   /**
    * Compaction mode:
-   * - "count": uses legacy message count
    * - "tokens": uses token awareness
    * - "summary": uses LLM to generate summary (default)
    */
-  compactionMode?: "count" | "tokens" | "summary" | undefined;
+  compactionMode?: "tokens" | "summary" | undefined;
   /** Compaction target utilization ratio (0-1), defaults to 0.5 */
   compactionTargetRatio?: number | undefined;
   /** Minimum messages to keep, defaults to 10 */