diff --git a/apps/claude-sdk-cli/src/AgentMessageHandler.ts b/apps/claude-sdk-cli/src/AgentMessageHandler.ts
new file mode 100644
index 0000000..afddfd1
--- /dev/null
+++ b/apps/claude-sdk-cli/src/AgentMessageHandler.ts
@@ -0,0 +1,62 @@
+import type { SdkMessage } from '@shellicar/claude-sdk';
+import type { AppLayout } from './AppLayout.js';
+import type { logger } from './logger.js';
+
+/**
+ * Handles the stateless SdkMessage cases: routes each message to the
+ * appropriate layout call. No accumulated state here.
+ *
+ * Stateful cases (tool_approval_request, tool_error, message_usage) stay
+ * in runAgent.ts until step 4b, when usageBeforeTools tracking and the
+ * async tool approval flow move here too.
+ *
+ * NOTE: message_compaction currently omits the "compacted at X/Y (Z%)"
+ * context-usage annotation. That annotation reads lastUsage, which is
+ * set by message_usage — a 4b case. The annotation is restored in 4b.
+ */
+export class AgentMessageHandler {
+  #layout: AppLayout;
+  #logger: typeof logger;
+
+  public constructor(layout: AppLayout, log: typeof logger) {
+    this.#layout = layout;
+    this.#logger = log;
+  }
+
+  public handle(msg: SdkMessage): void {
+    switch (msg.type) {
+      case 'query_summary': {
+        const parts = [`${msg.systemPrompts} system`, `${msg.userMessages} user`, `${msg.assistantMessages} assistant`, ...(msg.thinkingBlocks > 0 ? [`${msg.thinkingBlocks} thinking`] : [])];
+        this.#layout.transitionBlock('meta');
+        this.#layout.appendStreaming(parts.join(' · '));
+        break;
+      }
+      case 'message_thinking':
+        this.#layout.transitionBlock('thinking');
+        this.#layout.appendStreaming(msg.text);
+        break;
+      case 'message_text':
+        this.#layout.transitionBlock('response');
+        this.#layout.appendStreaming(msg.text);
+        break;
+      case 'message_compaction_start':
+        this.#layout.transitionBlock('compaction');
+        break;
+      case 'message_compaction':
+        this.#layout.transitionBlock('compaction');
+        this.#layout.appendStreaming(msg.summary);
+        break;
+      case 'done':
+        this.#logger.info('done', { stopReason: msg.stopReason });
+        if (msg.stopReason !== 'end_turn') {
+          this.#layout.appendStreaming(`\n\n[stop: ${msg.stopReason}]`);
+        }
+        break;
+      case 'error':
+        this.#layout.transitionBlock('response');
+        this.#layout.appendStreaming(`\n\n[error: ${msg.message}]`);
+        this.#logger.error('error', { message: msg.message });
+        break;
+    }
+  }
+}
diff --git a/apps/claude-sdk-cli/src/runAgent.ts b/apps/claude-sdk-cli/src/runAgent.ts
index 589db5e..78723a2 100644
--- a/apps/claude-sdk-cli/src/runAgent.ts
+++ b/apps/claude-sdk-cli/src/runAgent.ts
@@ -16,6 +16,7 @@ import { createRef } from '@shellicar/claude-sdk-tools/Ref';
 import type { RefStore } from '@shellicar/claude-sdk-tools/RefStore';
 import { SearchFiles } from '@shellicar/claude-sdk-tools/SearchFiles';
 import { Tail } from '@shellicar/claude-sdk-tools/Tail';
+import { AgentMessageHandler } from './AgentMessageHandler.js';
 import type { AppLayout, PendingTool } from './AppLayout.js';
 import { logger } from './logger.js';
 import { getPermission, PermissionAction } from './permissions.js';
@@ -105,6 +106,8 @@ export async function runAgent(agent: IAnthropicAgent, prompt: string, layout: A
     return result;
   };
 
+  const handler = new AgentMessageHandler(layout, logger);
+
   layout.startStreaming(prompt);
 
   const model = 'claude-sonnet-4-6';
@@ -192,19 +195,6 @@ export async function runAgent(agent: IAnthropicAgent, prompt: string, layout: A
         layout.transitionBlock('tools');
         layout.appendStreaming(`${msg.name} error\n\`\`\`json\n${JSON.stringify(msg.input, null, 2)}\n\`\`\`\n\n${msg.error}\n`);
         break;
-      case 'message_compaction_start':
-        layout.transitionBlock('compaction');
-        break;
-      case 'message_compaction':
-        layout.transitionBlock('compaction');
-        layout.appendStreaming(msg.summary);
-        if (lastUsage) {
-          const used = lastUsage.inputTokens + lastUsage.cacheCreationTokens + lastUsage.cacheReadTokens;
-          const pct = ((used / lastUsage.contextWindow) * 100).toFixed(1);
-          const fmt = (n: number) => (n >= 1000 ? `${(n / 1000).toFixed(1)}k` : String(n));
-          layout.appendStreaming(`\n\n[compacted at ${fmt(used)} / ${fmt(lastUsage.contextWindow)} (${pct}%)]`);
-        }
-        break;
       case 'message_usage': {
         // Annotate the (now-sealed) tools block with how many tokens this batch added to the
         // context window: delta = (input+cacheCreate+cacheRead at N+1) - (same at N).
@@ -237,17 +227,8 @@ export async function runAgent(agent: IAnthropicAgent, prompt: string, layout: A
         layout.updateUsage(msg);
         break;
       }
-      case 'done':
-        logger.info('done', { stopReason: msg.stopReason });
-        if (msg.stopReason !== 'end_turn') {
-          layout.appendStreaming(`\n\n[stop: ${msg.stopReason}]`);
-        }
-        break;
-      case 'error':
-        layout.transitionBlock('response');
-        layout.appendStreaming(`\n\n[error: ${msg.message}]`);
-        logger.error('error', { message: msg.message });
-        break;
+      default:
+        handler.handle(msg);
     }
   });
 
diff --git a/apps/claude-sdk-cli/test/AgentMessageHandler.spec.ts b/apps/claude-sdk-cli/test/AgentMessageHandler.spec.ts
new file mode 100644
index 0000000..13860e1
--- /dev/null
+++ b/apps/claude-sdk-cli/test/AgentMessageHandler.spec.ts
@@ -0,0 +1,189 @@
+import { describe, expect, it, vi } from 'vitest';
+import { AgentMessageHandler } from '../src/AgentMessageHandler.js';
+import type { AppLayout } from '../src/AppLayout.js';
+import { logger } from '../src/logger.js';
+
+// ---------------------------------------------------------------------------
+// Helpers
+// ---------------------------------------------------------------------------
+
+function makeLayout() {
+  return {
+    transitionBlock: vi.fn(),
+    appendStreaming: vi.fn(),
+  } as unknown as AppLayout;
+}
+
+function makeHandler(layout: AppLayout) {
+  return new AgentMessageHandler(layout, logger);
+}
+
+// ---------------------------------------------------------------------------
+// query_summary
+// ---------------------------------------------------------------------------
+
+describe('AgentMessageHandler — query_summary', () => {
+  it('transitions to meta block', () => {
+    const layout = makeLayout();
+    makeHandler(layout).handle({ type: 'query_summary', systemPrompts: 1, userMessages: 2, assistantMessages: 1, thinkingBlocks: 0 });
+    const expected = 'meta';
+    const actual = vi.mocked(layout.transitionBlock).mock.calls[0]?.[0];
+    expect(actual).toBe(expected);
+  });
+
+  it('streams the parts joined by ·', () => {
+    const layout = makeLayout();
+    makeHandler(layout).handle({ type: 'query_summary', systemPrompts: 1, userMessages: 2, assistantMessages: 1, thinkingBlocks: 0 });
+    const expected = '1 system · 2 user · 1 assistant';
+    const actual = vi.mocked(layout.appendStreaming).mock.calls[0]?.[0];
+    expect(actual).toBe(expected);
+  });
+
+  it('includes thinking block count when non-zero', () => {
+    const layout = makeLayout();
+    makeHandler(layout).handle({ type: 'query_summary', systemPrompts: 1, userMessages: 1, assistantMessages: 1, thinkingBlocks: 3 });
+    const expected = true;
+    const actual = (vi.mocked(layout.appendStreaming).mock.calls[0]?.[0] ?? '').includes('3 thinking');
+    expect(actual).toBe(expected);
+  });
+
+  it('omits thinking count when zero', () => {
+    const layout = makeLayout();
+    makeHandler(layout).handle({ type: 'query_summary', systemPrompts: 1, userMessages: 1, assistantMessages: 1, thinkingBlocks: 0 });
+    const expected = false;
+    const actual = (vi.mocked(layout.appendStreaming).mock.calls[0]?.[0] ?? '').includes('thinking');
+    expect(actual).toBe(expected);
+  });
+});
+
+// ---------------------------------------------------------------------------
+// message_thinking
+// ---------------------------------------------------------------------------
+
+describe('AgentMessageHandler — message_thinking', () => {
+  it('transitions to thinking block', () => {
+    const layout = makeLayout();
+    makeHandler(layout).handle({ type: 'message_thinking', text: 'hmm' });
+    const expected = 'thinking';
+    const actual = vi.mocked(layout.transitionBlock).mock.calls[0]?.[0];
+    expect(actual).toBe(expected);
+  });
+
+  it('streams the thinking text', () => {
+    const layout = makeLayout();
+    makeHandler(layout).handle({ type: 'message_thinking', text: 'hmm' });
+    const expected = 'hmm';
+    const actual = vi.mocked(layout.appendStreaming).mock.calls[0]?.[0];
+    expect(actual).toBe(expected);
+  });
+});
+
+// ---------------------------------------------------------------------------
+// message_text
+// ---------------------------------------------------------------------------
+
+describe('AgentMessageHandler — message_text', () => {
+  it('transitions to response block', () => {
+    const layout = makeLayout();
+    makeHandler(layout).handle({ type: 'message_text', text: 'hello' });
+    const expected = 'response';
+    const actual = vi.mocked(layout.transitionBlock).mock.calls[0]?.[0];
+    expect(actual).toBe(expected);
+  });
+
+  it('streams the text', () => {
+    const layout = makeLayout();
+    makeHandler(layout).handle({ type: 'message_text', text: 'hello' });
+    const expected = 'hello';
+    const actual = vi.mocked(layout.appendStreaming).mock.calls[0]?.[0];
+    expect(actual).toBe(expected);
+  });
+});
+
+// ---------------------------------------------------------------------------
+// message_compaction_start
+// ---------------------------------------------------------------------------
+
+describe('AgentMessageHandler — message_compaction_start', () => {
+  it('transitions to compaction block', () => {
+    const layout = makeLayout();
+    makeHandler(layout).handle({ type: 'message_compaction_start' });
+    const expected = 'compaction';
+    const actual = vi.mocked(layout.transitionBlock).mock.calls[0]?.[0];
+    expect(actual).toBe(expected);
+  });
+
+  it('does not stream any text', () => {
+    const layout = makeLayout();
+    makeHandler(layout).handle({ type: 'message_compaction_start' });
+    const expected = 0;
+    const actual = vi.mocked(layout.appendStreaming).mock.calls.length;
+    expect(actual).toBe(expected);
+  });
+});
+
+// ---------------------------------------------------------------------------
+// message_compaction
+// ---------------------------------------------------------------------------
+
+describe('AgentMessageHandler — message_compaction', () => {
+  it('transitions to compaction block', () => {
+    const layout = makeLayout();
+    makeHandler(layout).handle({ type: 'message_compaction', summary: 'context trimmed' });
+    const expected = 'compaction';
+    const actual = vi.mocked(layout.transitionBlock).mock.calls[0]?.[0];
+    expect(actual).toBe(expected);
+  });
+
+  it('streams the summary', () => {
+    const layout = makeLayout();
+    makeHandler(layout).handle({ type: 'message_compaction', summary: 'context trimmed' });
+    const expected = 'context trimmed';
+    const actual = vi.mocked(layout.appendStreaming).mock.calls[0]?.[0];
+    expect(actual).toBe(expected);
+  });
+});
+
+// ---------------------------------------------------------------------------
+// done
+// ---------------------------------------------------------------------------
+
+describe('AgentMessageHandler — done', () => {
+  it('does not stream anything on end_turn', () => {
+    const layout = makeLayout();
+    makeHandler(layout).handle({ type: 'done', stopReason: 'end_turn' });
+    const expected = 0;
+    const actual = vi.mocked(layout.appendStreaming).mock.calls.length;
+    expect(actual).toBe(expected);
+  });
+
+  it('streams a stop annotation for non-end_turn reasons', () => {
+    const layout = makeLayout();
+    makeHandler(layout).handle({ type: 'done', stopReason: 'max_tokens' });
+    const expected = true;
+    const actual = (vi.mocked(layout.appendStreaming).mock.calls[0]?.[0] ?? '').includes('[stop: max_tokens]');
+    expect(actual).toBe(expected);
+  });
+});
+
+// ---------------------------------------------------------------------------
+// error
+// ---------------------------------------------------------------------------
+
+describe('AgentMessageHandler — error', () => {
+  it('transitions to response block', () => {
+    const layout = makeLayout();
+    makeHandler(layout).handle({ type: 'error', message: 'oops' });
+    const expected = 'response';
+    const actual = vi.mocked(layout.transitionBlock).mock.calls[0]?.[0];
+    expect(actual).toBe(expected);
+  });
+
+  it('streams an error annotation', () => {
+    const layout = makeLayout();
+    makeHandler(layout).handle({ type: 'error', message: 'oops' });
+    const expected = true;
+    const actual = (vi.mocked(layout.appendStreaming).mock.calls[0]?.[0] ?? '').includes('[error: oops]');
+    expect(actual).toBe(expected);
+  });
+});