diff --git a/lib/hooks/image-analyzer-transformer.cjs b/lib/hooks/image-analyzer-transformer.cjs
index 09362bb6..b041dd17 100755
--- a/lib/hooks/image-analyzer-transformer.cjs
+++ b/lib/hooks/image-analyzer-transformer.cjs
@@ -761,14 +761,16 @@ process.stdin.on('error', () => {
 
 /**
  * Main hook processing logic
+ *
+ * Two-phase design: Phase 1 filters non-image Read calls silently (exit 0).
+ * Phase 2 only runs for confirmed image/PDF files, so error messages are
+ * always relevant and never confuse users reading code or text files.
  */
 async function processHook() {
+  // Phase 1: Fast bail-out for non-image files
+  // Any failure here → pass through silently to native Read
+  let filePath;
   try {
-    // Skip for native accounts or explicit disable
-    if (shouldSkipHook()) {
-      process.exit(0);
-    }
-
     const data = JSON.parse(input);
 
     // Only handle Read tool
@@ -776,23 +778,35 @@ async function processHook() {
       process.exit(0);
     }
 
-    const filePath = data.tool_input?.file_path || '';
+    filePath = data.tool_input?.file_path || '';
 
     if (!filePath) {
       process.exit(0);
     }
 
+    // Check file extension BEFORE any other processing — this is the key gate
+    // that ensures non-image Read calls never see hook errors
+    if (!isAnalyzableFile(filePath)) {
+      process.exit(0);
+    }
+  } catch {
+    // stdin parse failure or unexpected error → pass through silently
+    process.exit(0);
+  }
+
+  // Phase 2: Image/PDF file processing — errors here are relevant to the user
+  try {
+    // Skip for native accounts or explicit disable
+    if (shouldSkipHook()) {
+      process.exit(0);
+    }
+
     // Check if file exists
     if (!fs.existsSync(filePath)) {
       // Let native Read handle the error
       process.exit(0);
     }
 
-    // Check if file is analyzable
-    if (!isAnalyzableFile(filePath)) {
-      process.exit(0);
-    }
-
     // Check file size
     const stats = fs.statSync(filePath);
     if (stats.size >= MAX_FILE_SIZE_BYTES) {
@@ -843,14 +857,7 @@ async function processHook() {
       console.error('[CCS Hook] Error:', err.message);
     }
 
-    // Try to extract file path from parsed input
-    let filePath = 'unknown file';
-    try {
-      const data = JSON.parse(input);
-      filePath = data.tool_input?.file_path || 'unknown file';
-    } catch {
-      // Ignore parse errors
-    }
+    // filePath is guaranteed set by Phase 1 — only image files reach here
 
     // Categorize error by message pattern
     const errMsg = err.message || '';
diff --git a/src/cliproxy/executor/index.ts b/src/cliproxy/executor/index.ts
index 986868fb..807ca8b1 100644
--- a/src/cliproxy/executor/index.ts
+++ b/src/cliproxy/executor/index.ts
@@ -50,6 +50,7 @@ import {
   displayWebSearchStatus,
 } from '../../utils/websearch-manager';
 import { loadOrCreateUnifiedConfig } from '../../config/unified-config-loader';
+import { installImageAnalyzerHook } from '../../utils/hooks';
 import { HttpsTunnelProxy } from '../https-tunnel-proxy';
 
 // Import modular components
@@ -162,6 +163,9 @@ export async function execClaudeWithCLIProxy(
   installWebSearchHook();
   displayWebSearchStatus();
 
+  // Sync image analyzer hook from npm package to ~/.ccs/hooks/
+  installImageAnalyzerHook();
+
   const providerConfig = getProviderConfig(provider);
   log(`Provider: ${providerConfig.displayName}`);
 
diff --git a/tests/e2e/image-analyzer-hook.e2e.test.ts b/tests/e2e/image-analyzer-hook.e2e.test.ts
index de00aa62..8df3adfa 100644
--- a/tests/e2e/image-analyzer-hook.e2e.test.ts
+++ b/tests/e2e/image-analyzer-hook.e2e.test.ts
@@ -425,8 +425,8 @@ describe('Image Analyzer Hook', () => {
         },
       });
 
-      // Should exit with error (code 2)
-      expect(hookProcess.status).toBe(2);
+      // Should pass through silently (exit 0) — can't determine file type from malformed input
+      expect(hookProcess.status).toBe(0);
     });
   });