diff --git a/src/vision.ts b/src/vision.ts
index 7a088ec..1266232 100644
--- a/src/vision.ts
+++ b/src/vision.ts
@@ -7,47 +7,55 @@ export async function applyVisionInterceptor(messages: AnthropicMessage[]): Prom
     const config = getConfig();
     if (!config.vision?.enabled) return;
 
-    for (const msg of messages) {
-        if (msg.role !== 'user') continue;
-        if (!Array.isArray(msg.content)) continue;
-
-        let hasImages = false;
-        const newContent: AnthropicContentBlock[] = [];
-        const imagesToAnalyze: AnthropicContentBlock[] = [];
-
-        for (const block of msg.content) {
-            if (block.type === 'image') {
-                hasImages = true;
-                imagesToAnalyze.push(block);
-            } else {
-                newContent.push(block);
-            }
+    // ★ 仅处理最后一条 user 消息中的图片
+    //   历史消息的图片已在前几轮被转换为文本描述，无需重复处理
+    //   这避免了多轮对话中重复消耗 Vision API 配额和增加延迟
+    let lastUserMsg: AnthropicMessage | null = null;
+    for (let i = messages.length - 1; i >= 0; i--) {
+        if (messages[i].role === 'user') {
+            lastUserMsg = messages[i];
+            break;
         }
+    }
 
-        if (hasImages && imagesToAnalyze.length > 0) {
-            try {
-                let descriptions = '';
-                if (config.vision.mode === 'ocr') {
-                    descriptions = await processWithLocalOCR(imagesToAnalyze);
-                } else {
-                    descriptions = await callVisionAPI(imagesToAnalyze);
-                }
+    if (!lastUserMsg || !Array.isArray(lastUserMsg.content)) return;
 
-                // Add descriptions as a simulated system text block
-                newContent.push({
-                    type: 'text',
-                    text: `\n\n[System: The user attached ${imagesToAnalyze.length} image(s). Visual analysis/OCR extracted the following context:\n${descriptions}]\n\n`
-                });
+    let hasImages = false;
+    const newContent: AnthropicContentBlock[] = [];
+    const imagesToAnalyze: AnthropicContentBlock[] = [];
 
-                msg.content = newContent;
-            } catch (e) {
-                console.error("[Vision API Error]", e);
-                newContent.push({
-                    type: 'text',
-                    text: `\n\n[System: The user attached image(s), but the Vision interceptor failed to process them. Error: ${(e as Error).message}]\n\n`
-                });
-                msg.content = newContent;
+    for (const block of lastUserMsg.content) {
+        if (block.type === 'image') {
+            hasImages = true;
+            imagesToAnalyze.push(block);
+        } else {
+            newContent.push(block);
+        }
+    }
+
+    if (hasImages && imagesToAnalyze.length > 0) {
+        try {
+            let descriptions = '';
+            if (config.vision.mode === 'ocr') {
+                descriptions = await processWithLocalOCR(imagesToAnalyze);
+            } else {
+                descriptions = await callVisionAPI(imagesToAnalyze);
             }
+
+            // Add descriptions as a simulated system text block
+            newContent.push({
+                type: 'text',
+                text: `\n\n[System: The user attached ${imagesToAnalyze.length} image(s). Visual analysis/OCR extracted the following context:\n${descriptions}]\n\n`
+            });
+
+            lastUserMsg.content = newContent;
+        } catch (e) {
+            console.error("[Vision API Error]", e);
+            newContent.push({
+                type: 'text',
+                text: `\n\n[System: The user attached image(s), but the Vision interceptor failed to process them. Error: ${(e as Error).message}]\n\n`
+            });
+            lastUserMsg.content = newContent;
         }
     }
 }