@make-u-free/migi 0.5.10 → 0.5.12

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@make-u-free/migi",
-  "version": "0.5.10",
+  "version": "0.5.12",
   "description": "Your AI right-hand agent. Works anywhere, with any LLM API.",
   "type": "module",
   "bin": {

package/src/tools.js

@@ -8,7 +8,8 @@ import chalk from 'chalk'
 import xlsxPkg from 'xlsx'
 import { createRequire } from 'module'
 const require = createRequire(import.meta.url)
-const pdfParse = require('pdf-parse')
+const _pdfParseModule = require('pdf-parse')
+const pdfParse = typeof _pdfParseModule === 'function' ? _pdfParseModule : _pdfParseModule.default
 import AdmZip from 'adm-zip'
 import OpenAI from 'openai'
 import { httpsAgent } from './tls.js'
@@ -134,14 +135,22 @@ function extractImagesFromPdf(buf) {
   const images = []
   let i = 0
 
-  while (i < buf.length - 1) {
-    // JPEG: FF D8 で始まり FF D9 で終わる
-    if (buf[i] === 0xFF && buf[i + 1] === 0xD8) {
-      const eoiIdx = buf.indexOf(Buffer.from([0xFF, 0xD9]), i + 2)
-      if (eoiIdx === -1) break
-      images.push({ data: buf.slice(i, eoiIdx + 2), mime: 'image/jpeg' })
-      i = eoiIdx + 2
-      continue
+  while (i < buf.length - 3) {
+    // JPEG: FF D8 で始まり、直後に FF E0〜EF（APPマーカー）か FF DB（DQT）が続く本物だけ拾う
+    if (buf[i] === 0xFF && buf[i + 1] === 0xD8 && buf[i + 2] === 0xFF) {
+      const nextMarker = buf[i + 3]
+      const isApp = nextMarker >= 0xE0 && nextMarker <= 0xEF  // JFIF / EXIF 等
+      const isDqt = nextMarker === 0xDB                        // 量子化テーブル
+      if (isApp || isDqt) {
+        const eoiIdx = buf.indexOf(Buffer.from([0xFF, 0xD9]), i + 4)
+        if (eoiIdx === -1) break
+        const data = buf.slice(i, eoiIdx + 2)
+        if (data.length > 1024) {  // 1KB未満はアイコン等のゴミなので除外
+          images.push({ data, mime: 'image/jpeg' })
+        }
+        i = eoiIdx + 2
+        continue
+      }
     }
 
     // PNG: 89 50 4E 47 0D 0A 1A 0A で始まる
@@ -152,7 +161,10 @@ function extractImagesFromPdf(buf) {
     ) {
       const iend = buf.indexOf(Buffer.from([0x49, 0x45, 0x4E, 0x44, 0xAE, 0x42, 0x60, 0x82]), i + 8)
       if (iend === -1) break
-      images.push({ data: buf.slice(i, iend + 8), mime: 'image/png' })
+      const data = buf.slice(i, iend + 8)
+      if (data.length > 1024) {
+        images.push({ data, mime: 'image/png' })
+      }
       i = iend + 8
       continue
     }
@@ -222,11 +234,13 @@ export async function executeTool(name, args, opts = {}) {
         const buf = readFileSync(args.path)
 
         // Step 1: テキストPDFとして抽出を試みる
-        try {
-          const data = await pdfParse(buf)
-          const text = data.text?.trim()
-          if (text) return text
-        } catch (_) {}
+        if (typeof pdfParse === 'function') {
+          try {
+            const data = await pdfParse(buf)
+            const text = data.text?.trim()
+            if (text) return text
+          } catch (_) {}
+        }
 
         // Step 2: 画像PDFとしてVision APIでOCR（ネイティブ依存なし）
         if (!opts.apiKey) return '(テキストが抽出できませんでした)'