#971 微信Bot P2 - 接收侧媒体下载解密

实现CDN媒体下载解密(parse_aes_key支持base64(raw16B)和base64(hex32)两种编码)+extract_inbound_media(图片/语音/文件/视频)。接收循环检测媒体消息下载解密保存到/tmp。图片已保存供后续vision模型接入。

backend/python/app/api/im_bot_user.py

@@ -425,6 +425,42 @@ def _wechat_extract_text(msg: dict) -> str:
     return "\n".join(parts)
 
 
+def _extract_and_save_inbound_media(msg: dict) -> str:
+    """检测入站媒体消息(图片/语音/文件/视频)，下载解密保存，返回给 LLM 的提示文本。
+
+    无媒体或解密失败返回空字符串。图片已保存到本地，供后续 vision 模型接入。
+    """
+    try:
+        from app.api.im_ilink_media import extract_inbound_media
+        import os as _os
+        for it in (msg.get("item_list") or []):
+            if it.get("type") in (2, 3, 4, 5):
+                kind, data = extract_inbound_media(it)
+                if kind and data:
+                    _dir = "/tmp/wechat_inbound_media"
+                    _os.makedirs(_dir, exist_ok=True)
+                    _ext = {"image": "jpg", "voice": "silk", "file": "bin", "video": "mp4"}.get(kind, "bin")
+                    _path = _os.path.join(_dir, "%d_%s.%s" % (int(time.time() * 1000), kind, _ext))
+                    with open(_path, "wb") as _f:
+                        _f.write(data)
+                    _label = {"image": "图片", "voice": "语音", "file": "文件", "video": "视频"}.get(kind, kind)
+                    try:
+                        with open("/tmp/wechat_inbound_media.log", "a", encoding="utf-8") as _f:
+                            _f.write(_beijing_now().isoformat() + " %s saved=%s size=%d\n"
+                                     % (kind, _path, len(data)))
+                    except Exception:
+                        pass
+                    return "[用户发送了%s，已保存到 %s（%d 字节）]" % (_label, _path, len(data))
+        return ""
+    except Exception as _e:
+        try:
+            with open("/tmp/wechat_inbound_media.log", "a", encoding="utf-8") as _f:
+                _f.write(_beijing_now().isoformat() + " MEDIA_EXTRACT_ERR %s\n" % (_e,))
+        except Exception:
+            pass
+        return ""
+
+
 def _wechat_sender_id(m: dict) -> str:
     """从入站消息中提取发送者 id（兼容多套字段名）。"""
     for k in ("from_user_id", "from_user_name", "from_username", "user_id",
@@ -576,7 +612,10 @@ def _wechat_receiver_loop(account_id: int):
                     continue  # 仅跳过 bot 自己发出的回声，其余（含后续追问）均尝试处理
                 content = _wechat_extract_text(m)
                 if not content:
-                    continue
+                    # 入站媒体（图片/语音/文件/视频）：下载解密保存，供 LLM 感知
+                    content = _extract_and_save_inbound_media(m)
+                    if not content:
+                        continue
                 sender = _wechat_sender_id(m)
                 ctx = _wechat_context_token(m)
                 # 诊断：记录原始入站消息，定位字段名 / 第二轮不回复