#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

backend/python/app/api/im_ilink_media.py

@@ -19,6 +19,7 @@ import base64
 import hashlib
 import math
 import os
+import re
 from urllib.parse import quote
 
 import requests
@@ -212,3 +213,118 @@ def build_video_item(uploaded: dict) -> dict:
             "video_size": uploaded["filesize"],
         },
     }
+
+
+# ============================================================
+# 接收侧媒体下载解密（P2）
+# 参考 openclaw cdn/pic-decrypt.ts + media/media-download.ts
+# ============================================================
+
+def parse_aes_key(aes_key_base64: str) -> bytes:
+    """解析 CDNMedia.aes_key 为原始 16 字节 AES key。
+
+    两种真实编码（openclaw parseAesKey）：
+      - base64(raw 16 bytes)          → 图片 (media.aes_key)
+      - base64(hex 32 chars 字符串)    → 文件/语音/视频
+    第二种需 base64 解码后得到 32 个 ASCII hex 字符，再 hex 解码成 16 字节。
+    """
+    decoded = base64.b64decode(aes_key_base64)
+    if len(decoded) == 16:
+        return decoded
+    if len(decoded) == 32 and re.match(r"^[0-9a-fA-F]{32}$", decoded.decode("ascii", "ignore")):
+        return bytes.fromhex(decoded.decode("ascii"))
+    raise ValueError(
+        f"aes_key 解码后应为 16 字节或 32 字符 hex，实际 {len(decoded)} 字节"
+    )
+
+
+def download_and_decrypt_media(
+    encrypted_query_param: str,
+    aes_key_base64: str,
+    cdn_base_url: str = WECHAT_CDN_BASE,
+    full_url: str = None,
+) -> bytes:
+    """下载并 AES-128-ECB 解密 CDN 媒体文件，返回明文字节。"""
+    key = parse_aes_key(aes_key_base64)
+    if full_url:
+        url = full_url
+    else:
+        url = (
+            f"{cdn_base_url}/download"
+            f"?encrypted_query_param={quote(encrypted_query_param, safe='')}"
+        )
+    resp = requests.get(url, timeout=60)
+    if not resp.ok:
+        raise RuntimeError(f"CDN download failed HTTP {resp.status_code}: {resp.text[:150]}")
+    encrypted = resp.content
+    return decrypt_aes_ecb(encrypted, key)
+
+
+def extract_inbound_media(item: dict) -> tuple:
+    """从入站 MessageItem 提取媒体并下载解密。
+
+    返回 (media_kind, media_bytes)，不支持的类型或失败返回 (None, None)。
+    media_kind ∈ {"image", "voice", "file", "video"}
+    """
+    try:
+        t = item.get("type")
+        if t == 2:  # 图片
+            img = item.get("image_item") or {}
+            media = img.get("media") or {}
+            eqp = media.get("encrypt_query_param") or ""
+            full = media.get("full_url")
+            if not eqp and not full:
+                return None, None
+            # 图片 key 优先用 image_item.aeskey（hex string）转 base64
+            if img.get("aeskey"):
+                aes_key_b64 = base64.b64encode(bytes.fromhex(img["aeskey"])).decode()
+            else:
+                aes_key_b64 = media.get("aes_key") or ""
+            if aes_key_b64:
+                return "image", download_and_decrypt_media(eqp, aes_key_b64, full_url=full)
+            return "image", _download_plain(eqp, full)
+        elif t == 3:  # 语音
+            voice = item.get("voice_item") or {}
+            media = voice.get("media") or {}
+            eqp = media.get("encrypt_query_param") or ""
+            full = media.get("full_url")
+            aes_key_b64 = media.get("aes_key") or ""
+            if not aes_key_b64:
+                return None, None
+            return "voice", download_and_decrypt_media(eqp, aes_key_b64, full_url=full)
+        elif t == 4:  # 文件
+            fitem = item.get("file_item") or {}
+            media = fitem.get("media") or {}
+            eqp = media.get("encrypt_query_param") or ""
+            full = media.get("full_url")
+            aes_key_b64 = media.get("aes_key") or ""
+            if not aes_key_b64:
+                return None, None
+            return "file", download_and_decrypt_media(eqp, aes_key_b64, full_url=full)
+        elif t == 5:  # 视频
+            vitem = item.get("video_item") or {}
+            media = vitem.get("media") or {}
+            eqp = media.get("encrypt_query_param") or ""
+            full = media.get("full_url")
+            aes_key_b64 = media.get("aes_key") or ""
+            if not aes_key_b64:
+                return None, None
+            return "video", download_and_decrypt_media(eqp, aes_key_b64, full_url=full)
+    except Exception:
+        return None, None
+    return None, None
+
+
+def _download_plain(encrypted_query_param: str, full_url: str = None) -> bytes:
+    """下载未加密的 CDN 媒体（无 aes_key 时）。"""
+    if full_url:
+        url = full_url
+    else:
+        url = (
+            f"{WECHAT_CDN_BASE}/download"
+            f"?encrypted_query_param={quote(encrypted_query_param, safe='')}"
+        )
+    resp = requests.get(url, timeout=60)
+    if not resp.ok:
+        raise RuntimeError(f"CDN plain download failed HTTP {resp.status_code}")
+    return resp.content