fix(gateway): retry Weixin prepare failures without token

gateway/platforms/weixin.py

@@ -66,8 +66,11 @@ _FENCE_RE = re.compile(r"^```([^\n`]*)\s*$")
 
 
 def _is_stale_session_ret(ret: "Optional[int]", errcode: "Optional[int]", errmsg: "Optional[str]") -> bool:
-    """ret/errcode=-2 with 'unknown error' is a stale-session signal (like -14), not a real rate limit."""
-    return (ret == RATE_LIMIT_ERRCODE or errcode == RATE_LIMIT_ERRCODE) and (errmsg or "").lower() == "unknown error"
+    """Recognize stale-session variants of iLink's ``-2`` response, not real rate limits."""
+    return (ret == RATE_LIMIT_ERRCODE or errcode == RATE_LIMIT_ERRCODE) and (errmsg or "").lower() in {
+        "unknown error",
+        "prepare failed",
+    }
 
 
 def _is_session_expired(resp: Dict[str, Any], ret: Any, errcode: Any) -> bool:

tests/gateway/test_weixin.py

@@ -322,6 +322,21 @@ class TestWeixinChunkDelivery:
         assert send_message_mock.await_count == 2
         assert sleep_mock.await_count == 1
 
+    @pytest.mark.parametrize("error_field", ["ret", "errcode"])
+    @patch("gateway.platforms.weixin._send_message", new_callable=AsyncMock)
+    def test_prepare_failed_retries_without_context_token(self, send_message_mock, error_field):
+        adapter = self._connected_adapter()
+        adapter._rate_limit_circuit_threshold = 1
+        adapter._token_store._cache[adapter._token_store._key(adapter._account_id, "wxid_test123")] = "ctx-token"
+        prepare_failed = {error_field: weixin.RATE_LIMIT_ERRCODE, "errmsg": "prepare failed"}
+        send_message_mock.side_effect = [prepare_failed, {"ret": 0}]
+
+        result = asyncio.run(adapter.send("wxid_test123", "hello"))
+
+        assert result.success is True
+        assert [call.kwargs["context_token"] for call in send_message_mock.await_args_list] == ["ctx-token", None]
+        assert adapter._rate_limit_circuit_until == 0.0
+
 
 class TestWeixinOutboundMedia:
 
@@ -879,4 +894,3 @@ class TestWeixinVoiceGatewayHandoff:
             "VOICE event body leaked Tencent's STT text — runner would trust "
             "the wrong transcript instead of re-transcribing (#27300)."
         )
-