fix(gateway): emit Anthropic-standard SSE error events and failover body

Two follow-ups to PR #2066's failover-wrap fix:

1. Failover ResponseBody (`UpstreamFailoverError.ResponseBody`) was encoded
   as `{"error": "<msg>"}` (string field). `ExtractUpstreamErrorMessage`
   probes for `error.message`, `detail`, or top-level `message` only — so
   `handleFailoverExhausted` and downstream passthrough rules saw an empty
   message, losing the EOF root cause in ops logs. Re-encode as the
   Anthropic standard shape `{"type":"error","error":{"type":"upstream_disconnected","message":"..."}}`.
   (Addresses the inline review comment from copilot-pull-request-reviewer
   on Wei-Shaw/sub2api#2066.)

2. The streaming `event: error` SSE frame for `response_too_large`,
   `stream_read_error`, and `stream_timeout` was non-standard
   (`{"error":"<reason>"}`). Anthropic SDKs (and Claude Code) expect
   `{"type":"error","error":{"type":"...","message":"..."}}` and parse
   `error.type`/`error.message` accordingly. Refactor `sendErrorEvent` to
   take both reason and message, and emit the standard frame so client
   SDKs surface a real diagnostic message instead of a generic stream error.

This does not by itself prevent task interruption on long-stream EOF
(SSE has no resume; client-side retry remains the only complete fix), but
it gives both server-side ops logs and client-side error UIs a meaningful
upstream message so users know the next step is to retry.

Tests updated to assert the new body shape on both branches plus a new
assertion that `ExtractUpstreamErrorMessage` returns a non-empty string.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

backend/internal/service/gateway_service.go

@@ -6871,14 +6871,31 @@ func (s *GatewayService) handleStreamingResponse(ctx context.Context, resp *http
 	}
 	lastDataAt := time.Now()
 
-	// 仅发送一次错误事件，避免多次写入导致协议混乱（写失败时尽力通知客户端）
+	// 仅发送一次错误事件，避免多次写入导致协议混乱（写失败时尽力通知客户端）。
+	// 事件格式遵循 Anthropic SSE 标准：{"type":"error","error":{"type":<reason>,"message":<message>}}
+	// 这样 Anthropic SDK / Claude Code 等客户端能按标准 error 类型解析，UI 能显示具体错误文案，
+	// 服务端 ExtractUpstreamErrorMessage 也能从透传的 body 中提取 message。
 	errorEventSent := false
-	sendErrorEvent := func(reason string) {
+	sendErrorEvent := func(reason, message string) {
 		if errorEventSent {
 			return
 		}
 		errorEventSent = true
-		_, _ = fmt.Fprintf(w, "event: error\ndata: {\"error\":\"%s\"}\n\n", reason)
+		if message == "" {
+			message = reason
+		}
+		body, err := json.Marshal(map[string]any{
+			"type": "error",
+			"error": map[string]string{
+				"type":    reason,
+				"message": message,
+			},
+		})
+		if err != nil {
+			// json.Marshal 不可能在已知 string-only 输入上失败，保守 fallback
+			body = []byte(fmt.Sprintf(`{"type":"error","error":{"type":%q,"message":%q}}`, reason, message))
+		}
+		_, _ = fmt.Fprintf(w, "event: error\ndata: %s\n\n", body)
 		flusher.Flush()
 	}
 
@@ -7038,16 +7055,21 @@ func (s *GatewayService) handleStreamingResponse(ctx context.Context, resp *http
 				// 客户端未断开，正常的错误处理
 				if errors.Is(ev.err, bufio.ErrTooLong) {
 					logger.LegacyPrintf("service.gateway", "SSE line too long: account=%d max_size=%d error=%v", account.ID, maxLineSize, ev.err)
-					sendErrorEvent("response_too_large")
+					sendErrorEvent("response_too_large", fmt.Sprintf("upstream SSE line exceeded %d bytes", maxLineSize))
 					return &streamingResult{usage: usage, firstTokenMs: firstTokenMs}, ev.err
 				}
 				// 上游中途读错误（unexpected EOF / connection reset 等，常见于 HTTP/2 GOAWAY）：
 				// 若尚未向客户端写过任何字节，包成 UpstreamFailoverError 让 handler 层走 failover/重试。
 				// 已经开始写流时 SSE 协议无 resume，只能透传错误事件给客户端。
+				disconnectMsg := fmt.Sprintf("upstream stream disconnected: %s", ev.err)
 				if !c.Writer.Written() {
 					logger.LegacyPrintf("service.gateway", "Upstream stream read error before any client output (account=%d), failing over: %v", account.ID, ev.err)
-					body, _ := json.Marshal(map[string]string{
+					body, _ := json.Marshal(map[string]any{
-						"error": fmt.Sprintf("upstream stream disconnected: %s", ev.err),
+						"type": "error",
+						"error": map[string]string{
+							"type":    "upstream_disconnected",
+							"message": disconnectMsg,
+						},
 					})
 					return nil, &UpstreamFailoverError{
 						StatusCode:             http.StatusBadGateway,
@@ -7055,7 +7077,7 @@ func (s *GatewayService) handleStreamingResponse(ctx context.Context, resp *http
 						RetryableOnSameAccount: true,
 					}
 				}
-				sendErrorEvent("stream_read_error")
+				sendErrorEvent("stream_read_error", disconnectMsg)
 				return &streamingResult{usage: usage, firstTokenMs: firstTokenMs}, fmt.Errorf("stream read error: %w", ev.err)
 			}
 			line := ev.line
@@ -7114,7 +7136,7 @@ func (s *GatewayService) handleStreamingResponse(ctx context.Context, resp *http
 			if s.rateLimitService != nil {
 				s.rateLimitService.HandleStreamTimeout(ctx, account, originalModel)
 			}
-			sendErrorEvent("stream_timeout")
+			sendErrorEvent("stream_timeout", fmt.Sprintf("upstream stream idle for %s", streamInterval))
 			return &streamingResult{usage: usage, firstTokenMs: firstTokenMs}, fmt.Errorf("stream data interval timeout")
 
 		case <-keepaliveCh:

backend/internal/service/gateway_streaming_test.go

@@ -8,7 +8,6 @@ import (
 	"io"
 	"net/http"
 	"net/http/httptest"
-	"strings"
 	"testing"
 	"time"
 
@@ -246,7 +245,15 @@ func TestHandleStreamingResponse_StreamReadErrorBeforeOutput_TriggersFailover(t
 	require.True(t, errors.As(err, &failoverErr), "未输出过字节时 stream read error 必须包成 UpstreamFailoverError，期望: %v", err)
 	require.Equal(t, http.StatusBadGateway, failoverErr.StatusCode)
 	require.True(t, failoverErr.RetryableOnSameAccount, "GOAWAY 类错误应允许同账号重试")
-	require.Contains(t, string(failoverErr.ResponseBody), "upstream stream disconnected")
+
+	// ResponseBody 必须是 Anthropic 标准 error 格式：
+	// 1) ExtractUpstreamErrorMessage 能正确从 error.message 提取消息（被 handleFailoverExhausted / ops 日志依赖）
+	// 2) error.type 标记为 upstream_disconnected
+	extractedMsg := ExtractUpstreamErrorMessage(failoverErr.ResponseBody)
+	require.NotEmpty(t, extractedMsg, "ExtractUpstreamErrorMessage 必须从 ResponseBody 取到非空 message，否则 ops 日志会丢失诊断信息")
+	require.Contains(t, extractedMsg, "upstream stream disconnected")
+	require.Contains(t, string(failoverErr.ResponseBody), `"type":"error"`)
+	require.Contains(t, string(failoverErr.ResponseBody), `"upstream_disconnected"`)
 
 	// 客户端应收不到任何 stream_read_error 事件，由 handler 层根据 failover 结果再决定
 	require.NotContains(t, rec.Body.String(), "stream_read_error")
@@ -282,9 +289,11 @@ func TestHandleStreamingResponse_StreamReadErrorAfterOutput_PassesThrough(t *tes
 	var failoverErr *UpstreamFailoverError
 	require.False(t, errors.As(err, &failoverErr), "已经向客户端写过字节时不能再 failover")
 
-	// 客户端必须收到 stream_read_error 事件
+	// 客户端必须收到 Anthropic 标准格式的 SSE error 事件，error.type=stream_read_error，
+	// error.message 含具体根因（让 SDK 能解析、UI 能显示具体错误）
 	body := rec.Body.String()
-	require.True(t,
+	require.Contains(t, body, "event: error\n", "必须按 Anthropic SSE 标准发送 error 事件帧")
-		strings.Contains(body, "stream_read_error"),
+	require.Contains(t, body, `"type":"error"`, "data 必须含 type:error 顶层字段（Anthropic 标准）")
-		"已开始流后必须发送 stream_read_error 事件给客户端，实际响应: %q", body)
+	require.Contains(t, body, `"stream_read_error"`, "error.type 必须为 stream_read_error")
+	require.Contains(t, body, "upstream stream disconnected", "error.message 必须包含具体根因，Claude Code 等客户端才能显示有效错误文案")
 }