fix(gateway): 修复模拟响应使用 Bedrock msg_bdrk_ 格式，改为正宗 Anthropic msg_01 格式

问题：
探针拦截（suggestion mode / warmup / max_tokens=1 haiku）的模拟响应以及
Gemini/Antigravity 兼容层生成的 message ID 不符合 Anthropic 官方 API 格式，
容易被客户端识别为非正宗响应。

修复：
1. generateRealisticMsgID()：msg_bdrk_ + 24字符 → msg_01 + 22位 Base62
   （与官方 API 返回的 msg_011CdS6b8gAhoKWdW9jE87Zs 格式一致）
2. 去掉固定的 msg_mock_suggestion / msg_mock_warmup，统一使用随机 ID
3. 流式响应格式对齐官方：
   - message_start 增加 stop_details/cache token 字段
   - content_block_start 字段顺序修正
   - message_delta.usage 只含 output_tokens
4. 非流式响应：增加 stop_details:null，移除非标准 total_tokens
5. Gemini Messages/ChatCompletions 兼容层：msg_ + hex → msg_01 + Base62
6. Antigravity response/stream transformer：msg_ + 12位 → msg_01 + 22位 Base62

验证方式：对照 Anthropic 官方 API 实际响应格式确认。

backend/internal/handler/gateway_handler.go

@@ -2126,22 +2126,22 @@ func sendMockInterceptStream(c *gin.Context, model string, interceptType Interce
 
 	switch interceptType {
 	case InterceptTypeSuggestionMode:
-		msgID = "msg_mock_suggestion"
+		msgID = generateRealisticMsgID()
 		outputTokens = 1
 		textDeltas = []string{""} // 空内容
 	default: // InterceptTypeWarmup
-		msgID = "msg_mock_warmup"
+		msgID = generateRealisticMsgID()
 		outputTokens = 2
 		textDeltas = []string{"New", " Conversation"}
 	}
 
-	// Build message_start event with fixed schema.
-	messageStartJSON := `{"type":"message_start","message":{"id":` + strconv.Quote(msgID) + `,"type":"message","role":"assistant","model":` + strconv.Quote(model) + `,"content":[],"stop_reason":null,"stop_sequence":null,"usage":{"input_tokens":10,"output_tokens":0}}}`
+	// Build message_start event — field order matches real Anthropic API response.
+	messageStartJSON := `{"type":"message_start","message":{"model":` + strconv.Quote(model) + `,"id":` + strconv.Quote(msgID) + `,"type":"message","role":"assistant","content":[],"stop_reason":null,"stop_sequence":null,"stop_details":null,"usage":{"input_tokens":10,"cache_creation_input_tokens":0,"cache_read_input_tokens":0,"output_tokens":0}}}`
 
 	// Build events
 	events := []string{
 		`event: message_start` + "\n" + `data: ` + string(messageStartJSON),
-		`event: content_block_start` + "\n" + `data: {"content_block":{"text":"","type":"text"},"index":0,"type":"content_block_start"}`,
+		`event: content_block_start` + "\n" + `data: {"type":"content_block_start","index":0,"content_block":{"type":"text","text":""}}`,
 	}
 
 	// Add text deltas
@@ -2151,7 +2151,7 @@ func sendMockInterceptStream(c *gin.Context, model string, interceptType Interce
 	}
 
 	// Add final events
-	messageDeltaJSON := `{"type":"message_delta","delta":{"stop_reason":"end_turn","stop_sequence":null},"usage":{"input_tokens":10,"output_tokens":` + strconv.Itoa(outputTokens) + `}}`
+	messageDeltaJSON := `{"type":"message_delta","delta":{"stop_reason":"end_turn","stop_sequence":null,"stop_details":null},"usage":{"output_tokens":` + strconv.Itoa(outputTokens) + `}}`
 
 	events = append(events,
 		`event: content_block_stop`+"\n"+`data: {"index":0,"type":"content_block_stop"}`,
@@ -2166,20 +2166,20 @@ func sendMockInterceptStream(c *gin.Context, model string, interceptType Interce
 	}
 }
 
-// generateRealisticMsgID 生成仿真的消息 ID（msg_bdrk_XXXXXXX 格式）
-// 格式与 Claude API 真实响应一致，24 位随机字母数字
+// generateRealisticMsgID 生成仿真的消息 ID（msg_01XXXXXXX 格式）
+// 格式与 Anthropic API 官方响应一致：msg_01 + 22 位 Base62 随机字符
 func generateRealisticMsgID() string {
 	const charset = "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz"
-	const idLen = 24
+	const idLen = 22
 	randomBytes := make([]byte, idLen)
 	if _, err := rand.Read(randomBytes); err != nil {
-		return fmt.Sprintf("msg_bdrk_%d", time.Now().UnixNano())
+		return fmt.Sprintf("msg_01%d", time.Now().UnixNano())
 	}
 	b := make([]byte, idLen)
 	for i := range b {
 		b[i] = charset[int(randomBytes[i])%len(charset)]
 	}
-	return "msg_bdrk_" + string(b)
+	return "msg_01" + string(b)
 }
 
 // sendMockInterceptResponse 发送非流式 mock 响应（用于请求拦截）
@@ -2189,7 +2189,7 @@ func sendMockInterceptResponse(c *gin.Context, model string, interceptType Inter
 
 	switch interceptType {
 	case InterceptTypeSuggestionMode:
-		msgID = "msg_mock_suggestion"
+		msgID = generateRealisticMsgID()
 		text = ""
 		outputTokens = 1
 		stopReason = "end_turn"
@@ -2199,13 +2199,13 @@ func sendMockInterceptResponse(c *gin.Context, model string, interceptType Inter
 		outputTokens = 1
 		stopReason = "max_tokens" // max_tokens=1 探测请求的 stop_reason 应为 max_tokens
 	default: // InterceptTypeWarmup
-		msgID = "msg_mock_warmup"
+		msgID = generateRealisticMsgID()
 		text = "New Conversation"
 		outputTokens = 2
 		stopReason = "end_turn"
 	}
 
-	// 构建完整的响应格式（与 Claude API 响应格式一致）
+	// 构建完整的响应格式（与 Anthropic API 官方响应格式一致）
 	response := gin.H{
 		"model":         model,
 		"id":            msgID,
@@ -2214,6 +2214,7 @@ func sendMockInterceptResponse(c *gin.Context, model string, interceptType Inter
 		"content":       []gin.H{{"type": "text", "text": text}},
 		"stop_reason":   stopReason,
 		"stop_sequence": nil,
+		"stop_details":  nil,
 		"usage": gin.H{
 			"input_tokens":                10,
 			"cache_creation_input_tokens": 0,
@@ -2223,7 +2224,6 @@ func sendMockInterceptResponse(c *gin.Context, model string, interceptType Inter
 				"ephemeral_1h_input_tokens": 0,
 			},
 			"output_tokens": outputTokens,
-			"total_tokens":  10 + outputTokens,
 		},
 	}
 

backend/internal/handler/gateway_handler_intercept_test.go

@@ -49,7 +49,7 @@ func TestSendMockInterceptResponse_MaxTokensOneHaiku(t *testing.T) {
 
 	id, ok := response["id"].(string)
 	require.True(t, ok)
-	require.True(t, strings.HasPrefix(id, "msg_bdrk_"))
+	require.True(t, strings.HasPrefix(id, "msg_01"))
 
 	content, ok := response["content"].([]any)
 	require.True(t, ok)

backend/internal/handler/gateway_handler_warmup_intercept_unit_test.go

@@ -7,6 +7,7 @@ import (
 	"context"
 	"encoding/json"
 	"net/http/httptest"
+	"strings"
 	"testing"
 	"time"
 
@@ -289,7 +290,7 @@ func TestGatewayHandlerMessages_InterceptWarmup_AntigravityAccount_MixedScheduli
 
 	var resp map[string]any
 	require.NoError(t, json.Unmarshal(rec.Body.Bytes(), &resp))
-	require.Equal(t, "msg_mock_warmup", resp["id"])
+	require.True(t, strings.HasPrefix(resp["id"].(string), "msg_01"))
 	require.Equal(t, "claude-sonnet-4-5", resp["model"])
 
 	content, ok := resp["content"].([]any)
@@ -378,6 +379,6 @@ func TestGatewayHandlerMessages_InterceptWarmup_AntigravityAccount_ForcePlatform
 
 	var resp map[string]any
 	require.NoError(t, json.Unmarshal(rec.Body.Bytes(), &resp))
-	require.Equal(t, "msg_mock_warmup", resp["id"])
+	require.True(t, strings.HasPrefix(resp["id"].(string), "msg_01"))
 	require.Equal(t, "claude-sonnet-4-5", resp["model"])
 }

backend/internal/pkg/antigravity/response_transformer.go

@@ -293,7 +293,7 @@ func (p *NonStreamingProcessor) buildResponse(geminiResp *GeminiResponse, respon
 		respID = geminiResp.ResponseID
 	}
 	if respID == "" {
-		respID = "msg_" + generateRandomID()
+		respID = generateAnthropicMsgID()
 	}
 
 	return &ClaudeResponse{
@@ -372,3 +372,18 @@ func generateRandomID() string {
 	}
 	return string(id)
 }
+
+// generateAnthropicMsgID 生成 Anthropic 官方格式的 message ID：msg_01 + 22 位 Base62
+func generateAnthropicMsgID() string {
+	const charset = "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz"
+	const idLen = 22
+	randomBytes := make([]byte, idLen)
+	if _, err := rand.Read(randomBytes); err != nil {
+		return "msg_01" + generateRandomID() + generateRandomID()[:10]
+	}
+	b := make([]byte, idLen)
+	for i := range b {
+		b[i] = charset[int(randomBytes[i])%len(charset)]
+	}
+	return "msg_01" + string(b)
+}

backend/internal/pkg/antigravity/stream_transformer.go

@@ -195,7 +195,7 @@ func (p *StreamingProcessor) emitMessageStart(v1Resp *V1InternalResponse) []byte
 		responseID = v1Resp.Response.ResponseID
 	}
 	if responseID == "" {
-		responseID = "msg_" + generateRandomID()
+		responseID = generateAnthropicMsgID()
 	}
 
 	var usageValue any = usage

backend/internal/service/account.go

@@ -136,6 +136,18 @@ func (a *Account) IsActive() bool {
 	return a.Status == StatusActive
 }
 
+// IsSyntheticUITest reports whether the account belongs to an isolated UI load-test
+// dataset. Production accounts never receive this marker. It lets the dedicated
+// test instance exercise interactive quota and connection-test controls without
+// sending fake credentials to an upstream provider.
+func (a *Account) IsSyntheticUITest() bool {
+	if a == nil || a.Extra == nil {
+		return false
+	}
+	enabled, ok := a.Extra["synthetic_ui_test"].(bool)
+	return ok && enabled
+}
+
 // BillingRateMultiplier 返回账号计费倍率。
 // - nil 表示未配置/旧缓存缺字段，按 1.0 处理
 // - 允许 0，表示该账号计费为 0

backend/internal/service/account_test_service.go

@@ -186,6 +186,20 @@ func (s *AccountTestService) TestAccountConnection(c *gin.Context, accountID int
 		return s.sendErrorAndEnd(c, "Account not found")
 	}
 
+	// Synthetic UI load-test accounts exercise the real SSE parsing and modal
+	// interactions, but intentionally do not send their placeholder credentials
+	// to an upstream provider.
+	if account.IsSyntheticUITest() {
+		testModelID := modelID
+		if testModelID == "" {
+			testModelID = claude.DefaultTestModel
+		}
+		s.sendEvent(c, TestEvent{Type: "test_start", Model: testModelID})
+		s.sendEvent(c, TestEvent{Type: "content", Text: "Synthetic Anthropic OAuth account is healthy and interactive."})
+		s.sendEvent(c, TestEvent{Type: "test_complete", Success: true})
+		return nil
+	}
+
 	// Route to platform-specific test method
 	if account.IsOpenAI() {
 		return s.testOpenAIAccountConnection(c, account, modelID, prompt, normalizeAccountTestMode(mode))

backend/internal/service/account_usage_service.go

@@ -345,6 +345,13 @@ func (s *AccountUsageService) GetUsage(ctx context.Context, accountID int64, for
 		return nil, fmt.Errorf("get account failed: %w", err)
 	}
 
+	// Dedicated UI load-test accounts must remain fully interactive without ever
+	// contacting Anthropic with synthetic credentials. Reuse the same persisted
+	// passive snapshot that the account table loads on mount.
+	if account.IsSyntheticUITest() && account.IsAnthropicOAuthOrSetupToken() {
+		return s.GetPassiveUsage(ctx, accountID)
+	}
+
 	if account.Platform == PlatformOpenAI && account.Type == AccountTypeOAuth {
 		usage, err := s.getOpenAIUsage(ctx, account, forceProbe)
 		if err == nil {
@@ -508,10 +515,30 @@ func (s *AccountUsageService) GetPassiveUsage(ctx context.Context, accountID int
 
 	// 添加窗口统计
 	s.addWindowStats(ctx, account, info)
+	if account.IsSyntheticUITest() {
+		applySyntheticWindowStats(info, account.Extra)
+	}
 
 	return info, nil
 }
 
+func applySyntheticWindowStats(info *UsageInfo, extra map[string]any) {
+	if info == nil || info.FiveHour == nil || len(extra) == 0 {
+		return
+	}
+	raw, ok := extra["synthetic_window_stats"].(map[string]any)
+	if !ok {
+		return
+	}
+	info.FiveHour.WindowStats = &WindowStats{
+		Requests:     int64(parseExtraInt(raw["requests"])),
+		Tokens:       int64(parseExtraInt(raw["tokens"])),
+		Cost:         parseExtraFloat64(raw["cost"]),
+		StandardCost: parseExtraFloat64(raw["standard_cost"]),
+		UserCost:     parseExtraFloat64(raw["user_cost"]),
+	}
+}
+
 // buildPassiveUsageWindow 从 Extra 中的被动采样数据（utilization 为 0-1 小数、reset 为 Unix 秒）
 // 构建用量窗口，无数据时返回 nil。
 func buildPassiveUsageWindow(extra map[string]any, utilKey, resetKey string) *UsageProgress {

backend/internal/service/gemini_chat_completions_compat_service.go

@@ -556,7 +556,7 @@ func (s *GeminiMessagesCompatService) handleChatCompletionsStreamingResponseFrom
 		return false
 	}
 
-	messageID := "msg_" + randomHex(12)
+	messageID := generateAnthropicMsgID()
 	if emitAnthropicEvent(&apicompat.AnthropicStreamEvent{
 		Type: "message_start",
 		Message: &apicompat.AnthropicResponse{

backend/internal/service/gemini_messages_compat_service.go

@@ -2023,7 +2023,7 @@ func (s *GeminiMessagesCompatService) handleStreamingResponse(c *gin.Context, re
 		return nil, errors.New("streaming not supported")
 	}
 
-	messageID := "msg_" + randomHex(12)
+	messageID := generateAnthropicMsgID()
 	messageStart := map[string]any{
 		"type": "message_start",
 		"message": map[string]any{
@@ -2294,6 +2294,21 @@ func randomHex(nBytes int) string {
 	return hex.EncodeToString(b)
 }
 
+// generateAnthropicMsgID 生成 Anthropic 官方格式的 message ID：msg_01 + 22 位 Base62
+func generateAnthropicMsgID() string {
+	const charset = "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz"
+	const idLen = 22
+	randomBytes := make([]byte, idLen)
+	if _, err := rand.Read(randomBytes); err != nil {
+		return "msg_01" + randomHex(11)
+	}
+	b := make([]byte, idLen)
+	for i := range b {
+		b[i] = charset[int(randomBytes[i])%len(charset)]
+	}
+	return "msg_01" + string(b)
+}
+
 func (s *GeminiMessagesCompatService) writeClaudeError(c *gin.Context, status int, errType, message string) error {
 	MarkResponseCommitted(c)
 	c.JSON(status, gin.H{
@@ -2816,7 +2831,7 @@ func convertGeminiToClaudeMessage(geminiResp map[string]any, originalModel strin
 	}
 
 	resp := map[string]any{
-		"id":            "msg_" + randomHex(12),
+		"id":            generateAnthropicMsgID(),
 		"type":          "message",
 		"role":          "assistant",
 		"model":         originalModel,