merge: 合并 test 分支到 test-dev，解决冲突

解决的冲突文件：
- wire_gen.go: 合并 ConcurrencyService/CRSSyncService 参数和 userAttributeHandler
- gateway_handler.go: 合并 pkg/errors 和 antigravity 导入
- gateway_service.go: 合并 validateUpstreamBaseURL 和 GetAvailableModels
- config.example.yaml: 合并 billing/turnstile 配置和额外 gateway 选项

🤖 Generated with [Claude Code](https://claude.com/claude-code

)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

backend/internal/handler/admin/gemini_oauth_handler.go

@@ -46,8 +46,8 @@ func (h *GeminiOAuthHandler) GenerateAuthURL(c *gin.Context) {
 	if oauthType == "" {
 		oauthType = "code_assist"
 	}
-	if oauthType != "code_assist" && oauthType != "ai_studio" {
-		response.BadRequest(c, "Invalid oauth_type: must be 'code_assist' or 'ai_studio'")
+	if oauthType != "code_assist" && oauthType != "google_one" && oauthType != "ai_studio" {
+		response.BadRequest(c, "Invalid oauth_type: must be 'code_assist', 'google_one', or 'ai_studio'")
 		return
 	}
 
@@ -92,8 +92,8 @@ func (h *GeminiOAuthHandler) ExchangeCode(c *gin.Context) {
 	if oauthType == "" {
 		oauthType = "code_assist"
 	}
-	if oauthType != "code_assist" && oauthType != "ai_studio" {
-		response.BadRequest(c, "Invalid oauth_type: must be 'code_assist' or 'ai_studio'")
+	if oauthType != "code_assist" && oauthType != "google_one" && oauthType != "ai_studio" {
+		response.BadRequest(c, "Invalid oauth_type: must be 'code_assist', 'google_one', or 'ai_studio'")
 		return
 	}
 

backend/internal/handler/admin/setting_handler.go

@@ -16,13 +16,15 @@ import (
 type SettingHandler struct {
 	settingService   *service.SettingService
 	emailService     *service.EmailService
+	turnstileService *service.TurnstileService
 }
 
 // NewSettingHandler 创建系统设置处理器
-func NewSettingHandler(settingService *service.SettingService, emailService *service.EmailService) *SettingHandler {
+func NewSettingHandler(settingService *service.SettingService, emailService *service.EmailService, turnstileService *service.TurnstileService) *SettingHandler {
 	return &SettingHandler{
 		settingService:   settingService,
 		emailService:     emailService,
+		turnstileService: turnstileService,
 	}
 }
 
@@ -118,6 +120,36 @@ func (h *SettingHandler) UpdateSettings(c *gin.Context) {
 		req.SmtpPort = 587
 	}
 
+	// Turnstile 参数验证
+	if req.TurnstileEnabled {
+		// 检查必填字段
+		if req.TurnstileSiteKey == "" {
+			response.BadRequest(c, "Turnstile Site Key is required when enabled")
+			return
+		}
+		if req.TurnstileSecretKey == "" {
+			response.BadRequest(c, "Turnstile Secret Key is required when enabled")
+			return
+		}
+
+		// 获取当前设置，检查参数是否有变化
+		currentSettings, err := h.settingService.GetAllSettings(c.Request.Context())
+		if err != nil {
+			response.ErrorFrom(c, err)
+			return
+		}
+
+		// 当 site_key 或 secret_key 任一变化时验证（避免配置错误导致无法登录）
+		siteKeyChanged := currentSettings.TurnstileSiteKey != req.TurnstileSiteKey
+		secretKeyChanged := currentSettings.TurnstileSecretKey != req.TurnstileSecretKey
+		if siteKeyChanged || secretKeyChanged {
+			if err := h.turnstileService.ValidateSecretKey(c.Request.Context(), req.TurnstileSecretKey); err != nil {
+				response.ErrorFrom(c, err)
+				return
+			}
+		}
+	}
+
 	settings := &service.SystemSettings{
 		RegistrationEnabled: req.RegistrationEnabled,
 		EmailVerifyEnabled:  req.EmailVerifyEnabled,

backend/internal/handler/admin/usage_handler.go

@@ -246,7 +246,7 @@ func (h *UsageHandler) SearchUsers(c *gin.Context) {
 	}
 
 	// Limit to 30 results
-	users, _, err := h.adminService.ListUsers(c.Request.Context(), 1, 30, "", "", keyword)
+	users, _, err := h.adminService.ListUsers(c.Request.Context(), 1, 30, service.UserListFilters{Search: keyword})
 	if err != nil {
 		response.ErrorFrom(c, err)
 		return

backend/internal/handler/admin/user_attribute_handler.go

+package admin
+
+import (
+	"strconv"
+
+	"github.com/Wei-Shaw/sub2api/internal/pkg/response"
+	"github.com/Wei-Shaw/sub2api/internal/service"
+
+	"github.com/gin-gonic/gin"
+)
+
+// UserAttributeHandler handles user attribute management
+type UserAttributeHandler struct {
+	attrService *service.UserAttributeService
+}
+
+// NewUserAttributeHandler creates a new handler
+func NewUserAttributeHandler(attrService *service.UserAttributeService) *UserAttributeHandler {
+	return &UserAttributeHandler{attrService: attrService}
+}
+
+// --- Request/Response DTOs ---
+
+// CreateAttributeDefinitionRequest represents create attribute definition request
+type CreateAttributeDefinitionRequest struct {
+	Key         string                          `json:"key" binding:"required,min=1,max=100"`
+	Name        string                          `json:"name" binding:"required,min=1,max=255"`
+	Description string                          `json:"description"`
+	Type        string                          `json:"type" binding:"required"`
+	Options     []service.UserAttributeOption   `json:"options"`
+	Required    bool                            `json:"required"`
+	Validation  service.UserAttributeValidation `json:"validation"`
+	Placeholder string                          `json:"placeholder"`
+	Enabled     bool                            `json:"enabled"`
+}
+
+// UpdateAttributeDefinitionRequest represents update attribute definition request
+type UpdateAttributeDefinitionRequest struct {
+	Name        *string                          `json:"name"`
+	Description *string                          `json:"description"`
+	Type        *string                          `json:"type"`
+	Options     *[]service.UserAttributeOption   `json:"options"`
+	Required    *bool                            `json:"required"`
+	Validation  *service.UserAttributeValidation `json:"validation"`
+	Placeholder *string                          `json:"placeholder"`
+	Enabled     *bool                            `json:"enabled"`
+}
+
+// ReorderRequest represents reorder attribute definitions request
+type ReorderRequest struct {
+	IDs []int64 `json:"ids" binding:"required"`
+}
+
+// UpdateUserAttributesRequest represents update user attributes request
+type UpdateUserAttributesRequest struct {
+	Values map[int64]string `json:"values" binding:"required"`
+}
+
+// BatchGetUserAttributesRequest represents batch get user attributes request
+type BatchGetUserAttributesRequest struct {
+	UserIDs []int64 `json:"user_ids" binding:"required"`
+}
+
+// BatchUserAttributesResponse represents batch user attributes response
+type BatchUserAttributesResponse struct {
+	// Map of userID -> map of attributeID -> value
+	Attributes map[int64]map[int64]string `json:"attributes"`
+}
+
+// AttributeDefinitionResponse represents attribute definition response
+type AttributeDefinitionResponse struct {
+	ID           int64                           `json:"id"`
+	Key          string                          `json:"key"`
+	Name         string                          `json:"name"`
+	Description  string                          `json:"description"`
+	Type         string                          `json:"type"`
+	Options      []service.UserAttributeOption   `json:"options"`
+	Required     bool                            `json:"required"`
+	Validation   service.UserAttributeValidation `json:"validation"`
+	Placeholder  string                          `json:"placeholder"`
+	DisplayOrder int                             `json:"display_order"`
+	Enabled      bool                            `json:"enabled"`
+	CreatedAt    string                          `json:"created_at"`
+	UpdatedAt    string                          `json:"updated_at"`
+}
+
+// AttributeValueResponse represents attribute value response
+type AttributeValueResponse struct {
+	ID          int64  `json:"id"`
+	UserID      int64  `json:"user_id"`
+	AttributeID int64  `json:"attribute_id"`
+	Value       string `json:"value"`
+	CreatedAt   string `json:"created_at"`
+	UpdatedAt   string `json:"updated_at"`
+}
+
+// --- Helpers ---
+
+func defToResponse(def *service.UserAttributeDefinition) *AttributeDefinitionResponse {
+	return &AttributeDefinitionResponse{
+		ID:           def.ID,
+		Key:          def.Key,
+		Name:         def.Name,
+		Description:  def.Description,
+		Type:         string(def.Type),
+		Options:      def.Options,
+		Required:     def.Required,
+		Validation:   def.Validation,
+		Placeholder:  def.Placeholder,
+		DisplayOrder: def.DisplayOrder,
+		Enabled:      def.Enabled,
+		CreatedAt:    def.CreatedAt.Format("2006-01-02T15:04:05Z07:00"),
+		UpdatedAt:    def.UpdatedAt.Format("2006-01-02T15:04:05Z07:00"),
+	}
+}
+
+func valueToResponse(val *service.UserAttributeValue) *AttributeValueResponse {
+	return &AttributeValueResponse{
+		ID:          val.ID,
+		UserID:      val.UserID,
+		AttributeID: val.AttributeID,
+		Value:       val.Value,
+		CreatedAt:   val.CreatedAt.Format("2006-01-02T15:04:05Z07:00"),
+		UpdatedAt:   val.UpdatedAt.Format("2006-01-02T15:04:05Z07:00"),
+	}
+}
+
+// --- Handlers ---
+
+// ListDefinitions lists all attribute definitions
+// GET /admin/user-attributes
+func (h *UserAttributeHandler) ListDefinitions(c *gin.Context) {
+	enabledOnly := c.Query("enabled") == "true"
+
+	defs, err := h.attrService.ListDefinitions(c.Request.Context(), enabledOnly)
+	if err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+
+	out := make([]*AttributeDefinitionResponse, 0, len(defs))
+	for i := range defs {
+		out = append(out, defToResponse(&defs[i]))
+	}
+
+	response.Success(c, out)
+}
+
+// CreateDefinition creates a new attribute definition
+// POST /admin/user-attributes
+func (h *UserAttributeHandler) CreateDefinition(c *gin.Context) {
+	var req CreateAttributeDefinitionRequest
+	if err := c.ShouldBindJSON(&req); err != nil {
+		response.BadRequest(c, "Invalid request: "+err.Error())
+		return
+	}
+
+	def, err := h.attrService.CreateDefinition(c.Request.Context(), service.CreateAttributeDefinitionInput{
+		Key:         req.Key,
+		Name:        req.Name,
+		Description: req.Description,
+		Type:        service.UserAttributeType(req.Type),
+		Options:     req.Options,
+		Required:    req.Required,
+		Validation:  req.Validation,
+		Placeholder: req.Placeholder,
+		Enabled:     req.Enabled,
+	})
+	if err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+
+	response.Success(c, defToResponse(def))
+}
+
+// UpdateDefinition updates an attribute definition
+// PUT /admin/user-attributes/:id
+func (h *UserAttributeHandler) UpdateDefinition(c *gin.Context) {
+	id, err := strconv.ParseInt(c.Param("id"), 10, 64)
+	if err != nil {
+		response.BadRequest(c, "Invalid attribute ID")
+		return
+	}
+
+	var req UpdateAttributeDefinitionRequest
+	if err := c.ShouldBindJSON(&req); err != nil {
+		response.BadRequest(c, "Invalid request: "+err.Error())
+		return
+	}
+
+	input := service.UpdateAttributeDefinitionInput{
+		Name:        req.Name,
+		Description: req.Description,
+		Options:     req.Options,
+		Required:    req.Required,
+		Validation:  req.Validation,
+		Placeholder: req.Placeholder,
+		Enabled:     req.Enabled,
+	}
+	if req.Type != nil {
+		t := service.UserAttributeType(*req.Type)
+		input.Type = &t
+	}
+
+	def, err := h.attrService.UpdateDefinition(c.Request.Context(), id, input)
+	if err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+
+	response.Success(c, defToResponse(def))
+}
+
+// DeleteDefinition deletes an attribute definition
+// DELETE /admin/user-attributes/:id
+func (h *UserAttributeHandler) DeleteDefinition(c *gin.Context) {
+	id, err := strconv.ParseInt(c.Param("id"), 10, 64)
+	if err != nil {
+		response.BadRequest(c, "Invalid attribute ID")
+		return
+	}
+
+	if err := h.attrService.DeleteDefinition(c.Request.Context(), id); err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+
+	response.Success(c, gin.H{"message": "Attribute definition deleted successfully"})
+}
+
+// ReorderDefinitions reorders attribute definitions
+// PUT /admin/user-attributes/reorder
+func (h *UserAttributeHandler) ReorderDefinitions(c *gin.Context) {
+	var req ReorderRequest
+	if err := c.ShouldBindJSON(&req); err != nil {
+		response.BadRequest(c, "Invalid request: "+err.Error())
+		return
+	}
+
+	// Convert IDs array to orders map (position in array = display_order)
+	orders := make(map[int64]int, len(req.IDs))
+	for i, id := range req.IDs {
+		orders[id] = i
+	}
+
+	if err := h.attrService.ReorderDefinitions(c.Request.Context(), orders); err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+
+	response.Success(c, gin.H{"message": "Reorder successful"})
+}
+
+// GetUserAttributes gets a user's attribute values
+// GET /admin/users/:id/attributes
+func (h *UserAttributeHandler) GetUserAttributes(c *gin.Context) {
+	userID, err := strconv.ParseInt(c.Param("id"), 10, 64)
+	if err != nil {
+		response.BadRequest(c, "Invalid user ID")
+		return
+	}
+
+	values, err := h.attrService.GetUserAttributes(c.Request.Context(), userID)
+	if err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+
+	out := make([]*AttributeValueResponse, 0, len(values))
+	for i := range values {
+		out = append(out, valueToResponse(&values[i]))
+	}
+
+	response.Success(c, out)
+}
+
+// UpdateUserAttributes updates a user's attribute values
+// PUT /admin/users/:id/attributes
+func (h *UserAttributeHandler) UpdateUserAttributes(c *gin.Context) {
+	userID, err := strconv.ParseInt(c.Param("id"), 10, 64)
+	if err != nil {
+		response.BadRequest(c, "Invalid user ID")
+		return
+	}
+
+	var req UpdateUserAttributesRequest
+	if err := c.ShouldBindJSON(&req); err != nil {
+		response.BadRequest(c, "Invalid request: "+err.Error())
+		return
+	}
+
+	inputs := make([]service.UpdateUserAttributeInput, 0, len(req.Values))
+	for attrID, value := range req.Values {
+		inputs = append(inputs, service.UpdateUserAttributeInput{
+			AttributeID: attrID,
+			Value:       value,
+		})
+	}
+
+	if err := h.attrService.UpdateUserAttributes(c.Request.Context(), userID, inputs); err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+
+	// Return updated values
+	values, err := h.attrService.GetUserAttributes(c.Request.Context(), userID)
+	if err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+
+	out := make([]*AttributeValueResponse, 0, len(values))
+	for i := range values {
+		out = append(out, valueToResponse(&values[i]))
+	}
+
+	response.Success(c, out)
+}
+
+// GetBatchUserAttributes gets attribute values for multiple users
+// POST /admin/user-attributes/batch
+func (h *UserAttributeHandler) GetBatchUserAttributes(c *gin.Context) {
+	var req BatchGetUserAttributesRequest
+	if err := c.ShouldBindJSON(&req); err != nil {
+		response.BadRequest(c, "Invalid request: "+err.Error())
+		return
+	}
+
+	if len(req.UserIDs) == 0 {
+		response.Success(c, BatchUserAttributesResponse{Attributes: map[int64]map[int64]string{}})
+		return
+	}
+
+	attrs, err := h.attrService.GetBatchUserAttributes(c.Request.Context(), req.UserIDs)
+	if err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+
+	response.Success(c, BatchUserAttributesResponse{Attributes: attrs})
+}

backend/internal/handler/admin/user_handler.go

@@ -27,7 +27,6 @@ type CreateUserRequest struct {
 	Email         string  `json:"email" binding:"required,email"`
 	Password      string  `json:"password" binding:"required,min=6"`
 	Username      string  `json:"username"`
-	Wechat        string  `json:"wechat"`
 	Notes         string  `json:"notes"`
 	Balance       float64 `json:"balance"`
 	Concurrency   int     `json:"concurrency"`
@@ -40,7 +39,6 @@ type UpdateUserRequest struct {
 	Email         string   `json:"email" binding:"omitempty,email"`
 	Password      string   `json:"password" binding:"omitempty,min=6"`
 	Username      *string  `json:"username"`
-	Wechat        *string  `json:"wechat"`
 	Notes         *string  `json:"notes"`
 	Balance       *float64 `json:"balance"`
 	Concurrency   *int     `json:"concurrency"`
@@ -57,13 +55,22 @@ type UpdateBalanceRequest struct {
 
 // List handles listing all users with pagination
 // GET /api/v1/admin/users
+// Query params:
+//   - status: filter by user status
+//   - role: filter by user role
+//   - search: search in email, username
+//   - attr[{id}]: filter by custom attribute value, e.g. attr[1]=company
 func (h *UserHandler) List(c *gin.Context) {
 	page, pageSize := response.ParsePagination(c)
-	status := c.Query("status")
-	role := c.Query("role")
-	search := c.Query("search")
 
-	users, total, err := h.adminService.ListUsers(c.Request.Context(), page, pageSize, status, role, search)
+	filters := service.UserListFilters{
+		Status:     c.Query("status"),
+		Role:       c.Query("role"),
+		Search:     c.Query("search"),
+		Attributes: parseAttributeFilters(c),
+	}
+
+	users, total, err := h.adminService.ListUsers(c.Request.Context(), page, pageSize, filters)
 	if err != nil {
 		response.ErrorFrom(c, err)
 		return
@@ -76,6 +83,29 @@ func (h *UserHandler) List(c *gin.Context) {
 	response.Paginated(c, out, total, page, pageSize)
 }
 
+// parseAttributeFilters extracts attribute filters from query params
+// Format: attr[{attributeID}]=value, e.g. attr[1]=company&attr[2]=developer
+func parseAttributeFilters(c *gin.Context) map[int64]string {
+	result := make(map[int64]string)
+
+	// Get all query params and look for attr[*] pattern
+	for key, values := range c.Request.URL.Query() {
+		if len(values) == 0 || values[0] == "" {
+			continue
+		}
+		// Check if key matches pattern attr[{id}]
+		if len(key) > 5 && key[:5] == "attr[" && key[len(key)-1] == ']' {
+			idStr := key[5 : len(key)-1]
+			id, err := strconv.ParseInt(idStr, 10, 64)
+			if err == nil && id > 0 {
+				result[id] = values[0]
+			}
+		}
+	}
+
+	return result
+}
+
 // GetByID handles getting a user by ID
 // GET /api/v1/admin/users/:id
 func (h *UserHandler) GetByID(c *gin.Context) {
@@ -107,7 +137,6 @@ func (h *UserHandler) Create(c *gin.Context) {
 		Email:         req.Email,
 		Password:      req.Password,
 		Username:      req.Username,
-		Wechat:        req.Wechat,
 		Notes:         req.Notes,
 		Balance:       req.Balance,
 		Concurrency:   req.Concurrency,
@@ -141,7 +170,6 @@ func (h *UserHandler) Update(c *gin.Context) {
 		Email:         req.Email,
 		Password:      req.Password,
 		Username:      req.Username,
-		Wechat:        req.Wechat,
 		Notes:         req.Notes,
 		Balance:       req.Balance,
 		Concurrency:   req.Concurrency,

backend/internal/handler/dto/mappers.go

@@ -10,7 +10,6 @@ func UserFromServiceShallow(u *service.User) *User {
 		ID:            u.ID,
 		Email:         u.Email,
 		Username:      u.Username,
-		Wechat:        u.Wechat,
 		Notes:         u.Notes,
 		Role:          u.Role,
 		Balance:       u.Balance,

backend/internal/handler/dto/types.go

@@ -6,7 +6,6 @@ type User struct {
 	ID            int64     `json:"id"`
 	Email         string    `json:"email"`
 	Username      string    `json:"username"`
-	Wechat        string    `json:"wechat"`
 	Notes         string    `json:"notes"`
 	Role          string    `json:"role"`
 	Balance       float64   `json:"balance"`

backend/internal/handler/gateway_handler.go

@@ -11,7 +11,8 @@ import (
 	"strings"
 	"time"
 
-	infraerrors "github.com/Wei-Shaw/sub2api/internal/infrastructure/errors"
+	"github.com/Wei-Shaw/sub2api/internal/pkg/antigravity"
+	pkgerrors "github.com/Wei-Shaw/sub2api/internal/pkg/errors"
 	"github.com/Wei-Shaw/sub2api/internal/pkg/claude"
 	"github.com/Wei-Shaw/sub2api/internal/pkg/openai"
 	middleware2 "github.com/Wei-Shaw/sub2api/internal/server/middleware"
@@ -143,6 +144,10 @@ func (h *GatewayHandler) Messages(c *gin.Context) {
 	} else if apiKey.Group != nil {
 		platform = apiKey.Group.Platform
 	}
+	sessionKey := sessionHash
+	if platform == service.PlatformGemini && sessionHash != "" {
+		sessionKey = "gemini:" + sessionHash
+	}
 
 	if platform == service.PlatformGemini {
 		const maxAccountSwitches = 3
@@ -151,7 +156,7 @@ func (h *GatewayHandler) Messages(c *gin.Context) {
 		lastFailoverStatus := 0
 
 		for {
-			account, err := h.geminiCompatService.SelectAccountForModelWithExclusions(c.Request.Context(), apiKey.GroupID, sessionHash, reqModel, failedAccountIDs)
+			selection, err := h.gatewayService.SelectAccountWithLoadAwareness(c.Request.Context(), apiKey.GroupID, sessionKey, reqModel, failedAccountIDs)
 			if err != nil {
 				if len(failedAccountIDs) == 0 {
 					h.handleStreamingAwareError(c, http.StatusServiceUnavailable, "api_error", "No available accounts: "+err.Error(), streamStarted)
@@ -160,9 +165,13 @@ func (h *GatewayHandler) Messages(c *gin.Context) {
 				h.handleFailoverExhausted(c, lastFailoverStatus, streamStarted)
 				return
 			}
+			account := selection.Account
 
 			// 检查预热请求拦截（在账号选择后、转发前检查）
 			if account.IsInterceptWarmupEnabled() && isWarmupRequest(body) {
+				if selection.Acquired && selection.ReleaseFunc != nil {
+					selection.ReleaseFunc()
+				}
 				if reqStream {
 					sendMockWarmupStream(c, reqModel)
 				} else {
@@ -172,12 +181,47 @@ func (h *GatewayHandler) Messages(c *gin.Context) {
 			}
 
 			// 3. 获取账号并发槽位
-			accountReleaseFunc, err := h.concurrencyHelper.AcquireAccountSlotWithWait(c, account.ID, account.Concurrency, reqStream, &streamStarted)
+			accountReleaseFunc := selection.ReleaseFunc
+			var accountWaitRelease func()
+			if !selection.Acquired {
+				if selection.WaitPlan == nil {
+					h.handleStreamingAwareError(c, http.StatusServiceUnavailable, "api_error", "No available accounts", streamStarted)
+					return
+				}
+				canWait, err := h.concurrencyHelper.IncrementAccountWaitCount(c.Request.Context(), account.ID, selection.WaitPlan.MaxWaiting)
+				if err != nil {
+					log.Printf("Increment account wait count failed: %v", err)
+				} else if !canWait {
+					log.Printf("Account wait queue full: account=%d", account.ID)
+					h.handleStreamingAwareError(c, http.StatusTooManyRequests, "rate_limit_error", "Too many pending requests, please retry later", streamStarted)
+					return
+				} else {
+					// Only set release function if increment succeeded
+					accountWaitRelease = func() {
+						h.concurrencyHelper.DecrementAccountWaitCount(c.Request.Context(), account.ID)
+					}
+				}
+
+				accountReleaseFunc, err = h.concurrencyHelper.AcquireAccountSlotWithWaitTimeout(
+					c,
+					account.ID,
+					selection.WaitPlan.MaxConcurrency,
+					selection.WaitPlan.Timeout,
+					reqStream,
+					&streamStarted,
+				)
 				if err != nil {
+					if accountWaitRelease != nil {
+						accountWaitRelease()
+					}
 					log.Printf("Account concurrency acquire failed: %v", err)
 					h.handleConcurrencyError(c, err, "account", streamStarted)
 					return
 				}
+				if err := h.gatewayService.BindStickySession(c.Request.Context(), sessionKey, account.ID); err != nil {
+					log.Printf("Bind sticky session failed: %v", err)
+				}
+			}
 
 			// 转发请求 - 根据账号平台分流
 			var result *service.ForwardResult
@@ -189,6 +233,9 @@ func (h *GatewayHandler) Messages(c *gin.Context) {
 			if accountReleaseFunc != nil {
 				accountReleaseFunc()
 			}
+			if accountWaitRelease != nil {
+				accountWaitRelease()
+			}
 			if err != nil {
 				var failoverErr *service.UpstreamFailoverError
 				if errors.As(err, &failoverErr) {
@@ -233,7 +280,7 @@ func (h *GatewayHandler) Messages(c *gin.Context) {
 
 	for {
 		// 选择支持该模型的账号
-		account, err := h.gatewayService.SelectAccountForModelWithExclusions(c.Request.Context(), apiKey.GroupID, sessionHash, reqModel, failedAccountIDs)
+		selection, err := h.gatewayService.SelectAccountWithLoadAwareness(c.Request.Context(), apiKey.GroupID, sessionKey, reqModel, failedAccountIDs)
 		if err != nil {
 			if len(failedAccountIDs) == 0 {
 				h.handleStreamingAwareError(c, http.StatusServiceUnavailable, "api_error", "No available accounts: "+err.Error(), streamStarted)
@@ -242,9 +289,13 @@ func (h *GatewayHandler) Messages(c *gin.Context) {
 			h.handleFailoverExhausted(c, lastFailoverStatus, streamStarted)
 			return
 		}
+		account := selection.Account
 
 		// 检查预热请求拦截（在账号选择后、转发前检查）
 		if account.IsInterceptWarmupEnabled() && isWarmupRequest(body) {
+			if selection.Acquired && selection.ReleaseFunc != nil {
+				selection.ReleaseFunc()
+			}
 			if reqStream {
 				sendMockWarmupStream(c, reqModel)
 			} else {
@@ -254,12 +305,47 @@ func (h *GatewayHandler) Messages(c *gin.Context) {
 		}
 
 		// 3. 获取账号并发槽位
-		accountReleaseFunc, err := h.concurrencyHelper.AcquireAccountSlotWithWait(c, account.ID, account.Concurrency, reqStream, &streamStarted)
+		accountReleaseFunc := selection.ReleaseFunc
+		var accountWaitRelease func()
+		if !selection.Acquired {
+			if selection.WaitPlan == nil {
+				h.handleStreamingAwareError(c, http.StatusServiceUnavailable, "api_error", "No available accounts", streamStarted)
+				return
+			}
+			canWait, err := h.concurrencyHelper.IncrementAccountWaitCount(c.Request.Context(), account.ID, selection.WaitPlan.MaxWaiting)
 			if err != nil {
+				log.Printf("Increment account wait count failed: %v", err)
+			} else if !canWait {
+				log.Printf("Account wait queue full: account=%d", account.ID)
+				h.handleStreamingAwareError(c, http.StatusTooManyRequests, "rate_limit_error", "Too many pending requests, please retry later", streamStarted)
+				return
+			} else {
+				// Only set release function if increment succeeded
+				accountWaitRelease = func() {
+					h.concurrencyHelper.DecrementAccountWaitCount(c.Request.Context(), account.ID)
+				}
+			}
+
+			accountReleaseFunc, err = h.concurrencyHelper.AcquireAccountSlotWithWaitTimeout(
+				c,
+				account.ID,
+				selection.WaitPlan.MaxConcurrency,
+				selection.WaitPlan.Timeout,
+				reqStream,
+				&streamStarted,
+			)
+			if err != nil {
+				if accountWaitRelease != nil {
+					accountWaitRelease()
+				}
 				log.Printf("Account concurrency acquire failed: %v", err)
 				h.handleConcurrencyError(c, err, "account", streamStarted)
 				return
 			}
+			if err := h.gatewayService.BindStickySession(c.Request.Context(), sessionKey, account.ID); err != nil {
+				log.Printf("Bind sticky session failed: %v", err)
+			}
+		}
 
 		// 转发请求 - 根据账号平台分流
 		var result *service.ForwardResult
@@ -271,6 +357,9 @@ func (h *GatewayHandler) Messages(c *gin.Context) {
 		if accountReleaseFunc != nil {
 			accountReleaseFunc()
 		}
+		if accountWaitRelease != nil {
+			accountWaitRelease()
+		}
 		if err != nil {
 			var failoverErr *service.UpstreamFailoverError
 			if errors.As(err, &failoverErr) {
@@ -310,12 +399,42 @@ func (h *GatewayHandler) Messages(c *gin.Context) {
 
 // Models handles listing available models
 // GET /v1/models
-// Returns different model lists based on the API key's group platform
+// Returns models based on account configurations (model_mapping whitelist)
+// Falls back to default models if no whitelist is configured
 func (h *GatewayHandler) Models(c *gin.Context) {
 	apiKey, _ := middleware2.GetApiKeyFromContext(c)
 
-	// Return OpenAI models for OpenAI platform groups
-	if apiKey != nil && apiKey.Group != nil && apiKey.Group.Platform == "openai" {
+	var groupID *int64
+	var platform string
+
+	if apiKey != nil && apiKey.Group != nil {
+		groupID = &apiKey.Group.ID
+		platform = apiKey.Group.Platform
+	}
+
+	// Get available models from account configurations (without platform filter)
+	availableModels := h.gatewayService.GetAvailableModels(c.Request.Context(), groupID, "")
+
+	if len(availableModels) > 0 {
+		// Build model list from whitelist
+		models := make([]claude.Model, 0, len(availableModels))
+		for _, modelID := range availableModels {
+			models = append(models, claude.Model{
+				ID:          modelID,
+				Type:        "model",
+				DisplayName: modelID,
+				CreatedAt:   "2024-01-01T00:00:00Z",
+			})
+		}
+		c.JSON(http.StatusOK, gin.H{
+			"object": "list",
+			"data":   models,
+		})
+		return
+	}
+
+	// Fallback to default models
+	if platform == "openai" {
 		c.JSON(http.StatusOK, gin.H{
 			"object": "list",
 			"data":   openai.DefaultModels,
@@ -323,13 +442,21 @@ func (h *GatewayHandler) Models(c *gin.Context) {
 		return
 	}
 
-	// Default: Claude models
 	c.JSON(http.StatusOK, gin.H{
 		"object": "list",
 		"data":   claude.DefaultModels,
 	})
 }
 
+// AntigravityModels 返回 Antigravity 支持的全部模型
+// GET /antigravity/models
+func (h *GatewayHandler) AntigravityModels(c *gin.Context) {
+	c.JSON(http.StatusOK, gin.H{
+		"object": "list",
+		"data":   antigravity.DefaultModels(),
+	})
+}
+
 // Usage handles getting account balance for CC Switch integration
 // GET /v1/usage
 func (h *GatewayHandler) Usage(c *gin.Context) {
@@ -648,13 +775,13 @@ func sendMockWarmupResponse(c *gin.Context, model string) {
 
 func billingErrorDetails(err error) (status int, code, message string) {
 	if errors.Is(err, service.ErrBillingServiceUnavailable) {
-		msg := infraerrors.Message(err)
+		msg := pkgerrors.Message(err)
 		if msg == "" {
 			msg = "Billing service temporarily unavailable. Please retry later."
 		}
 		return http.StatusServiceUnavailable, "billing_service_error", msg
 	}
-	msg := infraerrors.Message(err)
+	msg := pkgerrors.Message(err)
 	if msg == "" {
 		msg = err.Error()
 	}

backend/internal/handler/gateway_helper.go

@@ -83,6 +83,16 @@ func (h *ConcurrencyHelper) DecrementWaitCount(ctx context.Context, userID int64
 	h.concurrencyService.DecrementWaitCount(ctx, userID)
 }
 
+// IncrementAccountWaitCount increments the wait count for an account
+func (h *ConcurrencyHelper) IncrementAccountWaitCount(ctx context.Context, accountID int64, maxWait int) (bool, error) {
+	return h.concurrencyService.IncrementAccountWaitCount(ctx, accountID, maxWait)
+}
+
+// DecrementAccountWaitCount decrements the wait count for an account
+func (h *ConcurrencyHelper) DecrementAccountWaitCount(ctx context.Context, accountID int64) {
+	h.concurrencyService.DecrementAccountWaitCount(ctx, accountID)
+}
+
 // AcquireUserSlotWithWait acquires a user concurrency slot, waiting if necessary.
 // For streaming requests, sends ping events during the wait.
 // streamStarted is updated if streaming response has begun.
@@ -126,7 +136,12 @@ func (h *ConcurrencyHelper) AcquireAccountSlotWithWait(c *gin.Context, accountID
 // waitForSlotWithPing waits for a concurrency slot, sending ping events for streaming requests.
 // streamStarted pointer is updated when streaming begins (for proper error handling by caller).
 func (h *ConcurrencyHelper) waitForSlotWithPing(c *gin.Context, slotType string, id int64, maxConcurrency int, isStream bool, streamStarted *bool) (func(), error) {
-	ctx, cancel := context.WithTimeout(c.Request.Context(), maxConcurrencyWait)
+	return h.waitForSlotWithPingTimeout(c, slotType, id, maxConcurrency, maxConcurrencyWait, isStream, streamStarted)
+}
+
+// waitForSlotWithPingTimeout waits for a concurrency slot with a custom timeout.
+func (h *ConcurrencyHelper) waitForSlotWithPingTimeout(c *gin.Context, slotType string, id int64, maxConcurrency int, timeout time.Duration, isStream bool, streamStarted *bool) (func(), error) {
+	ctx, cancel := context.WithTimeout(c.Request.Context(), timeout)
 	defer cancel()
 
 	// Determine if ping is needed (streaming + ping format defined)
@@ -200,6 +215,11 @@ func (h *ConcurrencyHelper) waitForSlotWithPing(c *gin.Context, slotType string,
 	}
 }
 
+// AcquireAccountSlotWithWaitTimeout acquires an account slot with a custom timeout (keeps SSE ping).
+func (h *ConcurrencyHelper) AcquireAccountSlotWithWaitTimeout(c *gin.Context, accountID int64, maxConcurrency int, timeout time.Duration, isStream bool, streamStarted *bool) (func(), error) {
+	return h.waitForSlotWithPingTimeout(c, "account", accountID, maxConcurrency, timeout, isStream, streamStarted)
+}
+
 // nextBackoff 计算下一次退避时间
 // 性能优化：使用指数退避 + 随机抖动，避免惊群效应
 // current: 当前退避时间

backend/internal/handler/gemini_v1beta_handler.go

@@ -9,6 +9,7 @@ import (
 	"strings"
 	"time"
 
+	"github.com/Wei-Shaw/sub2api/internal/pkg/antigravity"
 	"github.com/Wei-Shaw/sub2api/internal/pkg/gemini"
 	"github.com/Wei-Shaw/sub2api/internal/pkg/googleapi"
 	"github.com/Wei-Shaw/sub2api/internal/server/middleware"
@@ -32,9 +33,9 @@ func (h *GatewayHandler) GeminiV1BetaListModels(c *gin.Context) {
 		return
 	}
 
-	// 强制 antigravity 模式：直接返回静态模型列表
+	// 强制 antigravity 模式：返回 antigravity 支持的模型列表
 	if forcePlatform == service.PlatformAntigravity {
-		c.JSON(http.StatusOK, gemini.FallbackModelsList())
+		c.JSON(http.StatusOK, antigravity.FallbackGeminiModelsList())
 		return
 	}
 
@@ -84,9 +85,9 @@ func (h *GatewayHandler) GeminiV1BetaGetModel(c *gin.Context) {
 		return
 	}
 
-	// 强制 antigravity 模式：直接返回静态模型信息
+	// 强制 antigravity 模式：返回 antigravity 模型信息
 	if forcePlatform == service.PlatformAntigravity {
-		c.JSON(http.StatusOK, gemini.FallbackModel(modelName))
+		c.JSON(http.StatusOK, antigravity.FallbackGeminiModel(modelName))
 		return
 	}
 
@@ -198,13 +199,17 @@ func (h *GatewayHandler) GeminiV1BetaModels(c *gin.Context) {
 	// 3) select account (sticky session based on request body)
 	parsedReq, _ := service.ParseGatewayRequest(body)
 	sessionHash := h.gatewayService.GenerateSessionHash(parsedReq)
+	sessionKey := sessionHash
+	if sessionHash != "" {
+		sessionKey = "gemini:" + sessionHash
+	}
 	const maxAccountSwitches = 3
 	switchCount := 0
 	failedAccountIDs := make(map[int64]struct{})
 	lastFailoverStatus := 0
 
 	for {
-		account, err := h.geminiCompatService.SelectAccountForModelWithExclusions(c.Request.Context(), apiKey.GroupID, sessionHash, modelName, failedAccountIDs)
+		selection, err := h.gatewayService.SelectAccountWithLoadAwareness(c.Request.Context(), apiKey.GroupID, sessionKey, modelName, failedAccountIDs)
 		if err != nil {
 			if len(failedAccountIDs) == 0 {
 				googleError(c, http.StatusServiceUnavailable, "No available Gemini accounts: "+err.Error())
@@ -213,13 +218,49 @@ func (h *GatewayHandler) GeminiV1BetaModels(c *gin.Context) {
 			handleGeminiFailoverExhausted(c, lastFailoverStatus)
 			return
 		}
+		account := selection.Account
 
 		// 4) account concurrency slot
-		accountReleaseFunc, err := geminiConcurrency.AcquireAccountSlotWithWait(c, account.ID, account.Concurrency, stream, &streamStarted)
+		accountReleaseFunc := selection.ReleaseFunc
+		var accountWaitRelease func()
+		if !selection.Acquired {
+			if selection.WaitPlan == nil {
+				googleError(c, http.StatusServiceUnavailable, "No available Gemini accounts")
+				return
+			}
+			canWait, err := geminiConcurrency.IncrementAccountWaitCount(c.Request.Context(), account.ID, selection.WaitPlan.MaxWaiting)
 			if err != nil {
+				log.Printf("Increment account wait count failed: %v", err)
+			} else if !canWait {
+				log.Printf("Account wait queue full: account=%d", account.ID)
+				googleError(c, http.StatusTooManyRequests, "Too many pending requests, please retry later")
+				return
+			} else {
+				// Only set release function if increment succeeded
+				accountWaitRelease = func() {
+					geminiConcurrency.DecrementAccountWaitCount(c.Request.Context(), account.ID)
+				}
+			}
+
+			accountReleaseFunc, err = geminiConcurrency.AcquireAccountSlotWithWaitTimeout(
+				c,
+				account.ID,
+				selection.WaitPlan.MaxConcurrency,
+				selection.WaitPlan.Timeout,
+				stream,
+				&streamStarted,
+			)
+			if err != nil {
+				if accountWaitRelease != nil {
+					accountWaitRelease()
+				}
 				googleError(c, http.StatusTooManyRequests, err.Error())
 				return
 			}
+			if err := h.gatewayService.BindStickySession(c.Request.Context(), sessionKey, account.ID); err != nil {
+				log.Printf("Bind sticky session failed: %v", err)
+			}
+		}
 
 		// 5) forward (根据平台分流)
 		var result *service.ForwardResult
@@ -231,6 +272,9 @@ func (h *GatewayHandler) GeminiV1BetaModels(c *gin.Context) {
 		if accountReleaseFunc != nil {
 			accountReleaseFunc()
 		}
+		if accountWaitRelease != nil {
+			accountWaitRelease()
+		}
 		if err != nil {
 			var failoverErr *service.UpstreamFailoverError
 			if errors.As(err, &failoverErr) {

backend/internal/handler/handler.go

@@ -20,6 +20,7 @@ type AdminHandlers struct {
 	System           *admin.SystemHandler
 	Subscription     *admin.SubscriptionHandler
 	Usage            *admin.UsageHandler
+	UserAttribute    *admin.UserAttributeHandler
 }
 
 // Handlers contains all HTTP handlers

backend/internal/handler/openai_gateway_handler.go

@@ -147,7 +147,7 @@ func (h *OpenAIGatewayHandler) Responses(c *gin.Context) {
 	for {
 		// Select account supporting the requested model
 		log.Printf("[OpenAI Handler] Selecting account: groupID=%v model=%s", apiKey.GroupID, reqModel)
-		account, err := h.gatewayService.SelectAccountForModelWithExclusions(c.Request.Context(), apiKey.GroupID, sessionHash, reqModel, failedAccountIDs)
+		selection, err := h.gatewayService.SelectAccountWithLoadAwareness(c.Request.Context(), apiKey.GroupID, sessionHash, reqModel, failedAccountIDs)
 		if err != nil {
 			log.Printf("[OpenAI Handler] SelectAccount failed: %v", err)
 			if len(failedAccountIDs) == 0 {
@@ -157,21 +157,60 @@ func (h *OpenAIGatewayHandler) Responses(c *gin.Context) {
 			h.handleFailoverExhausted(c, lastFailoverStatus, streamStarted)
 			return
 		}
+		account := selection.Account
 		log.Printf("[OpenAI Handler] Selected account: id=%d name=%s", account.ID, account.Name)
 
 		// 3. Acquire account concurrency slot
-		accountReleaseFunc, err := h.concurrencyHelper.AcquireAccountSlotWithWait(c, account.ID, account.Concurrency, reqStream, &streamStarted)
+		accountReleaseFunc := selection.ReleaseFunc
+		var accountWaitRelease func()
+		if !selection.Acquired {
+			if selection.WaitPlan == nil {
+				h.handleStreamingAwareError(c, http.StatusServiceUnavailable, "api_error", "No available accounts", streamStarted)
+				return
+			}
+			canWait, err := h.concurrencyHelper.IncrementAccountWaitCount(c.Request.Context(), account.ID, selection.WaitPlan.MaxWaiting)
+			if err != nil {
+				log.Printf("Increment account wait count failed: %v", err)
+			} else if !canWait {
+				log.Printf("Account wait queue full: account=%d", account.ID)
+				h.handleStreamingAwareError(c, http.StatusTooManyRequests, "rate_limit_error", "Too many pending requests, please retry later", streamStarted)
+				return
+			} else {
+				// Only set release function if increment succeeded
+				accountWaitRelease = func() {
+					h.concurrencyHelper.DecrementAccountWaitCount(c.Request.Context(), account.ID)
+				}
+			}
+
+			accountReleaseFunc, err = h.concurrencyHelper.AcquireAccountSlotWithWaitTimeout(
+				c,
+				account.ID,
+				selection.WaitPlan.MaxConcurrency,
+				selection.WaitPlan.Timeout,
+				reqStream,
+				&streamStarted,
+			)
 			if err != nil {
+				if accountWaitRelease != nil {
+					accountWaitRelease()
+				}
 				log.Printf("Account concurrency acquire failed: %v", err)
 				h.handleConcurrencyError(c, err, "account", streamStarted)
 				return
 			}
+			if err := h.gatewayService.BindStickySession(c.Request.Context(), sessionHash, account.ID); err != nil {
+				log.Printf("Bind sticky session failed: %v", err)
+			}
+		}
 
 		// Forward request
 		result, err := h.gatewayService.Forward(c.Request.Context(), c, account, body)
 		if accountReleaseFunc != nil {
 			accountReleaseFunc()
 		}
+		if accountWaitRelease != nil {
+			accountWaitRelease()
+		}
 		if err != nil {
 			var failoverErr *service.UpstreamFailoverError
 			if errors.As(err, &failoverErr) {

backend/internal/handler/user_handler.go

@@ -30,7 +30,6 @@ type ChangePasswordRequest struct {
 // UpdateProfileRequest represents the update profile request payload
 type UpdateProfileRequest struct {
 	Username *string `json:"username"`
-	Wechat   *string `json:"wechat"`
 }
 
 // GetProfile handles getting user profile
@@ -99,7 +98,6 @@ func (h *UserHandler) UpdateProfile(c *gin.Context) {
 
 	svcReq := service.UpdateProfileRequest{
 		Username: req.Username,
-		Wechat:   req.Wechat,
 	}
 	updatedUser, err := h.userService.UpdateProfile(c.Request.Context(), subject.UserID, svcReq)
 	if err != nil {

backend/internal/handler/wire.go

@@ -23,6 +23,7 @@ func ProvideAdminHandlers(
 	systemHandler *admin.SystemHandler,
 	subscriptionHandler *admin.SubscriptionHandler,
 	usageHandler *admin.UsageHandler,
+	userAttributeHandler *admin.UserAttributeHandler,
 ) *AdminHandlers {
 	return &AdminHandlers{
 		Dashboard:        dashboardHandler,
@@ -39,6 +40,7 @@ func ProvideAdminHandlers(
 		System:           systemHandler,
 		Subscription:     subscriptionHandler,
 		Usage:            usageHandler,
+		UserAttribute:    userAttributeHandler,
 	}
 }
 
@@ -107,6 +109,7 @@ var ProviderSet = wire.NewSet(
 	ProvideSystemHandler,
 	admin.NewSubscriptionHandler,
 	admin.NewUsageHandler,
+	admin.NewUserAttributeHandler,
 
 	// AdminHandlers and Handlers constructors
 	ProvideAdminHandlers,

backend/internal/infrastructure/wire.go

-package infrastructure
-
-import (
-	"database/sql"
-	"errors"
-
-	"github.com/Wei-Shaw/sub2api/ent"
-	"github.com/Wei-Shaw/sub2api/internal/config"
-
-	"github.com/google/wire"
-	"github.com/redis/go-redis/v9"
-
-	entsql "entgo.io/ent/dialect/sql"
-)
-
-// ProviderSet 是基础设施层的 Wire 依赖提供者集合。
-//
-// Wire 是 Google 开发的编译时依赖注入工具。ProviderSet 将相关的依赖提供函数
-// 组织在一起，便于在应用程序启动时自动组装依赖关系。
-//
-// 包含的提供者：
-//   - ProvideEnt: 提供 Ent ORM 客户端
-//   - ProvideSQLDB: 提供底层 SQL 数据库连接
-//   - ProvideRedis: 提供 Redis 客户端
-var ProviderSet = wire.NewSet(
-	ProvideEnt,
-	ProvideSQLDB,
-	ProvideRedis,
-)
-
-// ProvideEnt 为依赖注入提供 Ent 客户端。
-//
-// 该函数是 InitEnt 的包装器，符合 Wire 的依赖提供函数签名要求。
-// Wire 会在编译时分析依赖关系，自动生成初始化代码。
-//
-// 依赖：config.Config
-// 提供：*ent.Client
-func ProvideEnt(cfg *config.Config) (*ent.Client, error) {
-	client, _, err := InitEnt(cfg)
-	return client, err
-}
-
-// ProvideSQLDB 从 Ent 客户端提取底层的 *sql.DB 连接。
-//
-// 某些 Repository 需要直接执行原生 SQL（如复杂的批量更新、聚合查询），
-// 此时需要访问底层的 sql.DB 而不是通过 Ent ORM。
-//
-// 设计说明：
-//   - Ent 底层使用 sql.DB，通过 Driver 接口可以访问
-//   - 这种设计允许在同一事务中混用 Ent 和原生 SQL
-//
-// 依赖：*ent.Client
-// 提供：*sql.DB
-func ProvideSQLDB(client *ent.Client) (*sql.DB, error) {
-	if client == nil {
-		return nil, errors.New("nil ent client")
-	}
-	// 从 Ent 客户端获取底层驱动
-	drv, ok := client.Driver().(*entsql.Driver)
-	if !ok {
-		return nil, errors.New("ent driver does not expose *sql.DB")
-	}
-	// 返回驱动持有的 sql.DB 实例
-	return drv.DB(), nil
-}
-
-// ProvideRedis 为依赖注入提供 Redis 客户端。
-//
-// Redis 用于：
-//   - 分布式锁（如并发控制）
-//   - 缓存（如用户会话、API 响应缓存）
-//   - 速率限制
-//   - 实时统计数据
-//
-// 依赖：config.Config
-// 提供：*redis.Client
-func ProvideRedis(cfg *config.Config) *redis.Client {
-	return InitRedis(cfg)
-}

backend/internal/integration/e2e_gateway_test.go

@@ -57,6 +57,7 @@ var geminiModels = []string{
 	"gemini-2.5-flash-lite",
 	"gemini-3-flash",
 	"gemini-3-pro-low",
+	"gemini-3-pro-high",
 }
 
 func TestMain(m *testing.M) {
@@ -641,6 +642,37 @@ func testClaudeThinkingWithToolHistory(t *testing.T, model string) {
 	t.Logf("✅ thinking 模式工具调用测试通过, id=%v", result["id"])
 }
 
+// TestClaudeMessagesWithGeminiModel 测试在 Claude 端点使用 Gemini 模型
+// 验证：通过 /v1/messages 端点传入 gemini 模型名的场景（含前缀映射）
+// 仅在 Antigravity 模式下运行（ENDPOINT_PREFIX="/antigravity"）
+func TestClaudeMessagesWithGeminiModel(t *testing.T) {
+	if endpointPrefix != "/antigravity" {
+		t.Skip("仅在 Antigravity 模式下运行")
+	}
+
+	// 测试通过 Claude 端点调用 Gemini 模型
+	geminiViaClaude := []string{
+		"gemini-3-flash",       // 直接支持
+		"gemini-3-pro-low",     // 直接支持
+		"gemini-3-pro-high",    // 直接支持
+		"gemini-3-pro",         // 前缀映射 -> gemini-3-pro-high
+		"gemini-3-pro-preview", // 前缀映射 -> gemini-3-pro-high
+	}
+
+	for i, model := range geminiViaClaude {
+		if i > 0 {
+			time.Sleep(testInterval)
+		}
+		t.Run(model+"_通过Claude端点", func(t *testing.T) {
+			testClaudeMessage(t, model, false)
+		})
+		time.Sleep(testInterval)
+		t.Run(model+"_通过Claude端点_流式", func(t *testing.T) {
+			testClaudeMessage(t, model, true)
+		})
+	}
+}
+
 // TestClaudeMessagesWithNoSignature 测试历史 thinking block 不带 signature 的场景
 // 验证：Gemini 模型接受没有 signature 的 thinking block
 func TestClaudeMessagesWithNoSignature(t *testing.T) {
@@ -738,3 +770,30 @@ func testClaudeWithNoSignature(t *testing.T, model string) {
 	}
 	t.Logf("✅ 无 signature thinking 处理测试通过, id=%v", result["id"])
 }
+
+// TestGeminiEndpointWithClaudeModel 测试通过 Gemini 端点调用 Claude 模型
+// 仅在 Antigravity 模式下运行（ENDPOINT_PREFIX="/antigravity"）
+func TestGeminiEndpointWithClaudeModel(t *testing.T) {
+	if endpointPrefix != "/antigravity" {
+		t.Skip("仅在 Antigravity 模式下运行")
+	}
+
+	// 测试通过 Gemini 端点调用 Claude 模型
+	claudeViaGemini := []string{
+		"claude-sonnet-4-5",
+		"claude-opus-4-5-thinking",
+	}
+
+	for i, model := range claudeViaGemini {
+		if i > 0 {
+			time.Sleep(testInterval)
+		}
+		t.Run(model+"_通过Gemini端点", func(t *testing.T) {
+			testGeminiGenerate(t, model, false)
+		})
+		time.Sleep(testInterval)
+		t.Run(model+"_通过Gemini端点_流式", func(t *testing.T) {
+			testGeminiGenerate(t, model, true)
+		})
+	}
+}

backend/internal/pkg/antigravity/claude_types.go

@@ -37,12 +37,26 @@ type ClaudeMetadata struct {
 }
 
 // ClaudeTool Claude 工具定义
+// 支持两种格式：
+// 1. 标准格式: { "name": "...", "description": "...", "input_schema": {...} }
+// 2. Custom 格式 (MCP): { "type": "custom", "name": "...", "custom": { "description": "...", "input_schema": {...} } }
 type ClaudeTool struct {
+	Type        string          `json:"type,omitempty"` // "custom" 或空（标准格式）
 	Name        string          `json:"name"`
+	Description string          `json:"description,omitempty"`  // 标准格式使用
+	InputSchema map[string]any  `json:"input_schema,omitempty"` // 标准格式使用
+	Custom      *CustomToolSpec `json:"custom,omitempty"`       // custom 格式使用
+}
+
+// CustomToolSpec MCP custom 工具规格
+type CustomToolSpec struct {
 	Description string         `json:"description,omitempty"`
 	InputSchema map[string]any `json:"input_schema"`
 }
 
+// ClaudeCustomToolSpec 兼容旧命名（MCP custom 工具规格）
+type ClaudeCustomToolSpec = CustomToolSpec
+
 // SystemBlock system prompt 数组形式的元素
 type SystemBlock struct {
 	Type string `json:"type"`
@@ -124,3 +138,91 @@ type ErrorDetail struct {
 	Type    string `json:"type"`
 	Message string `json:"message"`
 }
+
+// modelDef Antigravity 模型定义（内部使用）
+type modelDef struct {
+	ID          string
+	DisplayName string
+	CreatedAt   string // 仅 Claude API 格式使用
+}
+
+// Antigravity 支持的 Claude 模型
+var claudeModels = []modelDef{
+	{ID: "claude-opus-4-5-thinking", DisplayName: "Claude Opus 4.5 Thinking", CreatedAt: "2025-11-01T00:00:00Z"},
+	{ID: "claude-sonnet-4-5", DisplayName: "Claude Sonnet 4.5", CreatedAt: "2025-09-29T00:00:00Z"},
+	{ID: "claude-sonnet-4-5-thinking", DisplayName: "Claude Sonnet 4.5 Thinking", CreatedAt: "2025-09-29T00:00:00Z"},
+}
+
+// Antigravity 支持的 Gemini 模型
+var geminiModels = []modelDef{
+	{ID: "gemini-2.5-flash", DisplayName: "Gemini 2.5 Flash", CreatedAt: "2025-01-01T00:00:00Z"},
+	{ID: "gemini-2.5-flash-lite", DisplayName: "Gemini 2.5 Flash Lite", CreatedAt: "2025-01-01T00:00:00Z"},
+	{ID: "gemini-2.5-flash-thinking", DisplayName: "Gemini 2.5 Flash Thinking", CreatedAt: "2025-01-01T00:00:00Z"},
+	{ID: "gemini-3-flash", DisplayName: "Gemini 3 Flash", CreatedAt: "2025-06-01T00:00:00Z"},
+	{ID: "gemini-3-pro-low", DisplayName: "Gemini 3 Pro Low", CreatedAt: "2025-06-01T00:00:00Z"},
+	{ID: "gemini-3-pro-high", DisplayName: "Gemini 3 Pro High", CreatedAt: "2025-06-01T00:00:00Z"},
+	{ID: "gemini-3-pro-preview", DisplayName: "Gemini 3 Pro Preview", CreatedAt: "2025-06-01T00:00:00Z"},
+	{ID: "gemini-3-pro-image", DisplayName: "Gemini 3 Pro Image", CreatedAt: "2025-06-01T00:00:00Z"},
+}
+
+// ========== Claude API 格式 (/v1/models) ==========
+
+// ClaudeModel Claude API 模型格式
+type ClaudeModel struct {
+	ID          string `json:"id"`
+	Type        string `json:"type"`
+	DisplayName string `json:"display_name"`
+	CreatedAt   string `json:"created_at"`
+}
+
+// DefaultModels 返回 Claude API 格式的模型列表（Claude + Gemini）
+func DefaultModels() []ClaudeModel {
+	all := append(claudeModels, geminiModels...)
+	result := make([]ClaudeModel, len(all))
+	for i, m := range all {
+		result[i] = ClaudeModel{ID: m.ID, Type: "model", DisplayName: m.DisplayName, CreatedAt: m.CreatedAt}
+	}
+	return result
+}
+
+// ========== Gemini v1beta 格式 (/v1beta/models) ==========
+
+// GeminiModel Gemini v1beta 模型格式
+type GeminiModel struct {
+	Name                       string   `json:"name"`
+	DisplayName                string   `json:"displayName,omitempty"`
+	SupportedGenerationMethods []string `json:"supportedGenerationMethods,omitempty"`
+}
+
+// GeminiModelsListResponse Gemini v1beta 模型列表响应
+type GeminiModelsListResponse struct {
+	Models []GeminiModel `json:"models"`
+}
+
+var defaultGeminiMethods = []string{"generateContent", "streamGenerateContent"}
+
+// DefaultGeminiModels 返回 Gemini v1beta 格式的模型列表（仅 Gemini 模型）
+func DefaultGeminiModels() []GeminiModel {
+	result := make([]GeminiModel, len(geminiModels))
+	for i, m := range geminiModels {
+		result[i] = GeminiModel{Name: "models/" + m.ID, DisplayName: m.DisplayName, SupportedGenerationMethods: defaultGeminiMethods}
+	}
+	return result
+}
+
+// FallbackGeminiModelsList 返回 Gemini v1beta 格式的模型列表响应
+func FallbackGeminiModelsList() GeminiModelsListResponse {
+	return GeminiModelsListResponse{Models: DefaultGeminiModels()}
+}
+
+// FallbackGeminiModel 返回单个模型信息（v1beta 格式）
+func FallbackGeminiModel(model string) GeminiModel {
+	if model == "" {
+		return GeminiModel{Name: "models/unknown", SupportedGenerationMethods: defaultGeminiMethods}
+	}
+	name := model
+	if len(model) < 7 || model[:7] != "models/" {
+		name = "models/" + model
+	}
+	return GeminiModel{Name: name, SupportedGenerationMethods: defaultGeminiMethods}
+}

backend/internal/pkg/antigravity/request_transformer.go

@@ -3,6 +3,7 @@ package antigravity
 import (
 	"encoding/json"
 	"fmt"
+	"log"
 	"strings"
 
 	"github.com/google/uuid"
@@ -13,13 +14,16 @@ func TransformClaudeToGemini(claudeReq *ClaudeRequest, projectID, mappedModel st
 	// 用于存储 tool_use id -> name 映射
 	toolIDToName := make(map[string]string)
 
-	// 检测是否启用 thinking
-	isThinkingEnabled := claudeReq.Thinking != nil && claudeReq.Thinking.Type == "enabled"
-
 	// 只有 Gemini 模型支持 dummy thought workaround
 	// Claude 模型通过 Vertex/Google API 需要有效的 thought signatures
 	allowDummyThought := strings.HasPrefix(mappedModel, "gemini-")
 
+	// 检测是否启用 thinking
+	requestedThinkingEnabled := claudeReq.Thinking != nil && claudeReq.Thinking.Type == "enabled"
+	// 为避免 Claude 模型的 thought signature/消息块约束导致 400（上游要求 thinking 块开头等），
+	// 非 Gemini 模型默认不启用 thinking（除非未来支持完整签名链路）。
+	isThinkingEnabled := requestedThinkingEnabled && allowDummyThought
+
 	// 1. 构建 contents
 	contents, err := buildContents(claudeReq.Messages, toolIDToName, isThinkingEnabled, allowDummyThought)
 	if err != nil {
@@ -30,7 +34,15 @@ func TransformClaudeToGemini(claudeReq *ClaudeRequest, projectID, mappedModel st
 	systemInstruction := buildSystemInstruction(claudeReq.System, claudeReq.Model)
 
 	// 3. 构建 generationConfig
-	generationConfig := buildGenerationConfig(claudeReq)
+	reqForGen := claudeReq
+	if requestedThinkingEnabled && !allowDummyThought {
+		log.Printf("[Warning] Disabling thinking for non-Gemini model in antigravity transform: model=%s", mappedModel)
+		// shallow copy to avoid mutating caller's request
+		clone := *claudeReq
+		clone.Thinking = nil
+		reqForGen = &clone
+	}
+	generationConfig := buildGenerationConfig(reqForGen)
 
 	// 4. 构建 tools
 	tools := buildTools(claudeReq.Tools)
@@ -149,6 +161,7 @@ func buildContents(messages []ClaudeMessage, toolIDToName map[string]string, isT
 				parts = append([]GeminiPart{{
 					Text:             "Thinking...",
 					Thought:          true,
+					ThoughtSignature: dummyThoughtSignature,
 				}}, parts...)
 			}
 		}
@@ -170,6 +183,34 @@ func buildContents(messages []ClaudeMessage, toolIDToName map[string]string, isT
 // 参考: https://ai.google.dev/gemini-api/docs/thought-signatures
 const dummyThoughtSignature = "skip_thought_signature_validator"
 
+// isValidThoughtSignature 验证 thought signature 是否有效
+// Claude API 要求 signature 必须是 base64 编码的字符串，长度至少 32 字节
+func isValidThoughtSignature(signature string) bool {
+	// 空字符串无效
+	if signature == "" {
+		return false
+	}
+
+	// signature 应该是 base64 编码，长度至少 40 个字符（约 30 字节）
+	// 参考 Claude API 文档和实际观察到的有效 signature
+	if len(signature) < 40 {
+		log.Printf("[Debug] Signature too short: len=%d", len(signature))
+		return false
+	}
+
+	// 检查是否是有效的 base64 字符
+	// base64 字符集: A-Z, a-z, 0-9, +, /, =
+	for i, c := range signature {
+		if (c < 'A' || c > 'Z') && (c < 'a' || c > 'z') &&
+			(c < '0' || c > '9') && c != '+' && c != '/' && c != '=' {
+			log.Printf("[Debug] Invalid base64 character at position %d: %c (code=%d)", i, c, c)
+			return false
+		}
+	}
+
+	return true
+}
+
 // buildParts 构建消息的 parts
 // allowDummyThought: 只有 Gemini 模型支持 dummy thought signature
 func buildParts(content json.RawMessage, toolIDToName map[string]string, allowDummyThought bool) ([]GeminiPart, error) {
@@ -198,15 +239,30 @@ func buildParts(content json.RawMessage, toolIDToName map[string]string, allowDu
 			}
 
 		case "thinking":
-			part := GeminiPart{
+			if allowDummyThought {
+				// Gemini 模型可以使用 dummy signature
+				parts = append(parts, GeminiPart{
 					Text:             block.Thinking,
 					Thought:          true,
+					ThoughtSignature: dummyThoughtSignature,
+				})
+				continue
 			}
-			// 保留原有 signature（Claude 模型需要有效的 signature）
-			if block.Signature != "" {
-				part.ThoughtSignature = block.Signature
+
+			// Claude 模型：仅在提供有效 signature 时保留 thinking block；否则跳过以避免上游校验失败。
+			signature := strings.TrimSpace(block.Signature)
+			if signature == "" || signature == dummyThoughtSignature {
+				log.Printf("[Warning] Skipping thinking block for Claude model (missing or dummy signature)")
+				continue
 			}
-			parts = append(parts, part)
+			if !isValidThoughtSignature(signature) {
+				log.Printf("[Debug] Thinking signature may be invalid (passing through anyway): len=%d", len(signature))
+			}
+			parts = append(parts, GeminiPart{
+				Text:             block.Thinking,
+				Thought:          true,
+				ThoughtSignature: signature,
+			})
 
 		case "image":
 			if block.Source != nil && block.Source.Type == "base64" {
@@ -231,10 +287,9 @@ func buildParts(content json.RawMessage, toolIDToName map[string]string, allowDu
 					ID:   block.ID,
 				},
 			}
-			// 保留原有 signature，或对 Gemini 模型使用 dummy signature
-			if block.Signature != "" {
-				part.ThoughtSignature = block.Signature
-			} else if allowDummyThought {
+			// 只有 Gemini 模型使用 dummy signature
+			// Claude 模型不设置 signature（避免验证问题）
+			if allowDummyThought {
 				part.ThoughtSignature = dummyThoughtSignature
 			}
 			parts = append(parts, part)
@@ -378,13 +433,53 @@ func buildTools(tools []ClaudeTool) []GeminiToolDeclaration {
 
 	// 普通工具
 	var funcDecls []GeminiFunctionDecl
-	for _, tool := range tools {
+	for i, tool := range tools {
+		// 跳过无效工具名称
+		if strings.TrimSpace(tool.Name) == "" {
+			log.Printf("Warning: skipping tool with empty name")
+			continue
+		}
+
+		var description string
+		var inputSchema map[string]any
+
+		// 检查是否为 custom 类型工具 (MCP)
+		if tool.Type == "custom" {
+			if tool.Custom == nil || tool.Custom.InputSchema == nil {
+				log.Printf("[Warning] Skipping invalid custom tool '%s': missing custom spec or input_schema", tool.Name)
+				continue
+			}
+			description = tool.Custom.Description
+			inputSchema = tool.Custom.InputSchema
+
+			// 调试日志：记录 custom 工具的 schema
+			if schemaJSON, err := json.Marshal(inputSchema); err == nil {
+				log.Printf("[Debug] Tool[%d] '%s' (custom) original schema: %s", i, tool.Name, string(schemaJSON))
+			}
+		} else {
+			// 标准格式: 从顶层字段获取
+			description = tool.Description
+			inputSchema = tool.InputSchema
+		}
+
 		// 清理 JSON Schema
-		params := cleanJSONSchema(tool.InputSchema)
+		params := cleanJSONSchema(inputSchema)
+		// 为 nil schema 提供默认值
+		if params == nil {
+			params = map[string]any{
+				"type":       "OBJECT",
+				"properties": map[string]any{},
+			}
+		}
+
+		// 调试日志：记录清理后的 schema
+		if paramsJSON, err := json.Marshal(params); err == nil {
+			log.Printf("[Debug] Tool[%d] '%s' cleaned schema: %s", i, tool.Name, string(paramsJSON))
+		}
 
 		funcDecls = append(funcDecls, GeminiFunctionDecl{
 			Name:        tool.Name,
-			Description: tool.Description,
+			Description: description,
 			Parameters:  params,
 		})
 	}
@@ -443,24 +538,54 @@ func cleanJSONSchema(schema map[string]any) map[string]any {
 }
 
 // excludedSchemaKeys 不支持的 schema 字段
+// 基于 Claude API (Vertex AI) 的实际支持情况
+// 支持: type, description, enum, properties, required, additionalProperties, items
+// 不支持: minItems, maxItems, minLength, maxLength, pattern, minimum, maximum 等验证字段
 var excludedSchemaKeys = map[string]bool{
+	// 元 schema 字段
 	"$schema": true,
 	"$id":     true,
 	"$ref":    true,
-	"additionalProperties": true,
+
+	// 字符串验证（Gemini 不支持）
 	"minLength": true,
 	"maxLength": true,
-	"minItems":             true,
-	"maxItems":             true,
-	"uniqueItems":          true,
+	"pattern":   true,
+
+	// 数字验证（Claude API 通过 Vertex AI 不支持这些字段）
 	"minimum":          true,
 	"maximum":          true,
 	"exclusiveMinimum": true,
 	"exclusiveMaximum": true,
-	"pattern":              true,
-	"format":               true,
+	"multipleOf":       true,
+
+	// 数组验证（Claude API 通过 Vertex AI 不支持这些字段）
+	"uniqueItems": true,
+	"minItems":    true,
+	"maxItems":    true,
+
+	// 组合 schema（Gemini 不支持）
+	"oneOf":       true,
+	"anyOf":       true,
+	"allOf":       true,
+	"not":         true,
+	"if":          true,
+	"then":        true,
+	"else":        true,
+	"$defs":       true,
+	"definitions": true,
+
+	// 对象验证（仅保留 properties/required/additionalProperties）
+	"minProperties":     true,
+	"maxProperties":     true,
+	"patternProperties": true,
+	"propertyNames":     true,
+	"dependencies":      true,
+	"dependentSchemas":  true,
+	"dependentRequired": true,
+
+	// 其他不支持的字段
 	"default":          true,
-	"strict":               true,
 	"const":            true,
 	"examples":         true,
 	"deprecated":       true,
@@ -468,6 +593,9 @@ var excludedSchemaKeys = map[string]bool{
 	"writeOnly":        true,
 	"contentMediaType": true,
 	"contentEncoding":  true,
+
+	// Claude 特有字段
+	"strict": true,
 }
 
 // cleanSchemaValue 递归清理 schema 值
@@ -487,6 +615,31 @@ func cleanSchemaValue(value any) any {
 				continue
 			}
 
+			// 特殊处理 format 字段：只保留 Gemini 支持的 format 值
+			if k == "format" {
+				if formatStr, ok := val.(string); ok {
+					// Gemini 只支持 date-time, date, time
+					if formatStr == "date-time" || formatStr == "date" || formatStr == "time" {
+						result[k] = val
+					}
+					// 其他 format 值直接跳过
+				}
+				continue
+			}
+
+			// 特殊处理 additionalProperties：Claude API 只支持布尔值，不支持 schema 对象
+			if k == "additionalProperties" {
+				if boolVal, ok := val.(bool); ok {
+					result[k] = boolVal
+					log.Printf("[Debug] additionalProperties is bool: %v", boolVal)
+				} else {
+					// 如果是 schema 对象，转换为 false（更安全的默认值）
+					result[k] = false
+					log.Printf("[Debug] additionalProperties is not bool (type: %T), converting to false", val)
+				}
+				continue
+			}
+
 			// 递归清理所有值
 			result[k] = cleanSchemaValue(val)
 		}

backend/internal/pkg/antigravity/request_transformer_test.go

+package antigravity
+
+import (
+	"encoding/json"
+	"testing"
+)
+
+// TestBuildParts_ThinkingBlockWithoutSignature 测试thinking block无signature时的处理
+func TestBuildParts_ThinkingBlockWithoutSignature(t *testing.T) {
+	tests := []struct {
+		name              string
+		content           string
+		allowDummyThought bool
+		expectedParts     int
+		description       string
+	}{
+		{
+			name: "Claude model - skip thinking block without signature",
+			content: `[
+				{"type": "text", "text": "Hello"},
+				{"type": "thinking", "thinking": "Let me think...", "signature": ""},
+				{"type": "text", "text": "World"}
+			]`,
+			allowDummyThought: false,
+			expectedParts:     2, // 只有两个text block
+			description:       "Claude模型应该跳过无signature的thinking block",
+		},
+		{
+			name: "Claude model - keep thinking block with signature",
+			content: `[
+				{"type": "text", "text": "Hello"},
+				{"type": "thinking", "thinking": "Let me think...", "signature": "valid_sig"},
+				{"type": "text", "text": "World"}
+			]`,
+			allowDummyThought: false,
+			expectedParts:     3, // 三个block都保留
+			description:       "Claude模型应该保留有signature的thinking block",
+		},
+		{
+			name: "Gemini model - use dummy signature",
+			content: `[
+				{"type": "text", "text": "Hello"},
+				{"type": "thinking", "thinking": "Let me think...", "signature": ""},
+				{"type": "text", "text": "World"}
+			]`,
+			allowDummyThought: true,
+			expectedParts:     3, // 三个block都保留，thinking使用dummy signature
+			description:       "Gemini模型应该为无signature的thinking block使用dummy signature",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			toolIDToName := make(map[string]string)
+			parts, err := buildParts(json.RawMessage(tt.content), toolIDToName, tt.allowDummyThought)
+
+			if err != nil {
+				t.Fatalf("buildParts() error = %v", err)
+			}
+
+			if len(parts) != tt.expectedParts {
+				t.Errorf("%s: got %d parts, want %d parts", tt.description, len(parts), tt.expectedParts)
+			}
+		})
+	}
+}
+
+// TestBuildTools_CustomTypeTools 测试custom类型工具转换
+func TestBuildTools_CustomTypeTools(t *testing.T) {
+	tests := []struct {
+		name        string
+		tools       []ClaudeTool
+		expectedLen int
+		description string
+	}{
+		{
+			name: "Standard tool format",
+			tools: []ClaudeTool{
+				{
+					Name:        "get_weather",
+					Description: "Get weather information",
+					InputSchema: map[string]any{
+						"type": "object",
+						"properties": map[string]any{
+							"location": map[string]any{"type": "string"},
+						},
+					},
+				},
+			},
+			expectedLen: 1,
+			description: "标准工具格式应该正常转换",
+		},
+		{
+			name: "Custom type tool (MCP format)",
+			tools: []ClaudeTool{
+				{
+					Type: "custom",
+					Name: "mcp_tool",
+					Custom: &CustomToolSpec{
+						Description: "MCP tool description",
+						InputSchema: map[string]any{
+							"type": "object",
+							"properties": map[string]any{
+								"param": map[string]any{"type": "string"},
+							},
+						},
+					},
+				},
+			},
+			expectedLen: 1,
+			description: "Custom类型工具应该从Custom字段读取description和input_schema",
+		},
+		{
+			name: "Mixed standard and custom tools",
+			tools: []ClaudeTool{
+				{
+					Name:        "standard_tool",
+					Description: "Standard tool",
+					InputSchema: map[string]any{"type": "object"},
+				},
+				{
+					Type: "custom",
+					Name: "custom_tool",
+					Custom: &CustomToolSpec{
+						Description: "Custom tool",
+						InputSchema: map[string]any{"type": "object"},
+					},
+				},
+			},
+			expectedLen: 1, // 返回一个GeminiToolDeclaration，包含2个function declarations
+			description: "混合标准和custom工具应该都能正确转换",
+		},
+		{
+			name: "Invalid custom tool - nil Custom field",
+			tools: []ClaudeTool{
+				{
+					Type: "custom",
+					Name: "invalid_custom",
+					// Custom 为 nil
+				},
+			},
+			expectedLen: 0, // 应该被跳过
+			description: "Custom字段为nil的custom工具应该被跳过",
+		},
+		{
+			name: "Invalid custom tool - nil InputSchema",
+			tools: []ClaudeTool{
+				{
+					Type: "custom",
+					Name: "invalid_custom",
+					Custom: &CustomToolSpec{
+						Description: "Invalid",
+						// InputSchema 为 nil
+					},
+				},
+			},
+			expectedLen: 0, // 应该被跳过
+			description: "InputSchema为nil的custom工具应该被跳过",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := buildTools(tt.tools)
+
+			if len(result) != tt.expectedLen {
+				t.Errorf("%s: got %d tool declarations, want %d", tt.description, len(result), tt.expectedLen)
+			}
+
+			// 验证function declarations存在
+			if len(result) > 0 && result[0].FunctionDeclarations != nil {
+				if len(result[0].FunctionDeclarations) != len(tt.tools) {
+					t.Errorf("%s: got %d function declarations, want %d",
+						tt.description, len(result[0].FunctionDeclarations), len(tt.tools))
+				}
+			}
+		})
+	}
+}

backend/internal/pkg/claude/constants.go

@@ -16,6 +16,12 @@ const DefaultBetaHeader = BetaClaudeCode + "," + BetaOAuth + "," + BetaInterleav
 // HaikuBetaHeader Haiku 模型使用的 anthropic-beta header（不需要 claude-code beta）
 const HaikuBetaHeader = BetaOAuth + "," + BetaInterleavedThinking
 
+// ApiKeyBetaHeader API-key 账号建议使用的 anthropic-beta header（不包含 oauth）
+const ApiKeyBetaHeader = BetaClaudeCode + "," + BetaInterleavedThinking + "," + BetaFineGrainedToolStreaming
+
+// ApiKeyHaikuBetaHeader Haiku 模型在 API-key 账号下使用的 anthropic-beta header（不包含 oauth / claude-code）
+const ApiKeyHaikuBetaHeader = BetaInterleavedThinking
+
 // Claude Code 客户端默认请求头
 var DefaultHeaders = map[string]string{
 	"User-Agent":                                "claude-cli/2.0.62 (external, cli)",