Merge branch 'main' into test

backend/ent/errorpassthroughrule.go

@@ -44,6 +44,8 @@ type ErrorPassthroughRule struct {
 	PassthroughBody bool `json:"passthrough_body,omitempty"`
 	// CustomMessage holds the value of the "custom_message" field.
 	CustomMessage *string `json:"custom_message,omitempty"`
+	// SkipMonitoring holds the value of the "skip_monitoring" field.
+	SkipMonitoring bool `json:"skip_monitoring,omitempty"`
 	// Description holds the value of the "description" field.
 	Description  *string `json:"description,omitempty"`
 	selectValues sql.SelectValues
@@ -56,7 +58,7 @@ func (*ErrorPassthroughRule) scanValues(columns []string) ([]any, error) {
 		switch columns[i] {
 		case errorpassthroughrule.FieldErrorCodes, errorpassthroughrule.FieldKeywords, errorpassthroughrule.FieldPlatforms:
 			values[i] = new([]byte)
-		case errorpassthroughrule.FieldEnabled, errorpassthroughrule.FieldPassthroughCode, errorpassthroughrule.FieldPassthroughBody:
+		case errorpassthroughrule.FieldEnabled, errorpassthroughrule.FieldPassthroughCode, errorpassthroughrule.FieldPassthroughBody, errorpassthroughrule.FieldSkipMonitoring:
 			values[i] = new(sql.NullBool)
 		case errorpassthroughrule.FieldID, errorpassthroughrule.FieldPriority, errorpassthroughrule.FieldResponseCode:
 			values[i] = new(sql.NullInt64)
@@ -171,6 +173,12 @@ func (_m *ErrorPassthroughRule) assignValues(columns []string, values []any) err
 				_m.CustomMessage = new(string)
 				*_m.CustomMessage = value.String
 			}
+		case errorpassthroughrule.FieldSkipMonitoring:
+			if value, ok := values[i].(*sql.NullBool); !ok {
+				return fmt.Errorf("unexpected type %T for field skip_monitoring", values[i])
+			} else if value.Valid {
+				_m.SkipMonitoring = value.Bool
+			}
 		case errorpassthroughrule.FieldDescription:
 			if value, ok := values[i].(*sql.NullString); !ok {
 				return fmt.Errorf("unexpected type %T for field description", values[i])
@@ -257,6 +265,9 @@ func (_m *ErrorPassthroughRule) String() string {
 		builder.WriteString(*v)
 	}
 	builder.WriteString(", ")
+	builder.WriteString("skip_monitoring=")
+	builder.WriteString(fmt.Sprintf("%v", _m.SkipMonitoring))
+	builder.WriteString(", ")
 	if v := _m.Description; v != nil {
 		builder.WriteString("description=")
 		builder.WriteString(*v)

backend/ent/errorpassthroughrule/errorpassthroughrule.go

@@ -39,6 +39,8 @@ const (
 	FieldPassthroughBody = "passthrough_body"
 	// FieldCustomMessage holds the string denoting the custom_message field in the database.
 	FieldCustomMessage = "custom_message"
+	// FieldSkipMonitoring holds the string denoting the skip_monitoring field in the database.
+	FieldSkipMonitoring = "skip_monitoring"
 	// FieldDescription holds the string denoting the description field in the database.
 	FieldDescription = "description"
 	// Table holds the table name of the errorpassthroughrule in the database.
@@ -61,6 +63,7 @@ var Columns = []string{
 	FieldResponseCode,
 	FieldPassthroughBody,
 	FieldCustomMessage,
+	FieldSkipMonitoring,
 	FieldDescription,
 }
 
@@ -95,6 +98,8 @@ var (
 	DefaultPassthroughCode bool
 	// DefaultPassthroughBody holds the default value on creation for the "passthrough_body" field.
 	DefaultPassthroughBody bool
+	// DefaultSkipMonitoring holds the default value on creation for the "skip_monitoring" field.
+	DefaultSkipMonitoring bool
 )
 
 // OrderOption defines the ordering options for the ErrorPassthroughRule queries.
@@ -155,6 +160,11 @@ func ByCustomMessage(opts ...sql.OrderTermOption) OrderOption {
 	return sql.OrderByField(FieldCustomMessage, opts...).ToFunc()
 }
 
+// BySkipMonitoring orders the results by the skip_monitoring field.
+func BySkipMonitoring(opts ...sql.OrderTermOption) OrderOption {
+	return sql.OrderByField(FieldSkipMonitoring, opts...).ToFunc()
+}
+
 // ByDescription orders the results by the description field.
 func ByDescription(opts ...sql.OrderTermOption) OrderOption {
 	return sql.OrderByField(FieldDescription, opts...).ToFunc()

backend/ent/errorpassthroughrule/where.go

@@ -104,6 +104,11 @@ func CustomMessage(v string) predicate.ErrorPassthroughRule {
 	return predicate.ErrorPassthroughRule(sql.FieldEQ(FieldCustomMessage, v))
 }
 
+// SkipMonitoring applies equality check predicate on the "skip_monitoring" field. It's identical to SkipMonitoringEQ.
+func SkipMonitoring(v bool) predicate.ErrorPassthroughRule {
+	return predicate.ErrorPassthroughRule(sql.FieldEQ(FieldSkipMonitoring, v))
+}
+
 // Description applies equality check predicate on the "description" field. It's identical to DescriptionEQ.
 func Description(v string) predicate.ErrorPassthroughRule {
 	return predicate.ErrorPassthroughRule(sql.FieldEQ(FieldDescription, v))
@@ -544,6 +549,16 @@ func CustomMessageContainsFold(v string) predicate.ErrorPassthroughRule {
 	return predicate.ErrorPassthroughRule(sql.FieldContainsFold(FieldCustomMessage, v))
 }
 
+// SkipMonitoringEQ applies the EQ predicate on the "skip_monitoring" field.
+func SkipMonitoringEQ(v bool) predicate.ErrorPassthroughRule {
+	return predicate.ErrorPassthroughRule(sql.FieldEQ(FieldSkipMonitoring, v))
+}
+
+// SkipMonitoringNEQ applies the NEQ predicate on the "skip_monitoring" field.
+func SkipMonitoringNEQ(v bool) predicate.ErrorPassthroughRule {
+	return predicate.ErrorPassthroughRule(sql.FieldNEQ(FieldSkipMonitoring, v))
+}
+
 // DescriptionEQ applies the EQ predicate on the "description" field.
 func DescriptionEQ(v string) predicate.ErrorPassthroughRule {
 	return predicate.ErrorPassthroughRule(sql.FieldEQ(FieldDescription, v))

backend/ent/errorpassthroughrule_create.go

@@ -172,6 +172,20 @@ func (_c *ErrorPassthroughRuleCreate) SetNillableCustomMessage(v *string) *Error
 	return _c
 }
 
+// SetSkipMonitoring sets the "skip_monitoring" field.
+func (_c *ErrorPassthroughRuleCreate) SetSkipMonitoring(v bool) *ErrorPassthroughRuleCreate {
+	_c.mutation.SetSkipMonitoring(v)
+	return _c
+}
+
+// SetNillableSkipMonitoring sets the "skip_monitoring" field if the given value is not nil.
+func (_c *ErrorPassthroughRuleCreate) SetNillableSkipMonitoring(v *bool) *ErrorPassthroughRuleCreate {
+	if v != nil {
+		_c.SetSkipMonitoring(*v)
+	}
+	return _c
+}
+
 // SetDescription sets the "description" field.
 func (_c *ErrorPassthroughRuleCreate) SetDescription(v string) *ErrorPassthroughRuleCreate {
 	_c.mutation.SetDescription(v)
@@ -249,6 +263,10 @@ func (_c *ErrorPassthroughRuleCreate) defaults() {
 		v := errorpassthroughrule.DefaultPassthroughBody
 		_c.mutation.SetPassthroughBody(v)
 	}
+	if _, ok := _c.mutation.SkipMonitoring(); !ok {
+		v := errorpassthroughrule.DefaultSkipMonitoring
+		_c.mutation.SetSkipMonitoring(v)
+	}
 }
 
 // check runs all checks and user-defined validators on the builder.
@@ -287,6 +305,9 @@ func (_c *ErrorPassthroughRuleCreate) check() error {
 	if _, ok := _c.mutation.PassthroughBody(); !ok {
 		return &ValidationError{Name: "passthrough_body", err: errors.New(`ent: missing required field "ErrorPassthroughRule.passthrough_body"`)}
 	}
+	if _, ok := _c.mutation.SkipMonitoring(); !ok {
+		return &ValidationError{Name: "skip_monitoring", err: errors.New(`ent: missing required field "ErrorPassthroughRule.skip_monitoring"`)}
+	}
 	return nil
 }
 
@@ -366,6 +387,10 @@ func (_c *ErrorPassthroughRuleCreate) createSpec() (*ErrorPassthroughRule, *sqlg
 		_spec.SetField(errorpassthroughrule.FieldCustomMessage, field.TypeString, value)
 		_node.CustomMessage = &value
 	}
+	if value, ok := _c.mutation.SkipMonitoring(); ok {
+		_spec.SetField(errorpassthroughrule.FieldSkipMonitoring, field.TypeBool, value)
+		_node.SkipMonitoring = value
+	}
 	if value, ok := _c.mutation.Description(); ok {
 		_spec.SetField(errorpassthroughrule.FieldDescription, field.TypeString, value)
 		_node.Description = &value
@@ -608,6 +633,18 @@ func (u *ErrorPassthroughRuleUpsert) ClearCustomMessage() *ErrorPassthroughRuleU
 	return u
 }
 
+// SetSkipMonitoring sets the "skip_monitoring" field.
+func (u *ErrorPassthroughRuleUpsert) SetSkipMonitoring(v bool) *ErrorPassthroughRuleUpsert {
+	u.Set(errorpassthroughrule.FieldSkipMonitoring, v)
+	return u
+}
+
+// UpdateSkipMonitoring sets the "skip_monitoring" field to the value that was provided on create.
+func (u *ErrorPassthroughRuleUpsert) UpdateSkipMonitoring() *ErrorPassthroughRuleUpsert {
+	u.SetExcluded(errorpassthroughrule.FieldSkipMonitoring)
+	return u
+}
+
 // SetDescription sets the "description" field.
 func (u *ErrorPassthroughRuleUpsert) SetDescription(v string) *ErrorPassthroughRuleUpsert {
 	u.Set(errorpassthroughrule.FieldDescription, v)
@@ -888,6 +925,20 @@ func (u *ErrorPassthroughRuleUpsertOne) ClearCustomMessage() *ErrorPassthroughRu
 	})
 }
 
+// SetSkipMonitoring sets the "skip_monitoring" field.
+func (u *ErrorPassthroughRuleUpsertOne) SetSkipMonitoring(v bool) *ErrorPassthroughRuleUpsertOne {
+	return u.Update(func(s *ErrorPassthroughRuleUpsert) {
+		s.SetSkipMonitoring(v)
+	})
+}
+
+// UpdateSkipMonitoring sets the "skip_monitoring" field to the value that was provided on create.
+func (u *ErrorPassthroughRuleUpsertOne) UpdateSkipMonitoring() *ErrorPassthroughRuleUpsertOne {
+	return u.Update(func(s *ErrorPassthroughRuleUpsert) {
+		s.UpdateSkipMonitoring()
+	})
+}
+
 // SetDescription sets the "description" field.
 func (u *ErrorPassthroughRuleUpsertOne) SetDescription(v string) *ErrorPassthroughRuleUpsertOne {
 	return u.Update(func(s *ErrorPassthroughRuleUpsert) {
@@ -1337,6 +1388,20 @@ func (u *ErrorPassthroughRuleUpsertBulk) ClearCustomMessage() *ErrorPassthroughR
 	})
 }
 
+// SetSkipMonitoring sets the "skip_monitoring" field.
+func (u *ErrorPassthroughRuleUpsertBulk) SetSkipMonitoring(v bool) *ErrorPassthroughRuleUpsertBulk {
+	return u.Update(func(s *ErrorPassthroughRuleUpsert) {
+		s.SetSkipMonitoring(v)
+	})
+}
+
+// UpdateSkipMonitoring sets the "skip_monitoring" field to the value that was provided on create.
+func (u *ErrorPassthroughRuleUpsertBulk) UpdateSkipMonitoring() *ErrorPassthroughRuleUpsertBulk {
+	return u.Update(func(s *ErrorPassthroughRuleUpsert) {
+		s.UpdateSkipMonitoring()
+	})
+}
+
 // SetDescription sets the "description" field.
 func (u *ErrorPassthroughRuleUpsertBulk) SetDescription(v string) *ErrorPassthroughRuleUpsertBulk {
 	return u.Update(func(s *ErrorPassthroughRuleUpsert) {

backend/ent/errorpassthroughrule_update.go

@@ -227,6 +227,20 @@ func (_u *ErrorPassthroughRuleUpdate) ClearCustomMessage() *ErrorPassthroughRule
 	return _u
 }
 
+// SetSkipMonitoring sets the "skip_monitoring" field.
+func (_u *ErrorPassthroughRuleUpdate) SetSkipMonitoring(v bool) *ErrorPassthroughRuleUpdate {
+	_u.mutation.SetSkipMonitoring(v)
+	return _u
+}
+
+// SetNillableSkipMonitoring sets the "skip_monitoring" field if the given value is not nil.
+func (_u *ErrorPassthroughRuleUpdate) SetNillableSkipMonitoring(v *bool) *ErrorPassthroughRuleUpdate {
+	if v != nil {
+		_u.SetSkipMonitoring(*v)
+	}
+	return _u
+}
+
 // SetDescription sets the "description" field.
 func (_u *ErrorPassthroughRuleUpdate) SetDescription(v string) *ErrorPassthroughRuleUpdate {
 	_u.mutation.SetDescription(v)
@@ -387,6 +401,9 @@ func (_u *ErrorPassthroughRuleUpdate) sqlSave(ctx context.Context) (_node int, e
 	if _u.mutation.CustomMessageCleared() {
 		_spec.ClearField(errorpassthroughrule.FieldCustomMessage, field.TypeString)
 	}
+	if value, ok := _u.mutation.SkipMonitoring(); ok {
+		_spec.SetField(errorpassthroughrule.FieldSkipMonitoring, field.TypeBool, value)
+	}
 	if value, ok := _u.mutation.Description(); ok {
 		_spec.SetField(errorpassthroughrule.FieldDescription, field.TypeString, value)
 	}
@@ -611,6 +628,20 @@ func (_u *ErrorPassthroughRuleUpdateOne) ClearCustomMessage() *ErrorPassthroughR
 	return _u
 }
 
+// SetSkipMonitoring sets the "skip_monitoring" field.
+func (_u *ErrorPassthroughRuleUpdateOne) SetSkipMonitoring(v bool) *ErrorPassthroughRuleUpdateOne {
+	_u.mutation.SetSkipMonitoring(v)
+	return _u
+}
+
+// SetNillableSkipMonitoring sets the "skip_monitoring" field if the given value is not nil.
+func (_u *ErrorPassthroughRuleUpdateOne) SetNillableSkipMonitoring(v *bool) *ErrorPassthroughRuleUpdateOne {
+	if v != nil {
+		_u.SetSkipMonitoring(*v)
+	}
+	return _u
+}
+
 // SetDescription sets the "description" field.
 func (_u *ErrorPassthroughRuleUpdateOne) SetDescription(v string) *ErrorPassthroughRuleUpdateOne {
 	_u.mutation.SetDescription(v)
@@ -801,6 +832,9 @@ func (_u *ErrorPassthroughRuleUpdateOne) sqlSave(ctx context.Context) (_node *Er
 	if _u.mutation.CustomMessageCleared() {
 		_spec.ClearField(errorpassthroughrule.FieldCustomMessage, field.TypeString)
 	}
+	if value, ok := _u.mutation.SkipMonitoring(); ok {
+		_spec.SetField(errorpassthroughrule.FieldSkipMonitoring, field.TypeBool, value)
+	}
 	if value, ok := _u.mutation.Description(); ok {
 		_spec.SetField(errorpassthroughrule.FieldDescription, field.TypeString, value)
 	}

backend/ent/migrate/schema.go

@@ -325,6 +325,7 @@ var (
 		{Name: "response_code", Type: field.TypeInt, Nullable: true},
 		{Name: "passthrough_body", Type: field.TypeBool, Default: true},
 		{Name: "custom_message", Type: field.TypeString, Nullable: true, Size: 2147483647},
+		{Name: "skip_monitoring", Type: field.TypeBool, Default: false},
 		{Name: "description", Type: field.TypeString, Nullable: true, Size: 2147483647},
 	}
 	// ErrorPassthroughRulesTable holds the schema information for the "error_passthrough_rules" table.

backend/ent/mutation.go

@@ -5778,6 +5778,7 @@ type ErrorPassthroughRuleMutation struct {
 	addresponse_code  *int
 	passthrough_body  *bool
 	custom_message    *string
+	skip_monitoring   *bool
 	description       *string
 	clearedFields     map[string]struct{}
 	done              bool
@@ -6505,6 +6506,42 @@ func (m *ErrorPassthroughRuleMutation) ResetCustomMessage() {
 	delete(m.clearedFields, errorpassthroughrule.FieldCustomMessage)
 }
 
+// SetSkipMonitoring sets the "skip_monitoring" field.
+func (m *ErrorPassthroughRuleMutation) SetSkipMonitoring(b bool) {
+	m.skip_monitoring = &b
+}
+
+// SkipMonitoring returns the value of the "skip_monitoring" field in the mutation.
+func (m *ErrorPassthroughRuleMutation) SkipMonitoring() (r bool, exists bool) {
+	v := m.skip_monitoring
+	if v == nil {
+		return
+	}
+	return *v, true
+}
+
+// OldSkipMonitoring returns the old "skip_monitoring" field's value of the ErrorPassthroughRule entity.
+// If the ErrorPassthroughRule object wasn't provided to the builder, the object is fetched from the database.
+// An error is returned if the mutation operation is not UpdateOne, or the database query fails.
+func (m *ErrorPassthroughRuleMutation) OldSkipMonitoring(ctx context.Context) (v bool, err error) {
+	if !m.op.Is(OpUpdateOne) {
+		return v, errors.New("OldSkipMonitoring is only allowed on UpdateOne operations")
+	}
+	if m.id == nil || m.oldValue == nil {
+		return v, errors.New("OldSkipMonitoring requires an ID field in the mutation")
+	}
+	oldValue, err := m.oldValue(ctx)
+	if err != nil {
+		return v, fmt.Errorf("querying old value for OldSkipMonitoring: %w", err)
+	}
+	return oldValue.SkipMonitoring, nil
+}
+
+// ResetSkipMonitoring resets all changes to the "skip_monitoring" field.
+func (m *ErrorPassthroughRuleMutation) ResetSkipMonitoring() {
+	m.skip_monitoring = nil
+}
+
 // SetDescription sets the "description" field.
 func (m *ErrorPassthroughRuleMutation) SetDescription(s string) {
 	m.description = &s
@@ -6588,7 +6625,7 @@ func (m *ErrorPassthroughRuleMutation) Type() string {
 // order to get all numeric fields that were incremented/decremented, call
 // AddedFields().
 func (m *ErrorPassthroughRuleMutation) Fields() []string {
-	fields := make([]string, 0, 14)
+	fields := make([]string, 0, 15)
 	if m.created_at != nil {
 		fields = append(fields, errorpassthroughrule.FieldCreatedAt)
 	}
@@ -6628,6 +6665,9 @@ func (m *ErrorPassthroughRuleMutation) Fields() []string {
 	if m.custom_message != nil {
 		fields = append(fields, errorpassthroughrule.FieldCustomMessage)
 	}
+	if m.skip_monitoring != nil {
+		fields = append(fields, errorpassthroughrule.FieldSkipMonitoring)
+	}
 	if m.description != nil {
 		fields = append(fields, errorpassthroughrule.FieldDescription)
 	}
@@ -6665,6 +6705,8 @@ func (m *ErrorPassthroughRuleMutation) Field(name string) (ent.Value, bool) {
 		return m.PassthroughBody()
 	case errorpassthroughrule.FieldCustomMessage:
 		return m.CustomMessage()
+	case errorpassthroughrule.FieldSkipMonitoring:
+		return m.SkipMonitoring()
 	case errorpassthroughrule.FieldDescription:
 		return m.Description()
 	}
@@ -6702,6 +6744,8 @@ func (m *ErrorPassthroughRuleMutation) OldField(ctx context.Context, name string
 		return m.OldPassthroughBody(ctx)
 	case errorpassthroughrule.FieldCustomMessage:
 		return m.OldCustomMessage(ctx)
+	case errorpassthroughrule.FieldSkipMonitoring:
+		return m.OldSkipMonitoring(ctx)
 	case errorpassthroughrule.FieldDescription:
 		return m.OldDescription(ctx)
 	}
@@ -6804,6 +6848,13 @@ func (m *ErrorPassthroughRuleMutation) SetField(name string, value ent.Value) er
 		}
 		m.SetCustomMessage(v)
 		return nil
+	case errorpassthroughrule.FieldSkipMonitoring:
+		v, ok := value.(bool)
+		if !ok {
+			return fmt.Errorf("unexpected type %T for field %s", value, name)
+		}
+		m.SetSkipMonitoring(v)
+		return nil
 	case errorpassthroughrule.FieldDescription:
 		v, ok := value.(string)
 		if !ok {
@@ -6965,6 +7016,9 @@ func (m *ErrorPassthroughRuleMutation) ResetField(name string) error {
 	case errorpassthroughrule.FieldCustomMessage:
 		m.ResetCustomMessage()
 		return nil
+	case errorpassthroughrule.FieldSkipMonitoring:
+		m.ResetSkipMonitoring()
+		return nil
 	case errorpassthroughrule.FieldDescription:
 		m.ResetDescription()
 		return nil

backend/ent/runtime/runtime.go

@@ -327,6 +327,10 @@ func init() {
 	errorpassthroughruleDescPassthroughBody := errorpassthroughruleFields[9].Descriptor()
 	// errorpassthroughrule.DefaultPassthroughBody holds the default value on creation for the passthrough_body field.
 	errorpassthroughrule.DefaultPassthroughBody = errorpassthroughruleDescPassthroughBody.Default.(bool)
+	// errorpassthroughruleDescSkipMonitoring is the schema descriptor for skip_monitoring field.
+	errorpassthroughruleDescSkipMonitoring := errorpassthroughruleFields[11].Descriptor()
+	// errorpassthroughrule.DefaultSkipMonitoring holds the default value on creation for the skip_monitoring field.
+	errorpassthroughrule.DefaultSkipMonitoring = errorpassthroughruleDescSkipMonitoring.Default.(bool)
 	groupMixin := schema.Group{}.Mixin()
 	groupMixinHooks1 := groupMixin[1].Hooks()
 	group.Hooks[0] = groupMixinHooks1[0]

backend/ent/schema/error_passthrough_rule.go

@@ -105,6 +105,12 @@ func (ErrorPassthroughRule) Fields() []ent.Field {
 			Optional().
 			Nillable(),
 
+		// skip_monitoring: 是否跳过运维监控记录
+		// true: 匹配此规则的错误不会被记录到 ops_error_logs
+		// false: 正常记录到运维监控（默认行为）
+		field.Bool("skip_monitoring").
+			Default(false),
+
 		// description: 规则描述，用于说明规则的用途
 		field.Text("description").
 			Optional().

backend/internal/handler/admin/antigravity_oauth_handler.go

@@ -65,3 +65,27 @@ func (h *AntigravityOAuthHandler) ExchangeCode(c *gin.Context) {
 
 	response.Success(c, tokenInfo)
 }
+
+// AntigravityRefreshTokenRequest represents the request for validating Antigravity refresh token
+type AntigravityRefreshTokenRequest struct {
+	RefreshToken string `json:"refresh_token" binding:"required"`
+	ProxyID      *int64 `json:"proxy_id"`
+}
+
+// RefreshToken validates an Antigravity refresh token and returns full token info
+// POST /api/v1/admin/antigravity/oauth/refresh-token
+func (h *AntigravityOAuthHandler) RefreshToken(c *gin.Context) {
+	var req AntigravityRefreshTokenRequest
+	if err := c.ShouldBindJSON(&req); err != nil {
+		response.BadRequest(c, "请求无效: "+err.Error())
+		return
+	}
+
+	tokenInfo, err := h.antigravityOAuthService.ValidateRefreshToken(c.Request.Context(), req.RefreshToken, req.ProxyID)
+	if err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+
+	response.Success(c, tokenInfo)
+}

backend/internal/handler/admin/error_passthrough_handler.go

@@ -32,6 +32,7 @@ type CreateErrorPassthroughRuleRequest struct {
 	ResponseCode    *int     `json:"response_code"`
 	PassthroughBody *bool    `json:"passthrough_body"`
 	CustomMessage   *string  `json:"custom_message"`
+	SkipMonitoring  *bool    `json:"skip_monitoring"`
 	Description     *string  `json:"description"`
 }
 
@@ -48,6 +49,7 @@ type UpdateErrorPassthroughRuleRequest struct {
 	ResponseCode    *int     `json:"response_code"`
 	PassthroughBody *bool    `json:"passthrough_body"`
 	CustomMessage   *string  `json:"custom_message"`
+	SkipMonitoring  *bool    `json:"skip_monitoring"`
 	Description     *string  `json:"description"`
 }
 
@@ -122,6 +124,9 @@ func (h *ErrorPassthroughHandler) Create(c *gin.Context) {
 	} else {
 		rule.PassthroughBody = true
 	}
+	if req.SkipMonitoring != nil {
+		rule.SkipMonitoring = *req.SkipMonitoring
+	}
 	rule.ResponseCode = req.ResponseCode
 	rule.CustomMessage = req.CustomMessage
 	rule.Description = req.Description
@@ -190,6 +195,7 @@ func (h *ErrorPassthroughHandler) Update(c *gin.Context) {
 		ResponseCode:    existing.ResponseCode,
 		PassthroughBody: existing.PassthroughBody,
 		CustomMessage:   existing.CustomMessage,
+		SkipMonitoring:  existing.SkipMonitoring,
 		Description:     existing.Description,
 	}
 
@@ -230,6 +236,9 @@ func (h *ErrorPassthroughHandler) Update(c *gin.Context) {
 	if req.Description != nil {
 		rule.Description = req.Description
 	}
+	if req.SkipMonitoring != nil {
+		rule.SkipMonitoring = *req.SkipMonitoring
+	}
 
 	// 确保切片不为 nil
 	if rule.ErrorCodes == nil {

backend/internal/handler/admin/redeem_handler.go

@@ -202,7 +202,7 @@ func (h *RedeemHandler) Export(c *gin.Context) {
 	writer := csv.NewWriter(&buf)
 
 	// Write header
-	if err := writer.Write([]string{"id", "code", "type", "value", "status", "used_by", "used_at", "created_at"}); err != nil {
+	if err := writer.Write([]string{"id", "code", "type", "value", "status", "used_by", "used_by_email", "used_at", "created_at"}); err != nil {
 		response.InternalError(c, "Failed to export redeem codes: "+err.Error())
 		return
 	}
@@ -213,6 +213,10 @@ func (h *RedeemHandler) Export(c *gin.Context) {
 		if code.UsedBy != nil {
 			usedBy = fmt.Sprintf("%d", *code.UsedBy)
 		}
+		usedByEmail := ""
+		if code.User != nil {
+			usedByEmail = code.User.Email
+		}
 		usedAt := ""
 		if code.UsedAt != nil {
 			usedAt = code.UsedAt.Format("2006-01-02 15:04:05")
@@ -224,6 +228,7 @@ func (h *RedeemHandler) Export(c *gin.Context) {
 			fmt.Sprintf("%.2f", code.Value),
 			code.Status,
 			usedBy,
+			usedByEmail,
 			usedAt,
 			code.CreatedAt.Format("2006-01-02 15:04:05"),
 		}); err != nil {

backend/internal/handler/gateway_handler.go

@@ -7,6 +7,7 @@ import (
 	"errors"
 	"fmt"
 	"io"
+	"log"
 	"net/http"
 	"strings"
 	"time"
@@ -247,6 +248,7 @@ func (h *GatewayHandler) Messages(c *gin.Context) {
 		maxAccountSwitches := h.maxAccountSwitchesGemini
 		switchCount := 0
 		failedAccountIDs := make(map[int64]struct{})
+		sameAccountRetryCount := make(map[int64]int) // 同账号重试计数
 		var lastFailoverErr *service.UpstreamFailoverError
 		var forceCacheBilling bool // 粘性会话切换时的缓存计费标记
 
@@ -376,11 +378,28 @@ func (h *GatewayHandler) Messages(c *gin.Context) {
 			if err != nil {
 				var failoverErr *service.UpstreamFailoverError
 				if errors.As(err, &failoverErr) {
-					failedAccountIDs[account.ID] = struct{}{}
 					lastFailoverErr = failoverErr
 					if needForceCacheBilling(hasBoundSession, failoverErr) {
 						forceCacheBilling = true
 					}
+
+					// 同账号重试：对 RetryableOnSameAccount 的临时性错误，先在同一账号上重试
+					if failoverErr.RetryableOnSameAccount && sameAccountRetryCount[account.ID] < maxSameAccountRetries {
+						sameAccountRetryCount[account.ID]++
+						log.Printf("Account %d: retryable error %d, same-account retry %d/%d",
+							account.ID, failoverErr.StatusCode, sameAccountRetryCount[account.ID], maxSameAccountRetries)
+						if !sleepSameAccountRetryDelay(c.Request.Context()) {
+							return
+						}
+						continue
+					}
+
+					// 同账号重试用尽，执行临时封禁并切换账号
+					if failoverErr.RetryableOnSameAccount {
+						h.gatewayService.TempUnscheduleRetryableError(c.Request.Context(), account.ID, failoverErr)
+					}
+
+					failedAccountIDs[account.ID] = struct{}{}
 					if switchCount >= maxAccountSwitches {
 						h.handleFailoverExhausted(c, failoverErr, service.PlatformGemini, streamStarted)
 						return
@@ -456,6 +475,7 @@ func (h *GatewayHandler) Messages(c *gin.Context) {
 		maxAccountSwitches := h.maxAccountSwitches
 		switchCount := 0
 		failedAccountIDs := make(map[int64]struct{})
+		sameAccountRetryCount := make(map[int64]int) // 同账号重试计数
 		var lastFailoverErr *service.UpstreamFailoverError
 		retryWithFallback := false
 		var forceCacheBilling bool // 粘性会话切换时的缓存计费标记
@@ -623,11 +643,28 @@ func (h *GatewayHandler) Messages(c *gin.Context) {
 				}
 				var failoverErr *service.UpstreamFailoverError
 				if errors.As(err, &failoverErr) {
-					failedAccountIDs[account.ID] = struct{}{}
 					lastFailoverErr = failoverErr
 					if needForceCacheBilling(hasBoundSession, failoverErr) {
 						forceCacheBilling = true
 					}
+
+					// 同账号重试：对 RetryableOnSameAccount 的临时性错误，先在同一账号上重试
+					if failoverErr.RetryableOnSameAccount && sameAccountRetryCount[account.ID] < maxSameAccountRetries {
+						sameAccountRetryCount[account.ID]++
+						log.Printf("Account %d: retryable error %d, same-account retry %d/%d",
+							account.ID, failoverErr.StatusCode, sameAccountRetryCount[account.ID], maxSameAccountRetries)
+						if !sleepSameAccountRetryDelay(c.Request.Context()) {
+							return
+						}
+						continue
+					}
+
+					// 同账号重试用尽，执行临时封禁并切换账号
+					if failoverErr.RetryableOnSameAccount {
+						h.gatewayService.TempUnscheduleRetryableError(c.Request.Context(), account.ID, failoverErr)
+					}
+
+					failedAccountIDs[account.ID] = struct{}{}
 					if switchCount >= maxAccountSwitches {
 						h.handleFailoverExhausted(c, failoverErr, account.Platform, streamStarted)
 						return
@@ -935,6 +972,23 @@ func needForceCacheBilling(hasBoundSession bool, failoverErr *service.UpstreamFa
 	return hasBoundSession || (failoverErr != nil && failoverErr.ForceCacheBilling)
 }
 
+const (
+	// maxSameAccountRetries 同账号重试次数上限（针对 RetryableOnSameAccount 错误）
+	maxSameAccountRetries = 2
+	// sameAccountRetryDelay 同账号重试间隔
+	sameAccountRetryDelay = 500 * time.Millisecond
+)
+
+// sleepSameAccountRetryDelay 同账号重试固定延时，返回 false 表示 context 已取消。
+func sleepSameAccountRetryDelay(ctx context.Context) bool {
+	select {
+	case <-ctx.Done():
+		return false
+	case <-time.After(sameAccountRetryDelay):
+		return true
+	}
+}
+
 // sleepFailoverDelay 账号切换线性递增延时：第1次0s、第2次1s、第3次2s…
 // 返回 false 表示 context 已取消。
 func sleepFailoverDelay(ctx context.Context, switchCount int) bool {
@@ -994,6 +1048,10 @@ func (h *GatewayHandler) handleFailoverExhausted(c *gin.Context, failoverErr *se
 				msg = *rule.CustomMessage
 			}
 
+			if rule.SkipMonitoring {
+				c.Set(service.OpsSkipPassthroughKey, true)
+			}
+
 			h.handleStreamingAwareError(c, respCode, "upstream_error", msg, streamStarted)
 			return
 		}

backend/internal/handler/gemini_v1beta_handler.go

@@ -598,6 +598,10 @@ func (h *GatewayHandler) handleGeminiFailoverExhausted(c *gin.Context, failoverE
 				msg = *rule.CustomMessage
 			}
 
+			if rule.SkipMonitoring {
+				c.Set(service.OpsSkipPassthroughKey, true)
+			}
+
 			googleError(c, respCode, msg)
 			return
 		}

backend/internal/handler/openai_gateway_handler.go

@@ -453,6 +453,10 @@ func (h *OpenAIGatewayHandler) handleFailoverExhausted(c *gin.Context, failoverE
 				msg = *rule.CustomMessage
 			}
 
+			if rule.SkipMonitoring {
+				c.Set(service.OpsSkipPassthroughKey, true)
+			}
+
 			h.handleStreamingAwareError(c, respCode, "upstream_error", msg, streamStarted)
 			return
 		}

backend/internal/handler/ops_error_logger.go

@@ -553,6 +553,13 @@ func OpsErrorLoggerMiddleware(ops *service.OpsService) gin.HandlerFunc {
 			// Store request headers/body only when an upstream error occurred to keep overhead minimal.
 			entry.RequestHeadersJSON = extractOpsRetryRequestHeaders(c)
 
+			// Skip logging if a passthrough rule with skip_monitoring=true matched.
+			if v, ok := c.Get(service.OpsSkipPassthroughKey); ok {
+				if skip, _ := v.(bool); skip {
+					return
+				}
+			}
+
 			enqueueOpsErrorLog(ops, entry, requestBody)
 			return
 		}
@@ -560,6 +567,13 @@ func OpsErrorLoggerMiddleware(ops *service.OpsService) gin.HandlerFunc {
 		body := w.buf.Bytes()
 		parsed := parseOpsErrorResponse(body)
 
+		// Skip logging if a passthrough rule with skip_monitoring=true matched.
+		if v, ok := c.Get(service.OpsSkipPassthroughKey); ok {
+			if skip, _ := v.(bool); skip {
+				return
+			}
+		}
+
 		// Skip logging if the error should be filtered based on settings
 		if shouldSkipOpsErrorLog(c.Request.Context(), ops, parsed.Message, string(body), c.Request.URL.Path) {
 			return

backend/internal/model/error_passthrough_rule.go

@@ -18,6 +18,7 @@ type ErrorPassthroughRule struct {
 	ResponseCode    *int      `json:"response_code"`    // 自定义状态码（passthrough_code=false 时使用）
 	PassthroughBody bool      `json:"passthrough_body"` // 是否透传原始错误信息
 	CustomMessage   *string   `json:"custom_message"`   // 自定义错误信息（passthrough_body=false 时使用）
+	SkipMonitoring  bool      `json:"skip_monitoring"`  // 是否跳过运维监控记录
 	Description     *string   `json:"description"`      // 规则描述
 	CreatedAt       time.Time `json:"created_at"`
 	UpdatedAt       time.Time `json:"updated_at"`

backend/internal/pkg/antigravity/claude_types.go

@@ -27,7 +27,7 @@ type ClaudeMessage struct {
 
 // ThinkingConfig Thinking 配置
 type ThinkingConfig struct {
-	Type         string `json:"type"`                    // "enabled" or "disabled"
+	Type         string `json:"type"`                    // "enabled" / "adaptive" / "disabled"
 	BudgetTokens int    `json:"budget_tokens,omitempty"` // thinking budget
 }
 

backend/internal/pkg/antigravity/client.go

@@ -115,6 +115,23 @@ type LoadCodeAssistResponse struct {
 	IneligibleTiers         []*IneligibleTier `json:"ineligibleTiers,omitempty"`
 }
 
+// OnboardUserRequest onboardUser 请求
+type OnboardUserRequest struct {
+	TierID   string `json:"tierId"`
+	Metadata struct {
+		IDEType    string `json:"ideType"`
+		Platform   string `json:"platform,omitempty"`
+		PluginType string `json:"pluginType,omitempty"`
+	} `json:"metadata"`
+}
+
+// OnboardUserResponse onboardUser 响应
+type OnboardUserResponse struct {
+	Name     string         `json:"name,omitempty"`
+	Done     bool           `json:"done"`
+	Response map[string]any `json:"response,omitempty"`
+}
+
 // GetTier 获取账户类型
 // 优先返回 paidTier（付费订阅级别），否则返回 currentTier
 func (r *LoadCodeAssistResponse) GetTier() string {
@@ -371,6 +388,117 @@ func (c *Client) LoadCodeAssist(ctx context.Context, accessToken string) (*LoadC
 	return nil, nil, lastErr
 }
 
+// OnboardUser 触发账号 onboarding，并返回 project_id
+// 说明：
+// 1) 部分账号 loadCodeAssist 不会立即返回 cloudaicompanionProject；
+// 2) 这时需要调用 onboardUser 完成初始化，之后才能拿到 project_id。
+func (c *Client) OnboardUser(ctx context.Context, accessToken, tierID string) (string, error) {
+	tierID = strings.TrimSpace(tierID)
+	if tierID == "" {
+		return "", fmt.Errorf("tier_id 为空")
+	}
+
+	reqBody := OnboardUserRequest{TierID: tierID}
+	reqBody.Metadata.IDEType = "ANTIGRAVITY"
+	reqBody.Metadata.Platform = "PLATFORM_UNSPECIFIED"
+	reqBody.Metadata.PluginType = "GEMINI"
+
+	bodyBytes, err := json.Marshal(reqBody)
+	if err != nil {
+		return "", fmt.Errorf("序列化请求失败: %w", err)
+	}
+
+	availableURLs := BaseURLs
+	var lastErr error
+
+	for urlIdx, baseURL := range availableURLs {
+		apiURL := baseURL + "/v1internal:onboardUser"
+
+		for attempt := 1; attempt <= 5; attempt++ {
+			req, err := http.NewRequestWithContext(ctx, http.MethodPost, apiURL, bytes.NewReader(bodyBytes))
+			if err != nil {
+				lastErr = fmt.Errorf("创建请求失败: %w", err)
+				break
+			}
+			req.Header.Set("Authorization", "Bearer "+accessToken)
+			req.Header.Set("Content-Type", "application/json")
+			req.Header.Set("User-Agent", UserAgent)
+
+			resp, err := c.httpClient.Do(req)
+			if err != nil {
+				lastErr = fmt.Errorf("onboardUser 请求失败: %w", err)
+				if shouldFallbackToNextURL(err, 0) && urlIdx < len(availableURLs)-1 {
+					log.Printf("[antigravity] onboardUser URL fallback: %s -> %s", baseURL, availableURLs[urlIdx+1])
+					break
+				}
+				return "", lastErr
+			}
+
+			respBodyBytes, err := io.ReadAll(resp.Body)
+			_ = resp.Body.Close()
+			if err != nil {
+				return "", fmt.Errorf("读取响应失败: %w", err)
+			}
+
+			if shouldFallbackToNextURL(nil, resp.StatusCode) && urlIdx < len(availableURLs)-1 {
+				log.Printf("[antigravity] onboardUser URL fallback (HTTP %d): %s -> %s", resp.StatusCode, baseURL, availableURLs[urlIdx+1])
+				break
+			}
+
+			if resp.StatusCode != http.StatusOK {
+				lastErr = fmt.Errorf("onboardUser 失败 (HTTP %d): %s", resp.StatusCode, string(respBodyBytes))
+				return "", lastErr
+			}
+
+			var onboardResp OnboardUserResponse
+			if err := json.Unmarshal(respBodyBytes, &onboardResp); err != nil {
+				lastErr = fmt.Errorf("onboardUser 响应解析失败: %w", err)
+				return "", lastErr
+			}
+
+			if onboardResp.Done {
+				if projectID := extractProjectIDFromOnboardResponse(onboardResp.Response); projectID != "" {
+					DefaultURLAvailability.MarkSuccess(baseURL)
+					return projectID, nil
+				}
+				lastErr = fmt.Errorf("onboardUser 完成但未返回 project_id")
+				return "", lastErr
+			}
+
+			// done=false 时等待后重试（与 CLIProxyAPI 行为一致）
+			select {
+			case <-time.After(2 * time.Second):
+			case <-ctx.Done():
+				return "", ctx.Err()
+			}
+		}
+	}
+
+	if lastErr != nil {
+		return "", lastErr
+	}
+	return "", fmt.Errorf("onboardUser 未返回 project_id")
+}
+
+func extractProjectIDFromOnboardResponse(resp map[string]any) string {
+	if len(resp) == 0 {
+		return ""
+	}
+
+	if v, ok := resp["cloudaicompanionProject"]; ok {
+		switch project := v.(type) {
+		case string:
+			return strings.TrimSpace(project)
+		case map[string]any:
+			if id, ok := project["id"].(string); ok {
+				return strings.TrimSpace(id)
+			}
+		}
+	}
+
+	return ""
+}
+
 // ModelQuotaInfo 模型配额信息
 type ModelQuotaInfo struct {
 	RemainingFraction float64 `json:"remainingFraction"`

backend/internal/pkg/antigravity/client_test.go

@@ -1655,3 +1655,74 @@ func TestClient_FetchAvailableModels_404Fallback_RealCall(t *testing.T) {
 		t.Error("应返回 fallback server 的模型 m1")
 	}
 }
+
+func TestExtractProjectIDFromOnboardResponse(t *testing.T) {
+	t.Parallel()
+
+	tests := []struct {
+		name string
+		resp map[string]any
+		want string
+	}{
+		{
+			name: "nil response",
+			resp: nil,
+			want: "",
+		},
+		{
+			name: "empty response",
+			resp: map[string]any{},
+			want: "",
+		},
+		{
+			name: "project as string",
+			resp: map[string]any{
+				"cloudaicompanionProject": "my-project-123",
+			},
+			want: "my-project-123",
+		},
+		{
+			name: "project as string with spaces",
+			resp: map[string]any{
+				"cloudaicompanionProject": "  my-project-123  ",
+			},
+			want: "my-project-123",
+		},
+		{
+			name: "project as map with id",
+			resp: map[string]any{
+				"cloudaicompanionProject": map[string]any{
+					"id": "proj-from-map",
+				},
+			},
+			want: "proj-from-map",
+		},
+		{
+			name: "project as map without id",
+			resp: map[string]any{
+				"cloudaicompanionProject": map[string]any{
+					"name": "some-name",
+				},
+			},
+			want: "",
+		},
+		{
+			name: "missing cloudaicompanionProject key",
+			resp: map[string]any{
+				"otherField": "value",
+			},
+			want: "",
+		},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+
+			got := extractProjectIDFromOnboardResponse(tc.resp)
+			if got != tc.want {
+				t.Fatalf("extractProjectIDFromOnboardResponse() = %q, want %q", got, tc.want)
+			}
+		})
+	}
+}

backend/internal/pkg/antigravity/gemini_types.go

@@ -155,6 +155,7 @@ type GeminiUsageMetadata struct {
 	CandidatesTokenCount    int `json:"candidatesTokenCount,omitempty"`
 	CachedContentTokenCount int `json:"cachedContentTokenCount,omitempty"`
 	TotalTokenCount         int `json:"totalTokenCount,omitempty"`
+	ThoughtsTokenCount      int `json:"thoughtsTokenCount,omitempty"` // thinking tokens（按输出价格计费）
 }
 
 // GeminiGroundingMetadata Gemini grounding 元数据（Web Search）

backend/internal/pkg/antigravity/request_transformer.go

@@ -64,6 +64,10 @@ const MaxTokensBudgetPadding = 1000
 // Gemini 2.5 Flash thinking budget 上限
 const Gemini25FlashThinkingBudgetLimit = 24576
 
+// 对于 Antigravity 的 Claude（budget-only）模型，该语义最终等价为 thinkingBudget=24576。
+// 这里复用相同数值以保持行为一致。
+const ClaudeAdaptiveHighThinkingBudgetTokens = Gemini25FlashThinkingBudgetLimit
+
 // ensureMaxTokensGreaterThanBudget 确保 max_tokens > budget_tokens
 // Claude API 要求启用 thinking 时，max_tokens 必须大于 thinking.budget_tokens
 // 返回调整后的 maxTokens 和是否进行了调整
@@ -96,7 +100,7 @@ func TransformClaudeToGeminiWithOptions(claudeReq *ClaudeRequest, projectID, map
 	}
 
 	// 检测是否启用 thinking
-	isThinkingEnabled := claudeReq.Thinking != nil && claudeReq.Thinking.Type == "enabled"
+	isThinkingEnabled := claudeReq.Thinking != nil && (claudeReq.Thinking.Type == "enabled" || claudeReq.Thinking.Type == "adaptive")
 
 	// 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
@@ -198,8 +202,7 @@ type modelInfo struct {
 
 // modelInfoMap 模型前缀 → 模型信息映射
 // 只有在此映射表中的模型才会注入身份提示词
-// 注意：当前 claude-opus-4-6 会被映射到 claude-opus-4-5-thinking，
-// 但保留此条目以便后续 Antigravity 上游支持 4.6 时快速切换
+// 注意：模型映射逻辑在网关层完成；这里仅用于按模型前缀判断是否注入身份提示词。
 var modelInfoMap = map[string]modelInfo{
 	"claude-opus-4-5":   {DisplayName: "Claude Opus 4.5", CanonicalID: "claude-opus-4-5-20250929"},
 	"claude-opus-4-6":   {DisplayName: "Claude Opus 4.6", CanonicalID: "claude-opus-4-6"},
@@ -593,6 +596,10 @@ func maxOutputTokensLimit(model string) int {
 	return maxOutputTokensUpperBound
 }
 
+func isAntigravityOpus46Model(model string) bool {
+	return strings.HasPrefix(strings.ToLower(model), "claude-opus-4-6")
+}
+
 func buildGenerationConfig(req *ClaudeRequest) *GeminiGenerationConfig {
 	maxLimit := maxOutputTokensLimit(req.Model)
 	config := &GeminiGenerationConfig{
@@ -606,25 +613,36 @@ func buildGenerationConfig(req *ClaudeRequest) *GeminiGenerationConfig {
 	}
 
 	// Thinking 配置
-	if req.Thinking != nil && req.Thinking.Type == "enabled" {
+	if req.Thinking != nil && (req.Thinking.Type == "enabled" || req.Thinking.Type == "adaptive") {
 		config.ThinkingConfig = &GeminiThinkingConfig{
 			IncludeThoughts: true,
 		}
+
+		// - thinking.type=enabled：budget_tokens>0 用显式预算
+		// - thinking.type=adaptive：仅在 Antigravity 的 Opus 4.6 上覆写为 （24576）
+		budget := -1
 		if req.Thinking.BudgetTokens > 0 {
-			budget := req.Thinking.BudgetTokens
+			budget = req.Thinking.BudgetTokens
+		}
+		if req.Thinking.Type == "adaptive" && isAntigravityOpus46Model(req.Model) {
+			budget = ClaudeAdaptiveHighThinkingBudgetTokens
+		}
+
+		// 正预算需要做上限与 max_tokens 约束；动态预算（-1）直接透传给上游。
+		if budget > 0 {
 			// gemini-2.5-flash 上限
 			if strings.Contains(req.Model, "gemini-2.5-flash") && budget > Gemini25FlashThinkingBudgetLimit {
 				budget = Gemini25FlashThinkingBudgetLimit
 			}
-			config.ThinkingConfig.ThinkingBudget = budget
 
-			// 自动修正：max_tokens 必须大于 budget_tokens
+			// 自动修正：max_tokens 必须大于 budget_tokens（Claude 上游要求）
 			if adjusted, ok := ensureMaxTokensGreaterThanBudget(config.MaxOutputTokens, budget); ok {
 				log.Printf("[Antigravity] Auto-adjusted max_tokens from %d to %d (must be > budget_tokens=%d)",
 					config.MaxOutputTokens, adjusted, budget)
 				config.MaxOutputTokens = adjusted
 			}
 		}
+		config.ThinkingConfig.ThinkingBudget = budget
 	}
 
 	if config.MaxOutputTokens > maxLimit {

backend/internal/pkg/antigravity/request_transformer_test.go

@@ -259,3 +259,93 @@ func TestBuildTools_CustomTypeTools(t *testing.T) {
 		})
 	}
 }
+
+func TestBuildGenerationConfig_ThinkingDynamicBudget(t *testing.T) {
+	tests := []struct {
+		name        string
+		model       string
+		thinking    *ThinkingConfig
+		wantBudget  int
+		wantPresent bool
+	}{
+		{
+			name:        "enabled without budget defaults to dynamic (-1)",
+			model:       "claude-opus-4-6-thinking",
+			thinking:    &ThinkingConfig{Type: "enabled"},
+			wantBudget:  -1,
+			wantPresent: true,
+		},
+		{
+			name:        "enabled with budget uses the provided value",
+			model:       "claude-opus-4-6-thinking",
+			thinking:    &ThinkingConfig{Type: "enabled", BudgetTokens: 1024},
+			wantBudget:  1024,
+			wantPresent: true,
+		},
+		{
+			name:        "enabled with -1 budget uses dynamic (-1)",
+			model:       "claude-opus-4-6-thinking",
+			thinking:    &ThinkingConfig{Type: "enabled", BudgetTokens: -1},
+			wantBudget:  -1,
+			wantPresent: true,
+		},
+		{
+			name:        "adaptive on opus4.6 maps to high budget (24576)",
+			model:       "claude-opus-4-6-thinking",
+			thinking:    &ThinkingConfig{Type: "adaptive", BudgetTokens: 20000},
+			wantBudget:  ClaudeAdaptiveHighThinkingBudgetTokens,
+			wantPresent: true,
+		},
+		{
+			name:        "adaptive on non-opus model keeps default dynamic (-1)",
+			model:       "claude-sonnet-4-5-thinking",
+			thinking:    &ThinkingConfig{Type: "adaptive"},
+			wantBudget:  -1,
+			wantPresent: true,
+		},
+		{
+			name:        "disabled does not emit thinkingConfig",
+			model:       "claude-opus-4-6-thinking",
+			thinking:    &ThinkingConfig{Type: "disabled", BudgetTokens: 1024},
+			wantBudget:  0,
+			wantPresent: false,
+		},
+		{
+			name:        "nil thinking does not emit thinkingConfig",
+			model:       "claude-opus-4-6-thinking",
+			thinking:    nil,
+			wantBudget:  0,
+			wantPresent: false,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			req := &ClaudeRequest{
+				Model:    tt.model,
+				Thinking: tt.thinking,
+			}
+			cfg := buildGenerationConfig(req)
+			if cfg == nil {
+				t.Fatalf("expected non-nil generationConfig")
+			}
+
+			if tt.wantPresent {
+				if cfg.ThinkingConfig == nil {
+					t.Fatalf("expected thinkingConfig to be present")
+				}
+				if !cfg.ThinkingConfig.IncludeThoughts {
+					t.Fatalf("expected includeThoughts=true")
+				}
+				if cfg.ThinkingConfig.ThinkingBudget != tt.wantBudget {
+					t.Fatalf("expected thinkingBudget=%d, got %d", tt.wantBudget, cfg.ThinkingConfig.ThinkingBudget)
+				}
+				return
+			}
+
+			if cfg.ThinkingConfig != nil {
+				t.Fatalf("expected thinkingConfig to be nil, got %+v", cfg.ThinkingConfig)
+			}
+		})
+	}
+}

backend/internal/pkg/antigravity/response_transformer.go

@@ -282,7 +282,7 @@ func (p *NonStreamingProcessor) buildResponse(geminiResp *GeminiResponse, respon
 	if geminiResp.UsageMetadata != nil {
 		cached := geminiResp.UsageMetadata.CachedContentTokenCount
 		usage.InputTokens = geminiResp.UsageMetadata.PromptTokenCount - cached
-		usage.OutputTokens = geminiResp.UsageMetadata.CandidatesTokenCount
+		usage.OutputTokens = geminiResp.UsageMetadata.CandidatesTokenCount + geminiResp.UsageMetadata.ThoughtsTokenCount
 		usage.CacheReadInputTokens = cached
 	}
 

backend/internal/pkg/antigravity/stream_transformer.go

@@ -85,7 +85,7 @@ func (p *StreamingProcessor) ProcessLine(line string) []byte {
 	if geminiResp.UsageMetadata != nil {
 		cached := geminiResp.UsageMetadata.CachedContentTokenCount
 		p.inputTokens = geminiResp.UsageMetadata.PromptTokenCount - cached
-		p.outputTokens = geminiResp.UsageMetadata.CandidatesTokenCount
+		p.outputTokens = geminiResp.UsageMetadata.CandidatesTokenCount + geminiResp.UsageMetadata.ThoughtsTokenCount
 		p.cacheReadTokens = cached
 	}
 
@@ -146,7 +146,7 @@ func (p *StreamingProcessor) emitMessageStart(v1Resp *V1InternalResponse) []byte
 	if v1Resp.Response.UsageMetadata != nil {
 		cached := v1Resp.Response.UsageMetadata.CachedContentTokenCount
 		usage.InputTokens = v1Resp.Response.UsageMetadata.PromptTokenCount - cached
-		usage.OutputTokens = v1Resp.Response.UsageMetadata.CandidatesTokenCount
+		usage.OutputTokens = v1Resp.Response.UsageMetadata.CandidatesTokenCount + v1Resp.Response.UsageMetadata.ThoughtsTokenCount
 		usage.CacheReadInputTokens = cached
 	}
 

backend/internal/pkg/logger/slog_handler_test.go

@@ -15,7 +15,6 @@ type captureState struct {
 }
 
 type capturedWrite struct {
-	entry  zapcore.Entry
 	fields []zapcore.Field
 }
 
@@ -51,7 +50,6 @@ func (c *captureCore) Write(entry zapcore.Entry, fields []zapcore.Field) error {
 	allFields = append(allFields, c.withFields...)
 	allFields = append(allFields, fields...)
 	c.state.writes = append(c.state.writes, capturedWrite{
-		entry:  entry,
 		fields: allFields,
 	})
 	return nil

backend/internal/repository/account_repo.go

@@ -448,7 +448,12 @@ func (r *accountRepository) ListWithFilters(ctx context.Context, params paginati
 		q = q.Where(dbaccount.TypeEQ(accountType))
 	}
 	if status != "" {
-		q = q.Where(dbaccount.StatusEQ(status))
+		switch status {
+		case "rate_limited":
+			q = q.Where(dbaccount.RateLimitResetAtGT(time.Now()))
+		default:
+			q = q.Where(dbaccount.StatusEQ(status))
+		}
 	}
 	if search != "" {
 		q = q.Where(dbaccount.NameContainsFold(search))

backend/internal/repository/error_passthrough_repo.go

@@ -54,7 +54,8 @@ func (r *errorPassthroughRepository) Create(ctx context.Context, rule *model.Err
 		SetPriority(rule.Priority).
 		SetMatchMode(rule.MatchMode).
 		SetPassthroughCode(rule.PassthroughCode).
-		SetPassthroughBody(rule.PassthroughBody)
+		SetPassthroughBody(rule.PassthroughBody).
+		SetSkipMonitoring(rule.SkipMonitoring)
 
 	if len(rule.ErrorCodes) > 0 {
 		builder.SetErrorCodes(rule.ErrorCodes)
@@ -90,7 +91,8 @@ func (r *errorPassthroughRepository) Update(ctx context.Context, rule *model.Err
 		SetPriority(rule.Priority).
 		SetMatchMode(rule.MatchMode).
 		SetPassthroughCode(rule.PassthroughCode).
-		SetPassthroughBody(rule.PassthroughBody)
+		SetPassthroughBody(rule.PassthroughBody).
+		SetSkipMonitoring(rule.SkipMonitoring)
 
 	// 处理可选字段
 	if len(rule.ErrorCodes) > 0 {
@@ -149,6 +151,7 @@ func (r *errorPassthroughRepository) toModel(e *ent.ErrorPassthroughRule) *model
 		Platforms:       e.Platforms,
 		PassthroughCode: e.PassthroughCode,
 		PassthroughBody: e.PassthroughBody,
+		SkipMonitoring:  e.SkipMonitoring,
 		CreatedAt:       e.CreatedAt,
 		UpdatedAt:       e.UpdatedAt,
 	}

backend/internal/repository/redeem_code_repo.go

@@ -6,6 +6,7 @@ import (
 
 	dbent "github.com/Wei-Shaw/sub2api/ent"
 	"github.com/Wei-Shaw/sub2api/ent/redeemcode"
+	"github.com/Wei-Shaw/sub2api/ent/user"
 	"github.com/Wei-Shaw/sub2api/internal/pkg/pagination"
 	"github.com/Wei-Shaw/sub2api/internal/service"
 )
@@ -106,7 +107,12 @@ func (r *redeemCodeRepository) ListWithFilters(ctx context.Context, params pagin
 		q = q.Where(redeemcode.StatusEQ(status))
 	}
 	if search != "" {
-		q = q.Where(redeemcode.CodeContainsFold(search))
+		q = q.Where(
+			redeemcode.Or(
+				redeemcode.CodeContainsFold(search),
+				redeemcode.HasUserWith(user.EmailContainsFold(search)),
+			),
+		)
 	}
 
 	total, err := q.Count(ctx)

backend/internal/repository/user_repo.go

@@ -10,6 +10,7 @@ import (
 	"time"
 
 	dbent "github.com/Wei-Shaw/sub2api/ent"
+	"github.com/Wei-Shaw/sub2api/ent/apikey"
 	dbuser "github.com/Wei-Shaw/sub2api/ent/user"
 	"github.com/Wei-Shaw/sub2api/ent/userallowedgroup"
 	"github.com/Wei-Shaw/sub2api/ent/usersubscription"
@@ -191,6 +192,7 @@ func (r *userRepository) ListWithFilters(ctx context.Context, params pagination.
 				dbuser.EmailContainsFold(filters.Search),
 				dbuser.UsernameContainsFold(filters.Search),
 				dbuser.NotesContainsFold(filters.Search),
+				dbuser.HasAPIKeysWith(apikey.KeyContainsFold(filters.Search)),
 			),
 		)
 	}

backend/internal/server/routes/admin.go

@@ -290,6 +290,7 @@ func registerAntigravityOAuthRoutes(admin *gin.RouterGroup, h *handler.Handlers)
 	{
 		antigravity.POST("/oauth/auth-url", h.Admin.AntigravityOAuth.GenerateAuthURL)
 		antigravity.POST("/oauth/exchange-code", h.Admin.AntigravityOAuth.ExchangeCode)
+		antigravity.POST("/oauth/refresh-token", h.Admin.AntigravityOAuth.RefreshToken)
 	}
 }
 

backend/internal/service/antigravity_gateway_service.go

@@ -8,6 +8,7 @@ import (
 	"errors"
 	"fmt"
 	"io"
+	"log"
 	"log/slog"
 	mathrand "math/rand"
 	"net"
@@ -15,6 +16,7 @@ import (
 	"os"
 	"strconv"
 	"strings"
+	"sync"
 	"sync/atomic"
 	"time"
 
@@ -41,6 +43,12 @@ const (
 	antigravitySmartRetryMaxAttempts    = 1                // 智能重试最大次数（仅重试 1 次，防止重复限流/长期等待）
 	antigravityDefaultRateLimitDuration = 30 * time.Second // 默认限流时间（无 retryDelay 时使用）
 
+	// MODEL_CAPACITY_EXHAUSTED 专用重试参数
+	// 模型容量不足时，所有账号共享同一容量池，切换账号无意义
+	// 使用固定 1s 间隔重试，最多重试 60 次
+	antigravityModelCapacityRetryMaxAttempts = 60
+	antigravityModelCapacityRetryWait        = 1 * time.Second
+
 	// Google RPC 状态和类型常量
 	googleRPCStatusResourceExhausted      = "RESOURCE_EXHAUSTED"
 	googleRPCStatusUnavailable            = "UNAVAILABLE"
@@ -61,6 +69,9 @@ const (
 	// 单账号 503 退避重试：原地重试的总累计等待时间上限
 	// 超过此上限将不再重试，直接返回 503
 	antigravitySingleAccountSmartRetryTotalMaxWait = 30 * time.Second
+
+	// MODEL_CAPACITY_EXHAUSTED 全局去重：重试全部失败后的 cooldown 时间
+	antigravityModelCapacityCooldown = 10 * time.Second
 )
 
 // antigravityPassthroughErrorMessages 透传给客户端的错误消息白名单（小写）
@@ -69,8 +80,15 @@ var antigravityPassthroughErrorMessages = []string{
 	"prompt is too long",
 }
 
+// MODEL_CAPACITY_EXHAUSTED 全局去重：避免多个并发请求同时对同一模型进行容量耗尽重试
+var (
+	modelCapacityExhaustedMu    sync.RWMutex
+	modelCapacityExhaustedUntil = make(map[string]time.Time) // modelName -> cooldown until
+)
+
 const (
 	antigravityBillingModelEnv    = "GATEWAY_ANTIGRAVITY_BILL_WITH_MAPPED_MODEL"
+	antigravityForwardBaseURLEnv  = "GATEWAY_ANTIGRAVITY_FORWARD_BASE_URL"
 	antigravityFallbackSecondsEnv = "GATEWAY_ANTIGRAVITY_FALLBACK_COOLDOWN_SECONDS"
 )
 
@@ -132,6 +150,20 @@ type antigravityRetryLoopResult struct {
 	resp *http.Response
 }
 
+// resolveAntigravityForwardBaseURL 解析转发用 base URL。
+// 默认使用 daily（ForwardBaseURLs 的首个地址）；当环境变量为 prod 时使用第二个地址。
+func resolveAntigravityForwardBaseURL() string {
+	baseURLs := antigravity.ForwardBaseURLs()
+	if len(baseURLs) == 0 {
+		return ""
+	}
+	mode := strings.ToLower(strings.TrimSpace(os.Getenv(antigravityForwardBaseURLEnv)))
+	if mode == "prod" && len(baseURLs) > 1 {
+		return baseURLs[1]
+	}
+	return baseURLs[0]
+}
+
 // smartRetryAction 智能重试的处理结果
 type smartRetryAction int
 
@@ -159,7 +191,7 @@ func (s *AntigravityGatewayService) handleSmartRetry(p antigravityRetryLoopParam
 	}
 
 	// 判断是否触发智能重试
-	shouldSmartRetry, shouldRateLimitModel, waitDuration, modelName := shouldTriggerAntigravitySmartRetry(p.account, respBody)
+	shouldSmartRetry, shouldRateLimitModel, waitDuration, modelName, isModelCapacityExhausted := shouldTriggerAntigravitySmartRetry(p.account, respBody)
 
 	// 情况1: retryDelay >= 阈值，限流模型并切换账号
 	if shouldRateLimitModel {
@@ -196,20 +228,48 @@ func (s *AntigravityGatewayService) handleSmartRetry(p antigravityRetryLoopParam
 		}
 	}
 
-	// 情况2: retryDelay < 阈值，智能重试（最多 antigravitySmartRetryMaxAttempts 次）
+	// 情况2: retryDelay < 阈值（或 MODEL_CAPACITY_EXHAUSTED），智能重试
 	if shouldSmartRetry {
 		var lastRetryResp *http.Response
 		var lastRetryBody []byte
 
-		for attempt := 1; attempt <= antigravitySmartRetryMaxAttempts; attempt++ {
-			logger.LegacyPrintf("service.antigravity_gateway", "%s status=%d oauth_smart_retry attempt=%d/%d delay=%v model=%s account=%d",
-				p.prefix, resp.StatusCode, attempt, antigravitySmartRetryMaxAttempts, waitDuration, modelName, p.account.ID)
+		// MODEL_CAPACITY_EXHAUSTED 使用独立的重试参数（60 次，固定 1s 间隔）
+		maxAttempts := antigravitySmartRetryMaxAttempts
+		if isModelCapacityExhausted {
+			maxAttempts = antigravityModelCapacityRetryMaxAttempts
+			waitDuration = antigravityModelCapacityRetryWait
 
+			// 全局去重：如果其他 goroutine 已在重试同一模型且尚在 cooldown 中，直接返回 503
+			if modelName != "" {
+				modelCapacityExhaustedMu.RLock()
+				cooldownUntil, exists := modelCapacityExhaustedUntil[modelName]
+				modelCapacityExhaustedMu.RUnlock()
+				if exists && time.Now().Before(cooldownUntil) {
+					log.Printf("%s status=%d model_capacity_exhausted_dedup model=%s account=%d cooldown_until=%v (skip retry)",
+						p.prefix, resp.StatusCode, modelName, p.account.ID, cooldownUntil.Format("15:04:05"))
+					return &smartRetryResult{
+						action: smartRetryActionBreakWithResp,
+						resp: &http.Response{
+							StatusCode: resp.StatusCode,
+							Header:     resp.Header.Clone(),
+							Body:       io.NopCloser(bytes.NewReader(respBody)),
+						},
+					}
+				}
+			}
+		}
+
+		for attempt := 1; attempt <= maxAttempts; attempt++ {
+			log.Printf("%s status=%d oauth_smart_retry attempt=%d/%d delay=%v model=%s account=%d",
+				p.prefix, resp.StatusCode, attempt, maxAttempts, waitDuration, modelName, p.account.ID)
+
+			timer := time.NewTimer(waitDuration)
 			select {
 			case <-p.ctx.Done():
-				logger.LegacyPrintf("service.antigravity_gateway", "%s status=context_canceled_during_smart_retry", p.prefix)
+				timer.Stop()
+				log.Printf("%s status=context_canceled_during_smart_retry", p.prefix)
 				return &smartRetryResult{action: smartRetryActionBreakWithResp, err: p.ctx.Err()}
-			case <-time.After(waitDuration):
+			case <-timer.C:
 			}
 
 			// 智能重试：创建新请求
@@ -229,13 +289,19 @@ func (s *AntigravityGatewayService) handleSmartRetry(p antigravityRetryLoopParam
 
 			retryResp, retryErr := p.httpUpstream.Do(retryReq, p.proxyURL, p.account.ID, p.account.Concurrency)
 			if retryErr == nil && retryResp != nil && retryResp.StatusCode != http.StatusTooManyRequests && retryResp.StatusCode != http.StatusServiceUnavailable {
-				logger.LegacyPrintf("service.antigravity_gateway", "%s status=%d smart_retry_success attempt=%d/%d", p.prefix, retryResp.StatusCode, attempt, antigravitySmartRetryMaxAttempts)
+				log.Printf("%s status=%d smart_retry_success attempt=%d/%d", p.prefix, retryResp.StatusCode, attempt, maxAttempts)
+				// 重试成功，清除 MODEL_CAPACITY_EXHAUSTED cooldown
+				if isModelCapacityExhausted && modelName != "" {
+					modelCapacityExhaustedMu.Lock()
+					delete(modelCapacityExhaustedUntil, modelName)
+					modelCapacityExhaustedMu.Unlock()
+				}
 				return &smartRetryResult{action: smartRetryActionBreakWithResp, resp: retryResp}
 			}
 
 			// 网络错误时，继续重试
 			if retryErr != nil || retryResp == nil {
-				logger.LegacyPrintf("service.antigravity_gateway", "%s status=smart_retry_network_error attempt=%d/%d error=%v", p.prefix, attempt, antigravitySmartRetryMaxAttempts, retryErr)
+				log.Printf("%s status=smart_retry_network_error attempt=%d/%d error=%v", p.prefix, attempt, maxAttempts, retryErr)
 				continue
 			}
 
@@ -245,13 +311,13 @@ func (s *AntigravityGatewayService) handleSmartRetry(p antigravityRetryLoopParam
 			}
 			lastRetryResp = retryResp
 			if retryResp != nil {
-				lastRetryBody, _ = io.ReadAll(io.LimitReader(retryResp.Body, 2<<20))
+				lastRetryBody, _ = io.ReadAll(io.LimitReader(retryResp.Body, 8<<10))
 				_ = retryResp.Body.Close()
 			}
 
-			// 解析新的重试信息，用于下次重试的等待时间
-			if attempt < antigravitySmartRetryMaxAttempts && lastRetryBody != nil {
-				newShouldRetry, _, newWaitDuration, _ := shouldTriggerAntigravitySmartRetry(p.account, lastRetryBody)
+			// 解析新的重试信息，用于下次重试的等待时间（MODEL_CAPACITY_EXHAUSTED 使用固定循环，跳过）
+			if !isModelCapacityExhausted && attempt < maxAttempts && lastRetryBody != nil {
+				newShouldRetry, _, newWaitDuration, _, _ := shouldTriggerAntigravitySmartRetry(p.account, lastRetryBody)
 				if newShouldRetry && newWaitDuration > 0 {
 					waitDuration = newWaitDuration
 				}
@@ -268,6 +334,27 @@ func (s *AntigravityGatewayService) handleSmartRetry(p antigravityRetryLoopParam
 			retryBody = respBody
 		}
 
+		// MODEL_CAPACITY_EXHAUSTED：模型容量不足，切换账号无意义
+		// 直接返回上游错误响应，不设置模型限流，不切换账号
+		if isModelCapacityExhausted {
+			// 设置 cooldown，让后续请求快速失败，避免重复重试
+			if modelName != "" {
+				modelCapacityExhaustedMu.Lock()
+				modelCapacityExhaustedUntil[modelName] = time.Now().Add(antigravityModelCapacityCooldown)
+				modelCapacityExhaustedMu.Unlock()
+			}
+			log.Printf("%s status=%d smart_retry_exhausted_model_capacity attempts=%d model=%s account=%d body=%s (model capacity exhausted, not switching account)",
+				p.prefix, resp.StatusCode, maxAttempts, modelName, p.account.ID, truncateForLog(retryBody, 200))
+			return &smartRetryResult{
+				action: smartRetryActionBreakWithResp,
+				resp: &http.Response{
+					StatusCode: resp.StatusCode,
+					Header:     resp.Header.Clone(),
+					Body:       io.NopCloser(bytes.NewReader(retryBody)),
+				},
+			}
+		}
+
 		// 单账号 503 退避重试模式：智能重试耗尽后不设限流、不切换账号，
 		// 直接返回 503 让 Handler 层的单账号退避循环做最终处理。
 		if resp.StatusCode == http.StatusServiceUnavailable && isSingleAccountRetry(p.ctx) {
@@ -283,8 +370,8 @@ func (s *AntigravityGatewayService) handleSmartRetry(p antigravityRetryLoopParam
 			}
 		}
 
-		logger.LegacyPrintf("service.antigravity_gateway", "%s status=%d smart_retry_exhausted attempts=%d model=%s account=%d upstream_retry_delay=%v body=%s (switch account)",
-			p.prefix, resp.StatusCode, antigravitySmartRetryMaxAttempts, modelName, p.account.ID, rateLimitDuration, truncateForLog(retryBody, 200))
+		log.Printf("%s status=%d smart_retry_exhausted attempts=%d model=%s account=%d upstream_retry_delay=%v body=%s (switch account)",
+			p.prefix, resp.StatusCode, maxAttempts, modelName, p.account.ID, rateLimitDuration, truncateForLog(retryBody, 200))
 
 		resetAt := time.Now().Add(rateLimitDuration)
 		if p.accountRepo != nil && modelName != "" {
@@ -368,11 +455,13 @@ func (s *AntigravityGatewayService) handleSingleAccountRetryInPlace(
 		logger.LegacyPrintf("service.antigravity_gateway", "%s status=%d single_account_503_retry attempt=%d/%d delay=%v total_waited=%v model=%s account=%d",
 			p.prefix, resp.StatusCode, attempt, antigravitySingleAccountSmartRetryMaxAttempts, waitDuration, totalWaited, modelName, p.account.ID)
 
+		timer := time.NewTimer(waitDuration)
 		select {
 		case <-p.ctx.Done():
+			timer.Stop()
 			logger.LegacyPrintf("service.antigravity_gateway", "%s status=context_canceled_during_single_account_retry", p.prefix)
 			return &smartRetryResult{action: smartRetryActionBreakWithResp, err: p.ctx.Err()}
-		case <-time.After(waitDuration):
+		case <-timer.C:
 		}
 		totalWaited += waitDuration
 
@@ -406,12 +495,12 @@ func (s *AntigravityGatewayService) handleSingleAccountRetryInPlace(
 			_ = lastRetryResp.Body.Close()
 		}
 		lastRetryResp = retryResp
-		lastRetryBody, _ = io.ReadAll(io.LimitReader(retryResp.Body, 2<<20))
+		lastRetryBody, _ = io.ReadAll(io.LimitReader(retryResp.Body, 8<<10))
 		_ = retryResp.Body.Close()
 
 		// 解析新的重试信息，更新下次等待时间
 		if attempt < antigravitySingleAccountSmartRetryMaxAttempts && lastRetryBody != nil {
-			_, _, newWaitDuration, _ := shouldTriggerAntigravitySmartRetry(p.account, lastRetryBody)
+			_, _, newWaitDuration, _, _ := shouldTriggerAntigravitySmartRetry(p.account, lastRetryBody)
 			if newWaitDuration > 0 {
 				waitDuration = newWaitDuration
 				if waitDuration > antigravitySingleAccountSmartRetryMaxWait {
@@ -467,10 +556,11 @@ func (s *AntigravityGatewayService) antigravityRetryLoop(p antigravityRetryLoopP
 		}
 	}
 
-	availableURLs := antigravity.DefaultURLAvailability.GetAvailableURLs()
-	if len(availableURLs) == 0 {
-		availableURLs = antigravity.BaseURLs
+	baseURL := resolveAntigravityForwardBaseURL()
+	if baseURL == "" {
+		return nil, errors.New("no antigravity forward base url configured")
 	}
+	availableURLs := []string{baseURL}
 
 	var resp *http.Response
 	var usedBaseURL string
@@ -908,11 +998,11 @@ func (s *AntigravityGatewayService) TestConnection(ctx context.Context, account
 		proxyURL = account.Proxy.URL()
 	}
 
-	// URL fallback 循环
-	availableURLs := antigravity.DefaultURLAvailability.GetAvailableURLs()
-	if len(availableURLs) == 0 {
-		availableURLs = antigravity.BaseURLs // 所有 URL 都不可用时，重试所有
+	baseURL := resolveAntigravityForwardBaseURL()
+	if baseURL == "" {
+		return nil, errors.New("no antigravity forward base url configured")
 	}
+	availableURLs := []string{baseURL}
 
 	var lastErr error
 	for urlIdx, baseURL := range availableURLs {
@@ -1217,7 +1307,7 @@ func (s *AntigravityGatewayService) Forward(ctx context.Context, c *gin.Context,
 		return nil, s.writeClaudeError(c, http.StatusForbidden, "permission_error", fmt.Sprintf("model %s not in whitelist", claudeReq.Model))
 	}
 	// 应用 thinking 模式自动后缀：如果 thinking 开启且目标是 claude-sonnet-4-5，自动改为 thinking 版本
-	thinkingEnabled := claudeReq.Thinking != nil && claudeReq.Thinking.Type == "enabled"
+	thinkingEnabled := claudeReq.Thinking != nil && (claudeReq.Thinking.Type == "enabled" || claudeReq.Thinking.Type == "adaptive")
 	mappedModel = applyThinkingModelSuffix(mappedModel, thinkingEnabled)
 
 	// 获取 access_token
@@ -1373,7 +1463,7 @@ func (s *AntigravityGatewayService) Forward(ctx context.Context, c *gin.Context,
 					break
 				}
 
-				retryBody, _ := io.ReadAll(io.LimitReader(retryResp.Body, 2<<20))
+				retryBody, _ := io.ReadAll(io.LimitReader(retryResp.Body, 8<<10))
 				_ = retryResp.Body.Close()
 				if retryResp.StatusCode == http.StatusTooManyRequests {
 					retryBaseURL := ""
@@ -1454,6 +1544,27 @@ func (s *AntigravityGatewayService) Forward(ctx context.Context, c *gin.Context,
 
 			s.handleUpstreamError(ctx, prefix, account, resp.StatusCode, resp.Header, respBody, originalModel, 0, "", isStickySession)
 
+			// 精确匹配服务端配置类 400 错误，触发同账号重试 + failover
+			if resp.StatusCode == http.StatusBadRequest {
+				msg := strings.ToLower(strings.TrimSpace(extractAntigravityErrorMessage(respBody)))
+				if isGoogleProjectConfigError(msg) {
+					upstreamMsg := sanitizeUpstreamErrorMessage(strings.TrimSpace(extractAntigravityErrorMessage(respBody)))
+					upstreamDetail := s.getUpstreamErrorDetail(respBody)
+					log.Printf("%s status=400 google_config_error failover=true upstream_message=%q account=%d", prefix, upstreamMsg, account.ID)
+					appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
+						Platform:           account.Platform,
+						AccountID:          account.ID,
+						AccountName:        account.Name,
+						UpstreamStatusCode: resp.StatusCode,
+						UpstreamRequestID:  resp.Header.Get("x-request-id"),
+						Kind:               "failover",
+						Message:            upstreamMsg,
+						Detail:             upstreamDetail,
+					})
+					return nil, &UpstreamFailoverError{StatusCode: resp.StatusCode, ResponseBody: respBody, RetryableOnSameAccount: true}
+				}
+			}
+
 			if s.shouldFailoverUpstreamError(resp.StatusCode) {
 				upstreamMsg := strings.TrimSpace(extractAntigravityErrorMessage(respBody))
 				upstreamMsg = sanitizeUpstreamErrorMessage(upstreamMsg)
@@ -1994,6 +2105,22 @@ func (s *AntigravityGatewayService) ForwardGemini(ctx context.Context, c *gin.Co
 		// Always record upstream context for Ops error logs, even when we will failover.
 		setOpsUpstreamError(c, resp.StatusCode, upstreamMsg, upstreamDetail)
 
+		// 精确匹配服务端配置类 400 错误，触发同账号重试 + failover
+		if resp.StatusCode == http.StatusBadRequest && isGoogleProjectConfigError(strings.ToLower(upstreamMsg)) {
+			log.Printf("%s status=400 google_config_error failover=true upstream_message=%q account=%d", prefix, upstreamMsg, account.ID)
+			appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
+				Platform:           account.Platform,
+				AccountID:          account.ID,
+				AccountName:        account.Name,
+				UpstreamStatusCode: resp.StatusCode,
+				UpstreamRequestID:  requestID,
+				Kind:               "failover",
+				Message:            upstreamMsg,
+				Detail:             upstreamDetail,
+			})
+			return nil, &UpstreamFailoverError{StatusCode: resp.StatusCode, ResponseBody: unwrappedForOps, RetryableOnSameAccount: true}
+		}
+
 		if s.shouldFailoverUpstreamError(resp.StatusCode) {
 			appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
 				Platform:           account.Platform,
@@ -2089,6 +2216,44 @@ func (s *AntigravityGatewayService) shouldFailoverUpstreamError(statusCode int)
 	}
 }
 
+// isGoogleProjectConfigError 判断（已提取的小写）错误消息是否属于 Google 服务端配置类问题。
+// 只精确匹配已知的服务端侧错误，避免对客户端请求错误做无意义重试。
+// 适用于所有走 Google 后端的平台（Antigravity、Gemini）。
+func isGoogleProjectConfigError(lowerMsg string) bool {
+	// Google 间歇性 Bug：Project ID 有效但被临时识别失败
+	return strings.Contains(lowerMsg, "invalid project resource name")
+}
+
+// googleConfigErrorCooldown 服务端配置类 400 错误的临时封禁时长
+const googleConfigErrorCooldown = 1 * time.Minute
+
+// tempUnscheduleGoogleConfigError 对服务端配置类 400 错误触发临时封禁，
+// 避免短时间内反复调度到同一个有问题的账号。
+func tempUnscheduleGoogleConfigError(ctx context.Context, repo AccountRepository, accountID int64, logPrefix string) {
+	until := time.Now().Add(googleConfigErrorCooldown)
+	reason := "400: invalid project resource name (auto temp-unschedule 1m)"
+	if err := repo.SetTempUnschedulable(ctx, accountID, until, reason); err != nil {
+		log.Printf("%s temp_unschedule_failed account=%d error=%v", logPrefix, accountID, err)
+	} else {
+		log.Printf("%s temp_unscheduled account=%d until=%v reason=%q", logPrefix, accountID, until.Format("15:04:05"), reason)
+	}
+}
+
+// emptyResponseCooldown 空流式响应的临时封禁时长
+const emptyResponseCooldown = 1 * time.Minute
+
+// tempUnscheduleEmptyResponse 对空流式响应触发临时封禁，
+// 避免短时间内反复调度到同一个返回空响应的账号。
+func tempUnscheduleEmptyResponse(ctx context.Context, repo AccountRepository, accountID int64, logPrefix string) {
+	until := time.Now().Add(emptyResponseCooldown)
+	reason := "empty stream response (auto temp-unschedule 1m)"
+	if err := repo.SetTempUnschedulable(ctx, accountID, until, reason); err != nil {
+		log.Printf("%s temp_unschedule_failed account=%d error=%v", logPrefix, accountID, err)
+	} else {
+		log.Printf("%s temp_unscheduled account=%d until=%v reason=%q", logPrefix, accountID, until.Format("15:04:05"), reason)
+	}
+}
+
 // sleepAntigravityBackoffWithContext 带 context 取消检查的退避等待
 // 返回 true 表示正常完成等待，false 表示 context 已取消
 func sleepAntigravityBackoffWithContext(ctx context.Context, attempt int) bool {
@@ -2105,10 +2270,12 @@ func sleepAntigravityBackoffWithContext(ctx context.Context, attempt int) bool {
 		sleepFor = 0
 	}
 
+	timer := time.NewTimer(sleepFor)
 	select {
 	case <-ctx.Done():
+		timer.Stop()
 		return false
-	case <-time.After(sleepFor):
+	case <-timer.C:
 		return true
 	}
 }
@@ -2153,8 +2320,9 @@ func antigravityFallbackCooldownSeconds() (time.Duration, bool) {
 
 // antigravitySmartRetryInfo 智能重试所需的信息
 type antigravitySmartRetryInfo struct {
-	RetryDelay time.Duration // 重试延迟时间
-	ModelName  string        // 限流的模型名称（如 "claude-sonnet-4-5"）
+	RetryDelay               time.Duration // 重试延迟时间
+	ModelName                string        // 限流的模型名称（如 "claude-sonnet-4-5"）
+	IsModelCapacityExhausted bool          // 是否为模型容量不足（MODEL_CAPACITY_EXHAUSTED）
 }
 
 // parseAntigravitySmartRetryInfo 解析 Google RPC RetryInfo 和 ErrorInfo 信息
@@ -2269,31 +2437,40 @@ func parseAntigravitySmartRetryInfo(body []byte) *antigravitySmartRetryInfo {
 	}
 
 	return &antigravitySmartRetryInfo{
-		RetryDelay: retryDelay,
-		ModelName:  modelName,
+		RetryDelay:               retryDelay,
+		ModelName:                modelName,
+		IsModelCapacityExhausted: hasModelCapacityExhausted,
 	}
 }
 
 // shouldTriggerAntigravitySmartRetry 判断是否应该触发智能重试
 // 返回：
-//   - shouldRetry: 是否应该智能重试（retryDelay < antigravityRateLimitThreshold）
-//   - shouldRateLimitModel: 是否应该限流模型（retryDelay >= antigravityRateLimitThreshold）
-//   - waitDuration: 等待时间（智能重试时使用，shouldRateLimitModel=true 时为 0）
+//   - shouldRetry: 是否应该智能重试（retryDelay < antigravityRateLimitThreshold，或 MODEL_CAPACITY_EXHAUSTED）
+//   - shouldRateLimitModel: 是否应该限流模型并切换账号（仅 RATE_LIMIT_EXCEEDED 且 retryDelay >= 阈值）
+//   - waitDuration: 等待时间
 //   - modelName: 限流的模型名称
-func shouldTriggerAntigravitySmartRetry(account *Account, respBody []byte) (shouldRetry bool, shouldRateLimitModel bool, waitDuration time.Duration, modelName string) {
+//   - isModelCapacityExhausted: 是否为模型容量不足（MODEL_CAPACITY_EXHAUSTED）
+func shouldTriggerAntigravitySmartRetry(account *Account, respBody []byte) (shouldRetry bool, shouldRateLimitModel bool, waitDuration time.Duration, modelName string, isModelCapacityExhausted bool) {
 	if account.Platform != PlatformAntigravity {
-		return false, false, 0, ""
+		return false, false, 0, "", false
 	}
 
 	info := parseAntigravitySmartRetryInfo(respBody)
 	if info == nil {
-		return false, false, 0, ""
+		return false, false, 0, "", false
 	}
 
+	// MODEL_CAPACITY_EXHAUSTED（模型容量不足）：所有账号共享同一模型容量池
+	// 切换账号无意义，使用固定 1s 间隔重试
+	if info.IsModelCapacityExhausted {
+		return true, false, antigravityModelCapacityRetryWait, info.ModelName, true
+	}
+
+	// RATE_LIMIT_EXCEEDED（账号级限流）：
 	// retryDelay >= 阈值：直接限流模型，不重试
 	// 注意：如果上游未提供 retryDelay，parseAntigravitySmartRetryInfo 已设置为默认 30s
 	if info.RetryDelay >= antigravityRateLimitThreshold {
-		return false, true, info.RetryDelay, info.ModelName
+		return false, true, info.RetryDelay, info.ModelName, false
 	}
 
 	// retryDelay < 阈值：智能重试
@@ -2302,7 +2479,7 @@ func shouldTriggerAntigravitySmartRetry(account *Account, respBody []byte) (shou
 		waitDuration = antigravitySmartRetryMinWait
 	}
 
-	return true, false, waitDuration, info.ModelName
+	return true, false, waitDuration, info.ModelName, false
 }
 
 // handleModelRateLimitParams 模型级限流处理参数
@@ -2328,8 +2505,9 @@ type handleModelRateLimitResult struct {
 
 // handleModelRateLimit 处理模型级限流（在原有逻辑之前调用）
 // 仅处理 429/503，解析模型名和 retryDelay
-// - retryDelay < antigravityRateLimitThreshold: 返回 ShouldRetry=true，由调用方等待后重试
-// - retryDelay >= antigravityRateLimitThreshold: 设置模型限流 + 清除粘性会话 + 返回 SwitchError
+// - MODEL_CAPACITY_EXHAUSTED: 返回 Handled=true（实际重试由 handleSmartRetry 处理）
+// - RATE_LIMIT_EXCEEDED + retryDelay < 阈值: 返回 ShouldRetry=true，由调用方等待后重试
+// - RATE_LIMIT_EXCEEDED + retryDelay >= 阈值: 设置模型限流 + 清除粘性会话 + 返回 SwitchError
 func (s *AntigravityGatewayService) handleModelRateLimit(p *handleModelRateLimitParams) *handleModelRateLimitResult {
 	if p.statusCode != 429 && p.statusCode != 503 {
 		return &handleModelRateLimitResult{Handled: false}
@@ -2340,7 +2518,17 @@ func (s *AntigravityGatewayService) handleModelRateLimit(p *handleModelRateLimit
 		return &handleModelRateLimitResult{Handled: false}
 	}
 
-	// < antigravityRateLimitThreshold: 等待后重试
+	// MODEL_CAPACITY_EXHAUSTED：模型容量不足，所有账号共享同一容量池
+	// 切换账号无意义，不设置模型限流（实际重试由 handleSmartRetry 处理）
+	if info.IsModelCapacityExhausted {
+		log.Printf("%s status=%d model_capacity_exhausted model=%s (not switching account, retry handled by smart retry)",
+			p.prefix, p.statusCode, info.ModelName)
+		return &handleModelRateLimitResult{
+			Handled: true,
+		}
+	}
+
+	// RATE_LIMIT_EXCEEDED: < antigravityRateLimitThreshold: 等待后重试
 	if info.RetryDelay < antigravityRateLimitThreshold {
 		logger.LegacyPrintf("service.antigravity_gateway", "%s status=%d model_rate_limit_wait model=%s wait=%v",
 			p.prefix, p.statusCode, info.ModelName, info.RetryDelay)
@@ -2351,7 +2539,7 @@ func (s *AntigravityGatewayService) handleModelRateLimit(p *handleModelRateLimit
 		}
 	}
 
-	// >= antigravityRateLimitThreshold: 设置限流 + 清除粘性会话 + 切换账号
+	// RATE_LIMIT_EXCEEDED: >= antigravityRateLimitThreshold: 设置限流 + 清除粘性会话 + 切换账号
 	s.setModelRateLimitAndClearSession(p, info)
 
 	return &handleModelRateLimitResult{
@@ -2903,9 +3091,14 @@ returnResponse:
 	// 选择最后一个有效响应
 	finalResponse := pickGeminiCollectResult(last, lastWithParts)
 
-	// 处理空响应情况
+	// 处理空响应情况 — 触发同账号重试 + failover 切换账号
 	if last == nil && lastWithParts == nil {
-		logger.LegacyPrintf("service.antigravity_gateway", "[antigravity-Forward] warning: empty stream response, no valid chunks received")
+		logger.LegacyPrintf("service.antigravity_gateway", "[antigravity-Forward] warning: empty stream response (gemini non-stream), triggering failover")
+		return nil, &UpstreamFailoverError{
+			StatusCode:             http.StatusBadGateway,
+			ResponseBody:           []byte(`{"error":"empty stream response from upstream"}`),
+			RetryableOnSameAccount: true,
+		}
 	}
 
 	// 如果收集到了图片 parts，需要合并到最终响应中
@@ -3123,6 +3316,21 @@ func (s *AntigravityGatewayService) writeMappedClaudeError(c *gin.Context, accou
 		logger.LegacyPrintf("service.antigravity_gateway", "[antigravity-Forward] upstream_error status=%d body=%s", upstreamStatus, truncateForLog(body, maxBytes))
 	}
 
+	// 检查错误透传规则
+	if ptStatus, ptErrType, ptErrMsg, matched := applyErrorPassthroughRule(
+		c, account.Platform, upstreamStatus, body,
+		0, "", "",
+	); matched {
+		c.JSON(ptStatus, gin.H{
+			"type":  "error",
+			"error": gin.H{"type": ptErrType, "message": ptErrMsg},
+		})
+		if upstreamMsg == "" {
+			return fmt.Errorf("upstream error: %d", upstreamStatus)
+		}
+		return fmt.Errorf("upstream error: %d message=%s", upstreamStatus, upstreamMsg)
+	}
+
 	var statusCode int
 	var errType, errMsg string
 
@@ -3320,10 +3528,14 @@ returnResponse:
 	// 选择最后一个有效响应
 	finalResponse := pickGeminiCollectResult(last, lastWithParts)
 
-	// 处理空响应情况
+	// 处理空响应情况 — 触发同账号重试 + failover 切换账号
 	if last == nil && lastWithParts == nil {
-		logger.LegacyPrintf("service.antigravity_gateway", "[antigravity-Forward] warning: empty stream response, no valid chunks received")
-		return nil, s.writeClaudeError(c, http.StatusBadGateway, "upstream_error", "Empty response from upstream")
+		logger.LegacyPrintf("service.antigravity_gateway", "[antigravity-Forward] warning: empty stream response (claude non-stream), triggering failover")
+		return nil, &UpstreamFailoverError{
+			StatusCode:             http.StatusBadGateway,
+			ResponseBody:           []byte(`{"error":"empty stream response from upstream"}`),
+			RetryableOnSameAccount: true,
+		}
 	}
 
 	// 将收集的所有 parts 合并到最终响应中

backend/internal/service/antigravity_gateway_service_test.go

@@ -592,6 +592,75 @@ func TestHandleClaudeStreamingResponse_NormalComplete(t *testing.T) {
 	require.NotContains(t, body, "event: error")
 }
 
+// TestHandleGeminiStreamingResponse_ThoughtsTokenCount
+// 验证：Gemini 流式转发时 thoughtsTokenCount 被计入 OutputTokens
+func TestHandleGeminiStreamingResponse_ThoughtsTokenCount(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	svc := newAntigravityTestService(&config.Config{
+		Gateway: config.GatewayConfig{MaxLineSize: defaultMaxLineSize},
+	})
+
+	rec := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(rec)
+	c.Request = httptest.NewRequest(http.MethodPost, "/", nil)
+
+	pr, pw := io.Pipe()
+	resp := &http.Response{StatusCode: http.StatusOK, Body: pr, Header: http.Header{}}
+
+	go func() {
+		defer func() { _ = pw.Close() }()
+		fmt.Fprintln(pw, `data: {"candidates":[{"content":{"parts":[{"text":"Hello"}]}}],"usageMetadata":{"promptTokenCount":100,"candidatesTokenCount":20,"thoughtsTokenCount":50}}`)
+		fmt.Fprintln(pw, "")
+		fmt.Fprintln(pw, `data: {"candidates":[{"content":{"parts":[{"text":" world"}]},"finishReason":"STOP"}],"usageMetadata":{"promptTokenCount":100,"candidatesTokenCount":30,"thoughtsTokenCount":80,"cachedContentTokenCount":10}}`)
+		fmt.Fprintln(pw, "")
+	}()
+
+	result, err := svc.handleGeminiStreamingResponse(c, resp, time.Now())
+	_ = pr.Close()
+
+	require.NoError(t, err)
+	require.NotNil(t, result)
+	require.NotNil(t, result.usage)
+	// promptTokenCount=100, cachedContentTokenCount=10 → InputTokens=90
+	require.Equal(t, 90, result.usage.InputTokens)
+	// candidatesTokenCount=30 + thoughtsTokenCount=80 → OutputTokens=110
+	require.Equal(t, 110, result.usage.OutputTokens)
+	require.Equal(t, 10, result.usage.CacheReadInputTokens)
+}
+
+// TestHandleClaudeStreamingResponse_ThoughtsTokenCount
+// 验证：Gemini→Claude 流式转换时 thoughtsTokenCount 被计入 OutputTokens
+func TestHandleClaudeStreamingResponse_ThoughtsTokenCount(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	svc := newAntigravityTestService(&config.Config{
+		Gateway: config.GatewayConfig{MaxLineSize: defaultMaxLineSize},
+	})
+
+	rec := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(rec)
+	c.Request = httptest.NewRequest(http.MethodPost, "/", nil)
+
+	pr, pw := io.Pipe()
+	resp := &http.Response{StatusCode: http.StatusOK, Body: pr, Header: http.Header{}}
+
+	go func() {
+		defer func() { _ = pw.Close() }()
+		fmt.Fprintln(pw, `data: {"response":{"candidates":[{"content":{"parts":[{"text":"Hi"}]},"finishReason":"STOP"}],"usageMetadata":{"promptTokenCount":50,"candidatesTokenCount":10,"thoughtsTokenCount":25}}}`)
+		fmt.Fprintln(pw, "")
+	}()
+
+	result, err := svc.handleClaudeStreamingResponse(c, resp, time.Now(), "gemini-2.5-pro")
+	_ = pr.Close()
+
+	require.NoError(t, err)
+	require.NotNil(t, result)
+	require.NotNil(t, result.usage)
+	// promptTokenCount=50 → InputTokens=50
+	require.Equal(t, 50, result.usage.InputTokens)
+	// candidatesTokenCount=10 + thoughtsTokenCount=25 → OutputTokens=35
+	require.Equal(t, 35, result.usage.OutputTokens)
+}
+
 // --- 流式客户端断开检测测试 ---
 
 // TestStreamUpstreamResponse_ClientDisconnectDrainsUsage

backend/internal/service/antigravity_oauth_service.go

@@ -192,6 +192,43 @@ func (s *AntigravityOAuthService) RefreshToken(ctx context.Context, refreshToken
 	return nil, fmt.Errorf("token 刷新失败 (重试后): %w", lastErr)
 }
 
+// ValidateRefreshToken 用 refresh token 验证并获取完整的 token 信息（含 email 和 project_id）
+func (s *AntigravityOAuthService) ValidateRefreshToken(ctx context.Context, refreshToken string, proxyID *int64) (*AntigravityTokenInfo, error) {
+	var proxyURL string
+	if proxyID != nil {
+		proxy, err := s.proxyRepo.GetByID(ctx, *proxyID)
+		if err == nil && proxy != nil {
+			proxyURL = proxy.URL()
+		}
+	}
+
+	// 刷新 token
+	tokenInfo, err := s.RefreshToken(ctx, refreshToken, proxyURL)
+	if err != nil {
+		return nil, err
+	}
+
+	// 获取用户信息（email）
+	client := antigravity.NewClient(proxyURL)
+	userInfo, err := client.GetUserInfo(ctx, tokenInfo.AccessToken)
+	if err != nil {
+		fmt.Printf("[AntigravityOAuth] 警告: 获取用户信息失败: %v\n", err)
+	} else {
+		tokenInfo.Email = userInfo.Email
+	}
+
+	// 获取 project_id（容错，失败不阻塞）
+	projectID, loadErr := s.loadProjectIDWithRetry(ctx, tokenInfo.AccessToken, proxyURL, 3)
+	if loadErr != nil {
+		fmt.Printf("[AntigravityOAuth] 警告: 获取 project_id 失败（重试后）: %v\n", loadErr)
+		tokenInfo.ProjectIDMissing = true
+	} else {
+		tokenInfo.ProjectID = projectID
+	}
+
+	return tokenInfo, nil
+}
+
 func isNonRetryableAntigravityOAuthError(err error) bool {
 	msg := err.Error()
 	nonRetryable := []string{
@@ -273,12 +310,21 @@ func (s *AntigravityOAuthService) loadProjectIDWithRetry(ctx context.Context, ac
 		}
 
 		client := antigravity.NewClient(proxyURL)
-		loadResp, _, err := client.LoadCodeAssist(ctx, accessToken)
+		loadResp, loadRaw, err := client.LoadCodeAssist(ctx, accessToken)
 
 		if err == nil && loadResp != nil && loadResp.CloudAICompanionProject != "" {
 			return loadResp.CloudAICompanionProject, nil
 		}
 
+		if err == nil {
+			if projectID, onboardErr := tryOnboardProjectID(ctx, client, accessToken, loadRaw); onboardErr == nil && projectID != "" {
+				return projectID, nil
+			} else if onboardErr != nil {
+				lastErr = onboardErr
+				continue
+			}
+		}
+
 		// 记录错误
 		if err != nil {
 			lastErr = err
@@ -292,6 +338,65 @@ func (s *AntigravityOAuthService) loadProjectIDWithRetry(ctx context.Context, ac
 	return "", fmt.Errorf("获取 project_id 失败 (重试 %d 次后): %w", maxRetries, lastErr)
 }
 
+func tryOnboardProjectID(ctx context.Context, client *antigravity.Client, accessToken string, loadRaw map[string]any) (string, error) {
+	tierID := resolveDefaultTierID(loadRaw)
+	if tierID == "" {
+		return "", fmt.Errorf("loadCodeAssist 未返回可用的默认 tier")
+	}
+
+	projectID, err := client.OnboardUser(ctx, accessToken, tierID)
+	if err != nil {
+		return "", fmt.Errorf("onboardUser 失败 (tier=%s): %w", tierID, err)
+	}
+	return projectID, nil
+}
+
+func resolveDefaultTierID(loadRaw map[string]any) string {
+	if len(loadRaw) == 0 {
+		return ""
+	}
+
+	rawTiers, ok := loadRaw["allowedTiers"]
+	if !ok {
+		return ""
+	}
+
+	tiers, ok := rawTiers.([]any)
+	if !ok {
+		return ""
+	}
+
+	for _, rawTier := range tiers {
+		tier, ok := rawTier.(map[string]any)
+		if !ok {
+			continue
+		}
+		if isDefault, _ := tier["isDefault"].(bool); !isDefault {
+			continue
+		}
+		if id, ok := tier["id"].(string); ok {
+			id = strings.TrimSpace(id)
+			if id != "" {
+				return id
+			}
+		}
+	}
+
+	return ""
+}
+
+// FillProjectID 仅获取 project_id，不刷新 OAuth token
+func (s *AntigravityOAuthService) FillProjectID(ctx context.Context, account *Account, accessToken string) (string, error) {
+	var proxyURL string
+	if account.ProxyID != nil {
+		proxy, err := s.proxyRepo.GetByID(ctx, *account.ProxyID)
+		if err == nil && proxy != nil {
+			proxyURL = proxy.URL()
+		}
+	}
+	return s.loadProjectIDWithRetry(ctx, accessToken, proxyURL, 3)
+}
+
 // BuildAccountCredentials 构建账户凭证
 func (s *AntigravityOAuthService) BuildAccountCredentials(tokenInfo *AntigravityTokenInfo) map[string]any {
 	creds := map[string]any{

backend/internal/service/antigravity_oauth_service_test.go

@@ -0,0 +1,82 @@
+package service
+
+import (
+	"testing"
+)
+
+func TestResolveDefaultTierID(t *testing.T) {
+	t.Parallel()
+
+	tests := []struct {
+		name    string
+		loadRaw map[string]any
+		want    string
+	}{
+		{
+			name:    "nil loadRaw",
+			loadRaw: nil,
+			want:    "",
+		},
+		{
+			name: "missing allowedTiers",
+			loadRaw: map[string]any{
+				"paidTier": map[string]any{"id": "g1-pro-tier"},
+			},
+			want: "",
+		},
+		{
+			name:    "empty allowedTiers",
+			loadRaw: map[string]any{"allowedTiers": []any{}},
+			want:    "",
+		},
+		{
+			name: "tier missing id field",
+			loadRaw: map[string]any{
+				"allowedTiers": []any{
+					map[string]any{"isDefault": true},
+				},
+			},
+			want: "",
+		},
+		{
+			name: "allowedTiers but no default",
+			loadRaw: map[string]any{
+				"allowedTiers": []any{
+					map[string]any{"id": "free-tier", "isDefault": false},
+					map[string]any{"id": "standard-tier", "isDefault": false},
+				},
+			},
+			want: "",
+		},
+		{
+			name: "default tier found",
+			loadRaw: map[string]any{
+				"allowedTiers": []any{
+					map[string]any{"id": "free-tier", "isDefault": true},
+					map[string]any{"id": "standard-tier", "isDefault": false},
+				},
+			},
+			want: "free-tier",
+		},
+		{
+			name: "default tier id with spaces",
+			loadRaw: map[string]any{
+				"allowedTiers": []any{
+					map[string]any{"id": "  standard-tier  ", "isDefault": true},
+				},
+			},
+			want: "standard-tier",
+		},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+
+			got := resolveDefaultTierID(tc.loadRaw)
+			if got != tc.want {
+				t.Fatalf("resolveDefaultTierID() = %q, want %q", got, tc.want)
+			}
+		})
+	}
+}

backend/internal/service/antigravity_rate_limit_test.go

@@ -92,7 +92,9 @@ func (s *stubAntigravityAccountRepo) SetModelRateLimit(ctx context.Context, id i
 	return nil
 }
 
-func TestAntigravityRetryLoop_URLFallback_UsesLatestSuccess(t *testing.T) {
+func TestAntigravityRetryLoop_NoURLFallback_UsesConfiguredBaseURL(t *testing.T) {
+	t.Setenv(antigravityForwardBaseURLEnv, "")
+
 	oldBaseURLs := append([]string(nil), antigravity.BaseURLs...)
 	oldAvailability := antigravity.DefaultURLAvailability
 	defer func() {
@@ -137,15 +139,16 @@ func TestAntigravityRetryLoop_URLFallback_UsesLatestSuccess(t *testing.T) {
 	require.NotNil(t, result)
 	require.NotNil(t, result.resp)
 	defer func() { _ = result.resp.Body.Close() }()
-	require.Equal(t, http.StatusOK, result.resp.StatusCode)
-	require.False(t, handleErrorCalled)
-	require.Len(t, upstream.calls, 2)
-	require.True(t, strings.HasPrefix(upstream.calls[0], base1))
-	require.True(t, strings.HasPrefix(upstream.calls[1], base2))
+	require.Equal(t, http.StatusTooManyRequests, result.resp.StatusCode)
+	require.True(t, handleErrorCalled)
+	require.Len(t, upstream.calls, antigravityMaxRetries)
+	for _, callURL := range upstream.calls {
+		require.True(t, strings.HasPrefix(callURL, base1))
+	}
 
 	available := antigravity.DefaultURLAvailability.GetAvailableURLs()
 	require.NotEmpty(t, available)
-	require.Equal(t, base2, available[0])
+	require.Equal(t, base1, available[0])
 }
 
 // TestHandleUpstreamError_429_ModelRateLimit 测试 429 模型限流场景
@@ -194,13 +197,14 @@ func TestHandleUpstreamError_429_NonModelRateLimit(t *testing.T) {
 	require.Equal(t, "claude-sonnet-4-5", repo.modelRateLimitCalls[0].modelKey)
 }
 
-// TestHandleUpstreamError_503_ModelRateLimit 测试 503 模型限流场景
-func TestHandleUpstreamError_503_ModelRateLimit(t *testing.T) {
+// TestHandleUpstreamError_503_ModelCapacityExhausted 测试 503 模型容量不足场景
+// MODEL_CAPACITY_EXHAUSTED 时应等待重试，不切换账号
+func TestHandleUpstreamError_503_ModelCapacityExhausted(t *testing.T) {
 	repo := &stubAntigravityAccountRepo{}
 	svc := &AntigravityGatewayService{accountRepo: repo}
 	account := &Account{ID: 3, Name: "acc-3", Platform: PlatformAntigravity}
 
-	// 503 + MODEL_CAPACITY_EXHAUSTED → 模型限流
+	// 503 + MODEL_CAPACITY_EXHAUSTED → 等待重试，不切换账号
 	body := []byte(`{
 		"error": {
 			"status": "UNAVAILABLE",
@@ -213,13 +217,13 @@ func TestHandleUpstreamError_503_ModelRateLimit(t *testing.T) {
 
 	result := svc.handleUpstreamError(context.Background(), "[test]", account, http.StatusServiceUnavailable, http.Header{}, body, "gemini-3-pro-high", 0, "", false)
 
-	// 应该触发模型限流
+	// MODEL_CAPACITY_EXHAUSTED 应该标记为已处理，不切换账号，不设置模型限流
+	// 实际重试由 handleSmartRetry 处理
 	require.NotNil(t, result)
 	require.True(t, result.Handled)
-	require.NotNil(t, result.SwitchError)
-	require.Equal(t, "gemini-3-pro-high", result.SwitchError.RateLimitedModel)
-	require.Len(t, repo.modelRateLimitCalls, 1)
-	require.Equal(t, "gemini-3-pro-high", repo.modelRateLimitCalls[0].modelKey)
+	require.False(t, result.ShouldRetry, "MODEL_CAPACITY_EXHAUSTED should not trigger retry from handleModelRateLimit path")
+	require.Nil(t, result.SwitchError, "MODEL_CAPACITY_EXHAUSTED should not trigger account switch")
+	require.Empty(t, repo.modelRateLimitCalls, "MODEL_CAPACITY_EXHAUSTED should not set model rate limit")
 }
 
 // TestHandleUpstreamError_503_NonModelRateLimit 测试 503 非模型限流场景（不处理）
@@ -307,11 +311,12 @@ func TestParseGeminiRateLimitResetTime_QuotaResetDelay_RoundsUp(t *testing.T) {
 
 func TestParseAntigravitySmartRetryInfo(t *testing.T) {
 	tests := []struct {
-		name          string
-		body          string
-		expectedDelay time.Duration
-		expectedModel string
-		expectedNil   bool
+		name                             string
+		body                             string
+		expectedDelay                    time.Duration
+		expectedModel                    string
+		expectedNil                      bool
+		expectedIsModelCapacityExhausted bool
 	}{
 		{
 			name: "valid complete response with RATE_LIMIT_EXCEEDED",
@@ -374,8 +379,9 @@ func TestParseAntigravitySmartRetryInfo(t *testing.T) {
 					"message": "No capacity available for model gemini-3-pro-high on the server"
 				}
 			}`,
-			expectedDelay: 39 * time.Second,
-			expectedModel: "gemini-3-pro-high",
+			expectedDelay:                    39 * time.Second,
+			expectedModel:                    "gemini-3-pro-high",
+			expectedIsModelCapacityExhausted: true,
 		},
 		{
 			name: "503 UNAVAILABLE without MODEL_CAPACITY_EXHAUSTED - should return nil",
@@ -486,6 +492,9 @@ func TestParseAntigravitySmartRetryInfo(t *testing.T) {
 			if result.ModelName != tt.expectedModel {
 				t.Errorf("ModelName = %q, want %q", result.ModelName, tt.expectedModel)
 			}
+			if result.IsModelCapacityExhausted != tt.expectedIsModelCapacityExhausted {
+				t.Errorf("IsModelCapacityExhausted = %v, want %v", result.IsModelCapacityExhausted, tt.expectedIsModelCapacityExhausted)
+			}
 		})
 	}
 }
@@ -497,13 +506,14 @@ func TestShouldTriggerAntigravitySmartRetry(t *testing.T) {
 	apiKeyAccount := &Account{Type: AccountTypeAPIKey}
 
 	tests := []struct {
-		name                    string
-		account                 *Account
-		body                    string
-		expectedShouldRetry     bool
-		expectedShouldRateLimit bool
-		minWait                 time.Duration
-		modelName               string
+		name                             string
+		account                          *Account
+		body                             string
+		expectedShouldRetry              bool
+		expectedShouldRateLimit          bool
+		expectedIsModelCapacityExhausted bool
+		minWait                          time.Duration
+		modelName                        string
 	}{
 		{
 			name:    "OAuth account with short delay (< 7s) - smart retry",
@@ -617,13 +627,14 @@ func TestShouldTriggerAntigravitySmartRetry(t *testing.T) {
 					]
 				}
 			}`,
-			expectedShouldRetry:     false,
-			expectedShouldRateLimit: true,
-			minWait:                 39 * time.Second,
-			modelName:               "gemini-3-pro-high",
+			expectedShouldRetry:              true,
+			expectedShouldRateLimit:          false,
+			expectedIsModelCapacityExhausted: true,
+			minWait:                          1 * time.Second,
+			modelName:                        "gemini-3-pro-high",
 		},
 		{
-			name:    "503 UNAVAILABLE with MODEL_CAPACITY_EXHAUSTED - no retryDelay - use default rate limit",
+			name:    "503 UNAVAILABLE with MODEL_CAPACITY_EXHAUSTED - no retryDelay - use fixed wait",
 			account: oauthAccount,
 			body: `{
 				"error": {
@@ -635,10 +646,11 @@ func TestShouldTriggerAntigravitySmartRetry(t *testing.T) {
 					"message": "No capacity available for model gemini-2.5-flash on the server"
 				}
 			}`,
-			expectedShouldRetry:     false,
-			expectedShouldRateLimit: true,
-			minWait:                 30 * time.Second,
-			modelName:               "gemini-2.5-flash",
+			expectedShouldRetry:              true,
+			expectedShouldRateLimit:          false,
+			expectedIsModelCapacityExhausted: true,
+			minWait:                          1 * time.Second,
+			modelName:                        "gemini-2.5-flash",
 		},
 		{
 			name:    "429 RESOURCE_EXHAUSTED with RATE_LIMIT_EXCEEDED - no retryDelay - use default rate limit",
@@ -662,13 +674,16 @@ func TestShouldTriggerAntigravitySmartRetry(t *testing.T) {
 
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
-			shouldRetry, shouldRateLimit, wait, model := shouldTriggerAntigravitySmartRetry(tt.account, []byte(tt.body))
+			shouldRetry, shouldRateLimit, wait, model, isModelCapacityExhausted := shouldTriggerAntigravitySmartRetry(tt.account, []byte(tt.body))
 			if shouldRetry != tt.expectedShouldRetry {
 				t.Errorf("shouldRetry = %v, want %v", shouldRetry, tt.expectedShouldRetry)
 			}
 			if shouldRateLimit != tt.expectedShouldRateLimit {
 				t.Errorf("shouldRateLimit = %v, want %v", shouldRateLimit, tt.expectedShouldRateLimit)
 			}
+			if isModelCapacityExhausted != tt.expectedIsModelCapacityExhausted {
+				t.Errorf("isModelCapacityExhausted = %v, want %v", isModelCapacityExhausted, tt.expectedIsModelCapacityExhausted)
+			}
 			if shouldRetry {
 				if wait < tt.minWait {
 					t.Errorf("wait = %v, want >= %v", wait, tt.minWait)
@@ -921,6 +936,22 @@ func TestIsAntigravityAccountSwitchError(t *testing.T) {
 	}
 }
 
+func TestResolveAntigravityForwardBaseURL_DefaultDaily(t *testing.T) {
+	t.Setenv(antigravityForwardBaseURLEnv, "")
+
+	oldBaseURLs := append([]string(nil), antigravity.BaseURLs...)
+	defer func() {
+		antigravity.BaseURLs = oldBaseURLs
+	}()
+
+	prodURL := "https://prod.test"
+	dailyURL := "https://daily.test"
+	antigravity.BaseURLs = []string{dailyURL, prodURL}
+
+	resolved := resolveAntigravityForwardBaseURL()
+	require.Equal(t, dailyURL, resolved)
+}
+
 func TestAntigravityAccountSwitchError_Error(t *testing.T) {
 	err := &AntigravityAccountSwitchError{
 		OriginalAccountID: 789,

backend/internal/service/antigravity_single_account_retry_test.go

@@ -153,13 +153,14 @@ func TestHandleSmartRetry_503_LongDelay_NoSingleAccountRetry_StillSwitches(t *te
 		Platform: PlatformAntigravity,
 	}
 
-	// 503 + 39s >= 7s 阈值
+	// 503 + 39s >= 7s 阈值（使用 RATE_LIMIT_EXCEEDED 而非 MODEL_CAPACITY_EXHAUSTED，
+	// 因为 MODEL_CAPACITY_EXHAUSTED 走独立的重试路径，不触发 shouldRateLimitModel）
 	respBody := []byte(`{
 		"error": {
 			"code": 503,
-			"status": "UNAVAILABLE",
+			"status": "RESOURCE_EXHAUSTED",
 			"details": [
-				{"@type": "type.googleapis.com/google.rpc.ErrorInfo", "metadata": {"model": "gemini-3-pro-high"}, "reason": "MODEL_CAPACITY_EXHAUSTED"},
+				{"@type": "type.googleapis.com/google.rpc.ErrorInfo", "metadata": {"model": "gemini-3-pro-high"}, "reason": "RATE_LIMIT_EXCEEDED"},
 				{"@type": "type.googleapis.com/google.rpc.RetryInfo", "retryDelay": "39s"}
 			]
 		}
@@ -339,13 +340,14 @@ func TestHandleSmartRetry_503_ShortDelay_SingleAccountRetry_NoRateLimit(t *testi
 
 // TestHandleSmartRetry_503_ShortDelay_NoSingleAccountRetry_SetsRateLimit
 // 对照组：503 + retryDelay < 7s + 无 SingleAccountRetry → 智能重试耗尽后照常设限流
+// 使用 RATE_LIMIT_EXCEEDED 而非 MODEL_CAPACITY_EXHAUSTED，因为后者走独立的 60 次重试路径
 func TestHandleSmartRetry_503_ShortDelay_NoSingleAccountRetry_SetsRateLimit(t *testing.T) {
 	failRespBody := `{
 		"error": {
 			"code": 503,
-			"status": "UNAVAILABLE",
+			"status": "RESOURCE_EXHAUSTED",
 			"details": [
-				{"@type": "type.googleapis.com/google.rpc.ErrorInfo", "metadata": {"model": "gemini-3-flash"}, "reason": "MODEL_CAPACITY_EXHAUSTED"},
+				{"@type": "type.googleapis.com/google.rpc.ErrorInfo", "metadata": {"model": "gemini-3-flash"}, "reason": "RATE_LIMIT_EXCEEDED"},
 				{"@type": "type.googleapis.com/google.rpc.RetryInfo", "retryDelay": "0.1s"}
 			]
 		}
@@ -371,9 +373,9 @@ func TestHandleSmartRetry_503_ShortDelay_NoSingleAccountRetry_SetsRateLimit(t *t
 	respBody := []byte(`{
 		"error": {
 			"code": 503,
-			"status": "UNAVAILABLE",
+			"status": "RESOURCE_EXHAUSTED",
 			"details": [
-				{"@type": "type.googleapis.com/google.rpc.ErrorInfo", "metadata": {"model": "gemini-3-flash"}, "reason": "MODEL_CAPACITY_EXHAUSTED"},
+				{"@type": "type.googleapis.com/google.rpc.ErrorInfo", "metadata": {"model": "gemini-3-flash"}, "reason": "RATE_LIMIT_EXCEEDED"},
 				{"@type": "type.googleapis.com/google.rpc.RetryInfo", "retryDelay": "0.1s"}
 			]
 		}

backend/internal/service/antigravity_smart_retry_test.go

@@ -294,8 +294,9 @@ func TestHandleSmartRetry_ShortDelay_SmartRetryFailed_ReturnsSwitchError(t *test
 	require.Len(t, upstream.calls, 1, "should have made one retry call (max attempts)")
 }
 
-// TestHandleSmartRetry_503_ModelCapacityExhausted_ReturnsSwitchError 测试 503 MODEL_CAPACITY_EXHAUSTED 返回 switchError
-func TestHandleSmartRetry_503_ModelCapacityExhausted_ReturnsSwitchError(t *testing.T) {
+// TestHandleSmartRetry_503_ModelCapacityExhausted_RetrySuccess 测试 503 MODEL_CAPACITY_EXHAUSTED 重试成功
+// MODEL_CAPACITY_EXHAUSTED 使用固定 1s 间隔重试，不切换账号
+func TestHandleSmartRetry_503_ModelCapacityExhausted_RetrySuccess(t *testing.T) {
 	repo := &stubAntigravityAccountRepo{}
 	account := &Account{
 		ID:       3,
@@ -304,7 +305,7 @@ func TestHandleSmartRetry_503_ModelCapacityExhausted_ReturnsSwitchError(t *testi
 		Platform: PlatformAntigravity,
 	}
 
-	// 503 + MODEL_CAPACITY_EXHAUSTED + 39s >= 7s 阈值
+	// 503 + MODEL_CAPACITY_EXHAUSTED + 39s（上游 retryDelay 应被忽略，使用固定 1s）
 	respBody := []byte(`{
 		"error": {
 			"code": 503,
@@ -322,6 +323,14 @@ func TestHandleSmartRetry_503_ModelCapacityExhausted_ReturnsSwitchError(t *testi
 		Body:       io.NopCloser(bytes.NewReader(respBody)),
 	}
 
+	// mock: 第 1 次重试返回 200 成功
+	upstream := &mockSmartRetryUpstream{
+		responses: []*http.Response{
+			{StatusCode: http.StatusOK, Header: http.Header{}, Body: io.NopCloser(strings.NewReader(`{"ok":true}`))},
+		},
+		errors: []error{nil},
+	}
+
 	params := antigravityRetryLoopParams{
 		ctx:             context.Background(),
 		prefix:          "[test]",
@@ -330,6 +339,7 @@ func TestHandleSmartRetry_503_ModelCapacityExhausted_ReturnsSwitchError(t *testi
 		action:          "generateContent",
 		body:            []byte(`{"input":"test"}`),
 		accountRepo:     repo,
+		httpUpstream:    upstream,
 		isStickySession: true,
 		handleError: func(ctx context.Context, prefix string, account *Account, statusCode int, headers http.Header, body []byte, requestedModel string, groupID int64, sessionHash string, isStickySession bool) *handleModelRateLimitResult {
 			return nil
@@ -343,16 +353,67 @@ func TestHandleSmartRetry_503_ModelCapacityExhausted_ReturnsSwitchError(t *testi
 
 	require.NotNil(t, result)
 	require.Equal(t, smartRetryActionBreakWithResp, result.action)
-	require.Nil(t, result.resp)
+	require.NotNil(t, result.resp, "should return successful response")
+	require.Equal(t, http.StatusOK, result.resp.StatusCode)
 	require.Nil(t, result.err)
-	require.NotNil(t, result.switchError, "should return switchError for 503 model capacity exhausted")
-	require.Equal(t, account.ID, result.switchError.OriginalAccountID)
-	require.Equal(t, "gemini-3-pro-high", result.switchError.RateLimitedModel)
-	require.True(t, result.switchError.IsStickySession)
+	require.Nil(t, result.switchError, "MODEL_CAPACITY_EXHAUSTED should not return switchError")
 
-	// 验证模型限流已设置
-	require.Len(t, repo.modelRateLimitCalls, 1)
-	require.Equal(t, "gemini-3-pro-high", repo.modelRateLimitCalls[0].modelKey)
+	// 不应设置模型限流
+	require.Empty(t, repo.modelRateLimitCalls, "MODEL_CAPACITY_EXHAUSTED should not set model rate limit")
+	require.Len(t, upstream.calls, 1, "should have made one retry call before success")
+}
+
+// TestHandleSmartRetry_503_ModelCapacityExhausted_ContextCancel 测试 MODEL_CAPACITY_EXHAUSTED 上下文取消
+func TestHandleSmartRetry_503_ModelCapacityExhausted_ContextCancel(t *testing.T) {
+	repo := &stubAntigravityAccountRepo{}
+	account := &Account{
+		ID:       3,
+		Name:     "acc-3",
+		Type:     AccountTypeOAuth,
+		Platform: PlatformAntigravity,
+	}
+
+	respBody := []byte(`{
+		"error": {
+			"code": 503,
+			"status": "UNAVAILABLE",
+			"details": [
+				{"@type": "type.googleapis.com/google.rpc.ErrorInfo", "metadata": {"model": "gemini-3-pro-high"}, "reason": "MODEL_CAPACITY_EXHAUSTED"},
+				{"@type": "type.googleapis.com/google.rpc.RetryInfo", "retryDelay": "39s"}
+			]
+		}
+	}`)
+	resp := &http.Response{
+		StatusCode: http.StatusServiceUnavailable,
+		Header:     http.Header{},
+		Body:       io.NopCloser(bytes.NewReader(respBody)),
+	}
+
+	// 立即取消上下文，验证重试循环能正确退出
+	ctx, cancel := context.WithCancel(context.Background())
+	cancel()
+
+	params := antigravityRetryLoopParams{
+		ctx:         ctx,
+		prefix:      "[test]",
+		account:     account,
+		accessToken: "token",
+		action:      "generateContent",
+		body:        []byte(`{"input":"test"}`),
+		accountRepo: repo,
+		handleError: func(ctx context.Context, prefix string, account *Account, statusCode int, headers http.Header, body []byte, requestedModel string, groupID int64, sessionHash string, isStickySession bool) *handleModelRateLimitResult {
+			return nil
+		},
+	}
+
+	svc := &AntigravityGatewayService{}
+	result := svc.handleSmartRetry(params, resp, respBody, "https://ag-1.test", 0, []string{"https://ag-1.test"})
+
+	require.NotNil(t, result)
+	require.Equal(t, smartRetryActionBreakWithResp, result.action)
+	require.Error(t, result.err, "should return context error")
+	require.Nil(t, result.switchError, "should not return switchError on context cancel")
+	require.Empty(t, repo.modelRateLimitCalls, "should not set model rate limit on context cancel")
 }
 
 // TestHandleSmartRetry_NonAntigravityAccount_ContinuesDefaultLogic 测试非 Antigravity 平台账号走默认逻辑
@@ -1129,20 +1190,20 @@ func TestHandleSmartRetry_ShortDelay_NetworkError_StickySession_ClearsSession(t
 }
 
 // TestHandleSmartRetry_ShortDelay_503_StickySession_FailedRetry_ClearsSession
-// 503 + 短延迟 + 粘性会话 + 重试失败 → 清除粘性绑定
+// 429 + 短延迟 + 粘性会话 + 重试失败 → 清除粘性绑定
 func TestHandleSmartRetry_ShortDelay_503_StickySession_FailedRetry_ClearsSession(t *testing.T) {
 	failRespBody := `{
 		"error": {
-			"code": 503,
-			"status": "UNAVAILABLE",
+			"code": 429,
+			"status": "RESOURCE_EXHAUSTED",
 			"details": [
-				{"@type": "type.googleapis.com/google.rpc.ErrorInfo", "metadata": {"model": "gemini-3-pro"}, "reason": "MODEL_CAPACITY_EXHAUSTED"},
+				{"@type": "type.googleapis.com/google.rpc.ErrorInfo", "metadata": {"model": "gemini-3-pro"}, "reason": "RATE_LIMIT_EXCEEDED"},
 				{"@type": "type.googleapis.com/google.rpc.RetryInfo", "retryDelay": "0.5s"}
 			]
 		}
 	}`
 	failResp := &http.Response{
-		StatusCode: http.StatusServiceUnavailable,
+		StatusCode: http.StatusTooManyRequests,
 		Header:     http.Header{},
 		Body:       io.NopCloser(strings.NewReader(failRespBody)),
 	}
@@ -1162,16 +1223,16 @@ func TestHandleSmartRetry_ShortDelay_503_StickySession_FailedRetry_ClearsSession
 
 	respBody := []byte(`{
 		"error": {
-			"code": 503,
-			"status": "UNAVAILABLE",
+			"code": 429,
+			"status": "RESOURCE_EXHAUSTED",
 			"details": [
-				{"@type": "type.googleapis.com/google.rpc.ErrorInfo", "metadata": {"model": "gemini-3-pro"}, "reason": "MODEL_CAPACITY_EXHAUSTED"},
+				{"@type": "type.googleapis.com/google.rpc.ErrorInfo", "metadata": {"model": "gemini-3-pro"}, "reason": "RATE_LIMIT_EXCEEDED"},
 				{"@type": "type.googleapis.com/google.rpc.RetryInfo", "retryDelay": "0.5s"}
 			]
 		}
 	}`)
 	resp := &http.Response{
-		StatusCode: http.StatusServiceUnavailable,
+		StatusCode: http.StatusTooManyRequests,
 		Header:     http.Header{},
 		Body:       io.NopCloser(bytes.NewReader(respBody)),
 	}

backend/internal/service/antigravity_token_provider.go

@@ -7,12 +7,14 @@ import (
 	"log/slog"
 	"strconv"
 	"strings"
+	"sync"
 	"time"
 )
 
 const (
 	antigravityTokenRefreshSkew = 3 * time.Minute
 	antigravityTokenCacheSkew   = 5 * time.Minute
+	antigravityBackfillCooldown = 5 * time.Minute
 )
 
 // AntigravityTokenCache Token 缓存接口（复用 GeminiTokenCache 接口定义）
@@ -23,6 +25,7 @@ type AntigravityTokenProvider struct {
 	accountRepo             AccountRepository
 	tokenCache              AntigravityTokenCache
 	antigravityOAuthService *AntigravityOAuthService
+	backfillCooldown        sync.Map // key: int64 (account.ID) → value: time.Time
 }
 
 func NewAntigravityTokenProvider(
@@ -93,13 +96,7 @@ func (p *AntigravityTokenProvider) GetAccessToken(ctx context.Context, account *
 				if err != nil {
 					return "", err
 				}
-				newCredentials := p.antigravityOAuthService.BuildAccountCredentials(tokenInfo)
-				for k, v := range account.Credentials {
-					if _, exists := newCredentials[k]; !exists {
-						newCredentials[k] = v
-					}
-				}
-				account.Credentials = newCredentials
+				p.mergeCredentials(account, tokenInfo)
 				if updateErr := p.accountRepo.Update(ctx, account); updateErr != nil {
 					log.Printf("[AntigravityTokenProvider] Failed to update account credentials: %v", updateErr)
 				}
@@ -113,6 +110,21 @@ func (p *AntigravityTokenProvider) GetAccessToken(ctx context.Context, account *
 		return "", errors.New("access_token not found in credentials")
 	}
 
+	// 如果账号还没有 project_id，尝试在线补齐，避免请求 daily/sandbox 时出现
+	// "Invalid project resource name projects/"。
+	// 仅调用 loadProjectIDWithRetry，不刷新 OAuth token；带冷却机制防止频繁重试。
+	if strings.TrimSpace(account.GetCredential("project_id")) == "" && p.antigravityOAuthService != nil {
+		if p.shouldAttemptBackfill(account.ID) {
+			p.markBackfillAttempted(account.ID)
+			if projectID, err := p.antigravityOAuthService.FillProjectID(ctx, account, accessToken); err == nil && projectID != "" {
+				account.Credentials["project_id"] = projectID
+				if updateErr := p.accountRepo.Update(ctx, account); updateErr != nil {
+					log.Printf("[AntigravityTokenProvider] project_id 补齐持久化失败: %v", updateErr)
+				}
+			}
+		}
+	}
+
 	// 3. 存入缓存（验证版本后再写入，避免异步刷新任务与请求线程的竞态条件）
 	if p.tokenCache != nil {
 		latestAccount, isStale := CheckTokenVersion(ctx, account, p.accountRepo)
@@ -144,6 +156,31 @@ func (p *AntigravityTokenProvider) GetAccessToken(ctx context.Context, account *
 	return accessToken, nil
 }
 
+// mergeCredentials 将 tokenInfo 构建的凭证合并到 account 中，保留原有未覆盖的字段
+func (p *AntigravityTokenProvider) mergeCredentials(account *Account, tokenInfo *AntigravityTokenInfo) {
+	newCredentials := p.antigravityOAuthService.BuildAccountCredentials(tokenInfo)
+	for k, v := range account.Credentials {
+		if _, exists := newCredentials[k]; !exists {
+			newCredentials[k] = v
+		}
+	}
+	account.Credentials = newCredentials
+}
+
+// shouldAttemptBackfill 检查是否应该尝试补齐 project_id（冷却期内不重复尝试）
+func (p *AntigravityTokenProvider) shouldAttemptBackfill(accountID int64) bool {
+	if v, ok := p.backfillCooldown.Load(accountID); ok {
+		if lastAttempt, ok := v.(time.Time); ok {
+			return time.Since(lastAttempt) > antigravityBackfillCooldown
+		}
+	}
+	return true
+}
+
+func (p *AntigravityTokenProvider) markBackfillAttempted(accountID int64) {
+	p.backfillCooldown.Store(accountID, time.Now())
+}
+
 func AntigravityTokenCacheKey(account *Account) string {
 	projectID := strings.TrimSpace(account.GetCredential("project_id"))
 	if projectID != "" {

backend/internal/service/error_passthrough_runtime.go

@@ -61,6 +61,11 @@ func applyErrorPassthroughRule(
 		errMsg = *rule.CustomMessage
 	}
 
+	// 命中 skip_monitoring 时在 context 中标记，供 ops_error_logger 跳过记录。
+	if rule.SkipMonitoring {
+		c.Set(OpsSkipPassthroughKey, true)
+	}
+
 	// 与现有 failover 场景保持一致：命中规则时统一返回 upstream_error。
 	errType = "upstream_error"
 	return status, errType, errMsg, true

backend/internal/service/error_passthrough_runtime_test.go

@@ -194,6 +194,63 @@ func TestGeminiWriteGeminiMappedError_AppliesRuleFor422(t *testing.T) {
 	assert.Equal(t, "Gemini上游失败", errField["message"])
 }
 
+func TestApplyErrorPassthroughRule_SkipMonitoringSetsContextKey(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	rec := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(rec)
+
+	rule := newNonFailoverPassthroughRule(http.StatusBadRequest, "prompt is too long", http.StatusBadRequest, "上下文超限")
+	rule.SkipMonitoring = true
+
+	ruleSvc := &ErrorPassthroughService{}
+	ruleSvc.setLocalCache([]*model.ErrorPassthroughRule{rule})
+	BindErrorPassthroughService(c, ruleSvc)
+
+	_, _, _, matched := applyErrorPassthroughRule(
+		c,
+		PlatformAnthropic,
+		http.StatusBadRequest,
+		[]byte(`{"error":{"message":"prompt is too long"}}`),
+		http.StatusBadGateway,
+		"upstream_error",
+		"Upstream request failed",
+	)
+
+	assert.True(t, matched)
+	v, exists := c.Get(OpsSkipPassthroughKey)
+	assert.True(t, exists, "OpsSkipPassthroughKey should be set when skip_monitoring=true")
+	boolVal, ok := v.(bool)
+	assert.True(t, ok, "value should be bool")
+	assert.True(t, boolVal)
+}
+
+func TestApplyErrorPassthroughRule_NoSkipMonitoringDoesNotSetContextKey(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	rec := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(rec)
+
+	rule := newNonFailoverPassthroughRule(http.StatusBadRequest, "prompt is too long", http.StatusBadRequest, "上下文超限")
+	rule.SkipMonitoring = false
+
+	ruleSvc := &ErrorPassthroughService{}
+	ruleSvc.setLocalCache([]*model.ErrorPassthroughRule{rule})
+	BindErrorPassthroughService(c, ruleSvc)
+
+	_, _, _, matched := applyErrorPassthroughRule(
+		c,
+		PlatformAnthropic,
+		http.StatusBadRequest,
+		[]byte(`{"error":{"message":"prompt is too long"}}`),
+		http.StatusBadGateway,
+		"upstream_error",
+		"Upstream request failed",
+	)
+
+	assert.True(t, matched)
+	_, exists := c.Get(OpsSkipPassthroughKey)
+	assert.False(t, exists, "OpsSkipPassthroughKey should NOT be set when skip_monitoring=false")
+}
+
 func newNonFailoverPassthroughRule(statusCode int, keyword string, respCode int, customMessage string) *model.ErrorPassthroughRule {
 	return &model.ErrorPassthroughRule{
 		ID:              1,

backend/internal/service/error_passthrough_service.go

@@ -45,10 +45,20 @@ type ErrorPassthroughService struct {
 	cache ErrorPassthroughCache
 
 	// 本地内存缓存，用于快速匹配
-	localCache   []*model.ErrorPassthroughRule
+	localCache   []*cachedPassthroughRule
 	localCacheMu sync.RWMutex
 }
 
+// cachedPassthroughRule 预计算的规则缓存，避免运行时重复 ToLower
+type cachedPassthroughRule struct {
+	*model.ErrorPassthroughRule
+	lowerKeywords  []string         // 预计算的小写关键词
+	lowerPlatforms []string         // 预计算的小写平台
+	errorCodeSet   map[int]struct{} // 预计算的 error code set
+}
+
+const maxBodyMatchLen = 8 << 10 // 8KB，错误信息不会在 8KB 之后才出现
+
 // NewErrorPassthroughService 创建错误透传规则服务
 func NewErrorPassthroughService(
 	repo ErrorPassthroughRepository,
@@ -150,17 +160,19 @@ func (s *ErrorPassthroughService) MatchRule(platform string, statusCode int, bod
 		return nil
 	}
 
-	bodyStr := strings.ToLower(string(body))
+	lowerPlatform := strings.ToLower(platform)
+	var bodyLower string // 延迟初始化，只在需要关键词匹配时计算
+	var bodyLowerDone bool
 
 	for _, rule := range rules {
 		if !rule.Enabled {
 			continue
 		}
-		if !s.platformMatches(rule, platform) {
+		if !s.platformMatchesCached(rule, lowerPlatform) {
 			continue
 		}
-		if s.ruleMatches(rule, statusCode, bodyStr) {
-			return rule
+		if s.ruleMatchesOptimized(rule, statusCode, body, &bodyLower, &bodyLowerDone) {
+			return rule.ErrorPassthroughRule
 		}
 	}
 
@@ -168,7 +180,7 @@ func (s *ErrorPassthroughService) MatchRule(platform string, statusCode int, bod
 }
 
 // getCachedRules 获取缓存的规则列表（按优先级排序）
-func (s *ErrorPassthroughService) getCachedRules() []*model.ErrorPassthroughRule {
+func (s *ErrorPassthroughService) getCachedRules() []*cachedPassthroughRule {
 	s.localCacheMu.RLock()
 	rules := s.localCache
 	s.localCacheMu.RUnlock()
@@ -223,17 +235,39 @@ func (s *ErrorPassthroughService) reloadRulesFromDB(ctx context.Context) error {
 	return nil
 }
 
-// setLocalCache 设置本地缓存
+// setLocalCache 设置本地缓存，预计算小写值和 set 以避免运行时重复计算
 func (s *ErrorPassthroughService) setLocalCache(rules []*model.ErrorPassthroughRule) {
+	cached := make([]*cachedPassthroughRule, len(rules))
+	for i, r := range rules {
+		cr := &cachedPassthroughRule{ErrorPassthroughRule: r}
+		if len(r.Keywords) > 0 {
+			cr.lowerKeywords = make([]string, len(r.Keywords))
+			for j, kw := range r.Keywords {
+				cr.lowerKeywords[j] = strings.ToLower(kw)
+			}
+		}
+		if len(r.Platforms) > 0 {
+			cr.lowerPlatforms = make([]string, len(r.Platforms))
+			for j, p := range r.Platforms {
+				cr.lowerPlatforms[j] = strings.ToLower(p)
+			}
+		}
+		if len(r.ErrorCodes) > 0 {
+			cr.errorCodeSet = make(map[int]struct{}, len(r.ErrorCodes))
+			for _, code := range r.ErrorCodes {
+				cr.errorCodeSet[code] = struct{}{}
+			}
+		}
+		cached[i] = cr
+	}
+
 	// 按优先级排序
-	sorted := make([]*model.ErrorPassthroughRule, len(rules))
-	copy(sorted, rules)
-	sort.Slice(sorted, func(i, j int) bool {
-		return sorted[i].Priority < sorted[j].Priority
+	sort.Slice(cached, func(i, j int) bool {
+		return cached[i].Priority < cached[j].Priority
 	})
 
 	s.localCacheMu.Lock()
-	s.localCache = sorted
+	s.localCache = cached
 	s.localCacheMu.Unlock()
 }
 
@@ -273,62 +307,79 @@ func (s *ErrorPassthroughService) invalidateAndNotify(ctx context.Context) {
 	}
 }
 
-// platformMatches 检查平台是否匹配
-func (s *ErrorPassthroughService) platformMatches(rule *model.ErrorPassthroughRule, platform string) bool {
-	// 如果没有配置平台限制，则匹配所有平台
-	if len(rule.Platforms) == 0 {
+// ensureBodyLower 延迟初始化 body 的小写版本，只做一次转换，限制 8KB
+func ensureBodyLower(body []byte, bodyLower *string, done *bool) string {
+	if *done {
+		return *bodyLower
+	}
+	b := body
+	if len(b) > maxBodyMatchLen {
+		b = b[:maxBodyMatchLen]
+	}
+	*bodyLower = strings.ToLower(string(b))
+	*done = true
+	return *bodyLower
+}
+
+// platformMatchesCached 使用预计算的小写平台检查是否匹配
+func (s *ErrorPassthroughService) platformMatchesCached(rule *cachedPassthroughRule, lowerPlatform string) bool {
+	if len(rule.lowerPlatforms) == 0 {
 		return true
 	}
-
-	platform = strings.ToLower(platform)
-	for _, p := range rule.Platforms {
-		if strings.ToLower(p) == platform {
+	for _, p := range rule.lowerPlatforms {
+		if p == lowerPlatform {
 			return true
 		}
 	}
-
 	return false
 }
 
-// ruleMatches 检查规则是否匹配
-func (s *ErrorPassthroughService) ruleMatches(rule *model.ErrorPassthroughRule, statusCode int, bodyLower string) bool {
-	hasErrorCodes := len(rule.ErrorCodes) > 0
-	hasKeywords := len(rule.Keywords) > 0
+// ruleMatchesOptimized 优化的规则匹配，支持短路和延迟 body 转换
+func (s *ErrorPassthroughService) ruleMatchesOptimized(rule *cachedPassthroughRule, statusCode int, body []byte, bodyLower *string, bodyLowerDone *bool) bool {
+	hasErrorCodes := len(rule.errorCodeSet) > 0
+	hasKeywords := len(rule.lowerKeywords) > 0
 
-	// 如果没有配置任何条件，不匹配
 	if !hasErrorCodes && !hasKeywords {
 		return false
 	}
 
-	codeMatch := !hasErrorCodes || s.containsInt(rule.ErrorCodes, statusCode)
-	keywordMatch := !hasKeywords || s.containsAnyKeyword(bodyLower, rule.Keywords)
+	codeMatch := !hasErrorCodes || s.containsIntSet(rule.errorCodeSet, statusCode)
 
 	if rule.MatchMode == model.MatchModeAll {
-		// "all" 模式：所有配置的条件都必须满足
-		return codeMatch && keywordMatch
+		// "all" 模式：所有配置的条件都必须满足，短路
+		if hasErrorCodes && !codeMatch {
+			return false
+		}
+		if hasKeywords {
+			return s.containsAnyKeywordCached(ensureBodyLower(body, bodyLower, bodyLowerDone), rule.lowerKeywords)
+		}
+		return codeMatch
 	}
 
-	// "any" 模式：任一条件满足即可
+	// "any" 模式：任一条件满足即可，短路
 	if hasErrorCodes && hasKeywords {
-		return codeMatch || keywordMatch
+		if codeMatch {
+			return true
+		}
+		return s.containsAnyKeywordCached(ensureBodyLower(body, bodyLower, bodyLowerDone), rule.lowerKeywords)
 	}
-	return codeMatch && keywordMatch
+	// 只配置了一种条件
+	if hasKeywords {
+		return s.containsAnyKeywordCached(ensureBodyLower(body, bodyLower, bodyLowerDone), rule.lowerKeywords)
+	}
+	return codeMatch
 }
 
-// containsInt 检查切片是否包含指定整数
-func (s *ErrorPassthroughService) containsInt(slice []int, val int) bool {
-	for _, v := range slice {
-		if v == val {
-			return true
-		}
-	}
-	return false
-}
-
-// containsAnyKeyword 检查字符串是否包含任一关键词（不区分大小写）
-func (s *ErrorPassthroughService) containsAnyKeyword(bodyLower string, keywords []string) bool {
-	for _, kw := range keywords {
-		if strings.Contains(bodyLower, strings.ToLower(kw)) {
+// containsIntSet 使用 map 查找替代线性扫描
+func (s *ErrorPassthroughService) containsIntSet(set map[int]struct{}, val int) bool {
+	_, ok := set[val]
+	return ok
+}
+
+// containsAnyKeywordCached 使用预计算的小写关键词检查匹配
+func (s *ErrorPassthroughService) containsAnyKeywordCached(bodyLower string, lowerKeywords []string) bool {
+	for _, kw := range lowerKeywords {
+		if strings.Contains(bodyLower, kw) {
 			return true
 		}
 	}

backend/internal/service/error_passthrough_service_test.go

@@ -145,32 +145,58 @@ func newTestService(rules []*model.ErrorPassthroughRule) *ErrorPassthroughServic
 	return svc
 }
 
+// newCachedRuleForTest 从 model.ErrorPassthroughRule 创建 cachedPassthroughRule（测试用）
+func newCachedRuleForTest(rule *model.ErrorPassthroughRule) *cachedPassthroughRule {
+	cr := &cachedPassthroughRule{ErrorPassthroughRule: rule}
+	if len(rule.Keywords) > 0 {
+		cr.lowerKeywords = make([]string, len(rule.Keywords))
+		for j, kw := range rule.Keywords {
+			cr.lowerKeywords[j] = strings.ToLower(kw)
+		}
+	}
+	if len(rule.Platforms) > 0 {
+		cr.lowerPlatforms = make([]string, len(rule.Platforms))
+		for j, p := range rule.Platforms {
+			cr.lowerPlatforms[j] = strings.ToLower(p)
+		}
+	}
+	if len(rule.ErrorCodes) > 0 {
+		cr.errorCodeSet = make(map[int]struct{}, len(rule.ErrorCodes))
+		for _, code := range rule.ErrorCodes {
+			cr.errorCodeSet[code] = struct{}{}
+		}
+	}
+	return cr
+}
+
 // =============================================================================
-// 测试 ruleMatches 核心匹配逻辑
+// 测试 ruleMatchesOptimized 核心匹配逻辑
 // =============================================================================
 
 func TestRuleMatches_NoConditions(t *testing.T) {
 	// 没有配置任何条件时，不应该匹配
 	svc := newTestService(nil)
-	rule := &model.ErrorPassthroughRule{
+	rule := newCachedRuleForTest(&model.ErrorPassthroughRule{
 		Enabled:    true,
 		ErrorCodes: []int{},
 		Keywords:   []string{},
 		MatchMode:  model.MatchModeAny,
-	}
+	})
 
-	assert.False(t, svc.ruleMatches(rule, 422, "some error message"),
+	var bodyLower string
+	var bodyLowerDone bool
+	assert.False(t, svc.ruleMatchesOptimized(rule, 422, []byte("some error message"), &bodyLower, &bodyLowerDone),
 		"没有配置条件时不应该匹配")
 }
 
 func TestRuleMatches_OnlyErrorCodes_AnyMode(t *testing.T) {
 	svc := newTestService(nil)
-	rule := &model.ErrorPassthroughRule{
+	rule := newCachedRuleForTest(&model.ErrorPassthroughRule{
 		Enabled:    true,
 		ErrorCodes: []int{422, 400},
 		Keywords:   []string{},
 		MatchMode:  model.MatchModeAny,
-	}
+	})
 
 	tests := []struct {
 		name       string
@@ -186,7 +212,9 @@ func TestRuleMatches_OnlyErrorCodes_AnyMode(t *testing.T) {
 
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
-			result := svc.ruleMatches(rule, tt.statusCode, tt.body)
+			var bodyLower string
+			var bodyLowerDone bool
+			result := svc.ruleMatchesOptimized(rule, tt.statusCode, []byte(tt.body), &bodyLower, &bodyLowerDone)
 			assert.Equal(t, tt.expected, result)
 		})
 	}
@@ -194,12 +222,12 @@ func TestRuleMatches_OnlyErrorCodes_AnyMode(t *testing.T) {
 
 func TestRuleMatches_OnlyKeywords_AnyMode(t *testing.T) {
 	svc := newTestService(nil)
-	rule := &model.ErrorPassthroughRule{
+	rule := newCachedRuleForTest(&model.ErrorPassthroughRule{
 		Enabled:    true,
 		ErrorCodes: []int{},
 		Keywords:   []string{"context limit", "model not supported"},
 		MatchMode:  model.MatchModeAny,
-	}
+	})
 
 	tests := []struct {
 		name       string
@@ -210,16 +238,14 @@ func TestRuleMatches_OnlyKeywords_AnyMode(t *testing.T) {
 		{"关键词匹配 context limit", 500, "error: context limit reached", true},
 		{"关键词匹配 model not supported", 400, "the model not supported here", true},
 		{"关键词不匹配", 422, "some other error", false},
-		// 注意：ruleMatches 接收的 body 参数应该是已经转换为小写的
-		// 实际使用时，MatchRule 会先将 body 转换为小写再传给 ruleMatches
-		{"关键词大小写 - 输入已小写", 500, "context limit exceeded", true},
+		{"关键词大小写 - 自动转换", 500, "Context Limit exceeded", true},
 	}
 
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
-			// 模拟 MatchRule 的行为：先转换为小写
-			bodyLower := strings.ToLower(tt.body)
-			result := svc.ruleMatches(rule, tt.statusCode, bodyLower)
+			var bodyLower string
+			var bodyLowerDone bool
+			result := svc.ruleMatchesOptimized(rule, tt.statusCode, []byte(tt.body), &bodyLower, &bodyLowerDone)
 			assert.Equal(t, tt.expected, result)
 		})
 	}
@@ -228,12 +254,12 @@ func TestRuleMatches_OnlyKeywords_AnyMode(t *testing.T) {
 func TestRuleMatches_BothConditions_AnyMode(t *testing.T) {
 	// any 模式：错误码 OR 关键词
 	svc := newTestService(nil)
-	rule := &model.ErrorPassthroughRule{
+	rule := newCachedRuleForTest(&model.ErrorPassthroughRule{
 		Enabled:    true,
 		ErrorCodes: []int{422, 400},
 		Keywords:   []string{"context limit"},
 		MatchMode:  model.MatchModeAny,
-	}
+	})
 
 	tests := []struct {
 		name       string
@@ -274,7 +300,9 @@ func TestRuleMatches_BothConditions_AnyMode(t *testing.T) {
 
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
-			result := svc.ruleMatches(rule, tt.statusCode, tt.body)
+			var bodyLower string
+			var bodyLowerDone bool
+			result := svc.ruleMatchesOptimized(rule, tt.statusCode, []byte(tt.body), &bodyLower, &bodyLowerDone)
 			assert.Equal(t, tt.expected, result, tt.reason)
 		})
 	}
@@ -283,12 +311,12 @@ func TestRuleMatches_BothConditions_AnyMode(t *testing.T) {
 func TestRuleMatches_BothConditions_AllMode(t *testing.T) {
 	// all 模式：错误码 AND 关键词
 	svc := newTestService(nil)
-	rule := &model.ErrorPassthroughRule{
+	rule := newCachedRuleForTest(&model.ErrorPassthroughRule{
 		Enabled:    true,
 		ErrorCodes: []int{422, 400},
 		Keywords:   []string{"context limit"},
 		MatchMode:  model.MatchModeAll,
-	}
+	})
 
 	tests := []struct {
 		name       string
@@ -329,14 +357,16 @@ func TestRuleMatches_BothConditions_AllMode(t *testing.T) {
 
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
-			result := svc.ruleMatches(rule, tt.statusCode, tt.body)
+			var bodyLower string
+			var bodyLowerDone bool
+			result := svc.ruleMatchesOptimized(rule, tt.statusCode, []byte(tt.body), &bodyLower, &bodyLowerDone)
 			assert.Equal(t, tt.expected, result, tt.reason)
 		})
 	}
 }
 
 // =============================================================================
-// 测试 platformMatches 平台匹配逻辑
+// 测试 platformMatchesCached 平台匹配逻辑
 // =============================================================================
 
 func TestPlatformMatches(t *testing.T) {
@@ -394,10 +424,10 @@ func TestPlatformMatches(t *testing.T) {
 
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
-			rule := &model.ErrorPassthroughRule{
+			rule := newCachedRuleForTest(&model.ErrorPassthroughRule{
 				Platforms: tt.rulePlatforms,
-			}
-			result := svc.platformMatches(rule, tt.requestPlatform)
+			})
+			result := svc.platformMatchesCached(rule, strings.ToLower(tt.requestPlatform))
 			assert.Equal(t, tt.expected, result)
 		})
 	}

backend/internal/service/gateway_request.go

@@ -105,8 +105,9 @@ func ParseGatewayRequest(body []byte, protocol string) (*ParsedRequest, error) {
 	// metadata.user_id: 直接路径提取，不需要严格类型校验
 	parsed.MetadataUserID = gjson.Get(jsonStr, "metadata.user_id").String()
 
-	// thinking.type: 直接路径提取
-	if gjson.Get(jsonStr, "thinking.type").String() == "enabled" {
+	// thinking.type: enabled/adaptive 都视为开启
+	thinkingType := gjson.Get(jsonStr, "thinking.type").String()
+	if thinkingType == "enabled" || thinkingType == "adaptive" {
 		parsed.ThinkingEnabled = true
 	}
 
@@ -193,9 +194,9 @@ func sliceRawFromBody(body []byte, r gjson.Result) []byte {
 // Returns filtered body or original body if filtering fails (fail-safe)
 // This prevents 400 errors from invalid thinking block signatures
 //
-// Strategy:
-//   - When thinking.type != "enabled": Remove all thinking blocks
-//   - When thinking.type == "enabled": Only remove thinking blocks without valid signatures
+// 策略：
+//   - 当 thinking.type 不是 "enabled"/"adaptive"：移除所有 thinking 相关块
+//   - 当 thinking.type 是 "enabled"/"adaptive"：仅移除缺失/无效 signature 的 thinking 块（避免 400）
 //     (blocks with missing/empty/dummy signatures that would cause 400 errors)
 func FilterThinkingBlocks(body []byte) []byte {
 	return filterThinkingBlocksInternal(body, false)
@@ -571,9 +572,9 @@ func FilterSignatureSensitiveBlocksForRetry(body []byte) []byte {
 }
 
 // filterThinkingBlocksInternal removes invalid thinking blocks from request
-// Strategy:
-//   - When thinking.type != "enabled": Remove all thinking blocks
-//   - When thinking.type == "enabled": Only remove thinking blocks without valid signatures
+// 策略：
+//   - 当 thinking.type 不是 "enabled"/"adaptive"：移除所有 thinking 相关块
+//   - 当 thinking.type 是 "enabled"/"adaptive"：仅移除缺失/无效 signature 的 thinking 块
 func filterThinkingBlocksInternal(body []byte, _ bool) []byte {
 	// Fast path: if body doesn't contain "thinking", skip parsing
 	if !bytes.Contains(body, []byte(`"type":"thinking"`)) &&
@@ -593,7 +594,7 @@ func filterThinkingBlocksInternal(body []byte, _ bool) []byte {
 	// Check if thinking is enabled
 	thinkingEnabled := false
 	if thinking, ok := req["thinking"].(map[string]any); ok {
-		if thinkType, ok := thinking["type"].(string); ok && thinkType == "enabled" {
+		if thinkType, ok := thinking["type"].(string); ok && (thinkType == "enabled" || thinkType == "adaptive") {
 			thinkingEnabled = true
 		}
 	}

backend/internal/service/gateway_request_test.go

@@ -33,6 +33,14 @@ func TestParseGatewayRequest_ThinkingEnabled(t *testing.T) {
 	require.True(t, parsed.ThinkingEnabled)
 }
 
+func TestParseGatewayRequest_ThinkingAdaptiveEnabled(t *testing.T) {
+	body := []byte(`{"model":"claude-sonnet-4-5","thinking":{"type":"adaptive"},"messages":[{"content":"hi"}]}`)
+	parsed, err := ParseGatewayRequest(body, "")
+	require.NoError(t, err)
+	require.Equal(t, "claude-sonnet-4-5", parsed.Model)
+	require.True(t, parsed.ThinkingEnabled)
+}
+
 func TestParseGatewayRequest_MaxTokens(t *testing.T) {
 	body := []byte(`{"model":"claude-haiku-4-5","max_tokens":1}`)
 	parsed, err := ParseGatewayRequest(body, "")
@@ -213,6 +221,16 @@ func TestFilterThinkingBlocks(t *testing.T) {
 			input:        `{"model":"claude-3-5-sonnet-20241022","messages":[{"role":"user","content":[{"type":"text","text":"Hello"},{"type":"thinking","thinking":"internal","signature":"invalid"},{"type":"text","text":"World"}]}]}`,
 			shouldFilter: true,
 		},
+		{
+			name:         "does not filter signed thinking blocks when thinking adaptive",
+			input:        `{"thinking":{"type":"adaptive"},"messages":[{"role":"assistant","content":[{"type":"thinking","thinking":"ok","signature":"sig_real_123"},{"type":"text","text":"B"}]}]}`,
+			shouldFilter: false,
+		},
+		{
+			name:         "filters unsigned thinking blocks when thinking adaptive",
+			input:        `{"thinking":{"type":"adaptive"},"messages":[{"role":"assistant","content":[{"type":"thinking","thinking":"internal","signature":""},{"type":"text","text":"B"}]}]}`,
+			shouldFilter: true,
+		},
 		{
 			name:         "handles no thinking blocks",
 			input:        `{"model":"claude-3-5-sonnet-20241022","messages":[{"role":"user","content":[{"type":"text","text":"Hello"}]}]}`,

backend/internal/service/gateway_service.go

@@ -372,15 +372,31 @@ type ForwardResult struct {
 
 // UpstreamFailoverError indicates an upstream error that should trigger account failover.
 type UpstreamFailoverError struct {
-	StatusCode        int
-	ResponseBody      []byte // 上游响应体，用于错误透传规则匹配
-	ForceCacheBilling bool   // Antigravity 粘性会话切换时设为 true
+	StatusCode             int
+	ResponseBody           []byte // 上游响应体，用于错误透传规则匹配
+	ForceCacheBilling      bool   // Antigravity 粘性会话切换时设为 true
+	RetryableOnSameAccount bool   // 临时性错误（如 Google 间歇性 400、空响应），应在同一账号上重试 N 次再切换
 }
 
 func (e *UpstreamFailoverError) Error() string {
 	return fmt.Sprintf("upstream error: %d (failover)", e.StatusCode)
 }
 
+// TempUnscheduleRetryableError 对 RetryableOnSameAccount 类型的 failover 错误触发临时封禁。
+// 由 handler 层在同账号重试全部用尽、切换账号时调用。
+func (s *GatewayService) TempUnscheduleRetryableError(ctx context.Context, accountID int64, failoverErr *UpstreamFailoverError) {
+	if failoverErr == nil || !failoverErr.RetryableOnSameAccount {
+		return
+	}
+	// 根据状态码选择封禁策略
+	switch failoverErr.StatusCode {
+	case http.StatusBadRequest:
+		tempUnscheduleGoogleConfigError(ctx, s.accountRepo, accountID, "[handler]")
+	case http.StatusBadGateway:
+		tempUnscheduleEmptyResponse(ctx, s.accountRepo, accountID, "[handler]")
+	}
+}
+
 // GatewayService handles API gateway operations
 type GatewayService struct {
 	accountRepo         AccountRepository
@@ -3644,7 +3660,8 @@ func requestNeedsBetaFeatures(body []byte) bool {
 	if tools.Exists() && tools.IsArray() && len(tools.Array()) > 0 {
 		return true
 	}
-	if strings.EqualFold(gjson.GetBytes(body, "thinking.type").String(), "enabled") {
+	thinkingType := gjson.GetBytes(body, "thinking.type").String()
+	if strings.EqualFold(thinkingType, "enabled") || strings.EqualFold(thinkingType, "adaptive") {
 		return true
 	}
 	return false

backend/internal/service/gemini_messages_compat_service.go

@@ -10,6 +10,7 @@ import (
 	"errors"
 	"fmt"
 	"io"
+	"log"
 	"math"
 	mathrand "math/rand"
 	"net/http"
@@ -881,6 +882,37 @@ func (s *GeminiMessagesCompatService) Forward(ctx context.Context, c *gin.Contex
 
 		// ErrorPolicyNone → 原有逻辑
 		s.handleGeminiUpstreamError(ctx, account, resp.StatusCode, resp.Header, respBody)
+		// 精确匹配服务端配置类 400 错误，触发 failover + 临时封禁
+		if resp.StatusCode == http.StatusBadRequest {
+			msg400 := strings.ToLower(strings.TrimSpace(extractUpstreamErrorMessage(respBody)))
+			if isGoogleProjectConfigError(msg400) {
+				upstreamReqID := resp.Header.Get(requestIDHeader)
+				if upstreamReqID == "" {
+					upstreamReqID = resp.Header.Get("x-goog-request-id")
+				}
+				upstreamMsg := sanitizeUpstreamErrorMessage(strings.TrimSpace(extractUpstreamErrorMessage(respBody)))
+				upstreamDetail := ""
+				if s.cfg != nil && s.cfg.Gateway.LogUpstreamErrorBody {
+					maxBytes := s.cfg.Gateway.LogUpstreamErrorBodyMaxBytes
+					if maxBytes <= 0 {
+						maxBytes = 2048
+					}
+					upstreamDetail = truncateString(string(respBody), maxBytes)
+				}
+				log.Printf("[Gemini] status=400 google_config_error failover=true upstream_message=%q account=%d", upstreamMsg, account.ID)
+				appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
+					Platform:           account.Platform,
+					AccountID:          account.ID,
+					AccountName:        account.Name,
+					UpstreamStatusCode: resp.StatusCode,
+					UpstreamRequestID:  upstreamReqID,
+					Kind:               "failover",
+					Message:            upstreamMsg,
+					Detail:             upstreamDetail,
+				})
+				return nil, &UpstreamFailoverError{StatusCode: resp.StatusCode, ResponseBody: respBody, RetryableOnSameAccount: true}
+			}
+		}
 		if s.shouldFailoverGeminiUpstreamError(resp.StatusCode) {
 			upstreamReqID := resp.Header.Get(requestIDHeader)
 			if upstreamReqID == "" {
@@ -1332,6 +1364,34 @@ func (s *GeminiMessagesCompatService) ForwardNative(ctx context.Context, c *gin.
 
 		// ErrorPolicyNone → 原有逻辑
 		s.handleGeminiUpstreamError(ctx, account, resp.StatusCode, resp.Header, respBody)
+		// 精确匹配服务端配置类 400 错误，触发 failover + 临时封禁
+		if resp.StatusCode == http.StatusBadRequest {
+			msg400 := strings.ToLower(strings.TrimSpace(extractUpstreamErrorMessage(respBody)))
+			if isGoogleProjectConfigError(msg400) {
+				evBody := unwrapIfNeeded(isOAuth, respBody)
+				upstreamMsg := sanitizeUpstreamErrorMessage(strings.TrimSpace(extractUpstreamErrorMessage(evBody)))
+				upstreamDetail := ""
+				if s.cfg != nil && s.cfg.Gateway.LogUpstreamErrorBody {
+					maxBytes := s.cfg.Gateway.LogUpstreamErrorBodyMaxBytes
+					if maxBytes <= 0 {
+						maxBytes = 2048
+					}
+					upstreamDetail = truncateString(string(evBody), maxBytes)
+				}
+				log.Printf("[Gemini] status=400 google_config_error failover=true upstream_message=%q account=%d", upstreamMsg, account.ID)
+				appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
+					Platform:           account.Platform,
+					AccountID:          account.ID,
+					AccountName:        account.Name,
+					UpstreamStatusCode: resp.StatusCode,
+					UpstreamRequestID:  requestID,
+					Kind:               "failover",
+					Message:            upstreamMsg,
+					Detail:             upstreamDetail,
+				})
+				return nil, &UpstreamFailoverError{StatusCode: resp.StatusCode, ResponseBody: evBody, RetryableOnSameAccount: true}
+			}
+		}
 		if s.shouldFailoverGeminiUpstreamError(resp.StatusCode) {
 			evBody := unwrapIfNeeded(isOAuth, respBody)
 			upstreamMsg := strings.TrimSpace(extractUpstreamErrorMessage(evBody))
@@ -2583,11 +2643,12 @@ func extractGeminiUsage(data []byte) *ClaudeUsage {
 	prompt := int(usage.Get("promptTokenCount").Int())
 	cand := int(usage.Get("candidatesTokenCount").Int())
 	cached := int(usage.Get("cachedContentTokenCount").Int())
+	thoughts := int(usage.Get("thoughtsTokenCount").Int())
 	// 注意：Gemini 的 promptTokenCount 包含 cachedContentTokenCount，
 	// 但 Claude 的 input_tokens 不包含 cache_read_input_tokens，需要减去
 	return &ClaudeUsage{
 		InputTokens:          prompt - cached,
-		OutputTokens:         cand,
+		OutputTokens:         cand + thoughts,
 		CacheReadInputTokens: cached,
 	}
 }

backend/internal/service/gemini_messages_compat_service_test.go

@@ -298,6 +298,26 @@ func TestExtractGeminiUsage(t *testing.T) {
 				CacheReadInputTokens: 20,
 			},
 		},
+		{
+			name:    "包含 thoughtsTokenCount",
+			input:   `{"usageMetadata":{"promptTokenCount":100,"candidatesTokenCount":20,"thoughtsTokenCount":50}}`,
+			wantNil: false,
+			wantUsage: &ClaudeUsage{
+				InputTokens:          100,
+				OutputTokens:         70,
+				CacheReadInputTokens: 0,
+			},
+		},
+		{
+			name:    "包含 thoughtsTokenCount 与缓存",
+			input:   `{"usageMetadata":{"promptTokenCount":100,"candidatesTokenCount":20,"cachedContentTokenCount":30,"thoughtsTokenCount":50}}`,
+			wantNil: false,
+			wantUsage: &ClaudeUsage{
+				InputTokens:          70,
+				OutputTokens:         70,
+				CacheReadInputTokens: 30,
+			},
+		},
 		{
 			name:    "缺失 cachedContentTokenCount",
 			input:   `{"usageMetadata":{"promptTokenCount":100,"candidatesTokenCount":50}}`,

backend/internal/service/openai_gateway_service_codex_cli_only_test.go

@@ -95,7 +95,7 @@ func TestGetAPIKeyIDFromContext(t *testing.T) {
 func TestLogCodexCLIOnlyDetection_NilSafety(t *testing.T) {
 	// 不校验日志内容，仅保证在 nil 入参下不会 panic。
 	require.NotPanics(t, func() {
-		logCodexCLIOnlyDetection(nil, nil, 0, CodexClientRestrictionDetectionResult{Enabled: true, Matched: false, Reason: "test"})
+		logCodexCLIOnlyDetection(context.TODO(), nil, 0, CodexClientRestrictionDetectionResult{Enabled: true, Matched: false, Reason: "test"})
 		logCodexCLIOnlyDetection(context.Background(), nil, 0, CodexClientRestrictionDetectionResult{Enabled: false, Matched: false, Reason: "disabled"})
 	})
 }

backend/internal/service/ops_upstream_context.go

@@ -27,6 +27,10 @@ const (
 	OpsUpstreamLatencyMsKey  = "ops_upstream_latency_ms"
 	OpsResponseLatencyMsKey  = "ops_response_latency_ms"
 	OpsTimeToFirstTokenMsKey = "ops_time_to_first_token_ms"
+
+	// OpsSkipPassthroughKey 由 applyErrorPassthroughRule 在命中 skip_monitoring=true 的规则时设置。
+	// ops_error_logger 中间件检查此 key，为 true 时跳过错误记录。
+	OpsSkipPassthroughKey = "ops_skip_passthrough"
 )
 
 func setOpsUpstreamRequestBody(c *gin.Context, body []byte) {
@@ -132,6 +136,37 @@ func appendOpsUpstreamError(c *gin.Context, ev OpsUpstreamErrorEvent) {
 	evCopy := ev
 	existing = append(existing, &evCopy)
 	c.Set(OpsUpstreamErrorsKey, existing)
+
+	checkSkipMonitoringForUpstreamEvent(c, &evCopy)
+}
+
+// checkSkipMonitoringForUpstreamEvent checks whether the upstream error event
+// matches a passthrough rule with skip_monitoring=true and, if so, sets the
+// OpsSkipPassthroughKey on the context.  This ensures intermediate retry /
+// failover errors (which never go through the final applyErrorPassthroughRule
+// path) can still suppress ops_error_logs recording.
+func checkSkipMonitoringForUpstreamEvent(c *gin.Context, ev *OpsUpstreamErrorEvent) {
+	if ev.UpstreamStatusCode == 0 {
+		return
+	}
+
+	svc := getBoundErrorPassthroughService(c)
+	if svc == nil {
+		return
+	}
+
+	// Use the best available body representation for keyword matching.
+	// Even when body is empty, MatchRule can still match rules that only
+	// specify ErrorCodes (no Keywords), so we always call it.
+	body := ev.Detail
+	if body == "" {
+		body = ev.Message
+	}
+
+	rule := svc.MatchRule(ev.Platform, ev.UpstreamStatusCode, []byte(body))
+	if rule != nil && rule.SkipMonitoring {
+		c.Set(OpsSkipPassthroughKey, true)
+	}
 }
 
 func marshalOpsUpstreamErrors(events []*OpsUpstreamErrorEvent) *string {

backend/migrations/053_add_skip_monitoring_to_error_passthrough.sql

@@ -0,0 +1,4 @@
+-- Add skip_monitoring field to error_passthrough_rules table
+-- When true, errors matching this rule will not be recorded in ops_error_logs
+ALTER TABLE error_passthrough_rules
+ADD COLUMN IF NOT EXISTS skip_monitoring BOOLEAN NOT NULL DEFAULT false;

frontend/src/api/admin/antigravity.ts

@@ -53,4 +53,18 @@ export async function exchangeCode(
   return data
 }
 
-export default { generateAuthUrl, exchangeCode }
+export async function refreshAntigravityToken(
+  refreshToken: string,
+  proxyId?: number | null
+): Promise<AntigravityTokenInfo> {
+  const payload: Record<string, any> = { refresh_token: refreshToken }
+  if (proxyId) payload.proxy_id = proxyId
+  
+  const { data } = await apiClient.post<AntigravityTokenInfo>(
+    '/admin/antigravity/oauth/refresh-token',
+    payload
+  )
+  return data
+}
+
+export default { generateAuthUrl, exchangeCode, refreshAntigravityToken }

frontend/src/api/admin/errorPassthrough.ts

@@ -21,6 +21,7 @@ export interface ErrorPassthroughRule {
   response_code: number | null
   passthrough_body: boolean
   custom_message: string | null
+  skip_monitoring: boolean
   description: string | null
   created_at: string
   updated_at: string
@@ -41,6 +42,7 @@ export interface CreateRuleRequest {
   response_code?: number | null
   passthrough_body?: boolean
   custom_message?: string | null
+  skip_monitoring?: boolean
   description?: string | null
 }
 
@@ -59,6 +61,7 @@ export interface UpdateRuleRequest {
   response_code?: number | null
   passthrough_body?: boolean
   custom_message?: string | null
+  skip_monitoring?: boolean
   description?: string | null
 }
 

frontend/src/components/account/CreateAccountModal.vue

@@ -1744,12 +1744,12 @@
         :show-proxy-warning="form.platform !== 'openai' && !!form.proxy_id"
         :allow-multiple="form.platform === 'anthropic'"
         :show-cookie-option="form.platform === 'anthropic'"
-        :show-refresh-token-option="form.platform === 'openai'"
+        :show-refresh-token-option="form.platform === 'openai' || form.platform === 'antigravity'"
         :platform="form.platform"
         :show-project-id="geminiOAuthType === 'code_assist'"
         @generate-url="handleGenerateUrl"
         @cookie-auth="handleCookieAuth"
-        @validate-refresh-token="handleOpenAIValidateRT"
+        @validate-refresh-token="handleValidateRefreshToken"
       />
 
     </div>
@@ -2948,6 +2948,14 @@ const handleGenerateUrl = async () => {
   }
 }
 
+const handleValidateRefreshToken = (rt: string) => {
+  if (form.platform === 'openai') {
+    handleOpenAIValidateRT(rt)
+  } else if (form.platform === 'antigravity') {
+    handleAntigravityValidateRT(rt)
+  }
+}
+
 const formatDateTimeLocal = formatDateTimeLocalInput
 const parseDateTimeLocal = parseDateTimeLocalInput
 
@@ -3165,6 +3173,95 @@ const handleOpenAIValidateRT = async (refreshTokenInput: string) => {
   }
 }
 
+// Antigravity 手动 RT 批量验证和创建
+const handleAntigravityValidateRT = async (refreshTokenInput: string) => {
+  if (!refreshTokenInput.trim()) return
+
+  // Parse multiple refresh tokens (one per line)
+  const refreshTokens = refreshTokenInput
+    .split('\n')
+    .map((rt) => rt.trim())
+    .filter((rt) => rt)
+
+  if (refreshTokens.length === 0) {
+    antigravityOAuth.error.value = t('admin.accounts.oauth.antigravity.pleaseEnterRefreshToken')
+    return
+  }
+
+  antigravityOAuth.loading.value = true
+  antigravityOAuth.error.value = ''
+
+  let successCount = 0
+  let failedCount = 0
+  const errors: string[] = []
+
+  try {
+    for (let i = 0; i < refreshTokens.length; i++) {
+      try {
+        const tokenInfo = await antigravityOAuth.validateRefreshToken(
+          refreshTokens[i],
+          form.proxy_id
+        )
+        if (!tokenInfo) {
+          failedCount++
+          errors.push(`#${i + 1}: ${antigravityOAuth.error.value || 'Validation failed'}`)
+          antigravityOAuth.error.value = ''
+          continue
+        }
+
+        const credentials = antigravityOAuth.buildCredentials(tokenInfo)
+        
+        // Generate account name with index for batch
+        const accountName = refreshTokens.length > 1 ? `${form.name} #${i + 1}` : form.name
+
+        // Note: Antigravity doesn't have buildExtraInfo, so we pass empty extra or rely on credentials
+        await adminAPI.accounts.create({
+          name: accountName,
+          notes: form.notes,
+          platform: 'antigravity',
+          type: 'oauth',
+          credentials,
+          extra: {},
+          proxy_id: form.proxy_id,
+          concurrency: form.concurrency,
+          priority: form.priority,
+          rate_multiplier: form.rate_multiplier,
+          group_ids: form.group_ids,
+          expires_at: form.expires_at,
+          auto_pause_on_expired: autoPauseOnExpired.value
+        })
+        successCount++
+      } catch (error: any) {
+        failedCount++
+        const errMsg = error.response?.data?.detail || error.message || 'Unknown error'
+        errors.push(`#${i + 1}: ${errMsg}`)
+      }
+    }
+
+    // Show results
+    if (successCount > 0 && failedCount === 0) {
+      appStore.showSuccess(
+        refreshTokens.length > 1
+          ? t('admin.accounts.oauth.batchSuccess', { count: successCount })
+          : t('admin.accounts.accountCreated')
+      )
+      emit('created')
+      handleClose()
+    } else if (successCount > 0 && failedCount > 0) {
+      appStore.showWarning(
+        t('admin.accounts.oauth.batchPartialSuccess', { success: successCount, failed: failedCount })
+      )
+      antigravityOAuth.error.value = errors.join('\n')
+      emit('created')
+    } else {
+      antigravityOAuth.error.value = errors.join('\n')
+      appStore.showError(t('admin.accounts.oauth.batchFailed'))
+    }
+  } finally {
+    antigravityOAuth.loading.value = false
+  }
+}
+
 // Gemini OAuth 授权码兑换
 const handleGeminiExchange = async (authCode: string) => {
   if (!authCode.trim() || !geminiOAuth.sessionId.value) return

frontend/src/components/account/OAuthAuthorizationFlow.vue

@@ -45,19 +45,19 @@
                 class="text-blue-600 focus:ring-blue-500"
               />
               <span class="text-sm text-blue-900 dark:text-blue-200">{{
-                t('admin.accounts.oauth.openai.refreshTokenAuth')
+                t(getOAuthKey('refreshTokenAuth'))
               }}</span>
             </label>
           </div>
         </div>
 
-        <!-- Refresh Token Input (OpenAI only) -->
+        <!-- Refresh Token Input (OpenAI / Antigravity) -->
         <div v-if="inputMethod === 'refresh_token'" class="space-y-4">
           <div
             class="rounded-lg border border-blue-300 bg-white/80 p-4 dark:border-blue-600 dark:bg-gray-800/80"
           >
             <p class="mb-3 text-sm text-blue-700 dark:text-blue-300">
-              {{ t('admin.accounts.oauth.openai.refreshTokenDesc') }}
+              {{ t(getOAuthKey('refreshTokenDesc')) }}
             </p>
 
             <!-- Refresh Token Input -->
@@ -78,7 +78,7 @@
                 v-model="refreshTokenInput"
                 rows="3"
                 class="input w-full resize-y font-mono text-sm"
-                :placeholder="t('admin.accounts.oauth.openai.refreshTokenPlaceholder')"
+                :placeholder="t(getOAuthKey('refreshTokenPlaceholder'))"
               ></textarea>
               <p
                 v-if="parsedRefreshTokenCount > 1"
@@ -128,8 +128,8 @@
               <Icon v-else name="sparkles" size="sm" class="mr-2" />
               {{
                 loading
-                  ? t('admin.accounts.oauth.openai.validating')
-                  : t('admin.accounts.oauth.openai.validateAndCreate')
+                  ? t(getOAuthKey('validating'))
+                  : t(getOAuthKey('validateAndCreate'))
               }}
             </button>
           </div>

frontend/src/components/admin/ErrorPassthroughRulesModal.vue

@@ -148,6 +148,16 @@
                       {{ rule.passthrough_body ? t('admin.errorPassthrough.passthrough') : t('admin.errorPassthrough.custom') }}
                     </span>
                   </div>
+                  <div v-if="rule.skip_monitoring" class="flex items-center gap-1">
+                    <Icon
+                      name="checkCircle"
+                      size="xs"
+                      class="text-yellow-500"
+                    />
+                    <span class="text-gray-600 dark:text-gray-400">
+                      {{ t('admin.errorPassthrough.skipMonitoring') }}
+                    </span>
+                  </div>
                 </div>
               </td>
               <td class="px-3 py-2">
@@ -366,6 +376,19 @@
           </div>
         </div>
 
+        <!-- Skip Monitoring -->
+        <div class="flex items-center gap-1.5">
+          <input
+            type="checkbox"
+            v-model="form.skip_monitoring"
+            class="h-3.5 w-3.5 rounded border-gray-300 text-yellow-600 focus:ring-yellow-500"
+          />
+          <span class="text-xs font-medium text-gray-700 dark:text-gray-300">
+            {{ t('admin.errorPassthrough.form.skipMonitoring') }}
+          </span>
+        </div>
+        <p class="input-hint text-xs -mt-3">{{ t('admin.errorPassthrough.form.skipMonitoringHint') }}</p>
+
         <!-- Enabled -->
         <div class="flex items-center gap-1.5">
           <input
@@ -453,6 +476,7 @@ const form = reactive({
   response_code: null as number | null,
   passthrough_body: true,
   custom_message: null as string | null,
+  skip_monitoring: false,
   description: null as string | null
 })
 
@@ -497,6 +521,7 @@ const resetForm = () => {
   form.response_code = null
   form.passthrough_body = true
   form.custom_message = null
+  form.skip_monitoring = false
   form.description = null
   errorCodesInput.value = ''
   keywordsInput.value = ''
@@ -520,6 +545,7 @@ const handleEdit = (rule: ErrorPassthroughRule) => {
   form.response_code = rule.response_code
   form.passthrough_body = rule.passthrough_body
   form.custom_message = rule.custom_message
+  form.skip_monitoring = rule.skip_monitoring
   form.description = rule.description
   errorCodesInput.value = rule.error_codes.join(', ')
   keywordsInput.value = rule.keywords.join('\n')
@@ -575,6 +601,7 @@ const handleSubmit = async () => {
       response_code: form.passthrough_code ? null : form.response_code,
       passthrough_body: form.passthrough_body,
       custom_message: form.passthrough_body ? null : form.custom_message,
+      skip_monitoring: form.skip_monitoring,
       description: form.description?.trim() || null
     }
 

frontend/src/components/admin/account/AccountTableFilters.vue

@@ -21,5 +21,5 @@ const updateType = (value: string | number | boolean | null) => { emit('update:f
 const updateStatus = (value: string | number | boolean | null) => { emit('update:filters', { ...props.filters, status: value }) }
 const pOpts = computed(() => [{ value: '', label: t('admin.accounts.allPlatforms') }, { value: 'anthropic', label: 'Anthropic' }, { value: 'openai', label: 'OpenAI' }, { value: 'gemini', label: 'Gemini' }, { value: 'antigravity', label: 'Antigravity' }, { value: 'sora', label: 'Sora' }])
 const tOpts = computed(() => [{ value: '', label: t('admin.accounts.allTypes') }, { value: 'oauth', label: t('admin.accounts.oauthType') }, { value: 'setup-token', label: t('admin.accounts.setupToken') }, { value: 'apikey', label: t('admin.accounts.apiKey') }])
-const sOpts = computed(() => [{ value: '', label: t('admin.accounts.allStatus') }, { value: 'active', label: t('admin.accounts.status.active') }, { value: 'inactive', label: t('admin.accounts.status.inactive') }, { value: 'error', label: t('admin.accounts.status.error') }])
+const sOpts = computed(() => [{ value: '', label: t('admin.accounts.allStatus') }, { value: 'active', label: t('admin.accounts.status.active') }, { value: 'inactive', label: t('admin.accounts.status.inactive') }, { value: 'error', label: t('admin.accounts.status.error') }, { value: 'rate_limited', label: t('admin.accounts.status.rateLimited') }])
 </script>

frontend/src/components/layout/AuthLayout.vue

@@ -29,17 +29,19 @@
       <!-- Logo/Brand -->
       <div class="mb-8 text-center">
         <!-- Custom Logo or Default Logo -->
-        <div
-          class="mb-4 inline-flex h-16 w-16 items-center justify-center overflow-hidden rounded-2xl shadow-lg shadow-primary-500/30"
-        >
-          <img :src="siteLogo || '/logo.png'" alt="Logo" class="h-full w-full object-contain" />
-        </div>
-        <h1 class="text-gradient mb-2 text-3xl font-bold">
-          {{ siteName }}
-        </h1>
-        <p class="text-sm text-gray-500 dark:text-dark-400">
-          {{ siteSubtitle }}
-        </p>
+        <template v-if="settingsLoaded">
+          <div
+            class="mb-4 inline-flex h-16 w-16 items-center justify-center overflow-hidden rounded-2xl shadow-lg shadow-primary-500/30"
+          >
+            <img :src="siteLogo || '/logo.png'" alt="Logo" class="h-full w-full object-contain" />
+          </div>
+          <h1 class="text-gradient mb-2 text-3xl font-bold">
+            {{ siteName }}
+          </h1>
+          <p class="text-sm text-gray-500 dark:text-dark-400">
+            {{ siteSubtitle }}
+          </p>
+        </template>
       </div>
 
       <!-- Card Container -->
@@ -61,25 +63,21 @@
 </template>
 
 <script setup lang="ts">
-import { ref, computed, onMounted } from 'vue'
-import { getPublicSettings } from '@/api/auth'
+import { computed, onMounted } from 'vue'
+import { useAppStore } from '@/stores'
 import { sanitizeUrl } from '@/utils/url'
 
-const siteName = ref('Sub2API')
-const siteLogo = ref('')
-const siteSubtitle = ref('Subscription to API Conversion Platform')
+const appStore = useAppStore()
+
+const siteName = computed(() => appStore.siteName || 'Sub2API')
+const siteLogo = computed(() => sanitizeUrl(appStore.siteLogo || '', { allowRelative: true, allowDataUrl: true }))
+const siteSubtitle = computed(() => appStore.cachedPublicSettings?.site_subtitle || 'Subscription to API Conversion Platform')
+const settingsLoaded = computed(() => appStore.publicSettingsLoaded)
 
 const currentYear = computed(() => new Date().getFullYear())
 
-onMounted(async () => {
-  try {
-    const settings = await getPublicSettings()
-    siteName.value = settings.site_name || 'Sub2API'
-    siteLogo.value = sanitizeUrl(settings.site_logo || '', { allowRelative: true })
-    siteSubtitle.value = settings.site_subtitle || 'Subscription to API Conversion Platform'
-  } catch (error) {
-    console.error('Failed to load public settings:', error)
-  }
+onMounted(() => {
+  appStore.fetchPublicSettings()
 })
 </script>
 

frontend/src/composables/useAntigravityOAuth.ts

@@ -83,6 +83,35 @@ export function useAntigravityOAuth() {
     }
   }
 
+  const validateRefreshToken = async (
+    refreshToken: string,
+    proxyId?: number | null
+  ): Promise<AntigravityTokenInfo | null> => {
+    if (!refreshToken.trim()) {
+      error.value = t('admin.accounts.oauth.antigravity.pleaseEnterRefreshToken')
+      return null
+    }
+
+    loading.value = true
+    error.value = ''
+
+    try {
+      const tokenInfo = await adminAPI.antigravity.refreshAntigravityToken(
+        refreshToken.trim(),
+        proxyId
+      )
+      return tokenInfo as AntigravityTokenInfo
+    } catch (err: any) {
+      error.value =
+        err.response?.data?.detail || t('admin.accounts.oauth.antigravity.failedToValidateRT')
+      // Don't show global error toast for batch validation to avoid spamming
+      // appStore.showError(error.value)
+      return null
+    } finally {
+      loading.value = false
+    }
+  }
+
   const buildCredentials = (tokenInfo: AntigravityTokenInfo): Record<string, unknown> => {
     let expiresAt: string | undefined
     if (typeof tokenInfo.expires_at === 'number' && Number.isFinite(tokenInfo.expires_at)) {
@@ -110,6 +139,7 @@ export function useAntigravityOAuth() {
     resetState,
     generateAuthUrl,
     exchangeAuthCode,
+    validateRefreshToken,
     buildCredentials
   }
 }

frontend/src/i18n/locales/en.ts

@@ -841,7 +841,7 @@ export default {
       createUser: 'Create User',
       editUser: 'Edit User',
       deleteUser: 'Delete User',
-      searchUsers: 'Search users...',
+      searchUsers: 'Search by email, username, notes, or API key...',
       allRoles: 'All Roles',
       allStatus: 'All Status',
       admin: 'Admin',
@@ -1798,13 +1798,20 @@ export default {
           authCode: 'Authorization URL or Code',
           authCodePlaceholder:
             'Option 1: Copy the complete URL\n(http://localhost:xxx/auth/callback?code=...)\nOption 2: Copy only the code parameter value',
-          authCodeHint: 'You can copy the entire URL or just the code parameter value, the system will auto-detect',
-          failedToGenerateUrl: 'Failed to generate Antigravity auth URL',
-          missingExchangeParams: 'Missing code, session ID, or state',
-          failedToExchangeCode: 'Failed to exchange Antigravity auth code'
-        }
-	      },
-      // Gemini specific (platform-wide)
+                    authCodeHint: 'You can copy the entire URL or just the code parameter value, the system will auto-detect',
+                    failedToGenerateUrl: 'Failed to generate Antigravity auth URL',
+                    missingExchangeParams: 'Missing code, session ID, or state',
+                    failedToExchangeCode: 'Failed to exchange Antigravity auth code',
+                    // Refresh Token auth
+                    refreshTokenAuth: 'Manual RT',
+                    refreshTokenDesc: 'Enter your existing Antigravity Refresh Token. Supports batch input (one per line). The system will automatically validate and create accounts.',
+                    refreshTokenPlaceholder: 'Paste your Antigravity Refresh Token...\nSupports multiple tokens, one per line',
+                    validating: 'Validating...',
+                    validateAndCreate: 'Validate & Create',
+                    pleaseEnterRefreshToken: 'Please enter Refresh Token',
+                    failedToValidateRT: 'Failed to validate Refresh Token'
+                  }
+                },      // Gemini specific (platform-wide)
       gemini: {
         helpButton: 'Help',
         helpDialog: {
@@ -2153,7 +2160,7 @@ export default {
       title: 'Redeem Code Management',
       description: 'Generate and manage redeem codes',
       generateCodes: 'Generate Codes',
-      searchCodes: 'Search codes...',
+      searchCodes: 'Search codes or email...',
       allTypes: 'All Types',
       allStatus: 'All Status',
       balance: 'Balance',
@@ -3399,6 +3406,7 @@ export default {
       custom: 'Custom',
       code: 'Code',
       body: 'Body',
+      skipMonitoring: 'Skip Monitoring',
 
       // Columns
       columns: {
@@ -3443,6 +3451,8 @@ export default {
         passthroughBody: 'Passthrough upstream error message',
         customMessage: 'Custom error message',
         customMessagePlaceholder: 'Error message to return to client...',
+        skipMonitoring: 'Skip monitoring',
+        skipMonitoringHint: 'When enabled, errors matching this rule will not be recorded in ops monitoring',
         enabled: 'Enable this rule'
       },
 

frontend/src/i18n/locales/zh.ts

@@ -865,8 +865,8 @@ export default {
       editUser: '编辑用户',
       deleteUser: '删除用户',
       deleteConfirmMessage: "确定要删除用户 '{email}' 吗？此操作无法撤销。",
-      searchPlaceholder: '搜索用户邮箱或用户名、备注、支持模糊查询...',
-      searchUsers: '搜索用户邮箱或用户名、备注、支持模糊查询',
+      searchPlaceholder: '邮箱/用户名/备注/API Key 模糊搜索...',
+      searchUsers: '邮箱/用户名/备注/API Key 模糊搜索',
       roleFilter: '角色筛选',
       allRoles: '全部角色',
       allStatus: '全部状态',
@@ -1936,7 +1936,15 @@ export default {
           authCodeHint: '您可以直接复制整个链接或仅复制 code 参数值，系统会自动识别',
           failedToGenerateUrl: '生成 Antigravity 授权链接失败',
           missingExchangeParams: '缺少 code / session_id / state',
-          failedToExchangeCode: 'Antigravity 授权码兑换失败'
+          failedToExchangeCode: 'Antigravity 授权码兑换失败',
+          // Refresh Token auth
+          refreshTokenAuth: '手动输入 RT',
+          refreshTokenDesc: '输入您已有的 Antigravity Refresh Token，支持批量输入（每行一个），系统将自动验证并创建账号。',
+          refreshTokenPlaceholder: '粘贴您的 Antigravity Refresh Token...\n支持多个，每行一个',
+          validating: '验证中...',
+          validateAndCreate: '验证并创建账号',
+          pleaseEnterRefreshToken: '请输入 Refresh Token',
+          failedToValidateRT: '验证 Refresh Token 失败'
         }
       },
       // Gemini specific (platform-wide)
@@ -2315,7 +2323,7 @@ export default {
       allStatus: '全部状态',
       unused: '未使用',
       used: '已使用',
-      searchCodes: '搜索兑换码...',
+      searchCodes: '搜索兑换码或邮箱...',
       exportCsv: '导出 CSV',
       deleteAllUnused: '删除全部未使用',
       deleteCodeConfirm: '确定要删除此兑换码吗？此操作无法撤销。',
@@ -3572,6 +3580,7 @@ export default {
       custom: '自定义',
       code: '状态码',
       body: '消息体',
+      skipMonitoring: '跳过监控',
 
       // Columns
       columns: {
@@ -3616,6 +3625,8 @@ export default {
         passthroughBody: '透传上游错误信息',
         customMessage: '自定义错误信息',
         customMessagePlaceholder: '返回给客户端的错误信息...',
+        skipMonitoring: '跳过运维监控记录',
+        skipMonitoringHint: '开启后，匹配此规则的错误不会被记录到运维监控中',
         enabled: '启用此规则'
       },
 

frontend/src/utils/url.ts

@@ -6,6 +6,7 @@
  */
 type SanitizeOptions = {
   allowRelative?: boolean
+  allowDataUrl?: boolean
 }
 
 export function sanitizeUrl(value: string, options: SanitizeOptions = {}): string {
@@ -18,6 +19,11 @@ export function sanitizeUrl(value: string, options: SanitizeOptions = {}): strin
     return trimmed
   }
 
+  // 允许 data:image/ 开头的 data URL（仅限图片类型）
+  if (options.allowDataUrl && trimmed.startsWith('data:image/')) {
+    return trimmed
+  }
+
   // 只接受绝对 URL，不使用 base URL 来避免相对路径被解析为当前域名
   // 检查是否以 http:// 或 https:// 开头
   if (!trimmed.match(/^https?:\/\//i)) {

frontend/src/views/admin/RedeemView.vue

@@ -117,9 +117,9 @@
             </span>
           </template>
 
-          <template #cell-used_by="{ value }">
+          <template #cell-used_by="{ value, row }">
             <span class="text-sm text-gray-500 dark:text-dark-400">
-              {{ value ? t('admin.redeem.userPrefix', { id: value }) : '-' }}
+              {{ row.user?.email || (value ? t('admin.redeem.userPrefix', { id: value }) : '-') }}
             </span>
           </template>