Update handlers, registry and service

- Update model definitions static data
- Update kiro gateway config
- Update SDK API handlers for claude, gemini, and openai
- Update cliproxy service

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

internal/registry/model_definitions_static_data.go

@@ -879,7 +879,7 @@ func GetKiroModels() []*ModelInfo {
 			Type:                "kiro",
 			DisplayName:         "Claude 3.7 Sonnet (Kiro)",
 			Description:         "Legacy model - hidden in Kiro API but still functional",
-			ContextLength:       128000,
+			ContextLength:       200000,
 			MaxCompletionTokens: 8192,
 			Thinking:            &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: false, DynamicAllowed: false},
 		},

kiro-gateway/kiro/config.py

@@ -37,45 +37,46 @@ load_dotenv()
 def _get_raw_env_value(var_name: str, env_file: str = ".env") -> Optional[str]:
     """
     Read variable value from .env file without processing escape sequences.
-    
+
     This is necessary for correct handling of Windows paths where backslashes
     (e.g., D:\\Projects\\file.json) may be incorrectly interpreted
     as escape sequences (\\a -> bell, \\n -> newline, etc.).
-    
+
     Args:
         var_name: Environment variable name
         env_file: Path to .env file (default ".env")
-    
+
     Returns:
         Raw variable value or None if not found
     """
     env_path = Path(env_file)
     if not env_path.exists():
         return None
-    
+
     try:
         # Read file as-is, without interpretation
         content = env_path.read_text(encoding="utf-8")
-        
+
         # Search for variable considering different formats:
         # VAR="value" or VAR='value' or VAR=value
         # Pattern captures value with or without quotes
         pattern = rf'^{re.escape(var_name)}=(["\']?)(.+?)\1\s*$'
-        
+
         for line in content.splitlines():
             line = line.strip()
             if line.startswith("#") or not line:
                 continue
-            
+
             match = re.match(pattern, line)
             if match:
                 # Return value as-is, without processing escape sequences
                 return match.group(2)
     except Exception:
         pass
-    
+
     return None
 
+
 # ==================================================================================================
 # Server Settings
 # ==================================================================================================
@@ -198,7 +199,6 @@ HIDDEN_MODELS: Dict[str, str] = {
     # Claude 3.7 Sonnet - legacy flagship model, still works!
     # Hidden in Kiro API but functional. Great for users who prefer it.
     "claude-3.7-sonnet": "CLAUDE_3_7_SONNET_20250219_V1_0",
-    
     # Add other hidden/experimental models here as discovered.
     # Example: "claude-secret-model": "INTERNAL_SECRET_MODEL_ID",
 }
@@ -220,6 +220,7 @@ FALLBACK_MODELS: List[Dict[str, str]] = [
     {"modelId": "claude-haiku-4.5"},
     {"modelId": "claude-sonnet-4.5"},
     {"modelId": "claude-opus-4.5"},
+    {"modelId": "claude-3.7-sonnet"},
 ]
 
 # ==================================================================================================
@@ -323,10 +324,11 @@ def _warn_deprecated_debug_setting():
     """
     if _DEBUG_LAST_REQUEST_RAW and not _DEBUG_MODE_RAW:
         import sys
+
         # ANSI escape codes: yellow text
         YELLOW = "\033[93m"
         RESET = "\033[0m"
-        
+
         warning_text = f"""
 {YELLOW}⚠️  DEPRECATED: DEBUG_LAST_REQUEST will be removed in future releases.
     Please use DEBUG_MODE instead:
@@ -344,16 +346,17 @@ def _warn_timeout_configuration():
     """
     Print warning if timeout configuration is suboptimal.
     Called at application startup.
-    
+
     FIRST_TOKEN_TIMEOUT should be less than STREAMING_READ_TIMEOUT:
     - FIRST_TOKEN_TIMEOUT: time to wait for model to START responding
     - STREAMING_READ_TIMEOUT: time to wait BETWEEN chunks during streaming
     """
     if FIRST_TOKEN_TIMEOUT >= STREAMING_READ_TIMEOUT:
         import sys
+
         YELLOW = "\033[93m"
         RESET = "\033[0m"
-        
+
         warning_text = f"""
 {YELLOW}⚠️  WARNING: Suboptimal timeout configuration detected.
     
@@ -371,6 +374,7 @@ def _warn_timeout_configuration():
 """
         print(warning_text, file=sys.stderr)
 
+
 # ==================================================================================================
 # Fake Reasoning Settings (Extended Thinking via Tag Injection)
 # ==================================================================================================
@@ -402,7 +406,9 @@ FAKE_REASONING_MAX_TOKENS: int = int(os.getenv("FAKE_REASONING_MAX_TOKENS", "400
 # - "strip_tags": Remove tags but keep thinking content in regular content
 #
 # Default: "as_reasoning_content"
-_FAKE_REASONING_HANDLING_RAW: str = os.getenv("FAKE_REASONING_HANDLING", "as_reasoning_content").lower()
+_FAKE_REASONING_HANDLING_RAW: str = os.getenv(
+    "FAKE_REASONING_HANDLING", "as_reasoning_content"
+).lower()
 if _FAKE_REASONING_HANDLING_RAW in ("as_reasoning_content", "remove", "pass", "strip_tags"):
     FAKE_REASONING_HANDLING: str = _FAKE_REASONING_HANDLING_RAW
 else:
@@ -447,4 +453,3 @@ def get_kiro_api_host(region: str) -> str:
 def get_kiro_q_host(region: str) -> str:
     """Return Q API host for the specified region."""
     return KIRO_Q_HOST_TEMPLATE.format(region=region)
-

sdk/api/handlers/claude/code_handlers.go

@@ -53,7 +53,31 @@ func (h *ClaudeCodeAPIHandler) HandlerType() string {
 func (h *ClaudeCodeAPIHandler) Models() []map[string]any {
 	// Get dynamic models from the global registry
 	modelRegistry := registry.GetGlobalRegistry()
-	return modelRegistry.GetAvailableModels("claude")
+	models := modelRegistry.GetAvailableModels("claude")
+
+	// Always include static Kiro models
+	kiroModels := registry.GetKiroModels()
+	seen := make(map[string]bool)
+	for _, m := range models {
+		if id, ok := m["id"].(string); ok {
+			seen[id] = true
+		}
+	}
+	for _, km := range kiroModels {
+		if seen[km.ID] {
+			continue
+		}
+		models = append(models, map[string]any{
+			"id":         km.ID,
+			"object":     "model",
+			"created":    km.Created,
+			"owned_by":   "kiro",
+			"type":       "model",
+			"display_name": km.DisplayName,
+		})
+	}
+
+	return models
 }
 
 // ClaudeMessages handles Claude-compatible streaming chat completions.

sdk/api/handlers/gemini/gemini_handlers.go

@@ -42,7 +42,30 @@ func (h *GeminiAPIHandler) HandlerType() string {
 func (h *GeminiAPIHandler) Models() []map[string]any {
 	// Get dynamic models from the global registry
 	modelRegistry := registry.GetGlobalRegistry()
-	return modelRegistry.GetAvailableModels("gemini")
+	models := modelRegistry.GetAvailableModels("gemini")
+
+	// Always include static Kiro models (convert to Gemini format)
+	kiroModels := registry.GetKiroModels()
+	seen := make(map[string]bool)
+	for _, m := range models {
+		if name, ok := m["name"].(string); ok {
+			seen[name] = true
+		}
+	}
+	for _, km := range kiroModels {
+		name := "models/" + km.ID
+		if seen[name] {
+			continue
+		}
+		models = append(models, map[string]any{
+			"name":                       name,
+			"version":                    "1.0",
+			"displayName":                km.DisplayName,
+			"supportedGenerationMethods": []string{"generateContent"},
+		})
+	}
+
+	return models
 }
 
 // GeminiModels handles the Gemini models listing endpoint.

sdk/api/handlers/openai/openai_handlers.go

@@ -52,7 +52,29 @@ func (h *OpenAIAPIHandler) HandlerType() string {
 func (h *OpenAIAPIHandler) Models() []map[string]any {
 	// Get dynamic models from the global registry
 	modelRegistry := registry.GetGlobalRegistry()
-	return modelRegistry.GetAvailableModels("openai")
+	models := modelRegistry.GetAvailableModels("openai")
+
+	// Always include static Kiro models
+	kiroModels := registry.GetKiroModels()
+	seen := make(map[string]bool)
+	for _, m := range models {
+		if id, ok := m["id"].(string); ok {
+			seen[id] = true
+		}
+	}
+	for _, km := range kiroModels {
+		if seen[km.ID] {
+			continue
+		}
+		models = append(models, map[string]any{
+			"id":       km.ID,
+			"object":   "model",
+			"created":  km.Created,
+			"owned_by": "kiro",
+		})
+	}
+
+	return models
 }
 
 // OpenAIModels handles the /v1/models endpoint.

sdk/api/handlers/openai/openai_responses_handlers.go

@@ -49,7 +49,29 @@ func (h *OpenAIResponsesAPIHandler) HandlerType() string {
 func (h *OpenAIResponsesAPIHandler) Models() []map[string]any {
 	// Get dynamic models from the global registry
 	modelRegistry := registry.GetGlobalRegistry()
-	return modelRegistry.GetAvailableModels("openai")
+	models := modelRegistry.GetAvailableModels("openai")
+
+	// Always include static Kiro models
+	kiroModels := registry.GetKiroModels()
+	seen := make(map[string]bool)
+	for _, m := range models {
+		if id, ok := m["id"].(string); ok {
+			seen[id] = true
+		}
+	}
+	for _, km := range kiroModels {
+		if seen[km.ID] {
+			continue
+		}
+		models = append(models, map[string]any{
+			"id":       km.ID,
+			"object":   "model",
+			"created":  km.Created,
+			"owned_by": "kiro",
+		})
+	}
+
+	return models
 }
 
 // OpenAIResponsesModels handles the /v1/models endpoint.

sdk/cliproxy/service.go

@@ -774,6 +774,9 @@ func (s *Service) registerModelsForAuth(a *coreauth.Auth) {
 	case "iflow":
 		models = registry.GetIFlowModels()
 		models = applyExcludedModels(models, excluded)
+	case "kiro":
+		models = registry.GetKiroModels()
+		models = applyExcludedModels(models, excluded)
 	default:
 		// Handle OpenAI-compatibility providers by name using config
 		if s.cfg != nil {