Merge branch 'router-for-me:main' into main

Merge pull request #1435 from tianyicui/fix/haiku-4-5-thinking-support
fix: Enable extended thinking support for Claude Haiku 4.5
2026-03-19 07:21:58 +00:00 · 2026-02-06 05:45:21 +08:00 · 2026-02-06 05:44:14 +08:00 · 2026-02-06 05:43:32 +08:00 · 2026-02-06 05:39:41 +08:00 · 2026-02-06 05:39:41 +08:00
7 changed files with 66 additions and 13 deletions
--- a/internal/registry/model_definitions.go
+++ b/internal/registry/model_definitions.go
@@ -376,6 +376,18 @@ func GetKiroModels() []*ModelInfo {
 			MaxCompletionTokens: 64000,
 			Thinking:            &ThinkingSupport{Min: 1024, Max: 32000, ZeroAllowed: true, DynamicAllowed: true},
 		},
+		{
+			ID:                  "kiro-claude-opus-4-6",
+			Object:              "model",
+			Created:             1736899200, // 2025-01-15
+			OwnedBy:             "aws",
+			Type:                "kiro",
+			DisplayName:         "Kiro Claude Opus 4.6",
+			Description:         "Claude Opus 4.6 via Kiro (2.2x credit)",
+			ContextLength:       200000,
+			MaxCompletionTokens: 64000,
+			Thinking:            &ThinkingSupport{Min: 1024, Max: 32000, ZeroAllowed: true, DynamicAllowed: true},
+		},
 		{
 			ID:                  "kiro-claude-opus-4-5",
 			Object:              "model",
@@ -425,6 +437,18 @@ func GetKiroModels() []*ModelInfo {
 			Thinking:            &ThinkingSupport{Min: 1024, Max: 32000, ZeroAllowed: true, DynamicAllowed: true},
 		},
 		// --- Agentic Variants (Optimized for coding agents with chunked writes) ---
+		{
+			ID:                  "kiro-claude-opus-4-6-agentic",
+			Object:              "model",
+			Created:             1736899200, // 2025-01-15
+			OwnedBy:             "aws",
+			Type:                "kiro",
+			DisplayName:         "Kiro Claude Opus 4.6 (Agentic)",
+			Description:         "Claude Opus 4.6 optimized for coding agents (chunked writes)",
+			ContextLength:       200000,
+			MaxCompletionTokens: 64000,
+			Thinking:            &ThinkingSupport{Min: 1024, Max: 32000, ZeroAllowed: true, DynamicAllowed: true},
+		},
 		{
 			ID:                  "kiro-claude-opus-4-5-agentic",
 			Object:              "model",
--- a/internal/registry/model_definitions_static_data.go
+++ b/internal/registry/model_definitions_static_data.go
@@ -15,7 +15,7 @@ func GetClaudeModels() []*ModelInfo {
 			DisplayName:         "Claude 4.5 Haiku",
 			ContextLength:       200000,
 			MaxCompletionTokens: 64000,
-			// Thinking: not supported for Haiku models
+			Thinking:            &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: true, DynamicAllowed: false},
 		},
 		{
 			ID:                  "claude-sonnet-4-5-20250929",
@@ -29,15 +29,15 @@ func GetClaudeModels() []*ModelInfo {
 			Thinking:            &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: true, DynamicAllowed: false},
 		},
 		{
-			ID:                  "claude-opus-4-6-20260205",
+			ID:                  "claude-opus-4-6",
 			Object:              "model",
 			Created:             1770318000, // 2026-02-05
 			OwnedBy:             "anthropic",
 			Type:                "claude",
 			DisplayName:         "Claude 4.6 Opus",
 			Description:         "Premium model combining maximum intelligence with practical performance",
-			ContextLength:       200000,
-			MaxCompletionTokens: 64000,
+			ContextLength:       1000000,
+			MaxCompletionTokens: 128000,
 			Thinking:            &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: true, DynamicAllowed: false},
 		},
 		{
@@ -866,7 +866,7 @@ func GetAntigravityModelConfig() map[string]*AntigravityModelConfig {
 		"gemini-3-flash":             {Thinking: &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true, Levels: []string{"minimal", "low", "medium", "high"}}},
 		"claude-sonnet-4-5-thinking": {Thinking: &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: true, DynamicAllowed: true}, MaxCompletionTokens: 64000},
 		"claude-opus-4-5-thinking":   {Thinking: &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: true, DynamicAllowed: true}, MaxCompletionTokens: 64000},
-		"claude-opus-4-6-thinking":   {Thinking: &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: true, DynamicAllowed: true}, MaxCompletionTokens: 64000},
+		"claude-opus-4-6-thinking":   {Thinking: &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: true, DynamicAllowed: true}, MaxCompletionTokens: 128000},
 		"claude-sonnet-4-5":          {MaxCompletionTokens: 64000},
 		"gpt-oss-120b-medium":        {},
 		"tab_flash_lite_preview":     {},
--- a/internal/runtime/executor/codex_executor.go
+++ b/internal/runtime/executor/codex_executor.go
@@ -27,6 +27,11 @@ import (
 	"github.com/google/uuid"
 )

+const (
+	codexClientVersion = "0.98.0"
+	codexUserAgent     = "codex_cli_rs/0.98.0 (Mac OS 26.0.1; arm64) Apple_Terminal/464"
+)
+
 var dataTag = []byte("data:")

 // CodexExecutor is a stateless executor for Codex (OpenAI Responses API entrypoint).
@@ -637,10 +642,10 @@ func applyCodexHeaders(r *http.Request, auth *cliproxyauth.Auth, token string, s
 		ginHeaders = ginCtx.Request.Header
 	}

-	misc.EnsureHeader(r.Header, ginHeaders, "Version", "0.21.0")
+	misc.EnsureHeader(r.Header, ginHeaders, "Version", codexClientVersion)
 	misc.EnsureHeader(r.Header, ginHeaders, "Openai-Beta", "responses=experimental")
 	misc.EnsureHeader(r.Header, ginHeaders, "Session_id", uuid.NewString())
-	misc.EnsureHeader(r.Header, ginHeaders, "User-Agent", "codex_cli_rs/0.50.0 (Mac OS 26.0.1; arm64) Apple_Terminal/464")
+	misc.EnsureHeader(r.Header, ginHeaders, "User-Agent", codexUserAgent)

 	if stream {
 		r.Header.Set("Accept", "text/event-stream")
--- a/internal/runtime/executor/kiro_executor.go
+++ b/internal/runtime/executor/kiro_executor.go
@@ -1681,6 +1681,7 @@ func (e *KiroExecutor) mapModelToKiro(model string) string {
 	modelMap := map[string]string{
 		// Amazon Q format (amazonq- prefix) - same API as Kiro
 		"amazonq-auto":                       "auto",
+		"amazonq-claude-opus-4-6":            "claude-opus-4.6",
 		"amazonq-claude-opus-4-5":            "claude-opus-4.5",
 		"amazonq-claude-sonnet-4-5":          "claude-sonnet-4.5",
 		"amazonq-claude-sonnet-4-5-20250929": "claude-sonnet-4.5",
@@ -1688,6 +1689,7 @@ func (e *KiroExecutor) mapModelToKiro(model string) string {
 		"amazonq-claude-sonnet-4-20250514":   "claude-sonnet-4",
 		"amazonq-claude-haiku-4-5":           "claude-haiku-4.5",
 		// Kiro format (kiro- prefix) - valid model names that should be preserved
+		"kiro-claude-opus-4-6":            "claude-opus-4.6",
 		"kiro-claude-opus-4-5":            "claude-opus-4.5",
 		"kiro-claude-sonnet-4-5":          "claude-sonnet-4.5",
 		"kiro-claude-sonnet-4-5-20250929": "claude-sonnet-4.5",
@@ -1696,6 +1698,8 @@ func (e *KiroExecutor) mapModelToKiro(model string) string {
 		"kiro-claude-haiku-4-5":           "claude-haiku-4.5",
 		"kiro-auto":                       "auto",
 		// Native format (no prefix) - used by Kiro IDE directly
+		"claude-opus-4-6":            "claude-opus-4.6",
+		"claude-opus-4.6":            "claude-opus-4.6",
 		"claude-opus-4-5":            "claude-opus-4.5",
 		"claude-opus-4.5":            "claude-opus-4.5",
 		"claude-haiku-4-5":           "claude-haiku-4.5",
@@ -1707,10 +1711,12 @@ func (e *KiroExecutor) mapModelToKiro(model string) string {
 		"claude-sonnet-4-20250514":   "claude-sonnet-4",
 		"auto":                       "auto",
 		// Agentic variants (same backend model IDs, but with special system prompt)
+		"claude-opus-4.6-agentic":        "claude-opus-4.6",
 		"claude-opus-4.5-agentic":        "claude-opus-4.5",
 		"claude-sonnet-4.5-agentic":      "claude-sonnet-4.5",
 		"claude-sonnet-4-agentic":        "claude-sonnet-4",
 		"claude-haiku-4.5-agentic":       "claude-haiku-4.5",
+		"kiro-claude-opus-4-6-agentic":   "claude-opus-4.6",
 		"kiro-claude-opus-4-5-agentic":   "claude-opus-4.5",
 		"kiro-claude-sonnet-4-5-agentic": "claude-sonnet-4.5",
 		"kiro-claude-sonnet-4-agentic":   "claude-sonnet-4",
--- a/internal/translator/kiro/claude/kiro_claude_request.go
+++ b/internal/translator/kiro/claude/kiro_claude_request.go
@@ -883,8 +883,21 @@ func BuildAssistantMessageStruct(msg gjson.Result) KiroAssistantResponseMessage
 		contentBuilder.WriteString(content.String())
 	}

+	// CRITICAL FIX: Kiro API requires non-empty content for assistant messages
+	// This can happen with compaction requests where assistant messages have only tool_use
+	// (no text content). Without this fix, Kiro API returns "Improperly formed request" error.
+	finalContent := contentBuilder.String()
+	if strings.TrimSpace(finalContent) == "" {
+		if len(toolUses) > 0 {
+			finalContent = kirocommon.DefaultAssistantContentWithTools
+		} else {
+			finalContent = kirocommon.DefaultAssistantContent
+		}
+		log.Debugf("kiro: assistant content was empty, using default: %s", finalContent)
+	}
+
 	return KiroAssistantResponseMessage{
-		Content:  contentBuilder.String(),
+		Content:  finalContent,
 		ToolUses: toolUses,
 	}
 }
--- a/internal/translator/kiro/common/constants.go
+++ b/internal/translator/kiro/common/constants.go
@@ -29,6 +29,14 @@ const (
 	// InlineCodeMarker is the markdown inline code marker (backtick).
 	InlineCodeMarker = "`"

+	// DefaultAssistantContentWithTools is the fallback content for assistant messages
+	// that have tool_use but no text content. Kiro API requires non-empty content.
+	DefaultAssistantContentWithTools = "I'll help you with that."
+
+	// DefaultAssistantContent is the fallback content for assistant messages
+	// that have no content at all. Kiro API requires non-empty content.
+	DefaultAssistantContent = "I understand."
+
 	// KiroAgenticSystemPrompt is injected only for -agentic models to prevent timeouts on large writes.
 	// AWS Kiro API has a 2-3 minute timeout for large file write operations.
 	KiroAgenticSystemPrompt = `
--- a/internal/translator/kiro/openai/kiro_openai_request.go
+++ b/internal/translator/kiro/openai/kiro_openai_request.go
@@ -718,13 +718,10 @@ func buildAssistantMessageFromOpenAI(msg gjson.Result) KiroAssistantResponseMess
 	// This can happen with compaction requests or error recovery scenarios
 	finalContent := contentBuilder.String()
 	if strings.TrimSpace(finalContent) == "" {
-		const defaultAssistantContentWithTools = "I'll help you with that."
-		const defaultAssistantContent = "I understand."
-
 		if len(toolUses) > 0 {
-			finalContent = defaultAssistantContentWithTools
+			finalContent = kirocommon.DefaultAssistantContentWithTools
 		} else {
-			finalContent = defaultAssistantContent
+			finalContent = kirocommon.DefaultAssistantContent
 		}
 		log.Debugf("kiro-openai: assistant content was empty, using default: %s", finalContent)
 	}
Author	SHA1	Message	Date
Luis Pater	e7e3ca1efb	Merge branch 'router-for-me:main' into main	2026-02-06 05:45:21 +08:00
Luis Pater	4b00312fef	Merge pull request #1435 from tianyicui/fix/haiku-4-5-thinking-support fix: Enable extended thinking support for Claude Haiku 4.5	2026-02-06 05:44:14 +08:00
Luis Pater	c5fd3db01e	Merge pull request #1446 from qyhfrank/fix-claude-opus-4-6-model-metadata fix(registry): correct Claude Opus 4.6 model metadata	2026-02-06 05:43:32 +08:00
Luis Pater	e35ffaa925	Merge pull request #186 from taetaetae/fix/kiro-claude-compaction-empty-content fix(kiro): handle empty content in Claude format assistant messages	2026-02-06 05:39:41 +08:00
Frank Qing	f870a9d2a7	fix(registry): correct Claude Opus 4.6 model metadata	2026-02-06 05:39:41 +08:00
Luis Pater	165e03f3a7	Merge branch 'router-for-me:main' into main	2026-02-06 05:32:10 +08:00
Luis Pater	86bdb7808c	Merge pull request #189 from PancakeZik/main feat: add Claude Opus 4.6 support for Kiro	2026-02-06 05:31:54 +08:00
Luis Pater	b4e034be1c	refactor(executor): centralize Codex client version and user agent constants - Introduced `codexClientVersion` and `codexUserAgent` constants for better maintainability. - Updated `EnsureHeader` calls to use the new constants.	2026-02-06 05:30:28 +08:00
Joao	84fcebf538	feat: add Claude Opus 4.6 support for Kiro - Add kiro-claude-opus-4-6 and kiro-claude-opus-4-6-agentic to model registry - Add model ID mappings for claude-opus-4.6 variants - Support both kiro- prefix and native format (claude-opus-4.6) - Tested and working with Kiro API	2026-02-05 21:26:29 +00:00
taetaetae	14f044ce4f	refactor: extract default assistant content to shared constants Apply code review feedback from gemini-code-assist: - Move fallback strings to kirocommon package as exported constants - Update kiro_claude_request.go to use shared constants - Update kiro_openai_request.go to use shared constants - Improves maintainability and avoids duplication	2026-02-05 23:36:57 +09:00
taetaetae	88872baffc	fix(kiro): handle empty content in Claude format assistant messages Problem: - PR #181 fixed empty content for OpenAI format (kiro_openai_request.go) - But Claude format (kiro_claude_request.go) was not fixed - OpenCode uses Claude format (/v1/messages endpoint) - When assistant messages have only tool_use (no text), content becomes empty - This causes 'Improperly formed request' errors from Kiro API Example of problematic message format: { "role": "assistant", "content": [ {"type": "tool_use", "id": "...", "name": "todowrite", "input": {...}} ] } Solution: - Add empty content fallback in BuildAssistantMessageStruct (Claude format) - Same fix as PR #181 applied to kiro_openai_request.go Fixes compaction failures for OpenCode + Quotio + CLIProxyAPIPlus + Kiro stack	2026-02-05 23:27:35 +09:00
Tianyi Cui	706590c62a	fix: Enable extended thinking support for Claude Haiku 4.5 Claude Haiku 4.5 (claude-haiku-4-5-20251001) supports extended thinking according to Anthropic's official documentation: https://docs.anthropic.com/en/docs/build-with-claude/extended-thinking The model was incorrectly marked as not supporting thinking in the static model definitions. This fix adds ThinkingSupport with the same parameters as other Claude 4.5 models (Sonnet, Opus): - Min: 1024 tokens - Max: 128000 tokens - ZeroAllowed: true - DynamicAllowed: false	2026-02-05 19:03:23 +08:00