Fix/multiple issues (#24)

* fix(gemini): 修复 google_one OAuth 配置和 scopes 问题 - 修复 google_one 类型在 ExchangeCode 和 RefreshToken 中使用内置客户端 - 添加 DefaultGoogleOneScopes，包含 generative-language 和 drive.readonly 权限 - 在 EffectiveOAuthConfig 中为 google_one 类型使用专门的 scopes - 将 docker-compose.override.yml 重命名为 .example 并添加到 .gitignore - 完善 docker-compose.override.yml.example 示例文档解决问题： 1. google_one OAuth 授权后 API 调用返回 403 权限不足 2. 缺少访问 Gemini API 所需的 generative-language scope 3. 缺少获取 Drive 存储配额所需的 drive.readonly scope * fix(antigravity): 完全跳过 Claude 模型的所有 thinking 块问题分析： - 当前代码尝试保留有 signature 的 thinking 块 - 但 Vertex AI 的 signature 是完整性令牌，无法在本地验证 - 导致 400 错误：Invalid signature in thinking block 根本原因： 1. thinking 功能已对非 Gemini 模型禁用 (isThinkingEnabled=false) 2. Vertex AI 要求原样重放 (thinking, signature) 对或完全不发送 3. 本地无法复制 Vertex 的加密验证逻辑修复方案： - 对 Claude 模型完全跳过所有 thinking 块（无论是否有 signature） - 保持 Gemini 模型使用 dummy signature 的行为不变 - 更新测试用例以反映新的预期行为影响： - 消除 thinking 相关的 400 错误 - 与现有的 thinking 禁用策略保持一致 - 不影响 Gemini 模型的 thinking 功能测试： - ✅ TestBuildParts_ThinkingBlockWithoutSignature 全部通过 - ✅ TestBuildTools_CustomTypeTools 全部通过参考：Codex review 建议 * fix(gateway): 修复 count_tokens 端点 400 错误问题分析： - count_tokens 请求包含 thinking 块时返回 400 错误 - 原因：thinking 块未被过滤，直接转发到上游 API - 上游 API 拒绝无效的 thinking signature 根本原因： 1. /v1/messages 请求通过 TransformClaudeToGemini 过滤 thinking 块 2. count_tokens 请求绕过转换，直接转发原始请求体 3. 导致包含无效 signature 的 thinking 块被发送到上游修复方案： - 创建 FilterThinkingBlocks 工具函数 - 在 buildCountTokensRequest 中应用过滤（1 行修改） - 与 /v1/messages 行为保持一致实现细节： - FilterThinkingBlocks: 解析 JSON，过滤 thinking 块，重新序列化 - 失败安全：解析/序列化失败时返回原始请求体 - 性能优化：仅在发现 thinking 块时重新序列化测试： - ✅ 6 个单元测试全部通过 - ✅ 覆盖正常过滤、无 thinking 块、无效 JSON 等场景 - ✅ 现有测试不受影响影响： - 消除 count_tokens 的 400 错误 - 不影响 Antigravity 账号（仍返回模拟响应） - 适用于所有账号类型（OAuth、API Key）文件修改： - backend/internal/service/gateway_request.go: +62 行（新函数） - backend/internal/service/gateway_service.go: +2 行（应用过滤） - backend/internal/service/gateway_request_test.go: +62 行（测试） * fix(gateway): 增强 thinking 块过滤逻辑基于 Codex 分析和建议的改进：问题分析： - 新错误：signature: Field required（signature 字段缺失） - 旧错误：Invalid signature（signature 存在但无效） - 两者都说明 thinking 块在请求中是危险的 Codex 建议： - 保持 Option A：完全跳过所有 thinking 块 - 原因：thinking 块应该是只输出的，除非有服务端来源证明 - 在无状态代理中，无法安全区分上游来源 vs 客户端注入改进内容： 1. 增强 FilterThinkingBlocks 函数 - 过滤显式的 thinking 块：{"type":"thinking", ...} - 过滤无 type 的 thinking 对象：{"thinking": {...}} - 保留 tool_use 等其他类型块中的 thinking 字段 - 修复：只在实际过滤时更新 content 数组 2. 扩展过滤范围 - 将 FilterThinkingBlocks 应用到 /v1/messages 主路径 - 之前只应用于 count_tokens，现在两个端点都过滤 - 防止所有端点的 thinking 相关 400 错误 3. 改进测试 - 新增：过滤无 type discriminator 的 thinking 块 - 新增：不过滤 tool_use 中的 thinking 字段 - 使用 containsThinkingBlock 辅助函数验证测试： - ✅ 8 个测试用例全部通过 - ✅ 覆盖各种 thinking 块格式 - ✅ 确保不误伤其他类型的块影响： - 消除 signature required 和 invalid signature 错误 - 统一 /v1/messages 和 count_tokens 的行为 - 更健壮的 thinking 块检测逻辑参考：Codex review 和代码改进 * refactor: 根据 Codex 审查建议进行代码优化基于 Codex 代码审查的 P1 和 P2 改进： P1 改进（重要问题）： 1. 优化日志输出 - 移除 thinking 块跳过时的 log.Printf - 避免高频请求下的日志噪音 - 添加注释说明可通过指标监控 2. 清理遗留代码 - 删除未使用的 isValidThoughtSignature 函数（27行） - 该函数在改为完全跳过 thinking 块后不再需要 P2 改进（性能优化）： 3. 添加快速路径检查 - 在 FilterThinkingBlocks 中添加 bytes.Contains 预检查 - 如果请求体不包含 "thinking" 字符串，直接返回 - 避免不必要的 JSON 解析，提升性能技术细节： - request_transformer.go: -27行（删除函数），+1行（优化注释） - gateway_request.go: +5行（快速路径 + bytes 导入）测试： - ✅ TestBuildParts_ThinkingBlockWithoutSignature 全部通过 - ✅ TestFilterThinkingBlocks 全部通过（8个测试用例）影响： - 减少日志噪音 - 提升性能（快速路径） - 代码更简洁（删除未使用代码）参考：Codex 代码审查建议 * fix: 修复 golangci-lint 检查问题 - 格式化 gateway_request_test.go - 使用 switch 语句替代 if-else 链（staticcheck QF1003） * fix(antigravity): 修复 thinking signature 处理并实现 Auto 模式降级问题分析： 1. 原先代码错误地禁用了 Claude via Vertex 的 thinkingConfig 2. 历史 thinking 块的 signature 被完全跳过，导致验证失败 3. 跨模型混用时 dummy signature 会导致 400 错误修复内容： **request_transformer.go**： - 删除第 38-43 行的错误逻辑（禁用 thinkingConfig） - 引入 thoughtSignatureMode（Preserve/Dummy）策略 - Claude 模式：透传真实 signature，过滤空/dummy - Gemini 模式：使用 dummy signature - 支持 signature-only thinking 块 - tool_use 的 signature 也透传 **antigravity_gateway_service.go**： - 新增 isSignatureRelatedError() 检测 signature 相关错误 - 新增 stripThinkingFromClaudeRequest() 移除 thinking 块 - 实现 Auto 模式：检测 400 + signature 关键词时自动降级重试 - 重试时完全移除 thinking 配置和消息中的 thinking 块 - 最多重试一次，避免循环 **测试**： - 更新并新增测试覆盖 Claude preserve/Gemini dummy 模式 - 新增 tool_use signature 处理测试 - 所有测试通过（6/6）影响： - ✅ Claude via Vertex 可以正常使用 thinking 功能 - ✅ 历史 signature 正确透传，避免验证失败 - ✅ 跨模型混用时自动过滤无效 signature - ✅ 错误驱动降级，自动修复 signature 问题 - ✅ 不影响纯 Claude API 和其他渠道参考：Codex 深度分析和实现建议 * fix(lint): 修复 gofmt 格式问题 * fix(antigravity): 修复 stripThinkingFromClaudeRequest 遗漏 untyped thinking blocks 问题： - Codex 审查指出 stripThinkingFromClaudeRequest 只移除了 type="thinking" 的块 - 没有处理没有 type 字段的 thinking 对象（如 {"thinking": "...", "signature": "..."}） - 导致重试时仍包含无效 thinking 块，上游 400 错误持续修复： - 添加检查：跳过没有 type 但有 thinking 字段的块 - 现在会移除两种格式： 1. {"type": "thinking", "thinking": "...", "signature": "..."} 2. {"thinking": "...", "signature": "..."}（untyped）测试：所有测试通过参考：Codex P1 审查意见
2026-01-02 17:47:49 +08:00
parent 68671749d8
commit 7fdc2b2d29
12 changed files with 627 additions and 102 deletions
--- a/.gitignore
+++ b/.gitignore
@@ -118,3 +118,4 @@ docs/
 code-reviews/
 AGENTS.md
 backend/cmd/server/server
+deploy/docker-compose.override.yml
--- a/backend/internal/pkg/antigravity/request_transformer.go
+++ b/backend/internal/pkg/antigravity/request_transformer.go
@@ -20,12 +20,18 @@ func TransformClaudeToGemini(claudeReq *ClaudeRequest, projectID, mappedModel st

 	// 检测是否启用 thinking
 	requestedThinkingEnabled := claudeReq.Thinking != nil && claudeReq.Thinking.Type == "enabled"
-	// 为避免 Claude 模型的 thought signature/消息块约束导致 400（上游要求 thinking 块开头等），
-	// 非 Gemini 模型默认不启用 thinking（除非未来支持完整签名链路）。
-	isThinkingEnabled := requestedThinkingEnabled && allowDummyThought
+	// antigravity(v1internal) 下，Gemini 与 Claude 的 “thinking” 都可能涉及 thoughtSignature 链路：
+	// - Gemini：支持 dummy signature 跳过校验
+	// - Claude：需要透传上游签名（否则容易 400）
+	isThinkingEnabled := requestedThinkingEnabled
+
+	thoughtSignatureMode := thoughtSignatureModePreserve
+	if allowDummyThought {
+		thoughtSignatureMode = thoughtSignatureModeDummy
+	}

 	// 1. 构建 contents
-	contents, err := buildContents(claudeReq.Messages, toolIDToName, isThinkingEnabled, allowDummyThought)
+	contents, err := buildContents(claudeReq.Messages, toolIDToName, isThinkingEnabled, thoughtSignatureMode)
 	if err != nil {
 		return nil, fmt.Errorf("build contents: %w", err)
 	}
@@ -34,15 +40,7 @@ func TransformClaudeToGemini(claudeReq *ClaudeRequest, projectID, mappedModel st
 	systemInstruction := buildSystemInstruction(claudeReq.System, claudeReq.Model)

 	// 3. 构建 generationConfig
-	reqForGen := claudeReq
-	if requestedThinkingEnabled && !allowDummyThought {
-		log.Printf("[Warning] Disabling thinking for non-Gemini model in antigravity transform: model=%s", mappedModel)
-		// shallow copy to avoid mutating caller's request
-		clone := *claudeReq
-		clone.Thinking = nil
-		reqForGen = &clone
-	}
-	generationConfig := buildGenerationConfig(reqForGen)
+	generationConfig := buildGenerationConfig(claudeReq)

 	// 4. 构建 tools
 	tools := buildTools(claudeReq.Tools)
@@ -131,7 +129,7 @@ func buildSystemInstruction(system json.RawMessage, modelName string) *GeminiCon
 }

 // buildContents 构建 contents
-func buildContents(messages []ClaudeMessage, toolIDToName map[string]string, isThinkingEnabled, allowDummyThought bool) ([]GeminiContent, error) {
+func buildContents(messages []ClaudeMessage, toolIDToName map[string]string, isThinkingEnabled bool, thoughtSignatureMode thoughtSignatureMode) ([]GeminiContent, error) {
 	var contents []GeminiContent

 	for i, msg := range messages {
@@ -140,11 +138,13 @@ func buildContents(messages []ClaudeMessage, toolIDToName map[string]string, isT
 			role = "model"
 		}

-		parts, err := buildParts(msg.Content, toolIDToName, allowDummyThought)
+		parts, err := buildParts(msg.Content, toolIDToName, thoughtSignatureMode)
 		if err != nil {
 			return nil, fmt.Errorf("build parts for message %d: %w", i, err)
 		}

+		allowDummyThought := thoughtSignatureMode == thoughtSignatureModeDummy
+
 		// 只有 Gemini 模型支持 dummy thinking block workaround
 		// 只对最后一条 assistant 消息添加（Pre-fill 场景）
 		// 历史 assistant 消息不能添加没有 signature 的 dummy thinking block
@@ -183,37 +183,19 @@ func buildContents(messages []ClaudeMessage, toolIDToName map[string]string, isT
 // 参考: https://ai.google.dev/gemini-api/docs/thought-signatures
 const dummyThoughtSignature = "skip_thought_signature_validator"

-// isValidThoughtSignature 验证 thought signature 是否有效
-// Claude API 要求 signature 必须是 base64 编码的字符串，长度至少 32 字节
-func isValidThoughtSignature(signature string) bool {
-	// 空字符串无效
-	if signature == "" {
-		return false
-	}
+// buildParts 构建消息的 parts
+type thoughtSignatureMode int

-	// signature 应该是 base64 编码，长度至少 40 个字符（约 30 字节）
-	// 参考 Claude API 文档和实际观察到的有效 signature
-	if len(signature) < 40 {
-		log.Printf("[Debug] Signature too short: len=%d", len(signature))
-		return false
-	}
-
-	// 检查是否是有效的 base64 字符
-	// base64 字符集: A-Z, a-z, 0-9, +, /, =
-	for i, c := range signature {
-		if (c < 'A' || c > 'Z') && (c < 'a' || c > 'z') &&
-			(c < '0' || c > '9') && c != '+' && c != '/' && c != '=' {
-			log.Printf("[Debug] Invalid base64 character at position %d: %c (code=%d)", i, c, c)
-			return false
-		}
-	}
-
-	return true
-}
+const (
+	thoughtSignatureModePreserve thoughtSignatureMode = iota
+	thoughtSignatureModeDummy
+)

 // buildParts 构建消息的 parts
-// allowDummyThought: 只有 Gemini 模型支持 dummy thought signature
-func buildParts(content json.RawMessage, toolIDToName map[string]string, allowDummyThought bool) ([]GeminiPart, error) {
+// thoughtSignatureMode:
+// - dummy: 用 dummy signature 跳过 Gemini thoughtSignature 校验
+// - preserve: 透传输入中的 signature（主要用于 Claude via Vertex 的签名链路）
+func buildParts(content json.RawMessage, toolIDToName map[string]string, thoughtSignatureMode thoughtSignatureMode) ([]GeminiPart, error) {
 	var parts []GeminiPart

 	// 尝试解析为字符串
@@ -239,7 +221,9 @@ func buildParts(content json.RawMessage, toolIDToName map[string]string, allowDu
 			}

 		case "thinking":
-			if allowDummyThought {
+			signature := strings.TrimSpace(block.Signature)
+
+			if thoughtSignatureMode == thoughtSignatureModeDummy {
 				// Gemini 模型可以使用 dummy signature
 				parts = append(parts, GeminiPart{
 					Text:             block.Thinking,
@@ -249,20 +233,27 @@ func buildParts(content json.RawMessage, toolIDToName map[string]string, allowDu
 				continue
 			}

-			// Claude 模型：仅在提供有效 signature 时保留 thinking block；否则跳过以避免上游校验失败。
-			signature := strings.TrimSpace(block.Signature)
+			// Claude via Vertex：
+			// - signature 是上游返回的完整性令牌；本地不需要/无法验证，只能透传
+			// - 缺失/无效 signature（例如来自 Gemini 的 dummy signature）会导致上游 400
 			if signature == "" || signature == dummyThoughtSignature {
-				log.Printf("[Warning] Skipping thinking block for Claude model (missing or dummy signature)")
 				continue
 			}
-			if !isValidThoughtSignature(signature) {
-				log.Printf("[Debug] Thinking signature may be invalid (passing through anyway): len=%d", len(signature))
+
+			// 兼容：用 Claude 的 "thinking" 块承载两类东西
+			// 1) 真正的 thought 文本（thinking != ""）-> Gemini thought part
+			// 2) 仅承载 signature 的空 thinking 块（thinking == ""）-> Gemini signature-only part
+			if strings.TrimSpace(block.Thinking) == "" {
+				parts = append(parts, GeminiPart{
+					ThoughtSignature: signature,
+				})
+			} else {
+				parts = append(parts, GeminiPart{
+					Text:             block.Thinking,
+					Thought:          true,
+					ThoughtSignature: signature,
+				})
 			}
-			parts = append(parts, GeminiPart{
-				Text:             block.Thinking,
-				Thought:          true,
-				ThoughtSignature: signature,
-			})

 		case "image":
 			if block.Source != nil && block.Source.Type == "base64" {
@@ -287,10 +278,15 @@ func buildParts(content json.RawMessage, toolIDToName map[string]string, allowDu
 					ID:   block.ID,
 				},
 			}
-			// 只有 Gemini 模型使用 dummy signature
-			// Claude 模型不设置 signature（避免验证问题）
-			if allowDummyThought {
+			switch thoughtSignatureMode {
+			case thoughtSignatureModeDummy:
 				part.ThoughtSignature = dummyThoughtSignature
+			case thoughtSignatureModePreserve:
+				// Claude via Vertex：透传 tool_use 的 signature（如果有）
+				// 注意：跨模型混用时可能出现 dummy signature，这里直接丢弃以避免 400。
+				if sig := strings.TrimSpace(block.Signature); sig != "" && sig != dummyThoughtSignature {
+					part.ThoughtSignature = sig
+				}
 			}
 			parts = append(parts, part)

--- a/backend/internal/pkg/antigravity/request_transformer_test.go
+++ b/backend/internal/pkg/antigravity/request_transformer_test.go
@@ -8,11 +8,11 @@ import (
 // TestBuildParts_ThinkingBlockWithoutSignature 测试thinking block无signature时的处理
 func TestBuildParts_ThinkingBlockWithoutSignature(t *testing.T) {
 	tests := []struct {
-		name              string
-		content           string
-		allowDummyThought bool
-		expectedParts     int
-		description       string
+		name          string
+		content       string
+		thoughtMode   thoughtSignatureMode
+		expectedParts int
+		description   string
 	}{
 		{
 			name: "Claude model - skip thinking block without signature",
@@ -21,20 +21,20 @@ func TestBuildParts_ThinkingBlockWithoutSignature(t *testing.T) {
 				{"type": "thinking", "thinking": "Let me think...", "signature": ""},
 				{"type": "text", "text": "World"}
 			]`,
-			allowDummyThought: false,
-			expectedParts:     2, // 只有两个text block
-			description:       "Claude模型应该跳过无signature的thinking block",
+			thoughtMode:   thoughtSignatureModePreserve,
+			expectedParts: 2, // 只有两个text block
+			description:   "Claude模型应该跳过无signature的thinking block",
 		},
 		{
-			name: "Claude model - keep thinking block with signature",
+			name: "Claude model - preserve thinking block with signature",
 			content: `[
 				{"type": "text", "text": "Hello"},
-				{"type": "thinking", "thinking": "Let me think...", "signature": "valid_sig"},
+				{"type": "thinking", "thinking": "Let me think...", "signature": "sig_real_123"},
 				{"type": "text", "text": "World"}
 			]`,
-			allowDummyThought: false,
-			expectedParts:     3, // 三个block都保留
-			description:       "Claude模型应该保留有signature的thinking block",
+			thoughtMode:   thoughtSignatureModePreserve,
+			expectedParts: 3,
+			description:   "Claude模型应透传带 signature 的 thinking block（用于 Vertex 签名链路）",
 		},
 		{
 			name: "Gemini model - use dummy signature",
@@ -43,16 +43,27 @@ func TestBuildParts_ThinkingBlockWithoutSignature(t *testing.T) {
 				{"type": "thinking", "thinking": "Let me think...", "signature": ""},
 				{"type": "text", "text": "World"}
 			]`,
-			allowDummyThought: true,
-			expectedParts:     3, // 三个block都保留，thinking使用dummy signature
-			description:       "Gemini模型应该为无signature的thinking block使用dummy signature",
+			thoughtMode:   thoughtSignatureModeDummy,
+			expectedParts: 3, // 三个block都保留，thinking使用dummy signature
+			description:   "Gemini模型应该为无signature的thinking block使用dummy signature",
+		},
+		{
+			name: "Claude model - signature-only thinking block becomes signature-only part",
+			content: `[
+				{"type": "text", "text": "Hello"},
+				{"type": "thinking", "thinking": "", "signature": "sig_only_456"},
+				{"type": "text", "text": "World"}
+			]`,
+			thoughtMode:   thoughtSignatureModePreserve,
+			expectedParts: 3,
+			description:   "Claude模型应将空 thinking + signature 映射为 signature-only part，便于 roundtrip",
 		},
 	}

 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
 			toolIDToName := make(map[string]string)
-			parts, err := buildParts(json.RawMessage(tt.content), toolIDToName, tt.allowDummyThought)
+			parts, err := buildParts(json.RawMessage(tt.content), toolIDToName, tt.thoughtMode)

 			if err != nil {
 				t.Fatalf("buildParts() error = %v", err)
@@ -61,10 +72,71 @@ func TestBuildParts_ThinkingBlockWithoutSignature(t *testing.T) {
 			if len(parts) != tt.expectedParts {
 				t.Errorf("%s: got %d parts, want %d parts", tt.description, len(parts), tt.expectedParts)
 			}
+
+			switch tt.name {
+			case "Claude model - preserve thinking block with signature":
+				if len(parts) != 3 {
+					t.Fatalf("expected 3 parts, got %d", len(parts))
+				}
+				if !parts[1].Thought || parts[1].ThoughtSignature != "sig_real_123" {
+					t.Fatalf("expected thought part with signature sig_real_123, got thought=%v signature=%q",
+						parts[1].Thought, parts[1].ThoughtSignature)
+				}
+			case "Claude model - signature-only thinking block becomes signature-only part":
+				if len(parts) != 3 {
+					t.Fatalf("expected 3 parts, got %d", len(parts))
+				}
+				if parts[1].Thought || parts[1].Text != "" || parts[1].ThoughtSignature != "sig_only_456" {
+					t.Fatalf("expected signature-only part, got thought=%v text=%q signature=%q",
+						parts[1].Thought, parts[1].Text, parts[1].ThoughtSignature)
+				}
+			case "Gemini model - use dummy signature":
+				if len(parts) != 3 {
+					t.Fatalf("expected 3 parts, got %d", len(parts))
+				}
+				if !parts[1].Thought || parts[1].ThoughtSignature != dummyThoughtSignature {
+					t.Fatalf("expected dummy thought signature, got thought=%v signature=%q",
+						parts[1].Thought, parts[1].ThoughtSignature)
+				}
+			}
 		})
 	}
 }

+func TestBuildParts_ToolUseSignatureHandling(t *testing.T) {
+	content := `[
+		{"type": "tool_use", "id": "t1", "name": "Bash", "input": {"command": "ls"}, "signature": "sig_tool_abc"}
+	]`
+
+	t.Run("Claude preserve tool_use signature", func(t *testing.T) {
+		toolIDToName := make(map[string]string)
+		parts, err := buildParts(json.RawMessage(content), toolIDToName, thoughtSignatureModePreserve)
+		if err != nil {
+			t.Fatalf("buildParts() error = %v", err)
+		}
+		if len(parts) != 1 || parts[0].FunctionCall == nil {
+			t.Fatalf("expected 1 functionCall part, got %+v", parts)
+		}
+		if parts[0].ThoughtSignature != "sig_tool_abc" {
+			t.Fatalf("expected tool signature sig_tool_abc, got %q", parts[0].ThoughtSignature)
+		}
+	})
+
+	t.Run("Gemini uses dummy tool_use signature", func(t *testing.T) {
+		toolIDToName := make(map[string]string)
+		parts, err := buildParts(json.RawMessage(content), toolIDToName, thoughtSignatureModeDummy)
+		if err != nil {
+			t.Fatalf("buildParts() error = %v", err)
+		}
+		if len(parts) != 1 || parts[0].FunctionCall == nil {
+			t.Fatalf("expected 1 functionCall part, got %+v", parts)
+		}
+		if parts[0].ThoughtSignature != dummyThoughtSignature {
+			t.Fatalf("expected dummy tool signature %q, got %q", dummyThoughtSignature, parts[0].ThoughtSignature)
+		}
+	})
+}
+
 // TestBuildTools_CustomTypeTools 测试custom类型工具转换
 func TestBuildTools_CustomTypeTools(t *testing.T) {
 	tests := []struct {
--- a/backend/internal/pkg/geminicli/constants.go
+++ b/backend/internal/pkg/geminicli/constants.go
@@ -26,6 +26,10 @@ const (
 	// https://www.googleapis.com/auth/generative-language.retriever (often with cloud-platform).
 	DefaultAIStudioScopes = "https://www.googleapis.com/auth/cloud-platform https://www.googleapis.com/auth/generative-language.retriever"

+	// DefaultScopes for Google One (personal Google accounts with Gemini access)
+	// Includes generative-language for Gemini API access and drive.readonly for storage tier detection
+	DefaultGoogleOneScopes = "https://www.googleapis.com/auth/cloud-platform https://www.googleapis.com/auth/generative-language.retriever https://www.googleapis.com/auth/drive.readonly https://www.googleapis.com/auth/userinfo.email https://www.googleapis.com/auth/userinfo.profile"
+
 	// GeminiCLIRedirectURI is the redirect URI used by Gemini CLI for Code Assist OAuth.
 	GeminiCLIRedirectURI = "https://codeassist.google.com/authcode"

--- a/backend/internal/pkg/geminicli/oauth.go
+++ b/backend/internal/pkg/geminicli/oauth.go
@@ -172,14 +172,19 @@ func EffectiveOAuthConfig(cfg OAuthConfig, oauthType string) (OAuthConfig, error

 	if effective.Scopes == "" {
 		// Use different default scopes based on OAuth type
-		if oauthType == "ai_studio" {
+		switch oauthType {
+		case "ai_studio":
 			// Built-in client can't request some AI Studio scopes (notably generative-language).
 			if isBuiltinClient {
 				effective.Scopes = DefaultCodeAssistScopes
 			} else {
 				effective.Scopes = DefaultAIStudioScopes
 			}
-		} else {
+		case "google_one":
+			// Google One accounts need generative-language scope for Gemini API access
+			// and drive.readonly scope for storage tier detection
+			effective.Scopes = DefaultGoogleOneScopes
+		default:
 			// Default to Code Assist scopes
 			effective.Scopes = DefaultCodeAssistScopes
 		}
--- a/backend/internal/repository/gemini_oauth_client.go
+++ b/backend/internal/repository/gemini_oauth_client.go
@@ -30,13 +30,14 @@ func (c *geminiOAuthClient) ExchangeCode(ctx context.Context, oauthType, code, c

 	// Use different OAuth clients based on oauthType:
 	// - code_assist: always use built-in Gemini CLI OAuth client (public)
+	// - google_one: same as code_assist, uses built-in client for personal Google accounts
 	// - ai_studio: requires a user-provided OAuth client
 	oauthCfgInput := geminicli.OAuthConfig{
 		ClientID:     c.cfg.Gemini.OAuth.ClientID,
 		ClientSecret: c.cfg.Gemini.OAuth.ClientSecret,
 		Scopes:       c.cfg.Gemini.OAuth.Scopes,
 	}
-	if oauthType == "code_assist" {
+	if oauthType == "code_assist" || oauthType == "google_one" {
 		oauthCfgInput.ClientID = ""
 		oauthCfgInput.ClientSecret = ""
 	}
@@ -77,7 +78,7 @@ func (c *geminiOAuthClient) RefreshToken(ctx context.Context, oauthType, refresh
 		ClientSecret: c.cfg.Gemini.OAuth.ClientSecret,
 		Scopes:       c.cfg.Gemini.OAuth.Scopes,
 	}
-	if oauthType == "code_assist" {
+	if oauthType == "code_assist" || oauthType == "google_one" {
 		oauthCfgInput.ClientID = ""
 		oauthCfgInput.ClientSecret = ""
 	}
--- a/backend/internal/service/antigravity_gateway_service.go
+++ b/backend/internal/service/antigravity_gateway_service.go
@@ -307,6 +307,74 @@ func (s *AntigravityGatewayService) unwrapV1InternalResponse(body []byte) ([]byt
 	return body, nil
 }

+// isSignatureRelatedError 检测是否为 signature 相关的 400 错误
+func isSignatureRelatedError(statusCode int, body []byte) bool {
+	if statusCode != 400 {
+		return false
+	}
+
+	bodyStr := strings.ToLower(string(body))
+	keywords := []string{
+		"signature",
+		"thought_signature",
+		"thoughtsignature",
+		"thinking",
+		"invalid signature",
+		"signature validation",
+	}
+
+	for _, keyword := range keywords {
+		if strings.Contains(bodyStr, keyword) {
+			return true
+		}
+	}
+	return false
+}
+
+// stripThinkingFromClaudeRequest 从 Claude 请求中移除所有 thinking 相关内容
+func stripThinkingFromClaudeRequest(req *antigravity.ClaudeRequest) *antigravity.ClaudeRequest {
+	// 创建副本
+	stripped := *req
+
+	// 移除 thinking 配置
+	stripped.Thinking = nil
+
+	// 移除消息中的 thinking 块
+	if len(stripped.Messages) > 0 {
+		newMessages := make([]antigravity.ClaudeMessage, 0, len(stripped.Messages))
+		for _, msg := range stripped.Messages {
+			newMsg := msg
+
+			// 如果 content 是数组，过滤 thinking 块
+			var blocks []map[string]any
+			if err := json.Unmarshal(msg.Content, &blocks); err == nil {
+				filtered := make([]map[string]any, 0, len(blocks))
+				for _, block := range blocks {
+					// 跳过有 type="thinking" 的块
+					if blockType, ok := block["type"].(string); ok && blockType == "thinking" {
+						continue
+					}
+					// 跳过没有 type 但有 thinking 字段的块（untyped thinking blocks）
+					if _, hasType := block["type"]; !hasType {
+						if _, hasThinking := block["thinking"]; hasThinking {
+							continue
+						}
+					}
+					filtered = append(filtered, block)
+				}
+				if newContent, err := json.Marshal(filtered); err == nil {
+					newMsg.Content = newContent
+				}
+			}
+
+			newMessages = append(newMessages, newMsg)
+		}
+		stripped.Messages = newMessages
+	}
+
+	return &stripped
+}
+
 // Forward 转发 Claude 协议请求（Claude → Gemini 转换）
 func (s *AntigravityGatewayService) Forward(ctx context.Context, c *gin.Context, account *Account, body []byte) (*ForwardResult, error) {
 	startTime := time.Now()
@@ -414,11 +482,70 @@ func (s *AntigravityGatewayService) Forward(ctx context.Context, c *gin.Context,
 		respBody, _ := io.ReadAll(io.LimitReader(resp.Body, 2<<20))
 		s.handleUpstreamError(ctx, account, resp.StatusCode, resp.Header, respBody)

-		if s.shouldFailoverUpstreamError(resp.StatusCode) {
-			return nil, &UpstreamFailoverError{StatusCode: resp.StatusCode}
+		// Auto 模式：检测 signature 错误并自动降级重试
+		if isSignatureRelatedError(resp.StatusCode, respBody) && claudeReq.Thinking != nil {
+			log.Printf("[Antigravity] Detected signature-related error, retrying without thinking blocks (account: %s, model: %s)", account.Name, mappedModel)
+
+			// 关闭原始响应，释放连接（respBody 已读取到内存）
+			_ = resp.Body.Close()
+
+			// 移除 thinking 块并重试一次
+			strippedReq := stripThinkingFromClaudeRequest(&claudeReq)
+			strippedBody, err := antigravity.TransformClaudeToGemini(strippedReq, projectID, mappedModel)
+			if err != nil {
+				log.Printf("[Antigravity] Failed to transform stripped request: %v", err)
+				// 降级失败，返回原始错误
+				if s.shouldFailoverUpstreamError(resp.StatusCode) {
+					return nil, &UpstreamFailoverError{StatusCode: resp.StatusCode}
+				}
+				return nil, s.writeMappedClaudeError(c, resp.StatusCode, respBody)
+			}
+
+			// 发送降级请求
+			retryReq, err := antigravity.NewAPIRequest(ctx, action, accessToken, strippedBody)
+			if err != nil {
+				log.Printf("[Antigravity] Failed to create retry request: %v", err)
+				if s.shouldFailoverUpstreamError(resp.StatusCode) {
+					return nil, &UpstreamFailoverError{StatusCode: resp.StatusCode}
+				}
+				return nil, s.writeMappedClaudeError(c, resp.StatusCode, respBody)
+			}
+
+			retryResp, err := s.httpUpstream.Do(retryReq, proxyURL, account.ID, account.Concurrency)
+			if err != nil {
+				log.Printf("[Antigravity] Retry request failed: %v", err)
+				if s.shouldFailoverUpstreamError(resp.StatusCode) {
+					return nil, &UpstreamFailoverError{StatusCode: resp.StatusCode}
+				}
+				return nil, s.writeMappedClaudeError(c, resp.StatusCode, respBody)
+			}
+
+			// 如果重试成功，使用重试的响应（不要 return，让后面的代码处理响应）
+			if retryResp.StatusCode < 400 {
+				log.Printf("[Antigravity] Retry succeeded after stripping thinking blocks (account: %s, model: %s)", account.Name, mappedModel)
+				resp = retryResp
+			} else {
+				// 重试也失败，返回重试的错误
+				retryRespBody, _ := io.ReadAll(io.LimitReader(retryResp.Body, 2<<20))
+				_ = retryResp.Body.Close()
+				log.Printf("[Antigravity] Retry also failed with status %d: %s", retryResp.StatusCode, string(retryRespBody))
+				s.handleUpstreamError(ctx, account, retryResp.StatusCode, retryResp.Header, retryRespBody)
+
+				if s.shouldFailoverUpstreamError(retryResp.StatusCode) {
+					return nil, &UpstreamFailoverError{StatusCode: retryResp.StatusCode}
+				}
+				return nil, s.writeMappedClaudeError(c, retryResp.StatusCode, retryRespBody)
+			}
 		}

-		return nil, s.writeMappedClaudeError(c, resp.StatusCode, respBody)
+		// 不是 signature 错误，或者已经没有 thinking 块，直接返回错误
+		if resp.StatusCode >= 400 {
+			if s.shouldFailoverUpstreamError(resp.StatusCode) {
+				return nil, &UpstreamFailoverError{StatusCode: resp.StatusCode}
+			}
+
+			return nil, s.writeMappedClaudeError(c, resp.StatusCode, respBody)
+		}
 	}

 	requestID := resp.Header.Get("x-request-id")
--- a/backend/internal/service/gateway_request.go
+++ b/backend/internal/service/gateway_request.go
@@ -1,6 +1,7 @@
 package service

 import (
+	"bytes"
 	"encoding/json"
 	"fmt"
 )
@@ -70,3 +71,85 @@ func ParseGatewayRequest(body []byte) (*ParsedRequest, error) {

 	return parsed, nil
 }
+
+// FilterThinkingBlocks removes thinking blocks from request body
+// Returns filtered body or original body if filtering fails (fail-safe)
+// This prevents 400 errors from invalid thinking block signatures
+func FilterThinkingBlocks(body []byte) []byte {
+	// Fast path: if body doesn't contain "thinking", skip parsing
+	if !bytes.Contains(body, []byte("thinking")) {
+		return body
+	}
+
+	var req map[string]any
+	if err := json.Unmarshal(body, &req); err != nil {
+		return body // Return original on parse error
+	}
+
+	messages, ok := req["messages"].([]any)
+	if !ok {
+		return body // No messages array
+	}
+
+	filtered := false
+	for _, msg := range messages {
+		msgMap, ok := msg.(map[string]any)
+		if !ok {
+			continue
+		}
+
+		content, ok := msgMap["content"].([]any)
+		if !ok {
+			continue
+		}
+
+		// Filter thinking blocks from content array
+		newContent := make([]any, 0, len(content))
+		filteredThisMessage := false
+		for _, block := range content {
+			blockMap, ok := block.(map[string]any)
+			if !ok {
+				newContent = append(newContent, block)
+				continue
+			}
+
+			blockType, _ := blockMap["type"].(string)
+			// Explicit Anthropic-style thinking block: {"type":"thinking", ...}
+			if blockType == "thinking" {
+				filtered = true
+				filteredThisMessage = true
+				continue // Skip thinking blocks
+			}
+
+			// Some clients send the "thinking" object without a "type" discriminator.
+			// Vertex/Claude still expects a signature for any thinking block, so we drop it.
+			// We intentionally do not drop other typed blocks (e.g. tool_use) that might
+			// legitimately contain a "thinking" key inside their payload.
+			if blockType == "" {
+				if _, hasThinking := blockMap["thinking"]; hasThinking {
+					filtered = true
+					filteredThisMessage = true
+					continue // Skip thinking blocks
+				}
+			}
+
+			newContent = append(newContent, block)
+		}
+
+		if filteredThisMessage {
+			msgMap["content"] = newContent
+		}
+	}
+
+	if !filtered {
+		return body // No changes needed
+	}
+
+	// Re-serialize
+	newBody, err := json.Marshal(req)
+	if err != nil {
+		return body // Return original on marshal error
+	}
+
+	return newBody
+}
--- a/backend/internal/service/gateway_request_test.go
+++ b/backend/internal/service/gateway_request_test.go
@@ -1,6 +1,7 @@
 package service

 import (
+	"encoding/json"
 	"testing"

 	"github.com/stretchr/testify/require"
@@ -38,3 +39,115 @@ func TestParseGatewayRequest_InvalidStreamType(t *testing.T) {
 	_, err := ParseGatewayRequest(body)
 	require.Error(t, err)
 }
+
+func TestFilterThinkingBlocks(t *testing.T) {
+	containsThinkingBlock := func(body []byte) bool {
+		var req map[string]any
+		if err := json.Unmarshal(body, &req); err != nil {
+			return false
+		}
+		messages, ok := req["messages"].([]any)
+		if !ok {
+			return false
+		}
+		for _, msg := range messages {
+			msgMap, ok := msg.(map[string]any)
+			if !ok {
+				continue
+			}
+			content, ok := msgMap["content"].([]any)
+			if !ok {
+				continue
+			}
+			for _, block := range content {
+				blockMap, ok := block.(map[string]any)
+				if !ok {
+					continue
+				}
+				blockType, _ := blockMap["type"].(string)
+				if blockType == "thinking" {
+					return true
+				}
+				if blockType == "" {
+					if _, hasThinking := blockMap["thinking"]; hasThinking {
+						return true
+					}
+				}
+			}
+		}
+		return false
+	}
+
+	tests := []struct {
+		name         string
+		input        string
+		shouldFilter bool
+		expectError  bool
+	}{
+		{
+			name:         "filters thinking blocks",
+			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:         "handles no thinking blocks",
+			input:        `{"model":"claude-3-5-sonnet-20241022","messages":[{"role":"user","content":[{"type":"text","text":"Hello"}]}]}`,
+			shouldFilter: false,
+		},
+		{
+			name:         "handles invalid JSON gracefully",
+			input:        `{invalid json`,
+			shouldFilter: false,
+			expectError:  true,
+		},
+		{
+			name:         "handles multiple messages with thinking blocks",
+			input:        `{"messages":[{"role":"user","content":[{"type":"text","text":"A"}]},{"role":"assistant","content":[{"type":"thinking","thinking":"think"},{"type":"text","text":"B"}]}]}`,
+			shouldFilter: true,
+		},
+		{
+			name:         "filters thinking blocks without type discriminator",
+			input:        `{"messages":[{"role":"assistant","content":[{"thinking":{"text":"internal"}},{"type":"text","text":"B"}]}]}`,
+			shouldFilter: true,
+		},
+		{
+			name:         "does not filter tool_use input fields named thinking",
+			input:        `{"messages":[{"role":"user","content":[{"type":"tool_use","id":"t1","name":"foo","input":{"thinking":"keepme","x":1}},{"type":"text","text":"Hello"}]}]}`,
+			shouldFilter: false,
+		},
+		{
+			name:         "handles empty messages array",
+			input:        `{"messages":[]}`,
+			shouldFilter: false,
+		},
+		{
+			name:         "handles missing messages field",
+			input:        `{"model":"claude-3"}`,
+			shouldFilter: false,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := FilterThinkingBlocks([]byte(tt.input))
+
+			if tt.expectError {
+				// For invalid JSON, should return original
+				require.Equal(t, tt.input, string(result))
+				return
+			}
+
+			if tt.shouldFilter {
+				require.False(t, containsThinkingBlock(result))
+			} else {
+				// Ensure we don't rewrite JSON when no filtering is needed.
+				require.Equal(t, tt.input, string(result))
+			}
+
+			// Verify valid JSON returned (unless input was invalid)
+			var parsed map[string]any
+			err := json.Unmarshal(result, &parsed)
+			require.NoError(t, err)
+		})
+	}
+}
--- a/backend/internal/service/gateway_service.go
+++ b/backend/internal/service/gateway_service.go
@@ -1136,6 +1136,10 @@ func (s *GatewayService) buildUpstreamRequest(ctx context.Context, c *gin.Contex
 		}
 	}

+	// Filter thinking blocks from request body (prevents 400 errors from missing/invalid signatures).
+	// We apply this for the main /v1/messages path as well as count_tokens.
+	body = FilterThinkingBlocks(body)
+
 	req, err := http.NewRequestWithContext(ctx, "POST", targetURL, bytes.NewReader(body))
 	if err != nil {
 		return nil, err
@@ -1862,6 +1866,9 @@ func (s *GatewayService) buildCountTokensRequest(ctx context.Context, c *gin.Con
 		}
 	}

+	// Filter thinking blocks from request body (prevents 400 errors from invalid signatures)
+	body = FilterThinkingBlocks(body)
+
 	req, err := http.NewRequestWithContext(ctx, "POST", targetURL, bytes.NewReader(body))
 	if err != nil {
 		return nil, err
--- a/deploy/docker-compose.override.yml
+++ b/deploy/docker-compose.override.yml
@@ -1,21 +0,0 @@
-# =============================================================================
-# Docker Compose Override for Local Development
-# =============================================================================
-# This file automatically extends docker-compose-test.yml
-# Usage: docker-compose -f docker-compose-test.yml up -d
-# =============================================================================
-
-services:
-  # ===========================================================================
-  # PostgreSQL - 暴露端口用于本地开发
-  # ===========================================================================
-  postgres:
-    ports:
-      - "127.0.0.1:5432:5432"
-
-  # ===========================================================================
-  # Redis - 暴露端口用于本地开发
-  # ===========================================================================
-  redis:
-    ports:
-      - "127.0.0.1:6379:6379"
--- a/deploy/docker-compose.override.yml.example
+++ b/deploy/docker-compose.override.yml.example
@@ -0,0 +1,137 @@
+# =============================================================================
+# Docker Compose Override Configuration Example
+# =============================================================================
+# This file provides examples for customizing the Docker Compose setup.
+# Copy this file to docker-compose.override.yml and modify as needed.
+#
+# Usage:
+#   cp docker-compose.override.yml.example docker-compose.override.yml
+#   # Edit docker-compose.override.yml with your settings
+#   docker-compose up -d
+#
+# IMPORTANT: docker-compose.override.yml is gitignored and will not be committed.
+# =============================================================================
+
+# =============================================================================
+# Scenario 1: Use External Database and Redis (Recommended for Production)
+# =============================================================================
+# Use this when you have PostgreSQL and Redis running on the host machine
+# or on separate servers.
+#
+# Prerequisites:
+# - PostgreSQL running on host (accessible via host.docker.internal)
+# - Redis running on host (accessible via host.docker.internal)
+# - Update DATABASE_PORT and REDIS_PORT in .env file if using non-standard ports
+#
+# Security Notes:
+# - Ensure PostgreSQL pg_hba.conf allows connections from Docker network
+# - Use strong passwords for database and Redis
+# - Consider using SSL/TLS for database connections in production
+# =============================================================================
+
+services:
+  sub2api:
+    # Remove dependencies on containerized postgres/redis
+    depends_on: []
+
+    # Enable access to host machine services
+    extra_hosts:
+      - "host.docker.internal:host-gateway"
+
+    # Override database and Redis connection settings
+    environment:
+      # PostgreSQL Configuration
+      DATABASE_HOST: host.docker.internal
+      DATABASE_PORT: "5678"  # Change to your PostgreSQL port
+      # DATABASE_USER: postgres  # Uncomment to override
+      # DATABASE_PASSWORD: your_password  # Uncomment to override
+      # DATABASE_DBNAME: sub2api  # Uncomment to override
+
+      # Redis Configuration
+      REDIS_HOST: host.docker.internal
+      REDIS_PORT: "6379"  # Change to your Redis port
+      # REDIS_PASSWORD: your_redis_password  # Uncomment if Redis requires auth
+      # REDIS_DB: 0  # Uncomment to override
+
+  # Disable containerized PostgreSQL
+  postgres:
+    deploy:
+      replicas: 0
+    scale: 0
+
+  # Disable containerized Redis
+  redis:
+    deploy:
+      replicas: 0
+    scale: 0
+
+# =============================================================================
+# Scenario 2: Development with Local Services (Alternative)
+# =============================================================================
+# Uncomment this section if you want to use the containerized postgres/redis
+# but expose their ports for local development tools.
+#
+# Usage: Comment out Scenario 1 above and uncomment this section.
+# =============================================================================
+
+# services:
+#   sub2api:
+#     # Keep default dependencies
+#     pass
+#
+#   postgres:
+#     ports:
+#       - "127.0.0.1:5432:5432"  # Expose PostgreSQL on localhost
+#
+#   redis:
+#     ports:
+#       - "127.0.0.1:6379:6379"  # Expose Redis on localhost
+
+# =============================================================================
+# Scenario 3: Custom Network Configuration
+# =============================================================================
+# Uncomment if you need to connect to an existing Docker network
+# =============================================================================
+
+# networks:
+#   default:
+#     external: true
+#     name: your-existing-network
+
+# =============================================================================
+# Scenario 4: Resource Limits (Production)
+# =============================================================================
+# Uncomment to set resource limits for the sub2api container
+# =============================================================================
+
+# services:
+#   sub2api:
+#     deploy:
+#       resources:
+#         limits:
+#           cpus: '2.0'
+#           memory: 2G
+#         reservations:
+#           cpus: '1.0'
+#           memory: 1G
+
+# =============================================================================
+# Scenario 5: Custom Volumes
+# =============================================================================
+# Uncomment to mount additional volumes (e.g., for logs, backups)
+# =============================================================================
+
+# services:
+#   sub2api:
+#     volumes:
+#       - ./logs:/app/logs
+#       - ./backups:/app/backups
+
+# =============================================================================
+# Additional Notes
+# =============================================================================
+# - This file overrides settings in docker-compose.yml
+# - Environment variables in .env file take precedence
+# - For more information, see: https://docs.docker.com/compose/extends/
+# - Check the main README.md for detailed configuration instructions
+# =============================================================================