[UPDATE] 增强 Claude Thinking 模式支持与 Opus 4.6 动态预算适配

✨ feat(antigravity): 支持 thinking adaptive 类型并适配 Opus 4.6 动态预算 🧪 test(gateway): 增加 thinking 模式解析与签名块过滤的边界用例测试
2026-02-11 10:31:16 +08:00
parent ae6fed15cc
commit 19cca11e00
7 changed files with 148 additions and 20 deletions
--- a/backend/internal/pkg/antigravity/claude_types.go
+++ b/backend/internal/pkg/antigravity/claude_types.go
@@ -27,8 +27,8 @@ type ClaudeMessage struct {

 // ThinkingConfig Thinking 配置
 type ThinkingConfig struct {
-	Type         string `json:"type"`                    // "enabled" or "disabled"
-	BudgetTokens int    `json:"budget_tokens,omitempty"` // thinking budget
+	Type         string `json:"type"`                    // "enabled" / "adaptive" / "disabled"
+	BudgetTokens int    `json:"budget_tokens,omitempty"` // thinking budget（-1 表示动态预算）
 }

 // ClaudeMetadata 请求元数据
--- a/backend/internal/pkg/antigravity/request_transformer.go
+++ b/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 +613,37 @@ 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；否则使用动态预算（-1）
+		// - thinking.type=adaptive：仅在 Antigravity 的 Opus 4.6 上覆写为 （24576）
+		//   （CLIProxyAPI 先翻译为 thinkingLevel="high"，再按模型能力转换为预算值）。
+		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 {
--- a/backend/internal/pkg/antigravity/request_transformer_test.go
+++ b/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)
+			}
+		})
+	}
+}