Merge pull request #6175 from wucm667/feat/issue-6172-antigravity-token-clamp

fix(antigravity): clamp compatible token limits

backend/internal/service/antigravity_gateway_compat.go

@@ -29,6 +29,8 @@ const (
 	AntigravityCredentialRejectedReason GatewayFailureReason = "antigravity_oauth_credential_rejected"
 )
 
+const antigravityCompatMaxTokens = 64000
+
 type antigravityCompatRequest struct {
 	protocol        antigravityCompatProtocol
 	originalBody    []byte
@@ -158,7 +160,7 @@ func preserveChatCompletionTokenLimit(request *apicompat.ChatCompletionsRequest,
 		limit = request.MaxCompletionTokens
 	}
 	if limit != nil && *limit > 0 {
-		claudeRequest.MaxTokens = *limit
+		claudeRequest.MaxTokens = min(*limit, antigravityCompatMaxTokens)
 	}
 }
 

backend/internal/service/antigravity_gateway_compat_test.go

@@ -12,6 +12,7 @@ import (
 	"time"
 
 	"github.com/Wei-Shaw/sub2api/internal/config"
+	"github.com/Wei-Shaw/sub2api/internal/pkg/apicompat"
 	"github.com/gin-gonic/gin"
 	"github.com/stretchr/testify/require"
 	"github.com/tidwall/gjson"
@@ -285,6 +286,21 @@ func TestAntigravityCompatPreservesChatTokenLimit(t *testing.T) {
 			body: `{"model":"gemini-3.1-pro-high","messages":[{"role":"user","content":"ok"}],"max_tokens":8,"max_completion_tokens":13}`,
 			want: 13,
 		},
+		{
+			name: "max_tokens at safe ceiling is preserved",
+			body: `{"model":"gemini-3.1-pro-high","messages":[{"role":"user","content":"ok"}],"max_tokens":64000}`,
+			want: 64000,
+		},
+		{
+			name: "max_tokens above safe ceiling is clamped",
+			body: `{"model":"gemini-3.1-pro-high","messages":[{"role":"user","content":"ok"}],"max_tokens":64001}`,
+			want: 64000,
+		},
+		{
+			name: "precedence applies before clamping",
+			body: `{"model":"gemini-3.1-pro-high","messages":[{"role":"user","content":"ok"}],"max_tokens":8,"max_completion_tokens":64001}`,
+			want: 64000,
+		},
 	}
 
 	for _, tt := range tests {
@@ -310,6 +326,27 @@ func TestAntigravityCompatPreservesChatTokenLimit(t *testing.T) {
 	}
 }
 
+func TestPreserveChatCompletionTokenLimitIgnoresAbsentAndNonPositiveValues(t *testing.T) {
+	tests := []struct {
+		name    string
+		request apicompat.ChatCompletionsRequest
+	}{
+		{name: "absent"},
+		{name: "zero max_tokens", request: apicompat.ChatCompletionsRequest{MaxTokens: antigravityCompatIntPtr(0)}},
+		{name: "negative max_completion_tokens takes precedence", request: apicompat.ChatCompletionsRequest{MaxTokens: antigravityCompatIntPtr(12), MaxCompletionTokens: antigravityCompatIntPtr(-1)}},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			claudeRequest := &apicompat.AnthropicRequest{MaxTokens: 99}
+			preserveChatCompletionTokenLimit(&tt.request, claudeRequest)
+			require.Equal(t, 99, claudeRequest.MaxTokens)
+		})
+	}
+}
+
+func antigravityCompatIntPtr(v int) *int { return &v }
+
 func TestAntigravityCompatRoutesByMappedModelFamily(t *testing.T) {
 	gin.SetMode(gin.TestMode)
 	tests := []struct {