Merge branch 'main' into codex/pr270-content-addressed-read-images

# Conflicts:
#	frontend/src/i18n/generated/catalog.json

internal/backend/agent/model/anthropic.go

@@ -1143,10 +1143,20 @@ func isAnthropicCacheableBlock(block map[string]any) bool {
 	}
 }
 
+// anthropicThinkingCarrier 记录请求内最近一个有 reasoning+signature 的 assistant 轮次。
+// thinking 模式下上游要求每个 assistant 轮次都回传 thinking 块；当某轮次（如 DeepSeek
+// adaptive thinking 跳过思考的 tool-call 轮次）没有 reasoning 时，用 carrier 的
+// thinking+signature 兜底，避免上游 "thinking must be passed back" 400。
+type anthropicThinkingCarrier struct {
+	reasoning string
+	signature string
+}
+
 func normalizeAnthropicProviderMessages(input []Message, thinkingEnabled bool, relocateImages bool) ([]string, []anthropicMessage, error) {
 	systemParts := make([]string, 0, len(input))
 	messages := make([]anthropicMessage, 0, len(input))
 	pendingToolResults := make([]map[string]any, 0, 2)
+	var thinkingCarrier *anthropicThinkingCarrier
 	flushToolResults := func() {
 		if len(pendingToolResults) == 0 {
 			return
@@ -1192,7 +1202,15 @@ func normalizeAnthropicProviderMessages(input []Message, thinkingEnabled bool, r
 			})
 		case "user", "assistant":
 			flushToolResults()
-			contentBlocks, err := anthropicProviderContentBlocks(message, thinkingEnabled)
+			if thinkingEnabled && role == "assistant" {
+				if reasoning := strings.TrimSpace(message.ReasoningContent); reasoning != "" {
+					thinkingCarrier = &anthropicThinkingCarrier{
+						reasoning: reasoning,
+						signature: anthropicThinkingSignature(message),
+					}
+				}
+			}
+			contentBlocks, err := anthropicProviderContentBlocks(message, thinkingEnabled, thinkingCarrier)
 			if err != nil {
 				return nil, nil, err
 			}
@@ -1301,7 +1319,7 @@ func isAnthropicImageBlock(block map[string]any) bool {
 	return strings.TrimSpace(anthropicStringField(block, "type")) == "image"
 }
 
-func anthropicProviderContentBlocks(message Message, thinkingEnabled bool) ([]map[string]any, error) {
+func anthropicProviderContentBlocks(message Message, thinkingEnabled bool, carrier *anthropicThinkingCarrier) ([]map[string]any, error) {
 	blocks, err := anthropicContentBlocks(message)
 	if err != nil {
 		return nil, err
@@ -1310,11 +1328,17 @@ func anthropicProviderContentBlocks(message Message, thinkingEnabled bool) ([]ma
 		return blocks, nil
 	}
 
+	reasoning := strings.TrimSpace(message.ReasoningContent)
+	signature := anthropicThinkingSignature(message)
+	if reasoning == "" && carrier != nil {
+		reasoning = carrier.reasoning
+		signature = carrier.signature
+	}
 	thinkingBlock := map[string]any{
 		"type":     "thinking",
-		"thinking": message.ReasoningContent,
+		"thinking": reasoning,
 	}
-	if signature := anthropicThinkingSignature(message); signature != "" {
+	if signature != "" {
 		thinkingBlock["signature"] = signature
 	}
 	return append([]map[string]any{thinkingBlock}, blocks...), nil
@@ -1398,9 +1422,6 @@ func shouldIncludeAnthropicThinkingBlock(message Message, thinkingEnabled bool)
 	if strings.TrimSpace(message.Role) != "assistant" {
 		return false
 	}
-	if strings.TrimSpace(message.ReasoningContent) == "" {
-		return false
-	}
 	return true
 }
 

internal/backend/agent/model/anthropic_thinking_carrier_test.go

@@ -0,0 +1,164 @@
+package modeladapter
+
+import (
+	"strings"
+	"testing"
+)
+
+// TestNormalizeAnthropicProviderMessagesThinkingCarrier 验证 thinking 模式下，
+// 缺少 reasoning 的 assistant 轮次（如 DeepSeek adaptive thinking 跳过思考的
+// tool-call 轮次）会用请求内最近一个 carrier 的 thinking+signature 兜底，
+// 保证每个 assistant 轮次都有 thinking 块，避免上游 "thinking must be passed
+// back to the API" 400。
+func TestNormalizeAnthropicProviderMessagesThinkingCarrier(t *testing.T) {
+	carrierToolCall := []ToolCallDescriptor{{
+		ID:   "call-2",
+		Type: "function",
+		Function: ToolCallFunctionShape{
+			Name:      "read",
+			Arguments: `{}`,
+		},
+	}}
+	input := []Message{
+		{Role: "user", Content: "hello"},
+		{Role: "assistant", Content: "let me check", ReasoningContent: "R1", ReasoningSignature: "S1"},
+		{Role: "user", Content: "tool result 1"},
+		{Role: "assistant", ToolCalls: carrierToolCall}, // 无 reasoning → 用 carrier
+		{Role: "user", Content: "tool result 2"},
+		{Role: "assistant", Content: "done", ReasoningContent: "R2", ReasoningSignature: "S2"},
+	}
+
+	_, messages, err := normalizeAnthropicProviderMessages(input, true, false)
+	if err != nil {
+		t.Fatalf("normalize: %v", err)
+	}
+	if len(messages) != 6 {
+		t.Fatalf("expected 6 messages, got %d", len(messages))
+	}
+
+	// 第 2 条（有 reasoning）应保留自己的 thinking。
+	assertAnthropicThinkingBlock(t, messages[1], "R1", "S1")
+	// 第 4 条（无 reasoning 的 tool-call 轮次）应复用 carrier 的 thinking+signature。
+	assertAnthropicThinkingBlock(t, messages[3], "R1", "S1")
+	// 第 5 条应为 tool_result 消息（合并路径不适用时，tool-call 轮次独立成消息）。
+	if role := messages[4].Role; role != "user" {
+		t.Fatalf("expected messages[4] role=user, got %s", role)
+	}
+	// 第 6 条有自己的 thinking。
+	assertAnthropicThinkingBlock(t, messages[5], "R2", "S2")
+
+	// tool-call 轮次应包含 tool_use 块。
+	hasToolUse := false
+	for _, block := range messages[3].Content {
+		if strings.TrimSpace(anthropicStringField(block, "type")) == "tool_use" {
+			hasToolUse = true
+		}
+	}
+	if !hasToolUse {
+		t.Fatal("expected tool_use block on the carrier-fallback assistant message")
+	}
+}
+
+// TestNormalizeAnthropicProviderMessagesThinkingCarrierFirstTurn 验证请求内第一条
+// assistant 轮次就缺 reasoning 且无 carrier 时，兜底输出空 thinking 块。
+func TestNormalizeAnthropicProviderMessagesThinkingCarrierFirstTurn(t *testing.T) {
+	input := []Message{
+		{Role: "user", Content: "hello"},
+		{Role: "assistant", Content: "ok"},
+	}
+
+	_, messages, err := normalizeAnthropicProviderMessages(input, true, false)
+	if err != nil {
+		t.Fatalf("normalize: %v", err)
+	}
+	if len(messages) != 2 {
+		t.Fatalf("expected 2 messages, got %d", len(messages))
+	}
+	if got := anthropicStringField(messages[1].Content[0], "type"); got != "thinking" {
+		t.Fatalf("expected first block type=thinking, got %s", got)
+	}
+	if got := anthropicStringField(messages[1].Content[0], "thinking"); got != "" {
+		t.Fatalf("expected empty fallback thinking, got %q", got)
+	}
+}
+
+// TestNormalizeAnthropicProviderMessagesThinkingDisabled 验证 thinking 关闭时
+// 不输出任何 thinking 块（回归保护）。
+func TestNormalizeAnthropicProviderMessagesThinkingDisabled(t *testing.T) {
+	input := []Message{
+		{Role: "user", Content: "hello"},
+		{Role: "assistant", Content: "ok", ReasoningContent: "R1", ReasoningSignature: "S1"},
+		{Role: "assistant", ToolCalls: []ToolCallDescriptor{{
+			ID:   "call-2",
+			Type: "function",
+			Function: ToolCallFunctionShape{
+				Name:      "read",
+				Arguments: `{}`,
+			},
+		}}},
+	}
+
+	_, messages, err := normalizeAnthropicProviderMessages(input, false, false)
+	if err != nil {
+		t.Fatalf("normalize: %v", err)
+	}
+	for index, message := range messages {
+		for _, block := range message.Content {
+			if blockType := anthropicStringField(block, "type"); blockType == "thinking" {
+				t.Fatalf("unexpected thinking block at messages[%d]", index)
+			}
+		}
+	}
+}
+
+// TestNormalizeAnthropicProviderMessagesThinkingMerge 验证有 reasoning 的纯
+// tool-call 轮次仍按既有逻辑合并进上一条 assistant 消息（thinking 去重，无回归）。
+func TestNormalizeAnthropicProviderMessagesThinkingMerge(t *testing.T) {
+	input := []Message{
+		{Role: "user", Content: "hello"},
+		{Role: "assistant", Content: "let me check", ReasoningContent: "R1", ReasoningSignature: "S1"},
+		{Role: "assistant", ToolCalls: []ToolCallDescriptor{{
+			ID:   "call-2",
+			Type: "function",
+			Function: ToolCallFunctionShape{
+				Name:      "read",
+				Arguments: `{}`,
+			},
+		}}, ReasoningContent: "R1", ReasoningSignature: "S1"},
+	}
+
+	_, messages, err := normalizeAnthropicProviderMessages(input, true, false)
+	if err != nil {
+		t.Fatalf("normalize: %v", err)
+	}
+	if len(messages) != 2 {
+		t.Fatalf("expected 2 messages (tool-call merged), got %d", len(messages))
+	}
+	assertAnthropicThinkingBlock(t, messages[1], "R1", "S1")
+	hasToolUse := false
+	for _, block := range messages[1].Content {
+		if blockType := anthropicStringField(block, "type"); blockType == "tool_use" {
+			hasToolUse = true
+		}
+	}
+	if !hasToolUse {
+		t.Fatal("expected merged tool_use block on messages[1]")
+	}
+}
+
+func assertAnthropicThinkingBlock(t *testing.T, message anthropicMessage, wantThinking string, wantSignature string) {
+	t.Helper()
+	if len(message.Content) == 0 {
+		t.Fatalf("expected non-empty content for %s message", message.Role)
+	}
+	first := message.Content[0]
+	if blockType := anthropicStringField(first, "type"); blockType != "thinking" {
+		t.Fatalf("expected first block type=thinking, got %s", blockType)
+	}
+	if got := anthropicStringField(first, "thinking"); got != wantThinking {
+		t.Fatalf("expected thinking=%q, got %q", wantThinking, got)
+	}
+	if got := anthropicStringField(first, "signature"); got != wantSignature {
+		t.Fatalf("expected signature=%q, got %q", wantSignature, got)
+	}
+}

internal/backend/agent/model/openai.go

@@ -49,7 +49,8 @@ type openAIResponsesRequestBody struct {
 }
 
 type openAIResponsesReasoning struct {
-	Effort string `json:"effort,omitempty"`
+	Effort  string `json:"effort,omitempty"`
+	Summary string `json:"summary,omitempty"`
 }
 
 type openAIToolAccumulator struct {
@@ -839,7 +840,7 @@ func (adapter *OpenAIAdapter) streamChatCompletions(ctx context.Context, req Str
 			}
 		}
 
-		if choice.FinishReason != nil {
+		if choice.FinishReason != nil && strings.TrimSpace(*choice.FinishReason) != "" {
 			if err := flushTaggedContentTail(); err != nil {
 				return fail(err)
 			}
@@ -944,7 +945,7 @@ func (adapter *OpenAIAdapter) streamResponses(ctx context.Context, req StreamReq
 			requestBody.Tools = tools
 		}
 		if effort := strings.TrimSpace(req.ReasoningEffort); effort != "" {
-			requestBody.Reasoning = &openAIResponsesReasoning{Effort: effort}
+			requestBody.Reasoning = &openAIResponsesReasoning{Effort: effort, Summary: "auto"}
 			requestBody.Include = []string{"reasoning.encrypted_content"}
 		}
 		body = requestBody

internal/backend/agent/model/openai_stream_test.go

@@ -0,0 +1,223 @@
+package modeladapter
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"net/http"
+	"net/http/httptest"
+	"testing"
+)
+
+func TestOpenAIResponsesRequestsReasoningSummary(t *testing.T) {
+	var requestBody map[string]any
+	server := httptest.NewServer(http.HandlerFunc(func(writer http.ResponseWriter, request *http.Request) {
+		if err := json.NewDecoder(request.Body).Decode(&requestBody); err != nil {
+			http.Error(writer, err.Error(), http.StatusBadRequest)
+			return
+		}
+		writer.Header().Set("Content-Type", "text/event-stream")
+		_, _ = fmt.Fprint(writer, "data: {\"type\":\"response.reasoning_summary_text.delta\",\"delta\":\"check the request\"}\n\n")
+		_, _ = fmt.Fprint(writer, "data: {\"type\":\"response.completed\",\"response\":{\"id\":\"resp-1\",\"model\":\"gpt-5.6\",\"status\":\"completed\",\"output_text\":\"done\"}}\n\n")
+		_, _ = fmt.Fprint(writer, "data: [DONE]\n\n")
+	}))
+	defer server.Close()
+
+	adapter := &OpenAIAdapter{client: server.Client()}
+	events := make([]ModelEvent, 0, 4)
+	err := adapter.Stream(context.Background(), StreamRequest{
+		RequestID:       "request-1",
+		RunID:           "run-1",
+		ModelCallID:     "model-call-1",
+		BaseURL:         server.URL,
+		APIKey:          "test-key",
+		ProviderModelID: "gpt-5.6",
+		OpenAIEndpoint:  "/v1/responses",
+		ReasoningEffort: "high",
+		Messages:        []Message{{Role: "user", Content: "hello"}},
+		MaxTokens:       128,
+	}, func(event ModelEvent) error {
+		events = append(events, event)
+		return nil
+	})
+	if err != nil {
+		t.Fatalf("stream failed: %v", err)
+	}
+
+	reasoning, ok := requestBody["reasoning"].(map[string]any)
+	if !ok {
+		t.Fatalf("reasoning request body missing: %#v", requestBody)
+	}
+	if got := reasoning["effort"]; got != "high" {
+		t.Fatalf("reasoning.effort = %#v, want high", got)
+	}
+	if got := reasoning["summary"]; got != "auto" {
+		t.Fatalf("reasoning.summary = %#v, want auto", got)
+	}
+	include, ok := requestBody["include"].([]any)
+	if !ok || len(include) != 1 || include[0] != "reasoning.encrypted_content" {
+		t.Fatalf("reasoning include = %#v, want encrypted content", requestBody["include"])
+	}
+	assertOpenAIEventKindCount(t, events, ModelEventKindThinkingDelta, 1)
+	assertOpenAIEventKindCount(t, events, ModelEventKindThinkingCompleted, 1)
+	assertOpenAIEventKindCount(t, events, ModelEventKindTextDelta, 1)
+}
+
+func TestOpenAIResponsesOmitsReasoningWhenEffortBlank(t *testing.T) {
+	var requestBody map[string]any
+	server := httptest.NewServer(http.HandlerFunc(func(writer http.ResponseWriter, request *http.Request) {
+		if err := json.NewDecoder(request.Body).Decode(&requestBody); err != nil {
+			http.Error(writer, err.Error(), http.StatusBadRequest)
+			return
+		}
+		writer.Header().Set("Content-Type", "text/event-stream")
+		_, _ = fmt.Fprint(writer, "data: {\"type\":\"response.completed\",\"response\":{\"id\":\"resp-1\",\"model\":\"grok-composer-2.5-fast\",\"status\":\"completed\",\"output_text\":\"done\"}}\n\n")
+		_, _ = fmt.Fprint(writer, "data: [DONE]\n\n")
+	}))
+	defer server.Close()
+
+	adapter := &OpenAIAdapter{client: server.Client()}
+	err := adapter.Stream(context.Background(), StreamRequest{
+		RequestID:       "request-1",
+		RunID:           "run-1",
+		ModelCallID:     "model-call-1",
+		BaseURL:         server.URL,
+		APIKey:          "test-key",
+		ProviderModelID: "grok-composer-2.5-fast",
+		OpenAIEndpoint:  "/v1/responses",
+		Messages:        []Message{{Role: "user", Content: "hello"}},
+		MaxTokens:       128,
+	}, func(ModelEvent) error { return nil })
+	if err != nil {
+		t.Fatalf("stream failed: %v", err)
+	}
+
+	if _, exists := requestBody["reasoning"]; exists {
+		t.Fatalf("reasoning should be omitted when effort is blank: %#v", requestBody["reasoning"])
+	}
+	if _, exists := requestBody["reasoning_effort"]; exists {
+		t.Fatalf("reasoning_effort should be omitted when effort is blank: %#v", requestBody["reasoning_effort"])
+	}
+}
+
+func TestOpenAIChatCompletionsOmitsReasoningWhenEffortBlank(t *testing.T) {
+	var requestBody map[string]any
+	server := httptest.NewServer(http.HandlerFunc(func(writer http.ResponseWriter, request *http.Request) {
+		if err := json.NewDecoder(request.Body).Decode(&requestBody); err != nil {
+			http.Error(writer, err.Error(), http.StatusBadRequest)
+			return
+		}
+		writer.Header().Set("Content-Type", "text/event-stream")
+		_, _ = fmt.Fprint(writer, "data: {\"model\":\"grok-composer-2.5-fast\",\"choices\":[{\"delta\":{\"content\":\"done\"},\"finish_reason\":\"stop\"}]}\n\n")
+		_, _ = fmt.Fprint(writer, "data: [DONE]\n\n")
+	}))
+	defer server.Close()
+
+	adapter := &OpenAIAdapter{client: server.Client()}
+	err := adapter.Stream(context.Background(), StreamRequest{
+		RequestID:       "request-1",
+		RunID:           "run-1",
+		ModelCallID:     "model-call-1",
+		BaseURL:         server.URL,
+		APIKey:          "test-key",
+		ProviderModelID: "grok-composer-2.5-fast",
+		OpenAIEndpoint:  "/v1/chat/completions",
+		Messages:        []Message{{Role: "user", Content: "hello"}},
+		MaxTokens:       128,
+	}, func(ModelEvent) error { return nil })
+	if err != nil {
+		t.Fatalf("stream failed: %v", err)
+	}
+
+	for _, field := range []string{"reasoning_effort", "reasoning", "include"} {
+		if value, exists := requestBody[field]; exists {
+			t.Fatalf("%s should be omitted when effort is blank: %#v", field, value)
+		}
+	}
+}
+
+func TestOpenAIChatCompletionsIgnoresBlankFinishReason(t *testing.T) {
+	server := httptest.NewServer(http.HandlerFunc(func(writer http.ResponseWriter, request *http.Request) {
+		writer.Header().Set("Content-Type", "text/event-stream")
+		chunks := []string{
+			`{"model":"deepseek-v4-flash","choices":[{"delta":{"reasoning_content":"first"},"finish_reason":""}]}`,
+			`{"model":"deepseek-v4-flash","choices":[{"delta":{"reasoning_content":" second"},"finish_reason":""}]}`,
+			`{"model":"deepseek-v4-flash","choices":[{"delta":{"tool_calls":[{"index":0,"id":"call_1","type":"function","function":{"name":"Ls","arguments":""}}]},"finish_reason":""}]}`,
+			`{"model":"deepseek-v4-flash","choices":[{"delta":{"tool_calls":[{"index":0,"id":"","type":"function","function":{"name":"","arguments":"{\"path\":"}}]},"finish_reason":""}]}`,
+			`{"model":"deepseek-v4-flash","choices":[{"delta":{"tool_calls":[{"index":0,"id":"","type":"function","function":{"name":"","arguments":"\"/tmp\"}"}}]},"finish_reason":"tool_calls"}]}`,
+			`{"model":"deepseek-v4-flash","choices":[],"usage":{"prompt_tokens":12,"completion_tokens":7}}`,
+		}
+		for _, chunk := range chunks {
+			_, _ = fmt.Fprintf(writer, "data: %s\n\n", chunk)
+		}
+		_, _ = fmt.Fprint(writer, "data: [DONE]\n\n")
+	}))
+	defer server.Close()
+
+	adapter := &OpenAIAdapter{client: server.Client()}
+	events := make([]ModelEvent, 0, 8)
+	err := adapter.Stream(context.Background(), StreamRequest{
+		RequestID:       "request-1",
+		RunID:           "run-1",
+		ModelCallID:     "model-call-1",
+		BaseURL:         server.URL,
+		APIKey:          "test-key",
+		ProviderModelID: "deepseek-v4-flash",
+		OpenAIEndpoint:  "/v1/chat/completions",
+		Messages:        []Message{{Role: "user", Content: "list files"}},
+		MaxTokens:       128,
+		RequestKnobs:    map[string]any{},
+	}, func(event ModelEvent) error {
+		events = append(events, event)
+		return nil
+	})
+	if err != nil {
+		t.Fatalf("stream failed: %v", err)
+	}
+
+	assertOpenAIEventKindCount(t, events, ModelEventKindThinkingDelta, 2)
+	assertOpenAIEventKindCount(t, events, ModelEventKindThinkingCompleted, 1)
+	assertOpenAIEventKindCount(t, events, ModelEventKindToolLikeCompleted, 1)
+	assertOpenAIEventKindCount(t, events, ModelEventKindTurnFinished, 1)
+
+	toolEvent := firstOpenAIEventForTest(events, ModelEventKindToolLikeCompleted)
+	if toolEvent == nil || toolEvent.ToolInvocation == nil {
+		t.Fatalf("completed tool invocation missing: %#v", toolEvent)
+	}
+	if toolEvent.ToolInvocation.ToolName != "Ls" {
+		t.Fatalf("tool name = %q, want Ls", toolEvent.ToolInvocation.ToolName)
+	}
+	if got := string(toolEvent.ToolInvocation.ArgsJSON); got != `{"path":"/tmp"}` {
+		t.Fatalf("tool args = %q, want %q", got, `{"path":"/tmp"}`)
+	}
+
+	finished := firstOpenAIEventForTest(events, ModelEventKindTurnFinished)
+	if finished == nil || finished.FinishReason != "tool_calls" {
+		t.Fatalf("finish reason = %q, want tool_calls", finished.FinishReason)
+	}
+	if finished.InputTokens != 12 || finished.OutputTokens != 7 {
+		t.Fatalf("usage = input:%d output:%d, want input:12 output:7", finished.InputTokens, finished.OutputTokens)
+	}
+}
+
+func assertOpenAIEventKindCount(t *testing.T, events []ModelEvent, kind ModelEventKind, want int) {
+	t.Helper()
+	got := 0
+	for _, event := range events {
+		if event.Kind == kind {
+			got++
+		}
+	}
+	if got != want {
+		t.Fatalf("event kind %s count = %d, want %d; events=%#v", kind, got, want, events)
+	}
+}
+
+func firstOpenAIEventForTest(events []ModelEvent, kind ModelEventKind) *ModelEvent {
+	for index := range events {
+		if events[index].Kind == kind {
+			return &events[index]
+		}
+	}
+	return nil
+}

internal/backend/agent/model/router_test.go

@@ -1,10 +1,64 @@
 package modeladapter
 
 import (
+	"context"
 	"reflect"
 	"testing"
+	"time"
+
+	legacyruntime "cursor/internal/runtime"
 )
 
+type recordingModelAdapter struct {
+	request StreamRequest
+}
+
+func (adapter *recordingModelAdapter) Stream(_ context.Context, req StreamRequest, _ func(ModelEvent) error) error {
+	adapter.request = req
+	return nil
+}
+
+type staticChannelResolver struct {
+	channel *legacyruntime.ResolvedChannel
+}
+
+func (resolver staticChannelResolver) SelectChannelForModel(context.Context, string) (*legacyruntime.ResolvedChannel, error) {
+	return resolver.channel, nil
+}
+
+func (staticChannelResolver) ProviderStreamIdleTimeout(context.Context) time.Duration {
+	return time.Second
+}
+
+func TestRouterRuntimeDisabledClearsReasoningEffort(t *testing.T) {
+	openAI := &recordingModelAdapter{}
+	router := &Router{
+		openai: openAI,
+		resolver: staticChannelResolver{channel: &legacyruntime.ResolvedChannel{
+			ID:              "channel-a",
+			Provider:        "openai",
+			Model:           "grok-composer-2.5-fast",
+			ReasoningEffort: "medium",
+		}},
+	}
+	requestKnobs := map[string]any{"reasoning_effort": "medium"}
+
+	err := router.Stream(context.Background(), StreamRequest{
+		ModelID:        "channel-a",
+		ThinkingEffort: "disabled",
+		RequestKnobs:   requestKnobs,
+	}, func(ModelEvent) error { return nil })
+	if err != nil {
+		t.Fatalf("Stream returned error: %v", err)
+	}
+	if got := openAI.request.ReasoningEffort; got != "" {
+		t.Fatalf("ReasoningEffort = %q, want blank", got)
+	}
+	if _, exists := openAI.request.RequestKnobs["reasoning_effort"]; exists {
+		t.Fatalf("reasoning_effort knob should be removed: %#v", openAI.request.RequestKnobs)
+	}
+}
+
 func TestSanitizeProviderMessagesMergesLegacyAssistantTextAndToolCallTurnsIdempotently(t *testing.T) {
 	input := []Message{
 		{