feat: enhance OpenAI response rendering to include reasoning and improve tool call detection from thinking channel, and refactor testing scripts for unified unit test execution.

internal/format/openai/render.go

@@ -47,27 +47,46 @@ func BuildResponseObject(responseID, model, finalPrompt, finalThinking, finalTex
 	// produced a standalone structured payload. This prevents accidental
 	// empty output_text on normal prose that merely contains tool_call-like text.
 	detected := util.ParseStandaloneToolCalls(finalText, toolNames)
+	toolCallsFromThinking := false
+	if len(detected) == 0 && strings.TrimSpace(finalThinking) != "" {
+		detected = util.ParseStandaloneToolCalls(finalThinking, toolNames)
+		toolCallsFromThinking = len(detected) > 0
+	}
 	exposedOutputText := finalText
 	output := make([]any, 0, 2)
 	if len(detected) > 0 {
-		exposedOutputText = ""
+		if !toolCallsFromThinking || strings.TrimSpace(finalText) != "" {
+			exposedOutputText = ""
+		} else {
+			exposedOutputText = finalThinking
+		}
+		if strings.TrimSpace(finalThinking) != "" {
+			output = append(output, map[string]any{
+				"type": "reasoning",
+				"text": finalThinking,
+			})
+		}
 		output = append(output, map[string]any{
 			"type":       "tool_calls",
 			"tool_calls": util.FormatOpenAIToolCalls(detected),
 		})
 	} else {
-		content := []any{
-			map[string]any{
-				"type": "output_text",
-				"text": finalText,
-			},
-		}
+		content := make([]any, 0, 2)
 		if finalThinking != "" {
 			content = append([]any{map[string]any{
 				"type": "reasoning",
 				"text": finalThinking,
 			}}, content...)
 		}
+		if strings.TrimSpace(finalText) != "" {
+			content = append(content, map[string]any{
+				"type": "output_text",
+				"text": finalText,
+			})
+		}
+		if strings.TrimSpace(finalText) == "" && strings.TrimSpace(finalThinking) != "" {
+			exposedOutputText = finalThinking
+		}
 		output = append(output, map[string]any{
 			"type":    "message",
 			"id":      "msg_" + strings.ReplaceAll(uuid.NewString(), "-", ""),

internal/format/openai/render_test.go

@@ -87,3 +87,60 @@ func TestBuildResponseObjectKeepsOutputTextForMixedProse(t *testing.T) {
 		t.Fatalf("expected output type message, got %#v", first["type"])
 	}
 }
+
+func TestBuildResponseObjectReasoningOnlyFallsBackToOutputText(t *testing.T) {
+	obj := BuildResponseObject(
+		"resp_test",
+		"gpt-4o",
+		"prompt",
+		"internal thinking content",
+		"",
+		nil,
+	)
+
+	outputText, _ := obj["output_text"].(string)
+	if outputText == "" {
+		t.Fatalf("expected output_text fallback from reasoning when final text is empty")
+	}
+
+	output, _ := obj["output"].([]any)
+	if len(output) != 1 {
+		t.Fatalf("expected one output item, got %#v", obj["output"])
+	}
+	first, _ := output[0].(map[string]any)
+	if first["type"] != "message" {
+		t.Fatalf("expected output type message, got %#v", first["type"])
+	}
+	content, _ := first["content"].([]any)
+	if len(content) == 0 {
+		t.Fatalf("expected reasoning content, got %#v", first["content"])
+	}
+	block0, _ := content[0].(map[string]any)
+	if block0["type"] != "reasoning" {
+		t.Fatalf("expected first content block reasoning, got %#v", block0["type"])
+	}
+}
+
+func TestBuildResponseObjectDetectsToolCallFromThinkingChannel(t *testing.T) {
+	obj := BuildResponseObject(
+		"resp_test",
+		"gpt-4o",
+		"prompt",
+		`{"tool_calls":[{"name":"search","input":{"q":"from-thinking"}}]}`,
+		"",
+		[]string{"search"},
+	)
+
+	output, _ := obj["output"].([]any)
+	if len(output) != 2 {
+		t.Fatalf("expected reasoning + tool_calls outputs, got %#v", obj["output"])
+	}
+	first, _ := output[0].(map[string]any)
+	if first["type"] != "reasoning" {
+		t.Fatalf("expected first output reasoning, got %#v", first["type"])
+	}
+	second, _ := output[1].(map[string]any)
+	if second["type"] != "tool_calls" {
+		t.Fatalf("expected second output tool_calls, got %#v", second["type"])
+	}
+}