refactor: 实现 ConversionEngine 协议转换引擎，替代旧 protocol 包

- 新增 ConversionEngine 核心引擎，支持 OpenAI 和 Anthropic 协议转换
- 添加 stream decoder/encoder 实现
- 更新 provider client 支持新引擎
- 补充单元测试和集成测试
- 更新 specs 文档

backend/internal/conversion/anthropic/stream_decoder.go

@@ -251,7 +251,7 @@ func (d *StreamDecoder) processMessageDelta(data []byte) []canonical.CanonicalSt
 	}
 
 	if d.accumulatedUsage != nil {
-		d.accumulatedUsage.OutputTokens += raw.Usage.OutputTokens
+		d.accumulatedUsage.OutputTokens = raw.Usage.OutputTokens
 	}
 
 	return []canonical.CanonicalStreamEvent{

backend/internal/conversion/anthropic/stream_decoder_test.go

@@ -272,3 +272,218 @@ func TestStreamDecoder_RedactedDeltaSuppressed(t *testing.T) {
 	events := d.ProcessChunk(raw)
 	assert.Empty(t, events)
 }
+
+func TestStreamDecoder_ServerToolUse_Suppressed(t *testing.T) {
+	d := NewStreamDecoder()
+
+	payload := map[string]any{
+		"type":  "content_block_start",
+		"index": 2,
+		"content_block": map[string]any{
+			"type": "server_tool_use",
+			"id":   "server_tool_1",
+			"name": "web_search",
+		},
+	}
+	raw := makeAnthropicEvent("content_block_start", payload)
+	events := d.ProcessChunk(raw)
+	assert.Empty(t, events)
+	assert.True(t, d.redactedBlocks[2])
+}
+
+func TestStreamDecoder_WebSearchToolResult_Suppressed(t *testing.T) {
+	d := NewStreamDecoder()
+
+	payload := map[string]any{
+		"type":  "content_block_start",
+		"index": 3,
+		"content_block": map[string]any{
+			"type": "web_search_tool_result",
+			"tool_use_id": "search_1",
+		},
+	}
+	raw := makeAnthropicEvent("content_block_start", payload)
+	events := d.ProcessChunk(raw)
+	assert.Empty(t, events)
+	assert.True(t, d.redactedBlocks[3])
+}
+
+func TestStreamDecoder_CodeExecutionToolResult_Suppressed(t *testing.T) {
+	d := NewStreamDecoder()
+
+	payload := map[string]any{
+		"type":  "content_block_start",
+		"index": 4,
+		"content_block": map[string]any{
+			"type": "code_execution_tool_result",
+		},
+	}
+	raw := makeAnthropicEvent("content_block_start", payload)
+	events := d.ProcessChunk(raw)
+	assert.Empty(t, events)
+	assert.True(t, d.redactedBlocks[4])
+}
+
+func TestStreamDecoder_CitationsDelta_Discarded(t *testing.T) {
+	d := NewStreamDecoder()
+
+	payload := map[string]any{
+		"type":  "content_block_delta",
+		"index": 0,
+		"delta": map[string]any{
+			"type":        "citations_delta",
+			"citation":    map[string]any{"title": "ref1"},
+		},
+	}
+	raw := makeAnthropicEvent("content_block_delta", payload)
+	events := d.ProcessChunk(raw)
+	assert.Empty(t, events)
+}
+
+func TestStreamDecoder_SignatureDelta_Discarded(t *testing.T) {
+	d := NewStreamDecoder()
+
+	payload := map[string]any{
+		"type":  "content_block_delta",
+		"index": 0,
+		"delta": map[string]any{
+			"type":      "signature_delta",
+			"signature": "sig_123",
+		},
+	}
+	raw := makeAnthropicEvent("content_block_delta", payload)
+	events := d.ProcessChunk(raw)
+	assert.Empty(t, events)
+}
+
+func TestStreamDecoder_UnknownEventType(t *testing.T) {
+	d := NewStreamDecoder()
+
+	raw := makeAnthropicEvent("unknown_event", map[string]any{"type": "unknown_event"})
+	events := d.ProcessChunk(raw)
+	assert.Empty(t, events)
+}
+
+func TestStreamDecoder_InvalidJSON(t *testing.T) {
+	d := NewStreamDecoder()
+
+	raw := []byte("event: message_start\ndata: {invalid}\n\n")
+	events := d.ProcessChunk(raw)
+	assert.Empty(t, events)
+}
+
+func TestStreamDecoder_MultipleEventsInSingleChunk(t *testing.T) {
+	d := NewStreamDecoder()
+
+	startPayload := map[string]any{
+		"type": "message_start",
+		"message": map[string]any{
+			"id":    "msg_multi",
+			"model": "claude-3",
+		},
+	}
+	deltaPayload := map[string]any{
+		"type":  "content_block_delta",
+		"index": 0,
+		"delta": map[string]any{
+			"type": "text_delta",
+			"text": "Hello",
+		},
+	}
+	stopPayload := map[string]any{"type": "message_stop"}
+
+	var raw []byte
+	raw = append(raw, makeAnthropicEvent("message_start", startPayload)...)
+	raw = append(raw, makeAnthropicEvent("content_block_delta", deltaPayload)...)
+	raw = append(raw, makeAnthropicEvent("message_stop", stopPayload)...)
+
+	events := d.ProcessChunk(raw)
+	require.Len(t, events, 3)
+	assert.Equal(t, canonical.EventMessageStart, events[0].Type)
+	assert.Equal(t, canonical.EventContentBlockDelta, events[1].Type)
+	assert.Equal(t, canonical.EventMessageStop, events[2].Type)
+}
+
+func TestStreamDecoder_ErrorInvalidJSON(t *testing.T) {
+	d := NewStreamDecoder()
+
+	raw := []byte("event: error\ndata: {invalid}\n\n")
+	events := d.ProcessChunk(raw)
+	require.Len(t, events, 1)
+	assert.Equal(t, canonical.EventError, events[0].Type)
+	assert.Contains(t, events[0].Error.Message, "解析错误事件失败")
+}
+
+func TestStreamDecoder_MessageStartWithUsage(t *testing.T) {
+	d := NewStreamDecoder()
+
+	payload := map[string]any{
+		"type": "message_start",
+		"message": map[string]any{
+			"id":    "msg_usage",
+			"model": "claude-3",
+			"usage": map[string]any{"input_tokens": 25, "output_tokens": 0},
+		},
+	}
+	raw := makeAnthropicEvent("message_start", payload)
+
+	events := d.ProcessChunk(raw)
+	require.Len(t, events, 1)
+	assert.Equal(t, canonical.EventMessageStart, events[0].Type)
+	require.NotNil(t, events[0].Message.Usage)
+	assert.Equal(t, 25, events[0].Message.Usage.InputTokens)
+}
+
+func TestStreamDecoder_ThinkingBlockStart(t *testing.T) {
+	d := NewStreamDecoder()
+
+	payload := map[string]any{
+		"type":  "content_block_start",
+		"index": 0,
+		"content_block": map[string]any{
+			"type":     "thinking",
+			"thinking": "",
+		},
+	}
+	raw := makeAnthropicEvent("content_block_start", payload)
+
+	events := d.ProcessChunk(raw)
+	require.Len(t, events, 1)
+	assert.Equal(t, canonical.EventContentBlockStart, events[0].Type)
+	require.NotNil(t, events[0].ContentBlock)
+	assert.Equal(t, "thinking", events[0].ContentBlock.Type)
+}
+
+func TestStreamDecoder_MessageDelta_UsageNotAccumulated(t *testing.T) {
+	d := NewStreamDecoder()
+
+	startPayload := map[string]any{
+		"type": "message_start",
+		"message": map[string]any{
+			"id":    "msg_usage_test",
+			"model": "claude-3",
+			"usage": map[string]any{"input_tokens": 10, "output_tokens": 0},
+		},
+	}
+	deltaPayload1 := map[string]any{
+		"type": "message_delta",
+		"delta": map[string]any{"stop_reason": "end_turn"},
+		"usage": map[string]any{"output_tokens": 25},
+	}
+
+	d.ProcessChunk(makeAnthropicEvent("message_start", startPayload))
+	events := d.ProcessChunk(makeAnthropicEvent("message_delta", deltaPayload1))
+
+	require.Len(t, events, 1)
+	assert.Equal(t, 25, events[0].Usage.OutputTokens)
+
+	deltaPayload2 := map[string]any{
+		"type": "message_delta",
+		"delta": map[string]any{"stop_reason": "end_turn"},
+		"usage": map[string]any{"output_tokens": 30},
+	}
+	events = d.ProcessChunk(makeAnthropicEvent("message_delta", deltaPayload2))
+	require.Len(t, events, 1)
+	assert.Equal(t, 30, events[0].Usage.OutputTokens, "output_tokens should be replaced, not accumulated")
+	assert.Equal(t, 30, d.accumulatedUsage.OutputTokens, "accumulated usage should match last value")
+}

backend/internal/conversion/anthropic/supplemental_test.go

@@ -0,0 +1,477 @@
+package anthropic
+
+import (
+	"encoding/json"
+	"testing"
+	"time"
+
+	"nex/backend/internal/conversion"
+	"nex/backend/internal/conversion/canonical"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestDecodeTools(t *testing.T) {
+	body := []byte(`{
+		"model": "claude-3",
+		"max_tokens": 1024,
+		"messages": [{"role": "user", "content": "hi"}],
+		"tools": [
+			{"name": "search", "description": "Search", "input_schema": {"type":"object"}},
+			{"name": "calc", "input_schema": {"type":"object"}}
+		]
+	}`)
+	req, err := decodeRequest(body)
+	require.NoError(t, err)
+	assert.Len(t, req.Tools, 2)
+	assert.Equal(t, "search", req.Tools[0].Name)
+	assert.Equal(t, "Search", req.Tools[0].Description)
+	assert.Equal(t, "calc", req.Tools[1].Name)
+}
+
+func TestDecodeToolChoice(t *testing.T) {
+	tests := []struct {
+		name     string
+		jsonBody string
+		wantType string
+		wantName string
+	}{
+		{
+			"auto string",
+			`{"model":"claude-3","max_tokens":1024,"messages":[{"role":"user","content":"hi"}],"tool_choice":"auto"}`,
+			"auto", "",
+		},
+		{
+			"none string",
+			`{"model":"claude-3","max_tokens":1024,"messages":[{"role":"user","content":"hi"}],"tool_choice":"none"}`,
+			"none", "",
+		},
+		{
+			"any string",
+			`{"model":"claude-3","max_tokens":1024,"messages":[{"role":"user","content":"hi"}],"tool_choice":"any"}`,
+			"any", "",
+		},
+		{
+			"tool object",
+			`{"model":"claude-3","max_tokens":1024,"messages":[{"role":"user","content":"hi"}],"tool_choice":{"type":"tool","name":"search"}}`,
+			"tool", "search",
+		},
+		{
+			"auto object",
+			`{"model":"claude-3","max_tokens":1024,"messages":[{"role":"user","content":"hi"}],"tool_choice":{"type":"auto"}}`,
+			"auto", "",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			req, err := decodeRequest([]byte(tt.jsonBody))
+			require.NoError(t, err)
+			require.NotNil(t, req.ToolChoice)
+			assert.Equal(t, tt.wantType, req.ToolChoice.Type)
+			assert.Equal(t, tt.wantName, req.ToolChoice.Name)
+		})
+	}
+}
+
+func TestDecodeParameters_TopK(t *testing.T) {
+	topK := 10
+	body := []byte(`{
+		"model": "claude-3",
+		"max_tokens": 1024,
+		"messages": [{"role": "user", "content": "hi"}],
+		"top_k": 10,
+		"stop_sequences": ["STOP"]
+	}`)
+	req, err := decodeRequest(body)
+	require.NoError(t, err)
+	require.NotNil(t, req.Parameters.TopK)
+	assert.Equal(t, topK, *req.Parameters.TopK)
+	assert.Equal(t, []string{"STOP"}, req.Parameters.StopSequences)
+}
+
+func TestDecodeRequest_MetadataUserID(t *testing.T) {
+	body := []byte(`{
+		"model": "claude-3",
+		"max_tokens": 1024,
+		"messages": [{"role": "user", "content": "hi"}],
+		"metadata": {"user_id": "user-123"}
+	}`)
+	req, err := decodeRequest(body)
+	require.NoError(t, err)
+	assert.Equal(t, "user-123", req.UserID)
+}
+
+func TestDecodeSystem_Empty(t *testing.T) {
+	body := []byte(`{
+		"model": "claude-3",
+		"max_tokens": 1024,
+		"system": "",
+		"messages": [{"role": "user", "content": "hi"}]
+	}`)
+	req, err := decodeRequest(body)
+	require.NoError(t, err)
+	assert.Nil(t, req.System)
+}
+
+func TestDecodeSystem_Nil(t *testing.T) {
+	body := []byte(`{
+		"model": "claude-3",
+		"max_tokens": 1024,
+		"messages": [{"role": "user", "content": "hi"}]
+	}`)
+	req, err := decodeRequest(body)
+	require.NoError(t, err)
+	assert.Nil(t, req.System)
+}
+
+func TestDecodeThinking_WithEffort(t *testing.T) {
+	body := []byte(`{
+		"model": "claude-3",
+		"max_tokens": 1024,
+		"messages": [{"role": "user", "content": "hi"}],
+		"thinking": {"type": "enabled", "budget_tokens": 5000},
+		"output_config": {"effort": "high"}
+	}`)
+	req, err := decodeRequest(body)
+	require.NoError(t, err)
+	require.NotNil(t, req.Thinking)
+	assert.Equal(t, "enabled", req.Thinking.Type)
+	assert.Equal(t, "high", req.Thinking.Effort)
+}
+
+func TestDecodeOutputFormat_NilOutputConfig(t *testing.T) {
+	body := []byte(`{
+		"model": "claude-3",
+		"max_tokens": 1024,
+		"messages": [{"role": "user", "content": "hi"}]
+	}`)
+	req, err := decodeRequest(body)
+	require.NoError(t, err)
+	assert.Nil(t, req.OutputFormat)
+}
+
+func TestDecodeMessage_UserWithOnlyToolResults(t *testing.T) {
+	body := []byte(`{
+		"model": "claude-3",
+		"max_tokens": 1024,
+		"messages": [
+			{"role": "user", "content": "hi"},
+			{"role": "assistant", "content": [{"type": "tool_use", "id": "t1", "name": "fn", "input": {}}]},
+			{
+				"role": "user",
+				"content": [{"type": "tool_result", "tool_use_id": "t1", "content": "result"}]
+			}
+		]
+	}`)
+	req, err := decodeRequest(body)
+	require.NoError(t, err)
+	lastMsg := req.Messages[len(req.Messages)-1]
+	assert.Equal(t, canonical.RoleTool, lastMsg.Role)
+	assert.Equal(t, "t1", lastMsg.Content[0].ToolUseID)
+}
+
+func TestDecodeContentBlocks_Nil(t *testing.T) {
+	blocks := decodeContentBlocks(nil)
+	assert.Len(t, blocks, 1)
+	assert.Equal(t, "", blocks[0].Text)
+}
+
+func TestDecodeContentBlocks_String(t *testing.T) {
+	blocks := decodeContentBlocks("hello")
+	assert.Len(t, blocks, 1)
+	assert.Equal(t, "hello", blocks[0].Text)
+}
+
+func TestParseTimestamp(t *testing.T) {
+	tests := []struct {
+		name  string
+		input string
+		want  int64
+	}{
+		{"valid RFC3339", "2024-01-15T00:00:00Z", time.Date(2024, 1, 15, 0, 0, 0, 0, time.UTC).Unix()},
+		{"empty", "", 0},
+		{"invalid", "not-a-date", 0},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			assert.Equal(t, tt.want, parseTimestamp(tt.input))
+		})
+	}
+}
+
+func TestEncodeToolChoice(t *testing.T) {
+	tests := []struct {
+		name   string
+		choice *canonical.ToolChoice
+		want   map[string]any
+	}{
+		{"auto", canonical.NewToolChoiceAuto(), map[string]any{"type": "auto"}},
+		{"none", canonical.NewToolChoiceNone(), map[string]any{"type": "none"}},
+		{"any", canonical.NewToolChoiceAny(), map[string]any{"type": "any"}},
+		{"tool", canonical.NewToolChoiceNamed("search"), map[string]any{"type": "tool", "name": "search"}},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := encodeToolChoice(tt.choice)
+			assert.Equal(t, tt.want["type"], result.(map[string]any)["type"])
+			assert.Equal(t, tt.want["name"], result.(map[string]any)["name"])
+		})
+	}
+}
+
+func TestEncodeThinkingConfig(t *testing.T) {
+	budget := 5000
+	tests := []struct {
+		name string
+		cfg  *canonical.ThinkingConfig
+		want map[string]any
+	}{
+		{"enabled", &canonical.ThinkingConfig{Type: "enabled", BudgetTokens: &budget}, map[string]any{"type": "enabled", "budget_tokens": float64(5000)}},
+		{"disabled", &canonical.ThinkingConfig{Type: "disabled"}, map[string]any{"type": "disabled"}},
+		{"adaptive", &canonical.ThinkingConfig{Type: "adaptive"}, map[string]any{"type": "adaptive"}},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := encodeThinkingConfig(tt.cfg)
+			assert.Equal(t, tt.want["type"], result["type"])
+		})
+	}
+}
+
+func TestEncodeRequest_PublicFields(t *testing.T) {
+	maxTokens := 1024
+	parallel := false
+	req := &canonical.CanonicalRequest{
+		Model:           "claude-3",
+		Parameters:      canonical.RequestParameters{MaxTokens: &maxTokens},
+		Messages:        []canonical.CanonicalMessage{{Role: canonical.RoleUser, Content: []canonical.ContentBlock{canonical.NewTextBlock("hi")}}},
+		UserID:          "user-123",
+		ParallelToolUse: &parallel,
+	}
+	provider := conversion.NewTargetProvider("", "key", "model")
+
+	body, err := encodeRequest(req, provider)
+	require.NoError(t, err)
+
+	var result map[string]any
+	require.NoError(t, json.Unmarshal(body, &result))
+	assert.Equal(t, map[string]any{"user_id": "user-123"}, result["metadata"])
+	assert.Equal(t, true, result["disable_parallel_tool_use"])
+}
+
+func TestEncodeRequest_DefaultMaxTokens(t *testing.T) {
+	req := &canonical.CanonicalRequest{
+		Model:    "claude-3",
+		Messages: []canonical.CanonicalMessage{{Role: canonical.RoleUser, Content: []canonical.ContentBlock{canonical.NewTextBlock("hi")}}},
+	}
+	provider := conversion.NewTargetProvider("", "key", "model")
+
+	body, err := encodeRequest(req, provider)
+	require.NoError(t, err)
+
+	var result map[string]any
+	require.NoError(t, json.Unmarshal(body, &result))
+	assert.Equal(t, float64(4096), result["max_tokens"])
+}
+
+func TestEncodeRequest_TopK(t *testing.T) {
+	maxTokens := 1024
+	topK := 10
+	req := &canonical.CanonicalRequest{
+		Model:      "claude-3",
+		Parameters: canonical.RequestParameters{MaxTokens: &maxTokens, TopK: &topK},
+		Messages:   []canonical.CanonicalMessage{{Role: canonical.RoleUser, Content: []canonical.ContentBlock{canonical.NewTextBlock("hi")}}},
+	}
+	provider := conversion.NewTargetProvider("", "key", "model")
+
+	body, err := encodeRequest(req, provider)
+	require.NoError(t, err)
+
+	var result map[string]any
+	require.NoError(t, json.Unmarshal(body, &result))
+	assert.Equal(t, float64(10), result["top_k"])
+}
+
+func TestEncodeRequest_WithTools(t *testing.T) {
+	maxTokens := 1024
+	req := &canonical.CanonicalRequest{
+		Model:      "claude-3",
+		Parameters: canonical.RequestParameters{MaxTokens: &maxTokens},
+		Messages:   []canonical.CanonicalMessage{{Role: canonical.RoleUser, Content: []canonical.ContentBlock{canonical.NewTextBlock("hi")}}},
+		Tools: []canonical.CanonicalTool{
+			{Name: "search", Description: "Search things", InputSchema: json.RawMessage(`{"type":"object"}`)},
+		},
+		ToolChoice: canonical.NewToolChoiceAuto(),
+	}
+	provider := conversion.NewTargetProvider("", "key", "model")
+
+	body, err := encodeRequest(req, provider)
+	require.NoError(t, err)
+
+	var result map[string]any
+	require.NoError(t, json.Unmarshal(body, &result))
+	tools := result["tools"].([]any)
+	assert.Len(t, tools, 1)
+	tool := tools[0].(map[string]any)
+	assert.Equal(t, "search", tool["name"])
+	assert.Equal(t, "Search things", tool["description"])
+	tc := result["tool_choice"].(map[string]any)
+	assert.Equal(t, "auto", tc["type"])
+}
+
+func TestEncodeRequest_ThinkingWithEffort(t *testing.T) {
+	maxTokens := 1024
+	req := &canonical.CanonicalRequest{
+		Model:      "claude-3",
+		Parameters: canonical.RequestParameters{MaxTokens: &maxTokens},
+		Messages:   []canonical.CanonicalMessage{{Role: canonical.RoleUser, Content: []canonical.ContentBlock{canonical.NewTextBlock("hi")}}},
+		Thinking:   &canonical.ThinkingConfig{Type: "enabled", Effort: "high"},
+	}
+	provider := conversion.NewTargetProvider("", "key", "model")
+
+	body, err := encodeRequest(req, provider)
+	require.NoError(t, err)
+
+	var result map[string]any
+	require.NoError(t, json.Unmarshal(body, &result))
+	oc, ok := result["output_config"].(map[string]any)
+	require.True(t, ok)
+	assert.Equal(t, "high", oc["effort"])
+}
+
+func TestEncodeResponse_UsageWithCacheAndCreation(t *testing.T) {
+	cacheRead := 30
+	cacheCreation := 10
+	sr := canonical.StopReasonEndTurn
+	resp := &canonical.CanonicalResponse{
+		ID:         "msg-1",
+		Model:      "claude-3",
+		Content:    []canonical.ContentBlock{canonical.NewTextBlock("ok")},
+		StopReason: &sr,
+		Usage: canonical.CanonicalUsage{
+			InputTokens:        100,
+			OutputTokens:       50,
+			CacheReadTokens:    &cacheRead,
+			CacheCreationTokens: &cacheCreation,
+		},
+	}
+
+	body, err := encodeResponse(resp)
+	require.NoError(t, err)
+
+	var result map[string]any
+	require.NoError(t, json.Unmarshal(body, &result))
+	usage := result["usage"].(map[string]any)
+	assert.Equal(t, float64(100), usage["input_tokens"])
+	assert.Equal(t, float64(30), usage["cache_read_input_tokens"])
+	assert.Equal(t, float64(10), usage["cache_creation_input_tokens"])
+}
+
+func TestEncodeResponse_StopReasons(t *testing.T) {
+	tests := []struct {
+		name       string
+		stopReason canonical.StopReason
+		want       string
+	}{
+		{"end_turn", canonical.StopReasonEndTurn, "end_turn"},
+		{"max_tokens", canonical.StopReasonMaxTokens, "max_tokens"},
+		{"tool_use", canonical.StopReasonToolUse, "tool_use"},
+		{"stop_sequence", canonical.StopReasonStopSequence, "stop_sequence"},
+		{"refusal", canonical.StopReasonRefusal, "refusal"},
+		{"content_filter→end_turn", canonical.StopReasonContentFilter, "end_turn"},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			sr := tt.stopReason
+			resp := &canonical.CanonicalResponse{
+				ID:         "r1",
+				Model:      "claude-3",
+				Content:    []canonical.ContentBlock{canonical.NewTextBlock("ok")},
+				StopReason: &sr,
+			}
+			body, err := encodeResponse(resp)
+			require.NoError(t, err)
+
+			var result map[string]any
+			require.NoError(t, json.Unmarshal(body, &result))
+			assert.Equal(t, tt.want, result["stop_reason"])
+		})
+	}
+}
+
+func TestEncodeSystem_SystemBlocks(t *testing.T) {
+	result := encodeSystem([]canonical.SystemBlock{{Text: "part1"}, {Text: "part2"}})
+	blocks, ok := result.([]map[string]any)
+	require.True(t, ok)
+	assert.Len(t, blocks, 2)
+	assert.Equal(t, "part1", blocks[0]["text"])
+}
+
+func TestEncodeModelInfoResponse(t *testing.T) {
+	info := &canonical.CanonicalModelInfo{
+		ID:      "claude-3-opus",
+		Name:    "Claude 3 Opus",
+		Created: time.Date(2024, 1, 15, 0, 0, 0, 0, time.UTC).Unix(),
+	}
+
+	body, err := encodeModelInfoResponse(info)
+	require.NoError(t, err)
+
+	var result map[string]any
+	require.NoError(t, json.Unmarshal(body, &result))
+	assert.Equal(t, "claude-3-opus", result["id"])
+	assert.Equal(t, "Claude 3 Opus", result["display_name"])
+}
+
+func TestDecodeModelInfoResponse(t *testing.T) {
+	body := []byte(`{"id":"claude-3-opus","type":"model","display_name":"Claude 3 Opus","created_at":"2024-01-15T00:00:00Z"}`)
+	info, err := decodeModelInfoResponse(body)
+	require.NoError(t, err)
+	assert.Equal(t, "claude-3-opus", info.ID)
+	assert.Equal(t, "Claude 3 Opus", info.Name)
+	assert.NotEqual(t, int64(0), info.Created)
+}
+
+func TestDecodeResponse_PauseTurn(t *testing.T) {
+	body := []byte(`{
+		"id": "msg-1", "type": "message", "role": "assistant", "model": "claude-3",
+		"content": [{"type": "text", "text": "ok"}],
+		"stop_reason": "pause_turn",
+		"usage": {"input_tokens": 1, "output_tokens": 1}
+	}`)
+	resp, err := decodeResponse(body)
+	require.NoError(t, err)
+	assert.Equal(t, canonical.StopReason("pause_turn"), *resp.StopReason)
+}
+
+func TestEncodeResponse_NoStopReason(t *testing.T) {
+	resp := &canonical.CanonicalResponse{
+		ID:      "msg-1",
+		Model:   "claude-3",
+		Content: []canonical.ContentBlock{canonical.NewTextBlock("ok")},
+	}
+
+	body, err := encodeResponse(resp)
+	require.NoError(t, err)
+
+	var result map[string]any
+	require.NoError(t, json.Unmarshal(body, &result))
+	assert.Equal(t, "end_turn", result["stop_reason"])
+}
+
+func TestDecodeRequest_MaxTokensZero(t *testing.T) {
+	body := []byte(`{
+		"model": "claude-3",
+		"max_tokens": 0,
+		"messages": [{"role": "user", "content": "hi"}]
+	}`)
+	req, err := decodeRequest(body)
+	require.NoError(t, err)
+	assert.Nil(t, req.Parameters.MaxTokens)
+}

backend/internal/conversion/canonical/types_test.go

@@ -0,0 +1,114 @@
+package canonical
+
+import (
+	"encoding/json"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestGetSystemString(t *testing.T) {
+	tests := []struct {
+		name    string
+		system  any
+		want    string
+	}{
+		{"string", "hello", "hello"},
+		{"nil", nil, ""},
+		{"empty string", "", ""},
+		{"system blocks", []SystemBlock{{Text: "part1"}, {Text: "part2"}}, "part1\n\npart2"},
+		{"single block", []SystemBlock{{Text: "only"}}, "only"},
+		{"other type", 123, "123"},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			req := &CanonicalRequest{System: tt.system}
+			assert.Equal(t, tt.want, req.GetSystemString())
+		})
+	}
+}
+
+func TestSetSystemString(t *testing.T) {
+	req := &CanonicalRequest{}
+
+	req.SetSystemString("hello")
+	assert.Equal(t, "hello", req.System)
+
+	req.SetSystemString("")
+	assert.Nil(t, req.System)
+}
+
+func TestNewTextBlock(t *testing.T) {
+	b := NewTextBlock("hello")
+	assert.Equal(t, "text", b.Type)
+	assert.Equal(t, "hello", b.Text)
+}
+
+func TestNewToolUseBlock(t *testing.T) {
+	input := json.RawMessage(`{"key":"val"}`)
+	b := NewToolUseBlock("id-1", "tool_name", input)
+	assert.Equal(t, "tool_use", b.Type)
+	assert.Equal(t, "id-1", b.ID)
+	assert.Equal(t, "tool_name", b.Name)
+	assert.Equal(t, input, b.Input)
+}
+
+func TestNewToolResultBlock(t *testing.T) {
+	b := NewToolResultBlock("tool-1", "result", false)
+	assert.Equal(t, "tool_result", b.Type)
+	assert.Equal(t, "tool-1", b.ToolUseID)
+	assert.NotNil(t, b.IsError)
+	assert.False(t, *b.IsError)
+}
+
+func TestNewThinkingBlock(t *testing.T) {
+	b := NewThinkingBlock("thought")
+	assert.Equal(t, "thinking", b.Type)
+	assert.Equal(t, "thought", b.Thinking)
+}
+
+func TestNewToolChoice(t *testing.T) {
+	assert.Equal(t, &ToolChoice{Type: "auto"}, NewToolChoiceAuto())
+	assert.Equal(t, &ToolChoice{Type: "none"}, NewToolChoiceNone())
+	assert.Equal(t, &ToolChoice{Type: "any"}, NewToolChoiceAny())
+	assert.Equal(t, &ToolChoice{Type: "tool", Name: "fn"}, NewToolChoiceNamed("fn"))
+}
+
+func TestCanonicalRequest_RoundTrip(t *testing.T) {
+	req := &CanonicalRequest{
+		Model:    "gpt-4",
+		System:   "system prompt",
+		Messages: []CanonicalMessage{{Role: RoleUser, Content: []ContentBlock{NewTextBlock("hi")}}},
+		Stream:   true,
+	}
+
+	data, err := json.Marshal(req)
+	require.NoError(t, err)
+
+	var decoded CanonicalRequest
+	require.NoError(t, json.Unmarshal(data, &decoded))
+	assert.Equal(t, "gpt-4", decoded.Model)
+	assert.Equal(t, "system prompt", decoded.System)
+	assert.True(t, decoded.Stream)
+}
+
+func TestCanonicalResponse_RoundTrip(t *testing.T) {
+	sr := StopReasonEndTurn
+	resp := &CanonicalResponse{
+		ID:      "resp-1",
+		Model:   "gpt-4",
+		Content: []ContentBlock{NewTextBlock("hello")},
+		StopReason: &sr,
+		Usage:   CanonicalUsage{InputTokens: 10, OutputTokens: 5},
+	}
+
+	data, err := json.Marshal(resp)
+	require.NoError(t, err)
+
+	var decoded CanonicalResponse
+	require.NoError(t, json.Unmarshal(data, &decoded))
+	assert.Equal(t, "resp-1", decoded.ID)
+	assert.Equal(t, StopReasonEndTurn, *decoded.StopReason)
+}

backend/internal/conversion/engine_supplemental_test.go

@@ -0,0 +1,323 @@
+package conversion
+
+import (
+	"encoding/json"
+	"errors"
+	"testing"
+
+	"nex/backend/internal/conversion/canonical"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestConversionError_WithProviderProtocol(t *testing.T) {
+	err := NewConversionError(ErrorCodeInvalidInput, "test").WithProviderProtocol("anthropic")
+	assert.Equal(t, "anthropic", err.ProviderProtocol)
+}
+
+func TestConversionError_WithInterfaceType(t *testing.T) {
+	err := NewConversionError(ErrorCodeInvalidInput, "test").WithInterfaceType("CHAT")
+	assert.Equal(t, "CHAT", err.InterfaceType)
+}
+
+func TestConversionError_FullBuilder(t *testing.T) {
+	err := NewConversionError(ErrorCodeInvalidInput, "bad").
+		WithClientProtocol("openai").
+		WithProviderProtocol("anthropic").
+		WithInterfaceType("CHAT").
+		WithDetail("field", "model").
+		WithCause(errors.New("root"))
+
+	assert.Equal(t, ErrorCodeInvalidInput, err.Code)
+	assert.Equal(t, "openai", err.ClientProtocol)
+	assert.Equal(t, "anthropic", err.ProviderProtocol)
+	assert.Equal(t, "CHAT", err.InterfaceType)
+	assert.Equal(t, "model", err.Details["field"])
+	assert.Equal(t, "root", err.Cause.Error())
+}
+
+func TestEngine_Use(t *testing.T) {
+	registry := NewMemoryRegistry()
+	engine := NewConversionEngine(registry)
+	called := false
+	engine.Use(&testMiddleware{fn: func(req *canonical.CanonicalRequest, cp, pp string, ctx *ConversionContext) (*canonical.CanonicalRequest, error) {
+		called = true
+		return req, nil
+	}})
+
+	clientAdapter := newMockAdapter("client", false)
+	clientAdapter.decodeReqFn = func(raw []byte) (*canonical.CanonicalRequest, error) {
+		return &canonical.CanonicalRequest{Model: "test"}, nil
+	}
+	providerAdapter := newMockAdapter("provider", false)
+	providerAdapter.encodeReqFn = func(req *canonical.CanonicalRequest, p *TargetProvider) ([]byte, error) {
+		return json.Marshal(req)
+	}
+	_ = engine.RegisterAdapter(clientAdapter)
+	_ = engine.RegisterAdapter(providerAdapter)
+
+	_, err := engine.ConvertHttpRequest(HTTPRequestSpec{
+		URL: "/v1/chat/completions", Method: "POST", Body: []byte(`{}`),
+	}, "client", "provider", NewTargetProvider("https://example.com", "key", "model"))
+	require.NoError(t, err)
+	assert.True(t, called)
+}
+
+func TestConvertHttpRequest_DecodeError(t *testing.T) {
+	registry := NewMemoryRegistry()
+	engine := NewConversionEngine(registry)
+	clientAdapter := newMockAdapter("client", false)
+	clientAdapter.decodeReqFn = func(raw []byte) (*canonical.CanonicalRequest, error) {
+		return nil, errors.New("decode failed")
+	}
+	_ = engine.RegisterAdapter(clientAdapter)
+	_ = engine.RegisterAdapter(newMockAdapter("provider", false))
+
+	_, err := engine.ConvertHttpRequest(HTTPRequestSpec{
+		URL: "/v1/chat/completions", Method: "POST", Body: []byte(`{}`),
+	}, "client", "provider", NewTargetProvider("", "", ""))
+	assert.Error(t, err)
+}
+
+func TestConvertHttpRequest_EncodeError(t *testing.T) {
+	registry := NewMemoryRegistry()
+	engine := NewConversionEngine(registry)
+	_ = engine.RegisterAdapter(newMockAdapter("client", false))
+	providerAdapter := newMockAdapter("provider", false)
+	providerAdapter.encodeReqFn = func(req *canonical.CanonicalRequest, p *TargetProvider) ([]byte, error) {
+		return nil, errors.New("encode failed")
+	}
+	_ = engine.RegisterAdapter(providerAdapter)
+
+	_, err := engine.ConvertHttpRequest(HTTPRequestSpec{
+		URL: "/v1/chat/completions", Method: "POST", Body: []byte(`{}`),
+	}, "client", "provider", NewTargetProvider("", "", ""))
+	assert.Error(t, err)
+}
+
+func TestConvertHttpResponse_CrossProtocol(t *testing.T) {
+	registry := NewMemoryRegistry()
+	engine := NewConversionEngine(registry)
+
+	clientAdapter := newMockAdapter("client", false)
+	clientAdapter.encodeRespFn = func(resp *canonical.CanonicalResponse) ([]byte, error) {
+		return json.Marshal(map[string]string{"id": resp.ID})
+	}
+	providerAdapter := newMockAdapter("provider", false)
+	providerAdapter.decodeRespFn = func(raw []byte) (*canonical.CanonicalResponse, error) {
+		return &canonical.CanonicalResponse{ID: "resp-1", Model: "test"}, nil
+	}
+	_ = engine.RegisterAdapter(clientAdapter)
+	_ = engine.RegisterAdapter(providerAdapter)
+
+	result, err := engine.ConvertHttpResponse(HTTPResponseSpec{
+		StatusCode: 200, Body: []byte(`{"id":"resp-1"}`),
+	}, "client", "provider", InterfaceTypeChat)
+	require.NoError(t, err)
+	assert.Equal(t, 200, result.StatusCode)
+	assert.Contains(t, string(result.Body), "resp-1")
+}
+
+func TestConvertHttpResponse_DecodeError(t *testing.T) {
+	registry := NewMemoryRegistry()
+	engine := NewConversionEngine(registry)
+	providerAdapter := newMockAdapter("provider", false)
+	providerAdapter.decodeRespFn = func(raw []byte) (*canonical.CanonicalResponse, error) {
+		return nil, errors.New("decode error")
+	}
+	_ = engine.RegisterAdapter(providerAdapter)
+	_ = engine.RegisterAdapter(newMockAdapter("client", false))
+
+	_, err := engine.ConvertHttpResponse(HTTPResponseSpec{Body: []byte(`{}`)}, "client", "provider", InterfaceTypeChat)
+	assert.Error(t, err)
+}
+
+func TestConvertHttpRequest_EmbeddingInterface(t *testing.T) {
+	registry := NewMemoryRegistry()
+	engine := NewConversionEngine(registry)
+
+	clientAdapter := newMockAdapter("client", false)
+	clientAdapter.ifaceType = InterfaceTypeEmbeddings
+	clientAdapter.supportsIface = map[InterfaceType]bool{InterfaceTypeEmbeddings: true}
+	clientAdapter.decodeReqFn = func(raw []byte) (*canonical.CanonicalRequest, error) {
+		return &canonical.CanonicalRequest{Model: "test"}, nil
+	}
+	providerAdapter := newMockAdapter("provider", false)
+	providerAdapter.ifaceType = InterfaceTypeEmbeddings
+	providerAdapter.supportsIface = map[InterfaceType]bool{InterfaceTypeEmbeddings: true}
+	_ = engine.RegisterAdapter(clientAdapter)
+	_ = engine.RegisterAdapter(providerAdapter)
+
+	result, err := engine.ConvertHttpRequest(HTTPRequestSpec{
+		URL: "/v1/embeddings", Method: "POST", Body: []byte(`{"model":"text-embedding","input":"hello"}`),
+	}, "client", "provider", NewTargetProvider("https://example.com", "key", "model"))
+	require.NoError(t, err)
+	assert.NotNil(t, result)
+}
+
+func TestConvertHttpRequest_RerankInterface(t *testing.T) {
+	registry := NewMemoryRegistry()
+	engine := NewConversionEngine(registry)
+
+	clientAdapter := newMockAdapter("client", false)
+	clientAdapter.ifaceType = InterfaceTypeRerank
+	clientAdapter.supportsIface = map[InterfaceType]bool{InterfaceTypeRerank: true}
+	providerAdapter := newMockAdapter("provider", false)
+	providerAdapter.ifaceType = InterfaceTypeRerank
+	providerAdapter.supportsIface = map[InterfaceType]bool{InterfaceTypeRerank: true}
+	_ = engine.RegisterAdapter(clientAdapter)
+	_ = engine.RegisterAdapter(providerAdapter)
+
+	result, err := engine.ConvertHttpRequest(HTTPRequestSpec{
+		URL: "/v1/rerank", Method: "POST", Body: []byte(`{"model":"rerank","query":"test","documents":["a"]}`),
+	}, "client", "provider", NewTargetProvider("https://example.com", "key", "model"))
+	require.NoError(t, err)
+	assert.NotNil(t, result)
+}
+
+func TestConvertHttpResponse_EmbeddingInterface(t *testing.T) {
+	registry := NewMemoryRegistry()
+	engine := NewConversionEngine(registry)
+
+	clientAdapter := newMockAdapter("client", false)
+	clientAdapter.supportsIface = map[InterfaceType]bool{InterfaceTypeEmbeddings: true}
+	providerAdapter := newMockAdapter("provider", false)
+	providerAdapter.supportsIface = map[InterfaceType]bool{InterfaceTypeEmbeddings: true}
+	_ = engine.RegisterAdapter(clientAdapter)
+	_ = engine.RegisterAdapter(providerAdapter)
+
+	result, err := engine.ConvertHttpResponse(HTTPResponseSpec{
+		StatusCode: 200, Body: []byte(`{"object":"list","data":[],"model":"test"}`),
+	}, "client", "provider", InterfaceTypeEmbeddings)
+	require.NoError(t, err)
+	assert.NotNil(t, result)
+}
+
+func TestConvertHttpResponse_RerankInterface(t *testing.T) {
+	registry := NewMemoryRegistry()
+	engine := NewConversionEngine(registry)
+
+	clientAdapter := newMockAdapter("client", false)
+	clientAdapter.supportsIface = map[InterfaceType]bool{InterfaceTypeRerank: true}
+	providerAdapter := newMockAdapter("provider", false)
+	providerAdapter.supportsIface = map[InterfaceType]bool{InterfaceTypeRerank: true}
+	_ = engine.RegisterAdapter(clientAdapter)
+	_ = engine.RegisterAdapter(providerAdapter)
+
+	result, err := engine.ConvertHttpResponse(HTTPResponseSpec{
+		StatusCode: 200, Body: []byte(`{"results":[],"model":"test"}`),
+	}, "client", "provider", InterfaceTypeRerank)
+	require.NoError(t, err)
+	assert.NotNil(t, result)
+}
+
+func TestConvertHttpRequest_ModelsInterface_Passthrough(t *testing.T) {
+	registry := NewMemoryRegistry()
+	engine := NewConversionEngine(registry)
+	clientAdapter := newMockAdapter("client", false)
+	clientAdapter.ifaceType = InterfaceTypeModels
+	providerAdapter := newMockAdapter("provider", false)
+	providerAdapter.ifaceType = InterfaceTypeModels
+	_ = engine.RegisterAdapter(clientAdapter)
+	_ = engine.RegisterAdapter(providerAdapter)
+
+	body := []byte(`{"object":"list","data":[]}`)
+	result, err := engine.ConvertHttpRequest(HTTPRequestSpec{
+		URL: "/v1/models", Method: "GET", Body: body,
+	}, "client", "provider", NewTargetProvider("https://example.com", "key", ""))
+	require.NoError(t, err)
+	assert.Equal(t, body, result.Body)
+}
+
+func TestConvertHttpResponse_ModelsInterface(t *testing.T) {
+	registry := NewMemoryRegistry()
+	engine := NewConversionEngine(registry)
+	clientAdapter := newMockAdapter("client", false)
+	clientAdapter.supportsIface = map[InterfaceType]bool{InterfaceTypeModels: true}
+	providerAdapter := newMockAdapter("provider", false)
+	providerAdapter.supportsIface = map[InterfaceType]bool{InterfaceTypeModels: true}
+	_ = engine.RegisterAdapter(clientAdapter)
+	_ = engine.RegisterAdapter(providerAdapter)
+
+	result, err := engine.ConvertHttpResponse(HTTPResponseSpec{
+		StatusCode: 200, Body: []byte(`{"object":"list","data":[]}`),
+	}, "client", "provider", InterfaceTypeModels)
+	require.NoError(t, err)
+	assert.NotNil(t, result)
+}
+
+func TestConvertHttpResponse_ModelInfoInterface(t *testing.T) {
+	registry := NewMemoryRegistry()
+	engine := NewConversionEngine(registry)
+	clientAdapter := newMockAdapter("client", false)
+	clientAdapter.supportsIface = map[InterfaceType]bool{InterfaceTypeModelInfo: true}
+	providerAdapter := newMockAdapter("provider", false)
+	providerAdapter.supportsIface = map[InterfaceType]bool{InterfaceTypeModelInfo: true}
+	_ = engine.RegisterAdapter(clientAdapter)
+	_ = engine.RegisterAdapter(providerAdapter)
+
+	result, err := engine.ConvertHttpResponse(HTTPResponseSpec{
+		StatusCode: 200, Body: []byte(`{"id":"gpt-4","object":"model"}`),
+	}, "client", "provider", InterfaceTypeModelInfo)
+	require.NoError(t, err)
+	assert.NotNil(t, result)
+}
+
+func TestRegistry_ListProtocols(t *testing.T) {
+	registry := NewMemoryRegistry()
+	_ = registry.Register(newMockAdapter("openai", true))
+	_ = registry.Register(newMockAdapter("anthropic", true))
+
+	protocols := registry.ListProtocols()
+	assert.Len(t, protocols, 2)
+	assert.Contains(t, protocols, "openai")
+	assert.Contains(t, protocols, "anthropic")
+}
+
+func TestRegistry_ConcurrentAccess(t *testing.T) {
+	registry := NewMemoryRegistry()
+	done := make(chan bool, 2)
+
+	go func() {
+		for i := 0; i < 100; i++ {
+			_ = registry.Register(newMockAdapter("proto-"+string(rune(i)), true))
+		}
+		done <- true
+	}()
+
+	go func() {
+		for i := 0; i < 100; i++ {
+			_, _ = registry.Get("proto-" + string(rune(i)))
+		}
+		_ = registry.ListProtocols()
+		done <- true
+	}()
+
+	<-done
+	<-done
+}
+
+func TestNewConversionContext(t *testing.T) {
+	ctx := NewConversionContext(InterfaceTypeChat)
+	assert.NotEmpty(t, ctx.ConversionID)
+	assert.Equal(t, InterfaceTypeChat, ctx.InterfaceType)
+	assert.NotNil(t, ctx.Metadata)
+}
+
+type testMiddleware struct {
+	fn func(req *canonical.CanonicalRequest, clientProtocol, providerProtocol string, ctx *ConversionContext) (*canonical.CanonicalRequest, error)
+}
+
+func (m *testMiddleware) Intercept(req *canonical.CanonicalRequest, clientProtocol, providerProtocol string, ctx *ConversionContext) (*canonical.CanonicalRequest, error) {
+	if m.fn != nil {
+		return m.fn(req, clientProtocol, providerProtocol, ctx)
+	}
+	return req, nil
+}
+
+func (m *testMiddleware) InterceptStreamEvent(event *canonical.CanonicalStreamEvent, clientProtocol, providerProtocol string, ctx *ConversionContext) (*canonical.CanonicalStreamEvent, error) {
+	return event, nil
+}
+
+var _ = json.Marshal

backend/internal/conversion/openai/stream_decoder_test.go

@@ -353,3 +353,120 @@ func TestStreamDecoder_MultipleChunks_Text(t *testing.T) {
 	}
 	assert.Equal(t, []string{"你好", "世界"}, deltas)
 }
+
+func TestStreamDecoder_UTF8Truncation(t *testing.T) {
+	d := NewStreamDecoder()
+
+	chunk := map[string]any{
+		"id":     "chatcmpl-utf8",
+		"model":  "gpt-4",
+		"choices": []any{
+			map[string]any{
+				"index": 0,
+				"delta": map[string]any{"content": "你"},
+			},
+		},
+	}
+	data, _ := json.Marshal(chunk)
+	sseData := []byte("data: " + string(data) + "\n\n")
+
+	mid := len(sseData) - 5
+	part1 := sseData[:mid]
+	part2 := sseData[mid:]
+
+	events1 := d.ProcessChunk(part1)
+	for _, e := range events1 {
+		if e.Type == canonical.EventContentBlockDelta && e.Delta != nil {
+			assert.Equal(t, "你", e.Delta.Text)
+		}
+	}
+
+	events2 := d.ProcessChunk(part2)
+	_ = events2
+}
+
+func TestStreamDecoder_ToolCallSubsequentDelta(t *testing.T) {
+	d := NewStreamDecoder()
+
+	idx := 0
+	chunk1 := map[string]any{
+		"id":     "chatcmpl-tc",
+		"model":  "gpt-4",
+		"choices": []any{
+			map[string]any{
+				"index": 0,
+				"delta": map[string]any{
+					"tool_calls": []any{
+						map[string]any{
+							"index": &idx,
+							"id":    "call_1",
+							"type":  "function",
+							"function": map[string]any{
+								"name":      "get_weather",
+								"arguments": "",
+							},
+						},
+					},
+				},
+			},
+		},
+	}
+	chunk2 := map[string]any{
+		"id":     "chatcmpl-tc",
+		"model":  "gpt-4",
+		"choices": []any{
+			map[string]any{
+				"index": 0,
+				"delta": map[string]any{
+					"tool_calls": []any{
+						map[string]any{
+							"index": &idx,
+							"function": map[string]any{
+								"arguments": "{\"city\":\"Beijing\"}",
+							},
+						},
+					},
+				},
+			},
+		},
+	}
+
+	events1 := d.ProcessChunk(makeChunkSSE(chunk1))
+	require.NotEmpty(t, events1)
+
+	events2 := d.ProcessChunk(makeChunkSSE(chunk2))
+	require.NotEmpty(t, events2)
+
+	foundInputJSON := false
+	for _, e := range events2 {
+		if e.Type == canonical.EventContentBlockDelta && e.Delta != nil && e.Delta.Type == "input_json_delta" {
+			foundInputJSON = true
+			assert.Equal(t, "{\"city\":\"Beijing\"}", e.Delta.PartialJSON)
+		}
+	}
+	assert.True(t, foundInputJSON, "subsequent tool call delta should emit input_json_delta")
+}
+
+func TestStreamDecoder_InvalidJSON(t *testing.T) {
+	d := NewStreamDecoder()
+
+	raw := []byte("data: {invalid json}\n\n")
+	events := d.ProcessChunk(raw)
+	assert.Nil(t, events)
+}
+
+func TestStreamDecoder_NonDataLines(t *testing.T) {
+	d := NewStreamDecoder()
+
+	raw := []byte(": comment line\ndata: {\"id\":\"1\",\"choices\":[{\"delta\":{\"content\":\"hi\"}}]}\n\n")
+	events := d.ProcessChunk(raw)
+	require.NotEmpty(t, events)
+	found := false
+	for _, e := range events {
+		if e.Type == canonical.EventContentBlockDelta && e.Delta != nil {
+			found = true
+			assert.Equal(t, "hi", e.Delta.Text)
+		}
+	}
+	assert.True(t, found)
+}

backend/internal/conversion/openai/stream_encoder.go

@@ -137,15 +137,10 @@ func (e *StreamEncoder) encodeInputJSONDelta(event canonical.CanonicalStreamEven
 	}
 
 	// 后续 delta，仅含 arguments
-	// 通过 index 查找 tool call
+	// 使用 canonical 事件中的 index 直接映射到 OpenAI tool_calls index
 	tcIdx := 0
 	if event.Index != nil {
-		for id, idx := range e.toolCallIndexMap {
-			if idx == tcIdx {
-				_ = id
-				break
-			}
-		}
+		tcIdx = *event.Index
 	}
 	delta := map[string]any{
 		"tool_calls": []map[string]any{{

backend/internal/conversion/openai/stream_encoder_test.go

@@ -170,3 +170,116 @@ func TestStreamEncoder_MessageDelta_WithUsage(t *testing.T) {
 	assert.Contains(t, s, "usage")
 	assert.Contains(t, s, "prompt_tokens")
 }
+
+func TestStreamEncoder_InputJSONDelta_SubsequentDelta(t *testing.T) {
+	e := NewStreamEncoder()
+
+	e.EncodeEvent(canonical.NewContentBlockStartEvent(0, canonical.StreamContentBlock{
+		Type: "tool_use",
+		ID:   "call_1",
+		Name: "get_weather",
+	}))
+
+	e.EncodeEvent(canonical.NewContentBlockDeltaEvent(0, canonical.StreamDelta{
+		Type:        string(canonical.DeltaTypeInputJSON),
+		PartialJSON: "{\"city\":",
+	}))
+
+	event := canonical.NewContentBlockDeltaEvent(0, canonical.StreamDelta{
+		Type:        string(canonical.DeltaTypeInputJSON),
+		PartialJSON: "\"Beijing\"}",
+	})
+
+	chunks := e.EncodeEvent(event)
+	require.NotEmpty(t, chunks)
+
+	s := string(chunks[0])
+	assert.Contains(t, s, "tool_calls")
+	assert.Contains(t, s, "Beijing")
+}
+
+func TestStreamEncoder_MessageStart_NilMessage(t *testing.T) {
+	e := NewStreamEncoder()
+	event := canonical.CanonicalStreamEvent{Type: canonical.EventMessageStart}
+
+	chunks := e.EncodeEvent(event)
+	require.Len(t, chunks, 1)
+
+	s := string(chunks[0])
+	assert.Contains(t, s, "chat.completion.chunk")
+}
+
+func TestStreamEncoder_UnknownEvent_ReturnsNil(t *testing.T) {
+	e := NewStreamEncoder()
+	event := canonical.CanonicalStreamEvent{Type: "unknown_type"}
+	chunks := e.EncodeEvent(event)
+	assert.Nil(t, chunks)
+}
+
+func TestStreamEncoder_ContentBlockDelta_NilDelta(t *testing.T) {
+	e := NewStreamEncoder()
+	event := canonical.CanonicalStreamEvent{Type: canonical.EventContentBlockDelta}
+	chunks := e.EncodeEvent(event)
+	assert.Nil(t, chunks)
+}
+
+func TestStreamEncoder_MultiToolCall_IndexMapping(t *testing.T) {
+	e := NewStreamEncoder()
+
+	e.EncodeEvent(canonical.NewContentBlockStartEvent(0, canonical.StreamContentBlock{
+		Type: "tool_use",
+		ID:   "call_1",
+		Name: "get_weather",
+	}))
+
+	firstDelta := canonical.NewContentBlockDeltaEvent(0, canonical.StreamDelta{
+		Type:        string(canonical.DeltaTypeInputJSON),
+		PartialJSON: `{"city":"北京"}`,
+	})
+	chunks := e.EncodeEvent(firstDelta)
+	require.NotEmpty(t, chunks)
+	s := string(chunks[0])
+	assert.Contains(t, s, `"index":0`)
+	assert.Contains(t, s, "get_weather")
+	assert.Contains(t, s, "北京")
+
+	e.EncodeEvent(canonical.NewContentBlockStopEvent(0))
+
+	e.EncodeEvent(canonical.NewContentBlockStartEvent(1, canonical.StreamContentBlock{
+		Type: "tool_use",
+		ID:   "call_2",
+		Name: "get_time",
+	}))
+
+	secondDelta := canonical.NewContentBlockDeltaEvent(1, canonical.StreamDelta{
+		Type:        string(canonical.DeltaTypeInputJSON),
+		PartialJSON: `{"tz":"Asia/Shanghai"}`,
+	})
+	chunks = e.EncodeEvent(secondDelta)
+	require.NotEmpty(t, chunks)
+	s = string(chunks[0])
+	assert.Contains(t, s, `"index":1`)
+	assert.Contains(t, s, "get_time")
+	assert.Contains(t, s, "Asia/Shanghai")
+
+	subsequentDelta0 := canonical.NewContentBlockDeltaEvent(0, canonical.StreamDelta{
+		Type:        string(canonical.DeltaTypeInputJSON),
+		PartialJSON: `"more_data"`,
+	})
+	chunks = e.EncodeEvent(subsequentDelta0)
+	require.NotEmpty(t, chunks)
+	s = string(chunks[0])
+	assert.Contains(t, s, `"index":0`)
+	assert.NotContains(t, s, "get_weather")
+	assert.Contains(t, s, "more_data")
+
+	subsequentDelta1 := canonical.NewContentBlockDeltaEvent(1, canonical.StreamDelta{
+		Type:        string(canonical.DeltaTypeInputJSON),
+		PartialJSON: `"more_time"`,
+	})
+	chunks = e.EncodeEvent(subsequentDelta1)
+	require.NotEmpty(t, chunks)
+	s = string(chunks[0])
+	assert.Contains(t, s, `"index":1`)
+	assert.Contains(t, s, "more_time")
+}

backend/internal/conversion/openai/supplemental_test.go

@@ -0,0 +1,434 @@
+package openai
+
+import (
+	"encoding/json"
+	"testing"
+
+	"nex/backend/internal/conversion"
+	"nex/backend/internal/conversion/canonical"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestDecodeEmbeddingRequest(t *testing.T) {
+	body := []byte(`{"model":"text-embedding-3-small","input":"hello world","encoding_format":"float","dimensions":256}`)
+	req, err := decodeEmbeddingRequest(body)
+	require.NoError(t, err)
+	assert.Equal(t, "text-embedding-3-small", req.Model)
+	assert.Equal(t, "hello world", req.Input)
+	assert.Equal(t, "float", req.EncodingFormat)
+	require.NotNil(t, req.Dimensions)
+	assert.Equal(t, 256, *req.Dimensions)
+}
+
+func TestDecodeEmbeddingRequest_ArrayInput(t *testing.T) {
+	body := []byte(`{"model":"text-embedding","input":["hello","world"]}`)
+	req, err := decodeEmbeddingRequest(body)
+	require.NoError(t, err)
+	assert.Equal(t, "text-embedding", req.Model)
+	inputArr, ok := req.Input.([]any)
+	require.True(t, ok)
+	assert.Len(t, inputArr, 2)
+}
+
+func TestDecodeEmbeddingRequest_InvalidJSON(t *testing.T) {
+	_, err := decodeEmbeddingRequest([]byte(`invalid`))
+	assert.Error(t, err)
+}
+
+func TestDecodeEmbeddingResponse(t *testing.T) {
+	body := []byte(`{
+		"object": "list",
+		"data": [{"index": 0, "embedding": [0.1, 0.2, 0.3]}],
+		"model": "text-embedding-3-small",
+		"usage": {"prompt_tokens": 5, "total_tokens": 5}
+	}`)
+	resp, err := decodeEmbeddingResponse(body)
+	require.NoError(t, err)
+	assert.Equal(t, "text-embedding-3-small", resp.Model)
+	assert.Len(t, resp.Data, 1)
+	assert.Equal(t, 0, resp.Data[0].Index)
+	assert.Equal(t, 5, resp.Usage.PromptTokens)
+}
+
+func TestDecodeRerankRequest(t *testing.T) {
+	topN := 3
+	returnDocs := true
+	body := []byte(`{"model":"rerank-1","query":"what is AI","documents":["doc1","doc2"],"top_n":3,"return_documents":true}`)
+	req, err := decodeRerankRequest(body)
+	require.NoError(t, err)
+	assert.Equal(t, "rerank-1", req.Model)
+	assert.Equal(t, "what is AI", req.Query)
+	assert.Equal(t, []string{"doc1", "doc2"}, req.Documents)
+	require.NotNil(t, req.TopN)
+	assert.Equal(t, topN, *req.TopN)
+	require.NotNil(t, req.ReturnDocuments)
+	assert.Equal(t, returnDocs, *req.ReturnDocuments)
+}
+
+func TestDecodeRerankResponse(t *testing.T) {
+	doc := "relevant doc"
+	body := []byte(`{
+		"results": [{"index": 0, "relevance_score": 0.95, "document": "relevant doc"}],
+		"model": "rerank-1"
+	}`)
+	resp, err := decodeRerankResponse(body)
+	require.NoError(t, err)
+	assert.Equal(t, "rerank-1", resp.Model)
+	assert.Len(t, resp.Results, 1)
+	assert.Equal(t, 0, resp.Results[0].Index)
+	assert.InDelta(t, 0.95, resp.Results[0].RelevanceScore, 0.001)
+	require.NotNil(t, resp.Results[0].Document)
+	assert.Equal(t, doc, *resp.Results[0].Document)
+}
+
+func TestDecodeModelInfoResponse(t *testing.T) {
+	body := []byte(`{"id":"gpt-4","object":"model","created":1700000000,"owned_by":"openai"}`)
+	info, err := decodeModelInfoResponse(body)
+	require.NoError(t, err)
+	assert.Equal(t, "gpt-4", info.ID)
+	assert.Equal(t, int64(1700000000), info.Created)
+	assert.Equal(t, "openai", info.OwnedBy)
+}
+
+func TestEncodeEmbeddingRequest(t *testing.T) {
+	req := &canonical.CanonicalEmbeddingRequest{
+		Model:          "text-embedding-3-small",
+		Input:          "hello",
+		EncodingFormat: "float",
+	}
+	provider := conversion.NewTargetProvider("", "key", "my-embedding-model")
+
+	body, err := encodeEmbeddingRequest(req, provider)
+	require.NoError(t, err)
+
+	var result map[string]any
+	require.NoError(t, json.Unmarshal(body, &result))
+	assert.Equal(t, "my-embedding-model", result["model"])
+	assert.Equal(t, "hello", result["input"])
+	assert.Equal(t, "float", result["encoding_format"])
+}
+
+func TestEncodeEmbeddingRequest_WithDimensions(t *testing.T) {
+	dims := 256
+	req := &canonical.CanonicalEmbeddingRequest{
+		Model:      "text-embedding",
+		Input:      "test",
+		Dimensions: &dims,
+	}
+	provider := conversion.NewTargetProvider("", "key", "model")
+
+	body, err := encodeEmbeddingRequest(req, provider)
+	require.NoError(t, err)
+
+	var result map[string]any
+	require.NoError(t, json.Unmarshal(body, &result))
+	assert.Equal(t, float64(256), result["dimensions"])
+}
+
+func TestEncodeEmbeddingResponse(t *testing.T) {
+	resp := &canonical.CanonicalEmbeddingResponse{
+		Data:  []canonical.EmbeddingData{{Index: 0, Embedding: []float64{0.1, 0.2}}},
+		Model: "text-embedding",
+		Usage: canonical.EmbeddingUsage{PromptTokens: 3, TotalTokens: 3},
+	}
+
+	body, err := encodeEmbeddingResponse(resp)
+	require.NoError(t, err)
+
+	var result map[string]any
+	require.NoError(t, json.Unmarshal(body, &result))
+	assert.Equal(t, "list", result["object"])
+	assert.Equal(t, "text-embedding", result["model"])
+}
+
+func TestEncodeRerankRequest(t *testing.T) {
+	topN := 5
+	req := &canonical.CanonicalRerankRequest{
+		Model:     "rerank-1",
+		Query:     "what is AI",
+		Documents: []string{"doc1", "doc2"},
+		TopN:      &topN,
+	}
+	provider := conversion.NewTargetProvider("", "key", "my-rerank-model")
+
+	body, err := encodeRerankRequest(req, provider)
+	require.NoError(t, err)
+
+	var result map[string]any
+	require.NoError(t, json.Unmarshal(body, &result))
+	assert.Equal(t, "my-rerank-model", result["model"])
+	assert.Equal(t, "what is AI", result["query"])
+}
+
+func TestEncodeRerankResponse(t *testing.T) {
+	doc := "relevant passage"
+	resp := &canonical.CanonicalRerankResponse{
+		Results: []canonical.RerankResult{
+			{Index: 0, RelevanceScore: 0.95, Document: &doc},
+		},
+		Model: "rerank-1",
+	}
+
+	body, err := encodeRerankResponse(resp)
+	require.NoError(t, err)
+
+	var result map[string]any
+	require.NoError(t, json.Unmarshal(body, &result))
+	assert.Equal(t, "rerank-1", result["model"])
+	results := result["results"].([]any)
+	assert.Len(t, results, 1)
+}
+
+func TestEncodeModelInfoResponse(t *testing.T) {
+	info := &canonical.CanonicalModelInfo{
+		ID:      "gpt-4",
+		Name:    "GPT-4",
+		Created: 1700000000,
+		OwnedBy: "openai",
+	}
+
+	body, err := encodeModelInfoResponse(info)
+	require.NoError(t, err)
+
+	var result map[string]any
+	require.NoError(t, json.Unmarshal(body, &result))
+	assert.Equal(t, "gpt-4", result["id"])
+	assert.Equal(t, "model", result["object"])
+}
+
+func TestDecodeEmbeddingResponse_InvalidJSON(t *testing.T) {
+	_, err := decodeEmbeddingResponse([]byte(`invalid`))
+	assert.Error(t, err)
+}
+
+func TestDecodeRerankRequest_InvalidJSON(t *testing.T) {
+	_, err := decodeRerankRequest([]byte(`invalid`))
+	assert.Error(t, err)
+}
+
+func TestDecodeRerankResponse_InvalidJSON(t *testing.T) {
+	_, err := decodeRerankResponse([]byte(`invalid`))
+	assert.Error(t, err)
+}
+
+func TestDecodeModelInfoResponse_InvalidJSON(t *testing.T) {
+	_, err := decodeModelInfoResponse([]byte(`invalid`))
+	assert.Error(t, err)
+}
+
+func TestDecodeRequest_ThinkingNone(t *testing.T) {
+	body := []byte(`{"model":"gpt-4","messages":[{"role":"user","content":"hi"}],"reasoning_effort":"none"}`)
+	req, err := decodeRequest(body)
+	require.NoError(t, err)
+	require.NotNil(t, req.Thinking)
+	assert.Equal(t, "disabled", req.Thinking.Type)
+}
+
+func TestDecodeRequest_ThinkingMinimal(t *testing.T) {
+	body := []byte(`{"model":"gpt-4","messages":[{"role":"user","content":"hi"}],"reasoning_effort":"minimal"}`)
+	req, err := decodeRequest(body)
+	require.NoError(t, err)
+	require.NotNil(t, req.Thinking)
+	assert.Equal(t, "enabled", req.Thinking.Type)
+	assert.Equal(t, "low", req.Thinking.Effort)
+}
+
+func TestDecodeRequest_OutputFormat_Text(t *testing.T) {
+	body := []byte(`{"model":"gpt-4","messages":[{"role":"user","content":"hi"}],"response_format":{"type":"text"}}`)
+	req, err := decodeRequest(body)
+	require.NoError(t, err)
+	assert.Nil(t, req.OutputFormat)
+}
+
+func TestDecodeRequest_DeprecatedFunctionCall(t *testing.T) {
+	body := []byte(`{"model":"gpt-4","messages":[{"role":"user","content":"hi"}],"function_call":"auto","functions":[{"name":"fn1","parameters":{}}]}`)
+	req, err := decodeRequest(body)
+	require.NoError(t, err)
+	assert.Equal(t, "auto", req.ToolChoice.Type)
+	assert.Len(t, req.Tools, 1)
+}
+
+func TestDecodeRequest_FunctionMessage(t *testing.T) {
+	body := []byte(`{
+		"model": "gpt-4",
+		"messages": [
+			{"role": "user", "content": "hi"},
+			{"role": "function", "name": "get_weather", "content": "sunny"}
+		]
+	}`)
+	req, err := decodeRequest(body)
+	require.NoError(t, err)
+	assert.Len(t, req.Messages, 2)
+	assert.Equal(t, canonical.RoleTool, req.Messages[1].Role)
+}
+
+func TestDecodeRequest_StopString(t *testing.T) {
+	body := []byte(`{"model":"gpt-4","messages":[{"role":"user","content":"hi"}],"stop":"END"}`)
+	req, err := decodeRequest(body)
+	require.NoError(t, err)
+	assert.Equal(t, []string{"END"}, req.Parameters.StopSequences)
+}
+
+func TestDecodeRequest_StopEmptyString(t *testing.T) {
+	body := []byte(`{"model":"gpt-4","messages":[{"role":"user","content":"hi"}],"stop":""}`)
+	req, err := decodeRequest(body)
+	require.NoError(t, err)
+	assert.Nil(t, req.Parameters.StopSequences)
+}
+
+func TestDecodeResponse_EmptyChoices(t *testing.T) {
+	body := []byte(`{"id":"resp-1","model":"gpt-4","choices":[],"usage":{"prompt_tokens":0,"completion_tokens":0,"total_tokens":0}}`)
+	resp, err := decodeResponse(body)
+	require.NoError(t, err)
+	assert.Equal(t, "resp-1", resp.ID)
+	assert.Len(t, resp.Content, 1)
+	assert.Equal(t, "", resp.Content[0].Text)
+}
+
+func TestDecodeResponse_FunctionCallFinishReason(t *testing.T) {
+	body := []byte(`{
+		"id":"r1","model":"gpt-4",
+		"choices":[{"index":0,"message":{"role":"assistant","content":"ok"},"finish_reason":"function_call"}],
+		"usage":{"prompt_tokens":1,"completion_tokens":1,"total_tokens":2}
+	}`)
+	resp, err := decodeResponse(body)
+	require.NoError(t, err)
+	assert.Equal(t, canonical.StopReasonToolUse, *resp.StopReason)
+}
+
+func TestEncodeRequest_DisabledThinking(t *testing.T) {
+	req := &canonical.CanonicalRequest{
+		Model:    "gpt-4",
+		Messages: []canonical.CanonicalMessage{{Role: canonical.RoleUser, Content: []canonical.ContentBlock{canonical.NewTextBlock("hi")}}},
+		Thinking: &canonical.ThinkingConfig{Type: "disabled"},
+	}
+	provider := conversion.NewTargetProvider("", "key", "model")
+
+	body, err := encodeRequest(req, provider)
+	require.NoError(t, err)
+
+	var result map[string]any
+	require.NoError(t, json.Unmarshal(body, &result))
+	assert.Equal(t, "none", result["reasoning_effort"])
+}
+
+func TestEncodeRequest_OutputFormat_JSONObject(t *testing.T) {
+	req := &canonical.CanonicalRequest{
+		Model:        "gpt-4",
+		Messages:     []canonical.CanonicalMessage{{Role: canonical.RoleUser, Content: []canonical.ContentBlock{canonical.NewTextBlock("hi")}}},
+		OutputFormat: &canonical.OutputFormat{Type: "json_object"},
+	}
+	provider := conversion.NewTargetProvider("", "key", "model")
+
+	body, err := encodeRequest(req, provider)
+	require.NoError(t, err)
+
+	var result map[string]any
+	require.NoError(t, json.Unmarshal(body, &result))
+	rf, ok := result["response_format"].(map[string]any)
+	require.True(t, ok)
+	assert.Equal(t, "json_object", rf["type"])
+}
+
+func TestEncodeRequest_PublicFields(t *testing.T) {
+	parallel := true
+	req := &canonical.CanonicalRequest{
+		Model:           "gpt-4",
+		Messages:        []canonical.CanonicalMessage{{Role: canonical.RoleUser, Content: []canonical.ContentBlock{canonical.NewTextBlock("hi")}}},
+		UserID:          "user-123",
+		ParallelToolUse: &parallel,
+	}
+	provider := conversion.NewTargetProvider("", "key", "model")
+
+	body, err := encodeRequest(req, provider)
+	require.NoError(t, err)
+
+	var result map[string]any
+	require.NoError(t, json.Unmarshal(body, &result))
+	assert.Equal(t, "user-123", result["user"])
+	assert.Equal(t, true, result["parallel_tool_calls"])
+}
+
+func TestEncodeResponse_UsageWithCacheAndReasoning(t *testing.T) {
+	cache := 80
+	reasoning := 20
+	sr := canonical.StopReasonEndTurn
+	resp := &canonical.CanonicalResponse{
+		ID:      "r1",
+		Model:   "gpt-4",
+		Content: []canonical.ContentBlock{canonical.NewTextBlock("ok")},
+		StopReason: &sr,
+		Usage: canonical.CanonicalUsage{
+			InputTokens:     100,
+			OutputTokens:    50,
+			CacheReadTokens: &cache,
+			ReasoningTokens: &reasoning,
+		},
+	}
+
+	body, err := encodeResponse(resp)
+	require.NoError(t, err)
+
+	var result map[string]any
+	require.NoError(t, json.Unmarshal(body, &result))
+	usage := result["usage"].(map[string]any)
+	assert.Equal(t, float64(100), usage["prompt_tokens"])
+	ptd, ok := usage["prompt_tokens_details"].(map[string]any)
+	require.True(t, ok)
+	assert.Equal(t, float64(80), ptd["cached_tokens"])
+	ctd, ok := usage["completion_tokens_details"].(map[string]any)
+	require.True(t, ok)
+	assert.Equal(t, float64(20), ctd["reasoning_tokens"])
+}
+
+func TestEncodeResponse_StopReasons(t *testing.T) {
+	tests := []struct {
+		name       string
+		stopReason canonical.StopReason
+		want       string
+	}{
+		{"end_turn→stop", canonical.StopReasonEndTurn, "stop"},
+		{"max_tokens→length", canonical.StopReasonMaxTokens, "length"},
+		{"tool_use→tool_calls", canonical.StopReasonToolUse, "tool_calls"},
+		{"content_filter→content_filter", canonical.StopReasonContentFilter, "content_filter"},
+		{"stop_sequence→stop", canonical.StopReasonStopSequence, "stop"},
+		{"refusal→stop", canonical.StopReasonRefusal, "stop"},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			sr := tt.stopReason
+			resp := &canonical.CanonicalResponse{
+				ID:         "r1",
+				Model:      "gpt-4",
+				Content:    []canonical.ContentBlock{canonical.NewTextBlock("ok")},
+				StopReason: &sr,
+				Usage:      canonical.CanonicalUsage{},
+			}
+			body, err := encodeResponse(resp)
+			require.NoError(t, err)
+
+			var result map[string]any
+			require.NoError(t, json.Unmarshal(body, &result))
+			choices := result["choices"].([]any)
+			choice := choices[0].(map[string]any)
+			assert.Equal(t, tt.want, choice["finish_reason"])
+		})
+	}
+}
+
+func TestMapErrorCode_AllCodes(t *testing.T) {
+	assert.Equal(t, "invalid_request_error", mapErrorCode(conversion.ErrorCodeInvalidInput))
+	assert.Equal(t, "invalid_request_error", mapErrorCode(conversion.ErrorCodeMissingRequiredField))
+	assert.Equal(t, "invalid_request_error", mapErrorCode(conversion.ErrorCodeIncompatibleFeature))
+	assert.Equal(t, "invalid_request_error", mapErrorCode(conversion.ErrorCodeFieldMappingFailure))
+	assert.Equal(t, "invalid_request_error", mapErrorCode(conversion.ErrorCodeToolCallParseError))
+	assert.Equal(t, "invalid_request_error", mapErrorCode(conversion.ErrorCodeJSONParseError))
+	assert.Equal(t, "invalid_request_error", mapErrorCode(conversion.ErrorCodeProtocolConstraint))
+	assert.Equal(t, "server_error", mapErrorCode(conversion.ErrorCodeStreamStateError))
+	assert.Equal(t, "server_error", mapErrorCode(conversion.ErrorCodeUTF8DecodeError))
+	assert.Equal(t, "server_error", mapErrorCode(conversion.ErrorCodeEncodingFailure))
+	assert.Equal(t, "server_error", mapErrorCode(conversion.ErrorCodeInterfaceNotSupported))
+}

backend/internal/conversion/stream_test.go

@@ -1,6 +1,7 @@
 package conversion
 
 import (
+	"fmt"
 	"testing"
 
 	"nex/backend/internal/conversion/canonical"
@@ -128,3 +129,71 @@ func TestCanonicalStreamConverter_EmptyDecoder(t *testing.T) {
 
 	assert.Nil(t, result)
 }
+
+func TestCanonicalStreamConverter_MiddlewareError_Continue(t *testing.T) {
+	event := canonical.NewMessageStartEvent("id-1", "gpt-4")
+	decoder := &mockStreamDecoder{
+		chunks: [][]canonical.CanonicalStreamEvent{{event}},
+	}
+	encoder := &mockStreamEncoder{
+		events: [][]byte{[]byte("data: ok\n\n")},
+	}
+
+	chain := NewMiddlewareChain()
+	chain.Use(&errorMiddleware{})
+	ctx := NewConversionContext(InterfaceTypeChat)
+
+	converter := NewCanonicalStreamConverterWithMiddleware(decoder, encoder, chain, *ctx, "openai", "anthropic")
+	result := converter.ProcessChunk([]byte("raw"))
+
+	assert.Nil(t, result, "middleware error should cause the event to be skipped (continue)")
+}
+
+func TestCanonicalStreamConverter_Flush_MiddlewareError_Continue(t *testing.T) {
+	event := canonical.NewMessageStartEvent("id-1", "gpt-4")
+	decoder := &mockStreamDecoder{
+		flush: []canonical.CanonicalStreamEvent{event},
+	}
+	encoder := &mockStreamEncoder{
+		events: [][]byte{[]byte("data: ok\n\n")},
+		flush:  [][]byte{[]byte("data: encoder_flush\n\n")},
+	}
+
+	chain := NewMiddlewareChain()
+	chain.Use(&errorMiddleware{})
+	ctx := NewConversionContext(InterfaceTypeChat)
+
+	converter := NewCanonicalStreamConverterWithMiddleware(decoder, encoder, chain, *ctx, "openai", "anthropic")
+	result := converter.Flush()
+
+	assert.Len(t, result, 1)
+	assert.Equal(t, []byte("data: encoder_flush\n\n"), result[0])
+}
+
+func TestCanonicalStreamConverter_Flush_DecoderAndEncoderBothProduce(t *testing.T) {
+	event := canonical.NewMessageStartEvent("id-1", "gpt-4")
+	decoder := &mockStreamDecoder{
+		flush: []canonical.CanonicalStreamEvent{event},
+	}
+	encoder := &mockStreamEncoder{
+		events: [][]byte{[]byte("data: decoder_flush\n\n")},
+		flush:  [][]byte{[]byte("data: encoder_flush\n\n")},
+	}
+
+	converter := NewCanonicalStreamConverter(decoder, encoder)
+	result := converter.Flush()
+
+	assert.Len(t, result, 2)
+	assert.Equal(t, []byte("data: decoder_flush\n\n"), result[0])
+	assert.Equal(t, []byte("data: encoder_flush\n\n"), result[1])
+}
+
+type errorMiddleware struct{}
+
+func (m *errorMiddleware) Intercept(req *canonical.CanonicalRequest, clientProtocol, providerProtocol string, ctx *ConversionContext) (*canonical.CanonicalRequest, error) {
+	return nil, fmt.Errorf("middleware error")
+}
+
+func (m *errorMiddleware) InterceptStreamEvent(event *canonical.CanonicalStreamEvent, clientProtocol, providerProtocol string, ctx *ConversionContext) (*canonical.CanonicalStreamEvent, error) {
+	return nil, fmt.Errorf("stream middleware error")
+}

backend/internal/handler/handler_supplemental_test.go

@@ -0,0 +1,165 @@
+package handler
+
+import (
+	"bytes"
+	"encoding/json"
+	"net/http/httptest"
+	"testing"
+
+	"github.com/gin-gonic/gin"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+
+	"nex/backend/internal/domain"
+)
+
+func TestProviderHandler_CreateProvider_Success(t *testing.T) {
+	h := NewProviderHandler(&mockProviderService{})
+
+	body, _ := json.Marshal(map[string]string{
+		"id":      "p1",
+		"name":    "Test",
+		"api_key": "sk-test",
+		"base_url": "https://api.test.com",
+	})
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Request = httptest.NewRequest("POST", "/api/providers", bytes.NewReader(body))
+	c.Request.Header.Set("Content-Type", "application/json")
+
+	h.CreateProvider(c)
+	assert.Equal(t, 201, w.Code)
+
+	var result domain.Provider
+	require.NoError(t, json.Unmarshal(w.Body.Bytes(), &result))
+	assert.Equal(t, "p1", result.ID)
+	assert.Contains(t, result.APIKey, "***")
+}
+
+func TestProviderHandler_CreateProvider_WithProtocol(t *testing.T) {
+	h := NewProviderHandler(&mockProviderService{})
+
+	body, _ := json.Marshal(map[string]string{
+		"id":       "p1",
+		"name":     "Test",
+		"api_key":  "sk-test",
+		"base_url": "https://api.test.com",
+		"protocol": "anthropic",
+	})
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Request = httptest.NewRequest("POST", "/api/providers", bytes.NewReader(body))
+	c.Request.Header.Set("Content-Type", "application/json")
+
+	h.CreateProvider(c)
+	assert.Equal(t, 201, w.Code)
+}
+
+func TestProviderHandler_UpdateProvider(t *testing.T) {
+	h := NewProviderHandler(&mockProviderService{
+		provider: &domain.Provider{ID: "p1", Name: "Updated", APIKey: "***"},
+	})
+
+	body, _ := json.Marshal(map[string]string{"name": "Updated"})
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Params = gin.Params{{Key: "id", Value: "p1"}}
+	c.Request = httptest.NewRequest("PUT", "/api/providers/p1", bytes.NewReader(body))
+	c.Request.Header.Set("Content-Type", "application/json")
+
+	h.UpdateProvider(c)
+	assert.Equal(t, 200, w.Code)
+}
+
+func TestProviderHandler_UpdateProvider_InvalidBody(t *testing.T) {
+	h := NewProviderHandler(&mockProviderService{})
+
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Params = gin.Params{{Key: "id", Value: "p1"}}
+	c.Request = httptest.NewRequest("PUT", "/api/providers/p1", nil)
+
+	h.UpdateProvider(c)
+	assert.Equal(t, 400, w.Code)
+}
+
+func TestProviderHandler_DeleteProvider(t *testing.T) {
+	h := NewProviderHandler(&mockProviderService{})
+
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Params = gin.Params{{Key: "id", Value: "p1"}}
+	c.Request = httptest.NewRequest("DELETE", "/api/providers/p1", bytes.NewReader([]byte{}))
+	c.Request.Header.Set("Content-Type", "application/json")
+
+	h.DeleteProvider(c)
+	assert.True(t, w.Code == 204 || w.Code == 200)
+}
+
+func TestModelHandler_DeleteModel(t *testing.T) {
+	h := NewModelHandler(&mockModelService{})
+
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Params = gin.Params{{Key: "id", Value: "m1"}}
+	c.Request = httptest.NewRequest("DELETE", "/api/models/m1", bytes.NewReader([]byte{}))
+	c.Request.Header.Set("Content-Type", "application/json")
+
+	h.DeleteModel(c)
+	assert.True(t, w.Code == 204 || w.Code == 200)
+}
+
+func TestModelHandler_CreateModel_Success(t *testing.T) {
+	h := NewModelHandler(&mockModelService{})
+
+	body, _ := json.Marshal(map[string]string{
+		"id":          "m1",
+		"provider_id": "p1",
+		"model_name":  "gpt-4",
+	})
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Request = httptest.NewRequest("POST", "/api/models", bytes.NewReader(body))
+	c.Request.Header.Set("Content-Type", "application/json")
+
+	h.CreateModel(c)
+	assert.Equal(t, 201, w.Code)
+
+	var result domain.Model
+	require.NoError(t, json.Unmarshal(w.Body.Bytes(), &result))
+	assert.Equal(t, "m1", result.ID)
+}
+
+func TestModelHandler_GetModel(t *testing.T) {
+	h := NewModelHandler(&mockModelService{
+		model: &domain.Model{ID: "m1", ModelName: "gpt-4"},
+	})
+
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Params = gin.Params{{Key: "id", Value: "m1"}}
+	c.Request = httptest.NewRequest("GET", "/api/models/m1", nil)
+
+	h.GetModel(c)
+	assert.Equal(t, 200, w.Code)
+
+	var result domain.Model
+	require.NoError(t, json.Unmarshal(w.Body.Bytes(), &result))
+	assert.Equal(t, "gpt-4", result.ModelName)
+}
+
+func TestModelHandler_UpdateModel(t *testing.T) {
+	h := NewModelHandler(&mockModelService{
+		model: &domain.Model{ID: "m1", ModelName: "gpt-4o"},
+	})
+
+	body, _ := json.Marshal(map[string]string{"model_name": "gpt-4o"})
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Params = gin.Params{{Key: "id", Value: "m1"}}
+	c.Request = httptest.NewRequest("PUT", "/api/models/m1", bytes.NewReader(body))
+	c.Request.Header.Set("Content-Type", "application/json")
+
+	h.UpdateModel(c)
+	assert.Equal(t, 200, w.Code)
+}

backend/internal/handler/proxy_handler_test.go

@@ -0,0 +1,761 @@
+package handler
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"fmt"
+	"net/http/httptest"
+	"testing"
+	"time"
+
+	"github.com/gin-gonic/gin"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+
+	"nex/backend/internal/conversion"
+	"nex/backend/internal/conversion/anthropic"
+	"nex/backend/internal/conversion/openai"
+	"nex/backend/internal/domain"
+	"nex/backend/internal/provider"
+	appErrors "nex/backend/pkg/errors"
+)
+
+func init() {
+	gin.SetMode(gin.TestMode)
+}
+
+type mockProxyProviderClient struct {
+	sendFn      func(ctx context.Context, spec conversion.HTTPRequestSpec) (*conversion.HTTPResponseSpec, error)
+	sendStreamFn func(ctx context.Context, spec conversion.HTTPRequestSpec) (<-chan provider.StreamEvent, error)
+}
+
+func (m *mockProxyProviderClient) Send(ctx context.Context, spec conversion.HTTPRequestSpec) (*conversion.HTTPResponseSpec, error) {
+	if m.sendFn != nil {
+		return m.sendFn(ctx, spec)
+	}
+	return &conversion.HTTPResponseSpec{
+		StatusCode: 200,
+		Headers:    map[string]string{"Content-Type": "application/json"},
+		Body:       []byte(`{"id":"resp-1","model":"gpt-4","choices":[{"index":0,"message":{"role":"assistant","content":"hi"},"finish_reason":"stop"}],"usage":{"prompt_tokens":1,"completion_tokens":1,"total_tokens":2}}`),
+	}, nil
+}
+
+func (m *mockProxyProviderClient) SendStream(ctx context.Context, spec conversion.HTTPRequestSpec) (<-chan provider.StreamEvent, error) {
+	if m.sendStreamFn != nil {
+		return m.sendStreamFn(ctx, spec)
+	}
+	ch := make(chan provider.StreamEvent, 10)
+	go func() {
+		defer close(ch)
+		ch <- provider.StreamEvent{Data: []byte("data: {\"id\":\"1\",\"choices\":[{\"delta\":{\"content\":\"hi\"}}]}\n\n")}
+		ch <- provider.StreamEvent{Data: []byte("data: [DONE]\n\n")}
+		ch <- provider.StreamEvent{Done: true}
+	}()
+	return ch, nil
+}
+
+type mockProxyRoutingService struct {
+	result *domain.RouteResult
+	err    error
+}
+
+func (m *mockProxyRoutingService) Route(modelName string) (*domain.RouteResult, error) {
+	return m.result, m.err
+}
+
+type mockProxyProviderService struct {
+	providers []domain.Provider
+	err       error
+}
+
+func (m *mockProxyProviderService) Create(p *domain.Provider) error                              { return nil }
+func (m *mockProxyProviderService) Get(id string, maskKey bool) (*domain.Provider, error)        { return nil, nil }
+func (m *mockProxyProviderService) List() ([]domain.Provider, error)                             { return m.providers, m.err }
+func (m *mockProxyProviderService) Update(id string, updates map[string]interface{}) error        { return nil }
+func (m *mockProxyProviderService) Delete(id string) error                                       { return nil }
+
+type mockProxyStatsService struct{}
+
+func (m *mockProxyStatsService) Record(providerID, modelName string) error                                         { return nil }
+func (m *mockProxyStatsService) Get(providerID, modelName string, startDate, endDate *time.Time) ([]domain.UsageStats, error) { return nil, nil }
+func (m *mockProxyStatsService) Aggregate(stats []domain.UsageStats, groupBy string) []map[string]interface{}      { return nil }
+
+func setupProxyEngine(t *testing.T) *conversion.ConversionEngine {
+	t.Helper()
+	registry := conversion.NewMemoryRegistry()
+	engine := conversion.NewConversionEngine(registry)
+	require.NoError(t, registry.Register(openai.NewAdapter()))
+	return engine
+}
+
+func newTestProxyHandler(engine *conversion.ConversionEngine, client *mockProxyProviderClient, routingSvc *mockProxyRoutingService, providerSvc *mockProxyProviderService) *ProxyHandler {
+	return NewProxyHandler(
+		engine,
+		client,
+		routingSvc,
+		providerSvc,
+		&mockProxyStatsService{},
+	)
+}
+
+func TestProxyHandler_HandleProxy_MissingProtocol(t *testing.T) {
+	engine := setupProxyEngine(t)
+	h := newTestProxyHandler(engine, &mockProxyProviderClient{}, &mockProxyRoutingService{}, &mockProxyProviderService{})
+
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Request = httptest.NewRequest("POST", "/", bytes.NewReader([]byte(`{}`)))
+
+	h.HandleProxy(c)
+	assert.Equal(t, 400, w.Code)
+}
+
+func TestProxyHandler_HandleProxy_NonStreamSuccess(t *testing.T) {
+	engine := setupProxyEngine(t)
+	routingSvc := &mockProxyRoutingService{
+		result: &domain.RouteResult{
+			Provider: &domain.Provider{ID: "p1", Name: "Test", APIKey: "sk-test", BaseURL: "https://api.test.com", Protocol: "openai", Enabled: true},
+			Model:    &domain.Model{ID: "m1", ProviderID: "p1", ModelName: "gpt-4", Enabled: true},
+		},
+	}
+	client := &mockProxyProviderClient{
+		sendFn: func(ctx context.Context, spec conversion.HTTPRequestSpec) (*conversion.HTTPResponseSpec, error) {
+			return &conversion.HTTPResponseSpec{
+				StatusCode: 200,
+				Headers:    map[string]string{"Content-Type": "application/json"},
+				Body: []byte(`{"id":"resp-1","object":"chat.completion","model":"gpt-4","choices":[{"index":0,"message":{"role":"assistant","content":"Hello"},"finish_reason":"stop"}],"usage":{"prompt_tokens":5,"completion_tokens":3,"total_tokens":8}}`),
+			}, nil
+		},
+	}
+	h := newTestProxyHandler(engine, client, routingSvc, &mockProxyProviderService{})
+
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Params = gin.Params{{Key: "protocol", Value: "openai"}, {Key: "path", Value: "/chat/completions"}}
+	c.Request = httptest.NewRequest("POST", "/openai/v1/chat/completions", bytes.NewReader([]byte(`{"model":"gpt-4","messages":[{"role":"user","content":"hi"}]}`)))
+
+	h.HandleProxy(c)
+	assert.Equal(t, 200, w.Code)
+
+	var resp map[string]any
+	require.NoError(t, json.Unmarshal(w.Body.Bytes(), &resp))
+	assert.Equal(t, "resp-1", resp["id"])
+}
+
+func TestProxyHandler_HandleProxy_RoutingError_WithBody(t *testing.T) {
+	engine := setupProxyEngine(t)
+	routingSvc := &mockProxyRoutingService{err: appErrors.ErrModelNotFound}
+	h := newTestProxyHandler(engine, &mockProxyProviderClient{}, routingSvc, &mockProxyProviderService{})
+
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Params = gin.Params{{Key: "protocol", Value: "openai"}, {Key: "path", Value: "/chat/completions"}}
+	c.Request = httptest.NewRequest("POST", "/openai/v1/chat/completions", bytes.NewReader([]byte(`{"model":"unknown","messages":[{"role":"user","content":"hi"}]}`)))
+
+	h.HandleProxy(c)
+	assert.Equal(t, 404, w.Code)
+}
+
+func TestProxyHandler_HandleProxy_ConversionError(t *testing.T) {
+	engine := setupProxyEngine(t)
+	routingSvc := &mockProxyRoutingService{
+		result: &domain.RouteResult{
+			Provider: &domain.Provider{ID: "p1", Name: "Test", APIKey: "sk-test", BaseURL: "https://api.test.com", Protocol: "openai", Enabled: true},
+			Model:    &domain.Model{ID: "m1", ProviderID: "p1", ModelName: "gpt-4", Enabled: true},
+		},
+	}
+	client := &mockProxyProviderClient{
+		sendFn: func(ctx context.Context, spec conversion.HTTPRequestSpec) (*conversion.HTTPResponseSpec, error) {
+			return nil, context.DeadlineExceeded
+		},
+	}
+	h := newTestProxyHandler(engine, client, routingSvc, &mockProxyProviderService{})
+
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Params = gin.Params{{Key: "protocol", Value: "openai"}, {Key: "path", Value: "/chat/completions"}}
+	c.Request = httptest.NewRequest("POST", "/openai/v1/chat/completions", bytes.NewReader([]byte(`{"model":"gpt-4","messages":[{"role":"user","content":"hi"}]}`)))
+
+	h.HandleProxy(c)
+	assert.Equal(t, 500, w.Code)
+}
+
+func TestProxyHandler_HandleProxy_ClientSendError(t *testing.T) {
+	engine := setupProxyEngine(t)
+	routingSvc := &mockProxyRoutingService{
+		result: &domain.RouteResult{
+			Provider: &domain.Provider{ID: "p1", Name: "Test", APIKey: "sk-test", BaseURL: "https://api.test.com", Protocol: "openai", Enabled: true},
+			Model:    &domain.Model{ID: "m1", ProviderID: "p1", ModelName: "gpt-4", Enabled: true},
+		},
+	}
+	client := &mockProxyProviderClient{
+		sendFn: func(ctx context.Context, spec conversion.HTTPRequestSpec) (*conversion.HTTPResponseSpec, error) {
+			return nil, context.DeadlineExceeded
+		},
+	}
+	h := newTestProxyHandler(engine, client, routingSvc, &mockProxyProviderService{})
+
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Params = gin.Params{{Key: "protocol", Value: "openai"}, {Key: "path", Value: "/chat/completions"}}
+	c.Request = httptest.NewRequest("POST", "/openai/v1/chat/completions", bytes.NewReader([]byte(`{"model":"gpt-4","messages":[{"role":"user","content":"hi"}]}`)))
+
+	h.HandleProxy(c)
+	assert.Equal(t, 500, w.Code)
+}
+
+func TestProxyHandler_HandleProxy_StreamSuccess(t *testing.T) {
+	engine := setupProxyEngine(t)
+	routingSvc := &mockProxyRoutingService{
+		result: &domain.RouteResult{
+			Provider: &domain.Provider{ID: "p1", Name: "Test", APIKey: "sk-test", BaseURL: "https://api.test.com", Protocol: "openai", Enabled: true},
+			Model:    &domain.Model{ID: "m1", ProviderID: "p1", ModelName: "gpt-4", Enabled: true},
+		},
+	}
+	client := &mockProxyProviderClient{
+		sendStreamFn: func(ctx context.Context, spec conversion.HTTPRequestSpec) (<-chan provider.StreamEvent, error) {
+			ch := make(chan provider.StreamEvent, 10)
+			go func() {
+				defer close(ch)
+				ch <- provider.StreamEvent{Data: []byte("data: {\"id\":\"1\",\"object\":\"chat.completion.chunk\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\"}}]}\n\n")}
+				ch <- provider.StreamEvent{Data: []byte("data: {\"id\":\"1\",\"object\":\"chat.completion.chunk\",\"choices\":[{\"index\":0,\"delta\":{\"content\":\"Hello\"}}]}\n\n")}
+				ch <- provider.StreamEvent{Data: []byte("data: [DONE]\n\n")}
+				ch <- provider.StreamEvent{Done: true}
+			}()
+			return ch, nil
+		},
+	}
+	h := newTestProxyHandler(engine, client, routingSvc, &mockProxyProviderService{})
+
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Params = gin.Params{{Key: "protocol", Value: "openai"}, {Key: "path", Value: "/chat/completions"}}
+	c.Request = httptest.NewRequest("POST", "/openai/v1/chat/completions", bytes.NewReader([]byte(`{"model":"gpt-4","messages":[{"role":"user","content":"hi"}],"stream":true}`)))
+
+	h.HandleProxy(c)
+	assert.Equal(t, 200, w.Code)
+	assert.Equal(t, "text/event-stream", w.Header().Get("Content-Type"))
+	assert.Contains(t, w.Body.String(), "Hello")
+}
+
+func TestProxyHandler_HandleProxy_StreamError(t *testing.T) {
+	engine := setupProxyEngine(t)
+	routingSvc := &mockProxyRoutingService{
+		result: &domain.RouteResult{
+			Provider: &domain.Provider{ID: "p1", Name: "Test", APIKey: "sk-test", BaseURL: "https://api.test.com", Protocol: "openai", Enabled: true},
+			Model:    &domain.Model{ID: "m1", ProviderID: "p1", ModelName: "gpt-4", Enabled: true},
+		},
+	}
+	client := &mockProxyProviderClient{
+		sendStreamFn: func(ctx context.Context, spec conversion.HTTPRequestSpec) (<-chan provider.StreamEvent, error) {
+			return nil, context.DeadlineExceeded
+		},
+	}
+	h := newTestProxyHandler(engine, client, routingSvc, &mockProxyProviderService{})
+
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Params = gin.Params{{Key: "protocol", Value: "openai"}, {Key: "path", Value: "/chat/completions"}}
+	c.Request = httptest.NewRequest("POST", "/openai/v1/chat/completions", bytes.NewReader([]byte(`{"model":"gpt-4","messages":[{"role":"user","content":"hi"}],"stream":true}`)))
+
+	h.HandleProxy(c)
+	assert.Equal(t, 500, w.Code)
+}
+
+func TestProxyHandler_ForwardPassthrough_GET(t *testing.T) {
+	engine := setupProxyEngine(t)
+	routingSvc := &mockProxyRoutingService{err: appErrors.ErrModelNotFound}
+	providerSvc := &mockProxyProviderService{
+		providers: []domain.Provider{
+			{ID: "p1", Name: "Test", APIKey: "sk-test", BaseURL: "https://api.test.com", Protocol: "openai"},
+		},
+	}
+	client := &mockProxyProviderClient{
+		sendFn: func(ctx context.Context, spec conversion.HTTPRequestSpec) (*conversion.HTTPResponseSpec, error) {
+			return &conversion.HTTPResponseSpec{
+				StatusCode: 200,
+				Headers:    map[string]string{"Content-Type": "application/json"},
+				Body:       []byte(`{"object":"list","data":[{"id":"gpt-4","object":"model"}]}`),
+			}, nil
+		},
+	}
+	h := newTestProxyHandler(engine, client, routingSvc, providerSvc)
+
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Params = gin.Params{{Key: "protocol", Value: "openai"}, {Key: "path", Value: "/models"}}
+	c.Request = httptest.NewRequest("GET", "/openai/v1/models", nil)
+
+	h.HandleProxy(c)
+	assert.Equal(t, 200, w.Code)
+}
+
+func TestProxyHandler_ForwardPassthrough_UnsupportedProtocol(t *testing.T) {
+	engine := setupProxyEngine(t)
+	routingSvc := &mockProxyRoutingService{err: appErrors.ErrModelNotFound}
+	providerSvc := &mockProxyProviderService{}
+	h := newTestProxyHandler(engine, &mockProxyProviderClient{}, routingSvc, providerSvc)
+
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Params = gin.Params{{Key: "protocol", Value: "unknown"}, {Key: "path", Value: "/models"}}
+	c.Request = httptest.NewRequest("GET", "/unknown/v1/models", nil)
+
+	h.HandleProxy(c)
+	assert.Equal(t, 400, w.Code)
+}
+
+func TestProxyHandler_ForwardPassthrough_NoProviders(t *testing.T) {
+	engine := setupProxyEngine(t)
+	routingSvc := &mockProxyRoutingService{err: appErrors.ErrModelNotFound}
+	providerSvc := &mockProxyProviderService{providers: []domain.Provider{}}
+	h := newTestProxyHandler(engine, &mockProxyProviderClient{}, routingSvc, providerSvc)
+
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Params = gin.Params{{Key: "protocol", Value: "openai"}, {Key: "path", Value: "/models"}}
+	c.Request = httptest.NewRequest("GET", "/openai/v1/models", nil)
+
+	h.HandleProxy(c)
+	assert.Equal(t, 404, w.Code)
+}
+
+func TestExtractModelName(t *testing.T) {
+	tests := []struct {
+		name string
+		body string
+		want string
+	}{
+		{"basic", `{"model":"gpt-4","messages":[]}`, "gpt-4"},
+		{"nested", `{"stream":true,"model":"claude-3","messages":[]}`, "claude-3"},
+		{"no_model", `{"messages":[]}`, ""},
+		{"empty", "", ""},
+		{"escaped", `{"model":"gpt\"4","messages":[]}`, `gpt\"4`},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			got := extractModelName([]byte(tt.body))
+			assert.Equal(t, tt.want, got)
+		})
+	}
+}
+
+func TestExtractHeaders(t *testing.T) {
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Request = httptest.NewRequest("POST", "/", nil)
+	c.Request.Header.Set("Authorization", "Bearer test")
+	c.Request.Header.Set("Content-Type", "application/json")
+
+	headers := extractHeaders(c)
+	assert.Equal(t, "Bearer test", headers["Authorization"])
+	assert.Equal(t, "application/json", headers["Content-Type"])
+}
+
+func TestIsStreamRequest(t *testing.T) {
+	engine := setupProxyEngine(t)
+	h := newTestProxyHandler(engine, &mockProxyProviderClient{}, &mockProxyRoutingService{}, &mockProxyProviderService{})
+
+	tests := []struct {
+		name     string
+		body     string
+		path     string
+		expected bool
+	}{
+		{"stream true chat", `{"model":"gpt-4","stream":true}`, "/v1/chat/completions", true},
+		{"stream false chat", `{"model":"gpt-4","stream":false}`, "/v1/chat/completions", false},
+		{"no stream field", `{"model":"gpt-4"}`, "/v1/chat/completions", false},
+		{"stream true non-chat", `{"model":"gpt-4","stream":true}`, "/v1/models", false},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := h.isStreamRequest([]byte(tt.body), "openai", tt.path)
+			assert.Equal(t, tt.expected, result)
+		})
+	}
+}
+
+func TestProxyHandler_HandleProxy_ProviderProtocolDefault(t *testing.T) {
+	engine := setupProxyEngine(t)
+	routingSvc := &mockProxyRoutingService{
+		result: &domain.RouteResult{
+			Provider: &domain.Provider{ID: "p1", Name: "Test", APIKey: "sk-test", BaseURL: "https://api.test.com", Protocol: "", Enabled: true},
+			Model:    &domain.Model{ID: "m1", ProviderID: "p1", ModelName: "gpt-4", Enabled: true},
+		},
+	}
+	client := &mockProxyProviderClient{
+		sendFn: func(ctx context.Context, spec conversion.HTTPRequestSpec) (*conversion.HTTPResponseSpec, error) {
+			return &conversion.HTTPResponseSpec{
+				StatusCode: 200,
+				Body:       []byte(`{"id":"r1","object":"chat.completion","model":"gpt-4","choices":[{"index":0,"message":{"role":"assistant","content":"ok"},"finish_reason":"stop"}],"usage":{"prompt_tokens":1,"completion_tokens":1,"total_tokens":2}}`),
+			}, nil
+		},
+	}
+	h := newTestProxyHandler(engine, client, routingSvc, &mockProxyProviderService{})
+
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Params = gin.Params{{Key: "protocol", Value: "openai"}, {Key: "path", Value: "/chat/completions"}}
+	c.Request = httptest.NewRequest("POST", "/openai/v1/chat/completions", bytes.NewReader([]byte(`{"model":"gpt-4","messages":[{"role":"user","content":"hi"}]}`)))
+
+	h.HandleProxy(c)
+	assert.Equal(t, 200, w.Code)
+}
+
+func TestProxyHandler_WriteConversionError_NonConversionError(t *testing.T) {
+	engine := setupProxyEngine(t)
+	h := newTestProxyHandler(engine, &mockProxyProviderClient{}, &mockProxyRoutingService{}, &mockProxyProviderService{})
+
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Request = httptest.NewRequest("POST", "/", nil)
+
+	h.writeConversionError(c, context.DeadlineExceeded, "openai")
+	assert.Equal(t, 500, w.Code)
+}
+
+func TestProxyHandler_WriteConversionError_ConversionError(t *testing.T) {
+	engine := setupProxyEngine(t)
+	h := newTestProxyHandler(engine, &mockProxyProviderClient{}, &mockProxyRoutingService{}, &mockProxyProviderService{})
+
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Request = httptest.NewRequest("POST", "/", nil)
+
+	convErr := conversion.NewConversionError(conversion.ErrorCodeInvalidInput, "bad request")
+	h.writeConversionError(c, convErr, "openai")
+	assert.Equal(t, 500, w.Code)
+}
+
+func TestProxyHandler_HandleProxy_EmptyBody(t *testing.T) {
+	engine := setupProxyEngine(t)
+	routingSvc := &mockProxyRoutingService{err: appErrors.ErrModelNotFound}
+	providerSvc := &mockProxyProviderService{
+		providers: []domain.Provider{
+			{ID: "p1", Name: "Test", APIKey: "sk-test", BaseURL: "https://api.test.com", Protocol: "openai"},
+		},
+	}
+	client := &mockProxyProviderClient{
+		sendFn: func(ctx context.Context, spec conversion.HTTPRequestSpec) (*conversion.HTTPResponseSpec, error) {
+			return &conversion.HTTPResponseSpec{
+				StatusCode: 200,
+				Body:       []byte(`{"object":"list","data":[]}`),
+			}, nil
+		},
+	}
+	h := newTestProxyHandler(engine, client, routingSvc, providerSvc)
+
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Params = gin.Params{{Key: "protocol", Value: "openai"}, {Key: "path", Value: "/models"}}
+	c.Request = httptest.NewRequest("GET", "/openai/v1/models", nil)
+
+	h.HandleProxy(c)
+	assert.Equal(t, 200, w.Code)
+}
+
+func TestProxyHandler_HandleStream_MidStreamError(t *testing.T) {
+	engine := setupProxyEngine(t)
+	routingSvc := &mockProxyRoutingService{
+		result: &domain.RouteResult{
+			Provider: &domain.Provider{ID: "p1", Name: "Test", APIKey: "sk-test", BaseURL: "https://api.test.com", Protocol: "openai", Enabled: true},
+			Model:    &domain.Model{ID: "m1", ProviderID: "p1", ModelName: "gpt-4", Enabled: true},
+		},
+	}
+	client := &mockProxyProviderClient{
+		sendStreamFn: func(ctx context.Context, spec conversion.HTTPRequestSpec) (<-chan provider.StreamEvent, error) {
+			ch := make(chan provider.StreamEvent, 10)
+			go func() {
+				defer close(ch)
+				ch <- provider.StreamEvent{Data: []byte("data: {\"id\":\"1\",\"object\":\"chat.completion.chunk\",\"choices\":[{\"index\":0,\"delta\":{\"content\":\"Hello\"}}]}\n\n")}
+				ch <- provider.StreamEvent{Error: fmt.Errorf("connection reset by peer")}
+			}()
+			return ch, nil
+		},
+	}
+	h := newTestProxyHandler(engine, client, routingSvc, &mockProxyProviderService{})
+
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Params = gin.Params{{Key: "protocol", Value: "openai"}, {Key: "path", Value: "/chat/completions"}}
+	c.Request = httptest.NewRequest("POST", "/openai/v1/chat/completions", bytes.NewReader([]byte(`{"model":"gpt-4","messages":[{"role":"user","content":"hi"}],"stream":true}`)))
+
+	h.HandleProxy(c)
+	assert.Equal(t, 200, w.Code)
+	assert.Equal(t, "text/event-stream", w.Header().Get("Content-Type"))
+	body := w.Body.String()
+	assert.Contains(t, body, "Hello")
+}
+
+func TestProxyHandler_HandleStream_FlushOutput(t *testing.T) {
+	engine := setupProxyEngine(t)
+	routingSvc := &mockProxyRoutingService{
+		result: &domain.RouteResult{
+			Provider: &domain.Provider{ID: "p1", Name: "Test", APIKey: "sk-test", BaseURL: "https://api.test.com", Protocol: "openai", Enabled: true},
+			Model:    &domain.Model{ID: "m1", ProviderID: "p1", ModelName: "gpt-4", Enabled: true},
+		},
+	}
+	client := &mockProxyProviderClient{
+		sendStreamFn: func(ctx context.Context, spec conversion.HTTPRequestSpec) (<-chan provider.StreamEvent, error) {
+			ch := make(chan provider.StreamEvent, 10)
+			go func() {
+				defer close(ch)
+				ch <- provider.StreamEvent{Data: []byte("data: {\"id\":\"1\",\"object\":\"chat.completion.chunk\",\"choices\":[{\"index\":0,\"delta\":{\"content\":\"Hi\"}}]}\n\n")}
+				ch <- provider.StreamEvent{Data: []byte("data: [DONE]\n\n")}
+				ch <- provider.StreamEvent{Done: true}
+			}()
+			return ch, nil
+		},
+	}
+	h := newTestProxyHandler(engine, client, routingSvc, &mockProxyProviderService{})
+
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Params = gin.Params{{Key: "protocol", Value: "openai"}, {Key: "path", Value: "/chat/completions"}}
+	c.Request = httptest.NewRequest("POST", "/openai/v1/chat/completions", bytes.NewReader([]byte(`{"model":"gpt-4","messages":[{"role":"user","content":"hi"}],"stream":true}`)))
+
+	h.HandleProxy(c)
+	assert.Equal(t, 200, w.Code)
+	assert.Equal(t, "text/event-stream", w.Header().Get("Content-Type"))
+	assert.Equal(t, "no-cache", w.Header().Get("Cache-Control"))
+	assert.Equal(t, "keep-alive", w.Header().Get("Connection"))
+	body := w.Body.String()
+	assert.Contains(t, body, "Hi")
+	assert.Contains(t, body, "[DONE]")
+}
+
+func TestProxyHandler_HandleStream_CreateStreamConverterError(t *testing.T) {
+	registry := conversion.NewMemoryRegistry()
+	engine := conversion.NewConversionEngine(registry)
+	err := registry.Register(openai.NewAdapter())
+	require.NoError(t, err)
+
+	routingSvc := &mockProxyRoutingService{
+		result: &domain.RouteResult{
+			Provider: &domain.Provider{ID: "p1", Name: "Test", APIKey: "sk-test", BaseURL: "https://api.test.com", Protocol: "nonexistent", Enabled: true},
+			Model:    &domain.Model{ID: "m1", ProviderID: "p1", ModelName: "gpt-4", Enabled: true},
+		},
+	}
+	h := newTestProxyHandler(engine, &mockProxyProviderClient{}, routingSvc, &mockProxyProviderService{})
+
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Params = gin.Params{{Key: "protocol", Value: "openai"}, {Key: "path", Value: "/chat/completions"}}
+	c.Request = httptest.NewRequest("POST", "/openai/v1/chat/completions", bytes.NewReader([]byte(`{"model":"gpt-4","messages":[{"role":"user","content":"hi"}],"stream":true}`)))
+
+	h.HandleProxy(c)
+	assert.Equal(t, 500, w.Code)
+}
+
+func TestProxyHandler_HandleStream_ConvertRequestError(t *testing.T) {
+	registry := conversion.NewMemoryRegistry()
+	engine := conversion.NewConversionEngine(registry)
+	require.NoError(t, registry.Register(openai.NewAdapter()))
+
+	routingSvc := &mockProxyRoutingService{
+		result: &domain.RouteResult{
+			Provider: &domain.Provider{ID: "p1", Name: "Test", APIKey: "sk-test", BaseURL: "https://api.test.com", Protocol: "nonexistent", Enabled: true},
+			Model:    &domain.Model{ID: "m1", ProviderID: "p1", ModelName: "gpt-4", Enabled: true},
+		},
+	}
+	h := newTestProxyHandler(engine, &mockProxyProviderClient{}, routingSvc, &mockProxyProviderService{})
+
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Params = gin.Params{{Key: "protocol", Value: "openai"}, {Key: "path", Value: "/chat/completions"}}
+	c.Request = httptest.NewRequest("POST", "/openai/v1/chat/completions", bytes.NewReader([]byte(`{"model":"gpt-4","messages":[{"role":"user","content":"hi"}],"stream":true}`)))
+
+	h.HandleProxy(c)
+	assert.Equal(t, 500, w.Code)
+}
+
+func TestProxyHandler_HandleNonStream_ConvertResponseError(t *testing.T) {
+	registry := conversion.NewMemoryRegistry()
+	engine := conversion.NewConversionEngine(registry)
+	require.NoError(t, registry.Register(openai.NewAdapter()))
+	require.NoError(t, registry.Register(anthropic.NewAdapter()))
+
+	routingSvc := &mockProxyRoutingService{
+		result: &domain.RouteResult{
+			Provider: &domain.Provider{ID: "p1", Name: "Test", APIKey: "sk-test", BaseURL: "https://api.test.com", Protocol: "anthropic", Enabled: true},
+			Model:    &domain.Model{ID: "m1", ProviderID: "p1", ModelName: "claude-3", Enabled: true},
+		},
+	}
+	client := &mockProxyProviderClient{
+		sendFn: func(ctx context.Context, spec conversion.HTTPRequestSpec) (*conversion.HTTPResponseSpec, error) {
+			return &conversion.HTTPResponseSpec{
+				StatusCode: 200,
+				Headers:    map[string]string{"Content-Type": "application/json"},
+				Body:       []byte(`invalid json`),
+			}, nil
+		},
+	}
+	h := newTestProxyHandler(engine, client, routingSvc, &mockProxyProviderService{})
+
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Params = gin.Params{{Key: "protocol", Value: "openai"}, {Key: "path", Value: "/chat/completions"}}
+	c.Request = httptest.NewRequest("POST", "/openai/v1/chat/completions", bytes.NewReader([]byte(`{"model":"claude-3","messages":[{"role":"user","content":"hi"}]}`)))
+
+	h.HandleProxy(c)
+	assert.Equal(t, 500, w.Code)
+}
+
+func TestProxyHandler_HandleNonStream_ResponseHeaders(t *testing.T) {
+	engine := setupProxyEngine(t)
+	routingSvc := &mockProxyRoutingService{
+		result: &domain.RouteResult{
+			Provider: &domain.Provider{ID: "p1", Name: "Test", APIKey: "sk-test", BaseURL: "https://api.test.com", Protocol: "openai", Enabled: true},
+			Model:    &domain.Model{ID: "m1", ProviderID: "p1", ModelName: "gpt-4", Enabled: true},
+		},
+	}
+	client := &mockProxyProviderClient{
+		sendFn: func(ctx context.Context, spec conversion.HTTPRequestSpec) (*conversion.HTTPResponseSpec, error) {
+			return &conversion.HTTPResponseSpec{
+				StatusCode: 200,
+				Headers:    map[string]string{"Content-Type": "application/json", "X-Custom": "test-value"},
+				Body:       []byte(`{"id":"r1","object":"chat.completion","model":"gpt-4","choices":[{"index":0,"message":{"role":"assistant","content":"ok"},"finish_reason":"stop"}],"usage":{"prompt_tokens":1,"completion_tokens":1,"total_tokens":2}}`),
+			}, nil
+		},
+	}
+	h := newTestProxyHandler(engine, client, routingSvc, &mockProxyProviderService{})
+
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Params = gin.Params{{Key: "protocol", Value: "openai"}, {Key: "path", Value: "/chat/completions"}}
+	c.Request = httptest.NewRequest("POST", "/openai/v1/chat/completions", bytes.NewReader([]byte(`{"model":"gpt-4","messages":[{"role":"user","content":"hi"}]}`)))
+
+	h.HandleProxy(c)
+	assert.Equal(t, 200, w.Code)
+	assert.Equal(t, "test-value", w.Header().Get("X-Custom"))
+	assert.Equal(t, "application/json", w.Header().Get("Content-Type"))
+}
+
+func TestProxyHandler_ForwardPassthrough_CrossProtocol(t *testing.T) {
+	registry := conversion.NewMemoryRegistry()
+	engine := conversion.NewConversionEngine(registry)
+	require.NoError(t, registry.Register(openai.NewAdapter()))
+
+	anthropicAdapter := anthropic.NewAdapter()
+	require.NoError(t, registry.Register(anthropicAdapter))
+
+	routingSvc := &mockProxyRoutingService{err: appErrors.ErrModelNotFound}
+	providerSvc := &mockProxyProviderService{
+		providers: []domain.Provider{
+			{ID: "p1", Name: "Test", APIKey: "sk-test", BaseURL: "https://api.test.com", Protocol: "anthropic"},
+		},
+	}
+	client := &mockProxyProviderClient{
+		sendFn: func(ctx context.Context, spec conversion.HTTPRequestSpec) (*conversion.HTTPResponseSpec, error) {
+			return &conversion.HTTPResponseSpec{
+				StatusCode: 200,
+				Headers:    map[string]string{"Content-Type": "application/json"},
+				Body:       []byte(`{"object":"list","data":[]}`),
+			}, nil
+		},
+	}
+	h := newTestProxyHandler(engine, client, routingSvc, providerSvc)
+
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Params = gin.Params{{Key: "protocol", Value: "openai"}, {Key: "path", Value: "/models"}}
+	c.Request = httptest.NewRequest("GET", "/openai/v1/models", nil)
+
+	h.HandleProxy(c)
+	assert.Equal(t, 200, w.Code)
+}
+
+func TestProxyHandler_ForwardPassthrough_NoBody_NoModel(t *testing.T) {
+	engine := setupProxyEngine(t)
+	routingSvc := &mockProxyRoutingService{err: appErrors.ErrModelNotFound}
+	providerSvc := &mockProxyProviderService{
+		providers: []domain.Provider{
+			{ID: "p1", Name: "Test", APIKey: "sk-test", BaseURL: "https://api.test.com", Protocol: "openai"},
+		},
+	}
+	client := &mockProxyProviderClient{
+		sendFn: func(ctx context.Context, spec conversion.HTTPRequestSpec) (*conversion.HTTPResponseSpec, error) {
+			return &conversion.HTTPResponseSpec{
+				StatusCode: 200,
+				Headers:    map[string]string{"Content-Type": "application/json"},
+				Body:       []byte(`{"object":"list","data":[{"id":"gpt-4","object":"model"}]}`),
+			}, nil
+		},
+	}
+	h := newTestProxyHandler(engine, client, routingSvc, providerSvc)
+
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Params = gin.Params{{Key: "protocol", Value: "openai"}, {Key: "path", Value: "/models"}}
+	c.Request = httptest.NewRequest("GET", "/openai/v1/models", nil)
+
+	h.HandleProxy(c)
+	assert.Equal(t, 200, w.Code)
+}
+
+func TestIsStreamRequest_EdgeCases(t *testing.T) {
+	engine := setupProxyEngine(t)
+	h := newTestProxyHandler(engine, &mockProxyProviderClient{}, &mockProxyRoutingService{}, &mockProxyProviderService{})
+
+	tests := []struct {
+		name     string
+		body     string
+		path     string
+		expected bool
+	}{
+		{"stream at end of JSON", `{"messages":[],"stream":true}`, "/v1/chat/completions", true},
+		{"stream with spaces", `{"stream" :  true}`, "/v1/chat/completions", true},
+		{"stream embedded in string value", `{"model":"stream:true"}`, "/v1/chat/completions", false},
+		{"empty body", "", "/v1/chat/completions", false},
+		{"stream true embeddings", `{"model":"text-emb","stream":true}`, "/v1/embeddings", false},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := h.isStreamRequest([]byte(tt.body), "openai", tt.path)
+			assert.Equal(t, tt.expected, result)
+		})
+	}
+}
+
+func TestProxyHandler_WriteError_RouteError(t *testing.T) {
+	engine := setupProxyEngine(t)
+	h := newTestProxyHandler(engine, &mockProxyProviderClient{}, &mockProxyRoutingService{}, &mockProxyProviderService{})
+
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Request = httptest.NewRequest("POST", "/", nil)
+
+	h.writeError(c, fmt.Errorf("model not found"), "openai")
+	assert.Equal(t, 404, w.Code)
+}
+
+func TestProxyHandler_HandleProxy_RouteEmptyBody_NoModel(t *testing.T) {
+	engine := setupProxyEngine(t)
+	routingSvc := &mockProxyRoutingService{err: appErrors.ErrModelNotFound}
+	providerSvc := &mockProxyProviderService{
+		providers: []domain.Provider{
+			{ID: "p1", Name: "Test", APIKey: "sk-test", BaseURL: "https://api.test.com", Protocol: "openai"},
+		},
+	}
+	client := &mockProxyProviderClient{
+		sendFn: func(ctx context.Context, spec conversion.HTTPRequestSpec) (*conversion.HTTPResponseSpec, error) {
+			return &conversion.HTTPResponseSpec{
+				StatusCode: 200,
+				Body:       []byte(`{"object":"list","data":[]}`),
+			}, nil
+		},
+	}
+	h := newTestProxyHandler(engine, client, routingSvc, providerSvc)
+
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Params = gin.Params{{Key: "protocol", Value: "openai"}, {Key: "path", Value: "/models"}}
+	c.Request = httptest.NewRequest("GET", "/openai/v1/models", nil)
+
+	h.HandleProxy(c)
+	assert.Equal(t, 200, w.Code)
+}

backend/internal/provider/client.go

@@ -173,22 +173,22 @@ func (c *Client) readStream(ctx context.Context, cancel context.CancelFunc, body
 		}
 
 		n, err := body.Read(buf)
+		if n > 0 {
+			dataBuf = append(dataBuf, buf[:n]...)
+		}
 		if err != nil {
-			if err == io.EOF {
+			if err != io.EOF {
+				if isNetworkError(err) {
+					c.logger.Error("流网络错误", zap.String("error", err.Error()))
+					eventChan <- StreamEvent{Error: fmt.Errorf("网络错误: %w", err)}
+				} else {
+					c.logger.Error("流读取错误", zap.String("error", err.Error()))
+					eventChan <- StreamEvent{Error: fmt.Errorf("读取错误: %w", err)}
+				}
 				return
 			}
-			if isNetworkError(err) {
-				c.logger.Error("流网络错误", zap.String("error", err.Error()))
-				eventChan <- StreamEvent{Error: fmt.Errorf("网络错误: %w", err)}
-			} else {
-				c.logger.Error("流读取错误", zap.String("error", err.Error()))
-				eventChan <- StreamEvent{Error: fmt.Errorf("读取错误: %w", err)}
-			}
-			return
 		}
 
-		dataBuf = append(dataBuf, buf[:n]...)
-
 		if len(dataBuf) > bufSize/2 && bufSize < c.streamCfg.MaxBufferSize {
 			newSize := bufSize * 2
 			if newSize > c.streamCfg.MaxBufferSize {
@@ -214,6 +214,10 @@ func (c *Client) readStream(ctx context.Context, cancel context.CancelFunc, body
 
 			eventChan <- StreamEvent{Data: rawEvent}
 		}
+
+		if err == io.EOF {
+			return
+		}
 	}
 }
 

backend/internal/provider/client_test.go

@@ -6,6 +6,7 @@ import (
 	"net/http"
 	"net/http/httptest"
 	"testing"
+	"time"
 
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
@@ -129,6 +130,184 @@ func TestClient_SendStream_ErrorResponse(t *testing.T) {
 	assert.Error(t, err)
 }
 
+func TestClient_SendStream_SSEEvents(t *testing.T) {
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.Header().Set("Content-Type", "text/event-stream")
+		w.WriteHeader(http.StatusOK)
+		flusher, ok := w.(http.Flusher)
+		require.True(t, ok)
+		w.Write([]byte("data: {\"id\":\"1\",\"choices\":[{\"delta\":{\"content\":\"Hello\"}}]}\n\n"))
+		flusher.Flush()
+		w.Write([]byte("data: {\"id\":\"1\",\"choices\":[{\"delta\":{\"content\":\" World\"}}]}\n\n"))
+		flusher.Flush()
+		time.Sleep(50 * time.Millisecond)
+		w.Write([]byte("data: [DONE]\n\n"))
+		flusher.Flush()
+		time.Sleep(50 * time.Millisecond)
+	}))
+	defer server.Close()
+
+	client := NewClient()
+	spec := conversion.HTTPRequestSpec{
+		URL:     server.URL + "/v1/chat/completions",
+		Method:  "POST",
+		Headers: map[string]string{"Authorization": "Bearer test-key"},
+		Body:    []byte(`{"model":"gpt-4","messages":[],"stream":true}`),
+	}
+
+	eventChan, err := client.SendStream(context.Background(), spec)
+	require.NoError(t, err)
+
+	var dataEvents [][]byte
+	var doneEvents int
+	for event := range eventChan {
+		if event.Done {
+			doneEvents++
+		} else if event.Error != nil {
+			t.Fatalf("unexpected error: %v", event.Error)
+		} else {
+			dataEvents = append(dataEvents, event.Data)
+		}
+	}
+
+	assert.Equal(t, 2, len(dataEvents), "expected exactly 2 data events from SSE stream")
+	assert.Contains(t, string(dataEvents[0]), "Hello")
+	assert.Contains(t, string(dataEvents[1]), "World")
+	assert.Equal(t, 1, doneEvents)
+}
+
+func TestClient_SendStream_ContextCancellation(t *testing.T) {
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.Header().Set("Content-Type", "text/event-stream")
+		w.WriteHeader(http.StatusOK)
+		time.Sleep(10 * time.Second)
+	}))
+	defer server.Close()
+
+	ctx, cancel := context.WithCancel(context.Background())
+	client := NewClient()
+	spec := conversion.HTTPRequestSpec{
+		URL:     server.URL + "/v1/chat/completions",
+		Method:  "POST",
+		Headers: map[string]string{"Authorization": "Bearer test-key"},
+		Body:    []byte(`{}`),
+	}
+
+	eventChan, err := client.SendStream(ctx, spec)
+	require.NoError(t, err)
+
+	cancel()
+
+	var gotError bool
+	for event := range eventChan {
+		if event.Error != nil {
+			gotError = true
+		}
+	}
+	assert.True(t, gotError)
+}
+
+func TestClient_Send_EmptyBody(t *testing.T) {
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		assert.Equal(t, "GET", r.Method)
+		w.WriteHeader(http.StatusOK)
+		w.Write([]byte(`{"result":"ok"}`))
+	}))
+	defer server.Close()
+
+	client := NewClient()
+	spec := conversion.HTTPRequestSpec{
+		URL:     server.URL + "/v1/models",
+		Method:  "GET",
+		Headers: map[string]string{"Authorization": "Bearer test-key"},
+	}
+
+	result, err := client.Send(context.Background(), spec)
+	require.NoError(t, err)
+	assert.Equal(t, 200, result.StatusCode)
+	assert.Contains(t, string(result.Body), "ok")
+}
+
+func TestClient_SendStream_SlowSSE(t *testing.T) {
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.Header().Set("Content-Type", "text/event-stream")
+		w.WriteHeader(http.StatusOK)
+		flusher, ok := w.(http.Flusher)
+		require.True(t, ok)
+		w.Write([]byte("data: {\"id\":\"1\"}\n\n"))
+		flusher.Flush()
+		time.Sleep(100 * time.Millisecond)
+		w.Write([]byte("data: [DONE]\n\n"))
+		flusher.Flush()
+		time.Sleep(100 * time.Millisecond)
+	}))
+	defer server.Close()
+
+	client := NewClient()
+	spec := conversion.HTTPRequestSpec{
+		URL:     server.URL + "/v1/chat/completions",
+		Method:  "POST",
+		Headers: map[string]string{"Authorization": "Bearer test-key"},
+		Body:    []byte(`{}`),
+	}
+
+	eventChan, err := client.SendStream(context.Background(), spec)
+	require.NoError(t, err)
+
+	var dataCount int
+	var doneCount int
+	for event := range eventChan {
+		if event.Done {
+			doneCount++
+		} else if event.Error != nil {
+			t.Fatalf("unexpected error: %v", event.Error)
+		} else {
+			dataCount++
+		}
+	}
+	assert.Equal(t, 1, dataCount, "expected exactly 1 data event from slow SSE")
+	assert.Equal(t, 1, doneCount, "expected exactly 1 done event from slow SSE")
+}
+
+func TestClient_SendStream_SplitSSEEvents(t *testing.T) {
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.Header().Set("Content-Type", "text/event-stream")
+		w.WriteHeader(http.StatusOK)
+		flusher, ok := w.(http.Flusher)
+		require.True(t, ok)
+		w.Write([]byte("data: {\"id\":\"1\"}\n\ndata: {\"id\":\"2\"}\n\n"))
+		flusher.Flush()
+		time.Sleep(50 * time.Millisecond)
+		w.Write([]byte("data: [DONE]\n\n"))
+		flusher.Flush()
+		time.Sleep(50 * time.Millisecond)
+	}))
+	defer server.Close()
+
+	client := NewClient()
+	spec := conversion.HTTPRequestSpec{
+		URL:     server.URL + "/v1/chat/completions",
+		Method:  "POST",
+		Headers: map[string]string{"Authorization": "Bearer test-key"},
+		Body:    []byte(`{}`),
+	}
+
+	eventChan, err := client.SendStream(context.Background(), spec)
+	require.NoError(t, err)
+
+	var dataEvents int
+	var doneEvents int
+	for event := range eventChan {
+		if event.Done {
+			doneEvents++
+		} else {
+			dataEvents++
+		}
+	}
+	assert.Equal(t, 2, dataEvents, "expected exactly 2 data events from split SSE")
+	assert.Equal(t, 1, doneEvents)
+}
+
 func TestIsNetworkError(t *testing.T) {
 	tests := []struct {
 		input string
@@ -147,3 +326,42 @@ func TestIsNetworkError(t *testing.T) {
 		assert.Equal(t, tt.want, isNetworkError(err), "isNetworkError(%q)", tt.input)
 	}
 }
+
+func TestClient_SendStream_MidStreamNetworkError(t *testing.T) {
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.Header().Set("Content-Type", "text/event-stream")
+		w.WriteHeader(http.StatusOK)
+		flusher, ok := w.(http.Flusher)
+		require.True(t, ok)
+		w.Write([]byte("data: {\"id\":\"1\"}\n\n"))
+		flusher.Flush()
+		time.Sleep(50 * time.Millisecond)
+		if hijacker, ok := w.(http.Hijacker); ok {
+			conn, _, _ := hijacker.Hijack()
+			if conn != nil {
+				conn.Close()
+			}
+		}
+	}))
+	defer server.Close()
+
+	client := NewClient()
+	spec := conversion.HTTPRequestSpec{
+		URL:     server.URL + "/v1/chat/completions",
+		Method:  "POST",
+		Headers: map[string]string{"Authorization": "Bearer test-key"},
+		Body:    []byte(`{}`),
+	}
+
+	eventChan, err := client.SendStream(context.Background(), spec)
+	require.NoError(t, err)
+
+	var gotData bool
+	for event := range eventChan {
+		if event.Error != nil {
+		} else if !event.Done {
+			gotData = true
+		}
+	}
+	assert.True(t, gotData, "should have received at least one data event before error")
+}

backend/internal/service/service_supplemental_test.go

@@ -0,0 +1,154 @@
+package service
+
+import (
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+
+	"nex/backend/internal/domain"
+	"nex/backend/internal/repository"
+)
+
+func TestProviderService_Update(t *testing.T) {
+	db := setupServiceTestDB(t)
+	repo := repository.NewProviderRepository(db)
+	svc := NewProviderService(repo)
+
+	svc.Create(&domain.Provider{ID: "p1", Name: "Original", APIKey: "key", BaseURL: "https://test.com"})
+
+	err := svc.Update("p1", map[string]interface{}{"name": "Updated"})
+	require.NoError(t, err)
+
+	result, err := svc.Get("p1", false)
+	require.NoError(t, err)
+	assert.Equal(t, "Updated", result.Name)
+}
+
+func TestProviderService_Update_NotFound(t *testing.T) {
+	db := setupServiceTestDB(t)
+	repo := repository.NewProviderRepository(db)
+	svc := NewProviderService(repo)
+
+	err := svc.Update("nonexistent", map[string]interface{}{"name": "test"})
+	assert.Error(t, err)
+}
+
+func TestModelService_Get(t *testing.T) {
+	db := setupServiceTestDB(t)
+	providerRepo := repository.NewProviderRepository(db)
+	modelRepo := repository.NewModelRepository(db)
+	svc := NewModelService(modelRepo, providerRepo)
+
+	providerRepo.Create(&domain.Provider{ID: "p1", Name: "P1", APIKey: "key", BaseURL: "https://test.com"})
+	svc.Create(&domain.Model{ID: "m1", ProviderID: "p1", ModelName: "gpt-4"})
+
+	model, err := svc.Get("m1")
+	require.NoError(t, err)
+	assert.Equal(t, "gpt-4", model.ModelName)
+}
+
+func TestModelService_Update(t *testing.T) {
+	db := setupServiceTestDB(t)
+	providerRepo := repository.NewProviderRepository(db)
+	modelRepo := repository.NewModelRepository(db)
+	svc := NewModelService(modelRepo, providerRepo)
+
+	providerRepo.Create(&domain.Provider{ID: "p1", Name: "P1", APIKey: "key", BaseURL: "https://test.com"})
+	svc.Create(&domain.Model{ID: "m1", ProviderID: "p1", ModelName: "gpt-4"})
+
+	err := svc.Update("m1", map[string]interface{}{"model_name": "gpt-4o"})
+	require.NoError(t, err)
+
+	model, err := svc.Get("m1")
+	require.NoError(t, err)
+	assert.Equal(t, "gpt-4o", model.ModelName)
+}
+
+func TestModelService_Update_ProviderID_Invalid(t *testing.T) {
+	db := setupServiceTestDB(t)
+	providerRepo := repository.NewProviderRepository(db)
+	modelRepo := repository.NewModelRepository(db)
+	svc := NewModelService(modelRepo, providerRepo)
+
+	providerRepo.Create(&domain.Provider{ID: "p1", Name: "P1", APIKey: "key", BaseURL: "https://test.com"})
+	svc.Create(&domain.Model{ID: "m1", ProviderID: "p1", ModelName: "gpt-4"})
+
+	err := svc.Update("m1", map[string]interface{}{"provider_id": "nonexistent"})
+	assert.Error(t, err)
+}
+
+func TestModelService_Delete(t *testing.T) {
+	db := setupServiceTestDB(t)
+	providerRepo := repository.NewProviderRepository(db)
+	modelRepo := repository.NewModelRepository(db)
+	svc := NewModelService(modelRepo, providerRepo)
+
+	providerRepo.Create(&domain.Provider{ID: "p1", Name: "P1", APIKey: "key", BaseURL: "https://test.com"})
+	svc.Create(&domain.Model{ID: "m1", ProviderID: "p1", ModelName: "gpt-4"})
+
+	err := svc.Delete("m1")
+	require.NoError(t, err)
+
+	_, err = svc.Get("m1")
+	assert.Error(t, err)
+}
+
+func TestModelService_Delete_NotFound(t *testing.T) {
+	db := setupServiceTestDB(t)
+	modelRepo := repository.NewModelRepository(db)
+	providerRepo := repository.NewProviderRepository(db)
+	svc := NewModelService(modelRepo, providerRepo)
+
+	err := svc.Delete("nonexistent")
+	assert.Error(t, err)
+}
+
+func TestStatsService_Aggregate_ByModel(t *testing.T) {
+	statsRepo := repository.NewStatsRepository(nil)
+	svc := NewStatsService(statsRepo)
+
+	stats := []domain.UsageStats{
+		{ProviderID: "p1", ModelName: "gpt-4", RequestCount: 10},
+		{ProviderID: "p1", ModelName: "gpt-4", RequestCount: 5},
+		{ProviderID: "p2", ModelName: "gpt-4", RequestCount: 8},
+	}
+
+	result := svc.Aggregate(stats, "model")
+	assert.True(t, len(result) >= 1)
+
+	totalCount := 0
+	for _, r := range result {
+		totalCount += r["request_count"].(int)
+	}
+	assert.Equal(t, 23, totalCount)
+}
+
+func TestStatsService_Aggregate_Default(t *testing.T) {
+	statsRepo := repository.NewStatsRepository(nil)
+	svc := NewStatsService(statsRepo)
+
+	stats := []domain.UsageStats{
+		{ProviderID: "p1", RequestCount: 10},
+		{ProviderID: "p2", RequestCount: 5},
+	}
+
+	result := svc.Aggregate(stats, "")
+	assert.Len(t, result, 2)
+
+	totalCount := 0
+	for _, r := range result {
+		totalCount += r["request_count"].(int)
+	}
+	assert.Equal(t, 15, totalCount)
+}
+
+func TestModelService_Update_NotFound(t *testing.T) {
+	db := setupServiceTestDB(t)
+	modelRepo := repository.NewModelRepository(db)
+	providerRepo := repository.NewProviderRepository(db)
+	svc := NewModelService(modelRepo, providerRepo)
+
+	err := svc.Update("nonexistent", map[string]interface{}{"model_name": "test"})
+	assert.Error(t, err)
+}

openspec/changes/refactor-conversion-engine/.openspec.yaml

@@ -1,2 +0,0 @@
-schema: spec-driven
-created: 2026-04-19

openspec/changes/refactor-conversion-engine/design.md

@@ -1,288 +0,0 @@
-## Context
-
-### 现有架构
-
-当前后端协议转换以 OpenAI 类型为内部枢纽，整体结构：
-
-```
-Anthropic Handler ──▶ anthropic.ConvertRequest() ──▶ openai.ChatCompletionRequest
-                                                 │
-OpenAI Handler ──────────────────────────────────▶ openai.ChatCompletionRequest
-                                                 │
-                                                 ▼
-                                          ProviderClient
-                                          (硬编码 OpenAI Adapter)
-                                                 │
-                                                 ▼
-                                          上游 OpenAI 兼容 API
-```
-
-关键文件：
-- `internal/protocol/openai/types.go` — OpenAI 线路格式类型，兼作内部枢纽格式
-- `internal/protocol/anthropic/converter.go` — Anthropic→OpenAI 单向转换
-- `internal/protocol/anthropic/stream_converter.go` — OpenAI chunk→Anthropic SSE 单向流式转换
-- `internal/handler/openai_handler.go` — OpenAI 请求处理
-- `internal/handler/anthropic_handler.go` — Anthropic 请求处理，内含协议转换编排
-- `internal/provider/client.go` — HTTP 客户端，硬编码 `openai.Adapter` 做序列化/反序列化
-
-### 核心限制
-
-1. **单向转换**：只有 Anthropic→OpenAI，无反向能力
-2. **OpenAI 绑定**：上游通信只能走 OpenAI 协议
-3. **无透传**：即使 client==provider（同协议），仍走完整序列化/反序列化
-4. **无扩展性**：新增协议需修改多处代码，无统一接入点
-5. **仅 Chat**：只支持 `/v1/chat/completions` 和 `/v1/messages` 两个固定端点，无 Models/Embeddings/Rerank
-
-### 设计参考
-
-三份设计文档已完整定义目标架构和两个协议的适配细节：
-- `docs/conversion_design.md` — 整体架构（Hub-and-Spoke、Canonical Model、ProtocolAdapter 接口、ConversionEngine、流式管道、错误处理）
-- `docs/conversion_openai.md` — OpenAI 协议适配清单（字段映射、流式状态机、角色合并等）
-- `docs/conversion_anthropic.md` — Anthropic 协议适配清单（角色约束、thinking、流式命名事件等）
-
-## Goals / Non-Goals
-
-**Goals:**
-
-- 实现完整的 Hub-and-Spoke 协议转换架构，以 Canonical Model 为枢纽
-- 支持任意协议对的请求/响应双向转换（当前：OpenAI ↔ Anthropic）
-- 支持同协议透传（零语义损失、零序列化开销）
-- 支持流式 SSE 双向转换（含 Tool Calling、Thinking）
-- 支持 Chat 核心层 + Models/Embeddings/Rerank 扩展层 + 未知接口透传
-- ProviderClient 支持多协议上游通信
-- 统一代理入口，URL 路由支持协议前缀
-
-**Non-Goals:**
-
-- 本阶段不实现多模态（Image/Audio/Video），Canonical Model 仅预留扩展点
-- 不实现 Middleware 的具体业务逻辑（仅定义接口和 Chain）
-- 不实现新的协议 Adapter（除 OpenAI 和 Anthropic 外）
-- 不实现有状态特性（架构预留 StatefulMiddleware 接口）
-- 不实现前端管理界面的协议选择功能
-- 不修改前端代码（前端使用管理 API，代理 API 路由变更对前端透明）
-
-## Decisions
-
-### D1: Canonical Model 用独立 Go 结构体实现，不使用 `interface{}` 或 `map[string]any`
-
-**选择**：为 CanonicalRequest、CanonicalResponse、CanonicalStreamEvent 等定义强类型 Go struct，ContentBlock 使用 discriminated union 模式（type 字段 + 各类型嵌入）
-
-**理由**：
-- 编译期类型安全，IDE 自动补全和重构友好
-- 性能优于 `map[string]any`（无反射开销）
-- 与 Go 生态的习惯一致
-
-**替代方案**：
-- `map[string]any` — 灵活但无类型安全，重构时容易遗漏字段
-- 代码生成（如 protobuf）— 引入新依赖和构建步骤，过度工程化
-
-**实现细节**：
-
-```go
-type ContentBlock struct {
-    Type string `json:"type"`
-    // Text
-    Text string `json:"text,omitempty"`
-    // ToolUse
-    ID   string          `json:"id,omitempty"`
-    Name string          `json:"name,omitempty"`
-    Input json.RawMessage `json:"input,omitempty"`
-    // ToolResult
-    ToolUseID string `json:"tool_use_id,omitempty"`
-    IsError   *bool  `json:"is_error,omitempty"`
-    // Thinking
-    Thinking string `json:"thinking,omitempty"`
-}
-```
-
-使用 `json.RawMessage` 保留 Tool Input 的原始 JSON，避免不必要的 `map` 解析。
-
-### D2: ProtocolAdapter 接口集中定义所有方法，不用接口组合
-
-**选择**：一个大的 `ProtocolAdapter` 接口包含所有方法（Chat、流式、扩展层、错误编码），不拆分为小接口
-
-**理由**：
-- 对照 `docs/conversion_design.md` §5.2 的定义，接口集中便于明确所有应实现的内容
-- Adapter 实现者可一目了然看到所有方法
-- 不支持的功能直接返回 false（`supportsInterface`）或空实现
-- `detectInterfaceType` 由各协议 Adapter 实现，因为不同协议有不同的 URL 路径约定
-
-**替代方案**：
-- 接口组合（ChatAdapter + StreamAdapter + ExtendedAdapter）——增加类型复杂度，Adapter 注册和管理更繁琐
-- 用空接口 + 类型断言——丢失编译期检查
-- `detectInterfaceType` 放在 ConversionEngine 中——违反开闭原则，新增协议需要修改 Engine
-
-### D3: StreamDecoder 直接解析原始 SSE 字节流
-
-**选择**：`ProviderClient.SendStream()` 返回 `<-chan []byte`（原始 SSE 字节流），`StreamDecoder.processChunk()` 负责拆分 SSE event 并解析 JSON
-
-**理由**：
-- SSE 解析与协议语义紧密相关（不同协议的 SSE 格式不同：OpenAI 用 `data:` 无名事件，Anthropic 用命名事件 `event: xxx\ndata: xxx`）
-- 减少中间层，降低内存拷贝
-- ProviderClient 保持最简——只做 HTTP 请求和字节流读取
-
-**替代方案**：
-- ProviderClient 内做 SSE 解析——强制所有上游使用同一 SSE 格式，不符合多协议目标
-- 独立 SSE Parser 层——增加不必要的抽象，SSE 格式本身就是 Adapter 职责的一部分
-
-### D4: ProviderClient 接受 `HTTPRequestSpec`，返回 `*HTTPResponseSpec`
-
-**选择**：ConversionEngine 输出 `HTTPRequestSpec{URL, Method, Headers, Body []byte}`，ProviderClient 接收后发送 HTTP 请求；响应返回 `HTTPResponseSpec{StatusCode, Headers, Body []byte}`
-
-**理由**：
-- ProviderClient 完全不感知协议，只做 HTTP 通信
-- ConversionEngine 统一负责 URL 构建、Header 构建、Body 序列化
-- 同协议透传时，Engine 直接透传 body bytes，Client 不做任何序列化/反序列化
-
-**接口定义**：
-
-```go
-type HTTPRequestSpec struct {
-    URL     string
-    Method  string
-    Headers map[string]string
-    Body    []byte
-}
-
-type HTTPResponseSpec struct {
-    StatusCode int
-    Headers    map[string]string
-    Body       []byte
-}
-
-type ProviderClient interface {
-    Send(ctx context.Context, spec HTTPRequestSpec) (*HTTPResponseSpec, error)
-    SendStream(ctx context.Context, spec HTTPRequestSpec) (<-chan StreamEvent, error)
-}
-```
-
-### D5: 统一代理入口使用 `/{protocol}/v1/...` URL 前缀
-
-**选择**：新路由格式 `/{protocol}/v1/{path}`，handler 从 URL 提取 protocol 前缀作为 clientProtocol
-
-**理由**：
-- 符合 `docs/conversion_design.md` §2.2 的设计
-- 调用方通过 URL 前缀明确指定协议，无需额外配置
-- 统一入口简化 handler 数量
-
-**兼容路由**：不保留旧路由，客户端需迁移到新路由格式。
-
-**替代方案**：
-- 保持两个独立 handler——违背统一架构目标
-- 请求体嗅探协议——不可靠，且设计文档明确"协议识别是调用方职责"
-
-### D6: Provider 新增 `Protocol` 字段，存储在数据库
-
-**选择**：Provider 表新增 `protocol TEXT DEFAULT 'openai'` 列，用于标识上游供应商使用的协议
-
-**理由**：
-- 上游供应商可能是 OpenAI 兼容（大多数）或 Anthropic 原生
-- 路由时需要知道 providerProtocol 以选择正确的 Adapter
-- 默认值 `'openai'` 确保现有数据兼容
-
-### D7: 删除旧 `internal/protocol/` 包，在 `internal/conversion/` 中全新实现
-
-**选择**：直接删除 `internal/protocol/openai/` 和 `internal/protocol/anthropic/`，在 `internal/conversion/` 下对照设计文档全新编写所有代码
-
-**理由**：
-- 旧代码的设计模式（OpenAI 类型为枢纽）与新架构根本不同，无法复用
-- 保留旧代码容易导致混用两种模式，引入隐蔽 bug
-- 旧代码中的类型定义不迁移，直接根据设计文档重新定义，确保与新架构一致
-
-### D8: 目标包结构
-
-```
-internal/conversion/
-  canonical/
-    types.go                     # CanonicalRequest/Response/Message/ContentBlock/Tool/ToolChoice/ThinkingConfig/OutputFormat
-    stream.go                    # CanonicalStreamEvent 联合体 + 所有事件类型
-    extended.go                  # CanonicalModelList/ModelInfo/Embedding/Rerank
-  errors.go                      # ConversionError + ErrorCode 枚举
-  interface.go                   # InterfaceType 枚举
-  provider.go                    # TargetProvider struct
-  adapter.go                     # ProtocolAdapter 接口 + AdapterRegistry 接口和实现
-  stream.go                      # StreamDecoder/StreamEncoder/StreamConverter 接口 + Passthrough/Canonical 实现
-  middleware.go                   # ConversionMiddleware 接口 + MiddlewareChain
-  engine.go                      # ConversionEngine 门面 + HTTPRequestSpec/HTTPResponseSpec
-
-  openai/
-    types.go                     # OpenAI 线路格式类型（对照 conversion_openai.md 全新定义）
-    adapter.go                   # ProtocolAdapter 实现（detectInterfaceType/buildUrl/buildHeaders/supportsInterface/encodeError）
-    decoder.go                   # decodeRequest/decodeResponse/扩展层 decode 方法
-    encoder.go                   # encodeRequest/encodeResponse/扩展层 encode 方法
-    stream_decoder.go            # OpenAIStreamDecoder（delta chunk 状态机）
-    stream_encoder.go            # OpenAIStreamEncoder（缓冲策略）
-
-  anthropic/
-    types.go                     # Anthropic 线路格式类型（对照 conversion_anthropic.md 全新定义）
-    adapter.go                   # ProtocolAdapter 实现（detectInterfaceType/buildUrl/buildHeaders/supportsInterface/encodeError）
-    decoder.go                   # decodeRequest/decodeResponse/扩展层 decode 方法
-    encoder.go                   # encodeRequest/encodeResponse/扩展层 encode 方法
-    stream_decoder.go            # AnthropicStreamDecoder（命名事件 1:1 映射）
-    stream_encoder.go            # AnthropicStreamEncoder（直接映射，无缓冲）
-```
-
-## Risks / Trade-offs
-
-### R1: Anthropic 角色约束处理复杂度高
-**风险**：Anthropic 要求 user/assistant 严格交替、首消息必须为 user、tool_result 必须嵌入 user 消息。从 Canonical 编码为 Anthropic 时需要合并/拆分/注入消息，逻辑容易出错。
-**缓解**：
-- 编写详尽的测试用例覆盖所有边界情况（连续 tool 消息、首条 assistant 消息、空 user 消息注入等）
-- 将角色约束处理封装为独立函数，与内容编码逻辑分离
-
-### R2: OpenAI 流式状态机复杂
-**风险**：OpenAI 的 delta chunk 没有显式生命周期（无 start/stop），StreamDecoder 需要状态机推断 block 边界，管理工具调用索引映射和参数累积。
-**缓解**：
-- 严格对照 `docs/conversion_openai.md` §6.2-§6.3 的伪代码实现
-- 为每种 delta 类型编写独立测试（text、tool_calls、reasoning_content、refusal、usage chunk）
-- UTF-8 跨 chunk 截断使用 `utf8Remainder` 缓冲
-
-### R3: 全量重构影响范围大
-**风险**：同时删除旧代码、新建包、改造 handler/provider/domain，可能导致系统长时间不可用。
-**缓解**：
-- 旧代码在删除前确认新代码所有测试通过
-- Git 分支隔离开发，完成后再合并
-- 新路由 `/{protocol}/v1/...` 确保协议明确指定
-
-### R4: Canonical Model 字段演进
-**风险**：Canonical Model 的字段集反映当前已适配协议的公共语义，未来新增协议时可能需要频繁修改。
-**缓解**：
-- 字段晋升规范已在 `docs/conversion_design.md` 附录 C 中定义
-- `json:"-"` 标签控制序列化输出，新增可选字段不影响已有编解码
-- 协议特有字段不纳入 Canonical，通过同协议透传保留
-
-### R5: 性能——双重序列化开销
-**风险**：跨协议转换时经过 decode→canonical→encode 两次序列化/反序列化，相比直接转换多一次拷贝。
-**权衡**：接受此开销以换取架构清晰和可扩展性。同协议透传路径零开销补偿。实际 LLM API 延迟（数百毫秒到数秒）远大于 JSON 序列化开销（微秒级）。
-
-## Migration Plan
-
-### 步骤
-
-1. **创建 `internal/conversion/` 包**：实现 Layer 1-3（Canonical Model、接口定义、Engine），不改动现有代码
-2. **全新实现 OpenAI Adapter 和 Anthropic Adapter**：Layer 4-5，对照设计文档在 conversion 包内全新编写，不沿用旧 protocol 包代码
-3. **编写全面测试**：覆盖编解码、流式转换、错误处理、同协议透传
-4. **改造 `domain.Provider`**：新增 `Protocol` 字段
-5. **创建数据库迁移**：`ALTER TABLE providers ADD COLUMN protocol TEXT DEFAULT 'openai'`
-6. **改造 `ProviderClient`**：简化为接受 `HTTPRequestSpec` 的 HTTP 发送器
-7. **创建 `ProxyHandler`**：统一代理入口，集成 ConversionEngine
-8. **更新 `cmd/server/main.go`**：注册 Adapter、创建 Engine、配置新路由
-9. **删除旧 `internal/protocol/` 包**：直接删除，不迁移代码，确认新架构完全替代
-10. **更新 README.md**：项目结构、API 接口、路由说明
-
-### 兼容策略
-
-- 旧路由 `/v1/chat/completions` 和 `/v1/messages` 不再保留，客户端需迁移
-- 现有 Provider 数据通过 `DEFAULT 'openai'` 自动获得协议标识
-- 前端管理 API 不受影响
-
-### 回滚策略
-
-- Git 分支隔离：在新分支开发，合并前充分测试
-- 旧 `internal/protocol/` 包在删除前确认新架构所有测试通过，删除后不可恢复旧代码（从 git 历史仍可找回）
-- 数据库迁移向下兼容（仅 ADD COLUMN）
-
-## Open Questions
-
-- ~~是否需要为兼容路由 `/v1/chat/completions` 和 `/v1/messages` 设置 deprecation 期限？~~ → **决定**：不保留旧路由，客户端直接迁移到 `/{protocol}/v1/...`
-- ~~扩展层接口（Models/Embeddings/Rerank）在本阶段是否全部实现，还是先做 Models，其余后续迭代？~~ → **决定**：本阶段全部实现（对照三份文档的字段映射已在 spec 中完整定义），因为扩展层接口编解码逻辑量不大（轻量字段映射），且实现后能完整验证引擎的接口分层分发逻辑

openspec/changes/refactor-conversion-engine/proposal.md

@@ -1,45 +0,0 @@
-## Why
-
-当前后端协议转换层以 OpenAI 类型作为内部枢纽，Anthropic 请求单向转换为 OpenAI 格式后再发往上游。这种设计导致：无法支持 OpenAI→Anthropic 的反向转换、无法对接 Anthropic 协议的上游供应商、无法实现同协议透传的零开销转发、无法横向扩展新协议。重构为基于协议中立 Canonical Model 的 Hub-and-Spoke 架构（参考 `docs/conversion_design.md`），从根本上解决这些问题。
-
-## What Changes
-
-- **引入 Canonical Model**：定义协议无关的 `CanonicalRequest`、`CanonicalResponse`、`CanonicalStreamEvent` 等规范模型，作为所有协议间转换的统一枢纽
-- **引入 ConversionEngine**：无状态的转换引擎门面，协调 Adapter 注册、接口识别、透传判断、请求/响应转换、流式转换
-- **引入 ProtocolAdapter 接口**：统一适配器契约，每种协议实现完整的编解码（Chat 请求/响应、流式、扩展层接口、错误编码）
-- **实现 OpenAI Adapter**：对照 `docs/conversion_openai.md` 全新实现 OpenAI 协议的完整 Adapter（含状态机流式解码器/编码器），不沿用旧 `internal/protocol/openai/` 代码
-- **实现 Anthropic Adapter**：对照 `docs/conversion_anthropic.md` 全新实现 Anthropic 协议的完整 Adapter（含命名事件流式解码器/编码器），不沿用旧 `internal/protocol/anthropic/` 代码
-- **统一代理 Handler**：合并 `OpenAIHandler` 和 `AnthropicHandler` 为统一的 `ProxyHandler`，支持 `/{protocol}/v1/...` URL 前缀路由
-- **同协议透传**：client == provider 时跳过 Canonical 转换，仅重建 Header 后原样转发
-- **接口分层**：核心层（Chat）走 Canonical 深度转换，扩展层（Models/Embeddings/Rerank）走轻量映射，未知接口走透传
-- **ProviderClient 简化**：移除 OpenAI Adapter 硬编码，变为协议无关的 HTTP 发送器
-- **Provider 新增 Protocol 字段**：**BREAKING** — Provider 模型新增 `protocol` 字段标识上游协议类型
-- **删除旧 protocol 包**：移除 `internal/protocol/openai/` 和 `internal/protocol/anthropic/`，在 `internal/conversion/` 中全新实现
-- **URL 路由变更**：**BREAKING** — 代理端点从 `/v1/chat/completions` + `/v1/messages` 变更为 `/{protocol}/v1/...`，不保留旧路由
-
-## Capabilities
-
-### New Capabilities
-
-- `conversion-engine`: 协议转换引擎核心能力——Canonical Model 定义、ProtocolAdapter 接口与注册表、ConversionEngine 门面（请求/响应转换、流式转换、接口识别、透传判断）、StreamDecoder/Encoder 接口、Middleware 拦截链、ConversionError 错误体系
-- `protocol-adapter-openai`: OpenAI 协议适配器——完整的 ProtocolAdapter 实现（对照 conversion_openai.md），涵盖 Chat 请求/响应编解码、流式状态机解码器（OpenAI delta chunk → CanonicalStreamEvent）和编码器（反向）、扩展层接口编解码（Models/Embeddings/Rerank）、错误编码、同协议透传
-- `protocol-adapter-anthropic`: Anthropic 协议适配器——完整的 ProtocolAdapter 实现（对照 conversion_anthropic.md），涵盖 Chat 请求/响应编解码（含角色约束处理：tool→user 合并、user/assistant 交替保证）、流式解码器（命名 SSE 事件 → CanonicalStreamEvent）和编码器（反向）、扩展层接口编解码（Models）、错误编码、同协议透传
-- `unified-proxy-handler`: 统一代理入口——合并 OpenAI/Anthropic 双 Handler 为统一 ProxyHandler，支持 `/{protocol}/v1/...` URL 前缀路由、协议识别
-
-### Modified Capabilities
-
-- `openai-protocol-proxy`: URL 路由从硬编码 `/v1/chat/completions` 变更为 `/{protocol}/v1/...` 统一入口；请求处理从直接调用 ProviderClient 变更为经 ConversionEngine 转换；新增同协议透传能力；新增扩展层接口（Models/Embeddings/Rerank）代理
-- `anthropic-protocol-proxy`: 从单向 Anthropic→OpenAI 转换变更为双向任意协议互转；从 Handler 内直接调用 converter 变更为经 ConversionEngine；新增 Anthropic 作为上游供应商的能力；新增同协议透传能力；新增扩展层接口代理
-- `provider-management`: Provider 模型新增 `protocol` 字段（标识上游协议类型，默认 "openai"）；数据库迁移新增 protocol 列
-- `layered-architecture`: 新增 conversion 层（internal/conversion/）位于 handler 和 provider 之间；ProviderClient 接口简化为协议无关的 HTTP 发送器
-- `error-handling`: 新增 ConversionError 错误类型和 ErrorCode 枚举；转换失败时使用客户端协议格式编码错误响应
-- `request-validation`: 请求验证从 handler 层前移到 ProtocolAdapter 的 decodeRequest 中；验证规则按各协议规范独立定义
-
-## Impact
-
-- **代码结构**：新增 `internal/conversion/` 包（约 20+ 文件，全新编写），删除 `internal/protocol/` 包（不迁移，直接删除后重写），改造 `internal/handler/` 和 `internal/provider/`
-- **API 兼容性**：**BREAKING** — 代理端点 URL 变更（`/v1/chat/completions` → `/openai/v1/chat/completions`，`/v1/messages` → `/anthropic/v1/messages`），不保留旧路由
-- **数据库**：Provider 表新增 `protocol` 列，需数据库迁移
-- **依赖**：无新增外部依赖，复用现有 Go 标准库和已引入的包
-- **测试**：需为 conversion 包编写全面单元测试，覆盖每个 Adapter 的编解码、流式转换、错误处理、同协议透传
-- **文档**：需更新 README.md 中的项目结构、API 接口说明

openspec/changes/refactor-conversion-engine/specs/anthropic-protocol-proxy/spec.md

@@ -1,83 +0,0 @@
-## MODIFIED Requirements
-
-### Requirement: 支持 Anthropic Messages API 端点
-
-网关 SHALL 提供 Anthropic Messages API 端点供外部应用调用。
-
-#### Scenario: 成功的非流式请求
-
-- **WHEN** 应用发送 POST 请求到 `/anthropic/v1/messages`，携带有效的 Anthropic 请求格式（非流式）
-- **THEN** 网关 SHALL 通过 ConversionEngine 将 Anthropic 请求解码为 Canonical 格式
-- **THEN** 网关 SHALL 将 Canonical 请求编码为目标供应商协议格式
-- **THEN** 网关 SHALL 将供应商的响应通过 ConversionEngine 转换为 Anthropic 格式返回给应用
-
-#### Scenario: 成功的流式请求
-
-- **WHEN** 应用发送 POST 请求到 `/anthropic/v1/messages`，携带 `stream: true`
-- **THEN** 网关 SHALL 通过 ConversionEngine 创建 StreamConverter
-- **THEN** 网关 SHALL 将上游协议的 SSE 流转换为 Anthropic 命名事件格式
-- **THEN** 网关 SHALL 使用 `event: <type>\ndata: <json>\n\n` 格式流式返回给应用
-
-#### Scenario: 同协议透传（Anthropic → Anthropic Provider）
-
-- **WHEN** 客户端使用 Anthropic 协议且目标供应商也是 Anthropic 协议
-- **THEN** 网关 SHALL 跳过 Canonical 转换，仅重建认证 Header 后原样转发
-- **THEN** 请求和响应 Body SHALL 保持原样
-
-### Requirement: 双向协议转换
-
-网关 SHALL 支持 Anthropic 协议与任意已注册协议间的双向转换。
-
-#### Scenario: Anthropic 客户端 → OpenAI 供应商
-
-- **WHEN** 客户端使用 Anthropic 协议且供应商使用 OpenAI 协议
-- **THEN** SHALL 将 Anthropic MessagesRequest 解码为 CanonicalRequest
-- **THEN** SHALL 将 CanonicalRequest 编码为 OpenAI ChatCompletionRequest
-- **THEN** SHALL 将 OpenAI ChatCompletionResponse 解码为 CanonicalResponse
-- **THEN** SHALL 将 CanonicalResponse 编码为 Anthropic MessagesResponse
-
-#### Scenario: OpenAI 客户端 → Anthropic 供应商
-
-- **WHEN** 客户端使用 OpenAI 协议且供应商使用 Anthropic 协议
-- **THEN** SHALL 将 OpenAI ChatCompletionRequest 解码为 CanonicalRequest
-- **THEN** SHALL 将 CanonicalRequest 编码为 Anthropic MessagesRequest
-- **THEN** SHALL 将 Anthropic MessagesResponse 解码为 CanonicalResponse
-- **THEN** SHALL 将 CanonicalResponse 编码为 OpenAI ChatCompletionResponse
-
-### Requirement: 使用 service 层处理请求
-
-Handler SHALL 通过 service 层处理业务逻辑。
-
-#### Scenario: 调用 routing service
-
-- **WHEN** ProxyHandler 收到 Anthropic 协议请求
-- **THEN** SHALL 调用 RoutingService.Route() 获取路由结果
-- **THEN** SHALL 从路由结果获取 Provider（含 protocol 字段）
-
-#### Scenario: 调用 stats service
-
-- **WHEN** 请求成功完成
-- **THEN** SHALL 调用 StatsService.Record() 记录统计
-- **THEN** SHALL 异步记录统计（不阻塞响应）
-
-### Requirement: 使用结构化错误处理
-
-ProxyHandler SHALL 使用 ConversionError 和 Anthropic 的 encodeError 处理错误。
-
-#### Scenario: 协议转换错误
-
-- **WHEN** ConversionEngine 返回 ConversionError
-- **THEN** SHALL 使用 Anthropic 的 Adapter.encodeError 编码错误响应
-- **THEN** SHALL 使用 Anthropic 错误格式（`{type: "error", error: {type, message}}`）
-
-#### Scenario: 路由错误处理
-
-- **WHEN** RoutingService 返回错误
-- **THEN** SHALL 转换为 ConversionError
-- **THEN** SHALL 使用 Anthropic 错误格式返回
-
-#### Scenario: 供应商错误处理
-
-- **WHEN** ProviderClient 返回错误
-- **THEN** SHALL 包装为 ConversionError
-- **THEN** SHALL 使用 Anthropic 错误格式返回

openspec/changes/refactor-conversion-engine/specs/error-handling/spec.md

@@ -1,53 +0,0 @@
-## MODIFIED Requirements
-
-### Requirement: 统一错误响应
-
-系统 SHALL 统一错误响应格式，新增 ConversionError 支持。
-
-#### Scenario: OpenAI 协议错误响应
-
-- **WHEN** OpenAI 协议发生错误
-- **THEN** SHALL 返回标准 OpenAI 错误响应格式
-- **THEN** SHALL 包含 error.message、error.type、error.code 字段
-
-#### Scenario: Anthropic 协议错误响应
-
-- **WHEN** Anthropic 协议发生错误
-- **THEN** SHALL 返回标准 Anthropic 错误响应格式
-- **THEN** SHALL 包含 type、error.type、error.message 字段
-
-#### Scenario: 转换错误响应
-
-- **WHEN** ConversionEngine 在协议转换过程中产生 ConversionError
-- **THEN** SHALL 使用客户端协议的 Adapter.encodeError 编码错误响应
-- **THEN** 错误响应 SHALL 使用客户端可理解的协议格式
-
-#### Scenario: 管理 API 错误响应
-
-- **WHEN** 管理 API 发生错误
-- **THEN** SHALL 返回统一的错误响应格式
-- **THEN** SHALL 包含 code、message 字段
-- **THEN** SHALL 可选包含 details 字段（验证错误详情）
-
-## ADDED Requirements
-
-### Requirement: 定义 ConversionError 错误类型
-
-系统 SHALL 定义 ConversionError 结构体和 ErrorCode 枚举。
-
-#### Scenario: ConversionError 结构
-
-- **WHEN** 定义转换错误
-- **THEN** SHALL 包含 Code（ErrorCode 枚举）、Message 字段
-- **THEN** SHALL 可选包含 ClientProtocol、ProviderProtocol、InterfaceType、Details、Cause 字段
-
-#### Scenario: ErrorCode 枚举
-
-- **WHEN** 定义错误码
-- **THEN** SHALL 包含 INVALID_INPUT、MISSING_REQUIRED_FIELD、INCOMPATIBLE_FEATURE、FIELD_MAPPING_FAILURE、TOOL_CALL_PARSE_ERROR、JSON_PARSE_ERROR、STREAM_STATE_ERROR、UTF8_DECODE_ERROR、PROTOCOL_CONSTRAINT_VIOLATION、ENCODING_FAILURE、INTERFACE_NOT_SUPPORTED
-
-#### Scenario: 错误码到协议错误类型的映射
-
-- **WHEN** 使用 encodeError 编码错误
-- **THEN** ErrorCode SHALL 映射为各协议的错误类型字符串
-- **THEN** 例如 INVALID_INPUT → OpenAI "invalid_request_error"，Anthropic "invalid_request_error"

openspec/changes/refactor-conversion-engine/specs/layered-architecture/spec.md

@@ -1,118 +0,0 @@
-## MODIFIED Requirements
-
-### Requirement: 实现三层架构
-
-系统 SHALL 实现 handler → service → repository 三层架构，并在 handler 和 provider 之间新增 conversion 层。
-
-#### Scenario: Handler 层职责
-
-- **WHEN** 处理 HTTP 请求
-- **THEN** handler 层 SHALL 仅负责 HTTP 请求解析、URL 路由和响应写入
-- **THEN** handler 层 SHALL 调用 ConversionEngine 处理协议转换
-- **THEN** handler 层 SHALL 调用 service 层处理业务逻辑
-- **THEN** handler 层 SHALL NOT 直接访问数据库或执行协议转换逻辑
-
-#### Scenario: Conversion 层职责
-
-- **WHEN** 处理协议转换
-- **THEN** conversion 层 SHALL 包含 Canonical Model 定义
-- **THEN** conversion 层 SHALL 包含各协议的 ProtocolAdapter 实现
-- **THEN** conversion 层 SHALL 包含 ConversionEngine 门面
-- **THEN** conversion 层 SHALL NOT 依赖 handler 或 service 层
-
-#### Scenario: Service 层职责
-
-- **WHEN** 处理业务逻辑
-- **THEN** service 层 SHALL 包含业务规则和验证
-- **THEN** service 层 SHALL 调用 repository 层访问数据
-- **THEN** service 层 SHALL NOT 包含协议转换逻辑
-
-#### Scenario: Repository 层职责
-
-- **WHEN** 访问数据
-- **THEN** repository 层 SHALL 仅负责数据访问
-- **THEN** repository 层 SHALL 封装数据库操作
-- **THEN** repository 层 SHALL NOT 包含业务逻辑或协议转换逻辑
-
-### Requirement: 定义核心接口
-
-系统 SHALL 定义清晰的接口边界。
-
-#### Scenario: Service 接口定义
-
-- **WHEN** 定义 service 接口
-- **THEN** SHALL 定义 ProviderService、ModelService、RoutingService、StatsService 接口
-- **THEN** SHALL 定义清晰的业务方法签名
-- **THEN** SHALL 使用 domain 类型作为参数和返回值
-
-#### Scenario: Repository 接口定义
-
-- **WHEN** 定义 repository 接口
-- **THEN** SHALL 定义 ProviderRepository、ModelRepository、StatsRepository 接口
-- **THEN** SHALL 定义清晰的数据访问方法签名
-- **THEN** SHALL 使用 domain 类型作为参数和返回值
-
-#### Scenario: Provider Client 接口定义
-
-- **WHEN** 定义 provider client 接口
-- **THEN** SHALL 定义 ProviderClient 接口
-- **THEN** SHALL 包含 Send（非流式）和 SendStream（流式）方法
-- **THEN** SHALL 接受 HTTPRequestSpec 作为参数，不绑定特定协议
-- **THEN** SHALL 支持接口 Mock
-
-#### Scenario: Conversion 层接口定义
-
-- **WHEN** 定义 conversion 层接口
-- **THEN** SHALL 定义 ProtocolAdapter、StreamDecoder、StreamEncoder、StreamConverter、ConversionMiddleware 接口
-- **THEN** SHALL 定义 AdapterRegistry 用于 Adapter 注册和查询
-- **THEN** SHALL 定义 ConversionEngine 作为统一门面
-
-### Requirement: 实现依赖注入
-
-系统 SHALL 使用手动依赖注入。
-
-#### Scenario: Repository 注入
-
-- **WHEN** 初始化 service
-- **THEN** SHALL 通过构造函数注入 repository 依赖
-- **THEN** SHALL 使用接口类型而非具体类型
-
-#### Scenario: Service 注入
-
-- **WHEN** 初始化 handler
-- **THEN** SHALL 通过构造函数注入 service 依赖、ConversionEngine、ProviderClient
-- **THEN** SHALL 使用接口类型而非具体类型
-
-#### Scenario: Conversion 组装
-
-- **WHEN** 应用启动
-- **THEN** SHALL 创建 AdapterRegistry 并注册所有 ProtocolAdapter
-- **THEN** SHALL 创建 ConversionEngine（注入 registry 和 middleware chain）
-- **THEN** SHALL 将 ConversionEngine 注入到 ProxyHandler
-
-#### Scenario: 主函数组装
-
-- **WHEN** 应用启动
-- **THEN** main.go SHALL 按顺序构造所有依赖
-- **THEN** SHALL 先构造基础设施（logger、database）
-- **THEN** SHALL 再构造 repository、service
-- **THEN** SHALL 再构造 conversion 层（registry → engine）
-- **THEN** SHALL 最后构造 handler
-
-### Requirement: 定义 Domain 模型
-
-系统 SHALL 定义独立的 domain 模型。
-
-#### Scenario: Domain 模型定义
-
-- **WHEN** 定义领域模型
-- **THEN** SHALL 在 internal/domain/ 包中定义
-- **THEN** SHALL 包含 Provider、Model、UsageStats 等模型
-- **THEN** Provider SHALL 包含 Protocol 字段
-- **THEN** SHALL 与数据库模型分离
-
-#### Scenario: Domain 模型使用
-
-- **WHEN** service 和 repository 处理数据
-- **THEN** SHALL 使用 domain 模型
-- **THEN** SHALL NOT 使用数据库模型（GORM 模型）

openspec/changes/refactor-conversion-engine/specs/openai-protocol-proxy/spec.md

@@ -1,99 +0,0 @@
-## MODIFIED Requirements
-
-### Requirement: 支持 OpenAI Chat Completions API 端点
-
-网关 SHALL 提供 OpenAI Chat Completions API 端点供外部应用调用。
-
-#### Scenario: 成功的非流式请求
-
-- **WHEN** 应用发送 POST 请求到 `/openai/v1/chat/completions`，携带有效的 OpenAI 请求格式（非流式）
-- **THEN** 网关 SHALL 通过 ConversionEngine 转换请求
-- **THEN** 网关 SHALL 将转换后的请求转发到配置的供应商
-- **THEN** 网关 SHALL 将供应商的响应通过 ConversionEngine 转换为 OpenAI 格式返回给应用
-
-#### Scenario: 成功的流式请求
-
-- **WHEN** 应用发送 POST 请求到 `/openai/v1/chat/completions`，携带 `stream: true`
-- **THEN** 网关 SHALL 通过 ConversionEngine 创建 StreamConverter
-- **THEN** 网关 SHALL 使用 SSE 格式将转换后的响应流式返回给应用
-- **THEN** 网关 SHALL 在流完成时发送 `data: [DONE]`
-
-#### Scenario: 同协议透传（OpenAI → OpenAI Provider）
-
-- **WHEN** 客户端使用 OpenAI 协议且目标供应商也是 OpenAI 协议
-- **THEN** 网关 SHALL 跳过 Canonical 转换，仅重建认证 Header 后原样转发
-- **THEN** 请求和响应 Body SHALL 保持原样
-
-### Requirement: 根据模型名称路由请求
-
-网关 SHALL 根据请求中的 `model` 字段将请求路由到相应的供应商。
-
-#### Scenario: 有效模型路由
-
-- **WHEN** 请求包含存在于配置模型中的 `model` 字段
-- **AND** 该模型已启用
-- **THEN** 网关 SHALL 将请求路由到该模型关联的供应商
-- **THEN** 网关 SHALL 从供应商的 `protocol` 字段获取 providerProtocol
-
-#### Scenario: 模型未找到
-
-- **WHEN** 请求包含不存在于配置模型中的 `model` 字段
-- **THEN** 网关 SHALL 使用 OpenAI 格式返回错误响应
-
-#### Scenario: 模型已禁用
-
-- **WHEN** 请求包含已禁用模型的 `model` 字段
-- **THEN** 网关 SHALL 使用 OpenAI 格式返回错误响应
-
-### Requirement: 对 OpenAI 兼容供应商透明代理
-
-网关 SHALL 对 OpenAI 兼容供应商的请求和响应通过 ConversionEngine 进行转换处理。
-
-#### Scenario: 跨协议请求转发
-
-- **WHEN** 网关收到 OpenAI 协议请求且目标供应商使用不同协议
-- **THEN** 网关 SHALL 通过 ConversionEngine 将请求转换为目标协议格式
-- **THEN** 网关 SHALL 使用目标协议的 Adapter 构建 URL 和 Header
-
-#### Scenario: 扩展层接口代理
-
-- **WHEN** 网关收到 `/openai/v1/models` 等 GET 请求
-- **THEN** 网关 SHALL 通过 ConversionEngine 转换扩展层接口的响应格式
-
-### Requirement: 使用 service 层处理请求
-
-Handler SHALL 通过 service 层处理业务逻辑。
-
-#### Scenario: 调用 routing service
-
-- **WHEN** ProxyHandler 收到请求
-- **THEN** SHALL 调用 RoutingService.Route() 获取路由结果
-- **THEN** SHALL 从路由结果获取 Provider（含 protocol 字段）
-
-#### Scenario: 调用 stats service
-
-- **WHEN** 请求成功完成
-- **THEN** SHALL 调用 StatsService.Record() 记录统计
-- **THEN** SHALL 异步记录统计（不阻塞响应）
-
-### Requirement: 使用结构化错误处理
-
-ProxyHandler SHALL 使用 ConversionError 和协议对应的 encodeError 处理错误。
-
-#### Scenario: 转换错误
-
-- **WHEN** ConversionEngine 返回 ConversionError
-- **THEN** SHALL 使用 clientProtocol 的 Adapter.encodeError 编码错误响应
-- **THEN** SHALL 使用 OpenAI 错误格式（`{error: {message, type, code}}`）
-
-#### Scenario: 路由错误处理
-
-- **WHEN** RoutingService 返回错误
-- **THEN** SHALL 转换为 ConversionError
-- **THEN** SHALL 使用 OpenAI 错误格式返回
-
-#### Scenario: 供应商错误处理
-
-- **WHEN** ProviderClient 返回错误
-- **THEN** SHALL 包装为 ConversionError
-- **THEN** SHALL 使用 OpenAI 错误格式返回

openspec/changes/refactor-conversion-engine/specs/provider-management/spec.md

@@ -1,73 +0,0 @@
-## MODIFIED Requirements
-
-### Requirement: 创建供应商配置
-
-网关 SHALL 允许通过管理 API 创建新的供应商配置。
-
-#### Scenario: 使用有效数据创建供应商
-
-- **WHEN** 向 `/api/providers` 发送 POST 请求，携带有效的供应商数据（id, name, api_key, base_url, protocol）
-- **THEN** 网关 SHALL 在数据库中创建新的供应商记录
-- **THEN** 网关 SHALL 返回创建的供应商，状态码为 201
-- **THEN** 供应商 SHALL 默认启用
-- **THEN** protocol 字段 SHALL 默认为 "openai"
-
-#### Scenario: 使用重复 ID 创建供应商
-
-- **WHEN** 向 `/api/providers` 发送 POST 请求，携带已存在的 ID
-- **THEN** 网关 SHALL 返回错误，状态码为 409 (Conflict)
-
-#### Scenario: 创建供应商时缺少必需字段
-
-- **WHEN** 向 `/api/providers` 发送 POST 请求，缺少必需字段（id, name, api_key 或 base_url）
-- **THEN** 网关 SHALL 返回错误，状态码为 400 (Bad Request)
-- **THEN** 错误 SHALL 指示缺少哪些字段
-
-### Requirement: 列出所有供应商
-
-网关 SHALL 允许获取所有供应商配置。
-
-#### Scenario: 成功列出供应商
-
-- **WHEN** 向 `/api/providers` 发送 GET 请求
-- **THEN** 网关 SHALL 返回所有供应商的列表
-- **THEN** 每个供应商 SHALL 包含 id, name, api_key（已掩码）, base_url, protocol, enabled, created_at, updated_at
-- **THEN** api_key SHALL 被掩码（仅显示最后 4 个字符）
-
-### Requirement: 获取特定供应商
-
-网关 SHALL 允许通过 ID 获取特定供应商。
-
-#### Scenario: 获取存在的供应商
-
-- **WHEN** 向 `/api/providers/:id` 发送 GET 请求，携带有效的供应商 ID
-- **THEN** 网关 SHALL 返回供应商详情
-- **THEN** SHALL 包含 protocol 字段
-- **THEN** api_key SHALL 被掩码
-
-#### Scenario: 获取不存在的供应商
-
-- **WHEN** 向 `/api/providers/:id` 发送 GET 请求，携带不存在的 ID
-- **THEN** 网关 SHALL 返回错误，状态码为 404 (Not Found)
-
-### Requirement: 更新供应商配置
-
-网关 SHALL 允许更新现有供应商配置。
-
-#### Scenario: 使用有效数据更新供应商
-
-- **WHEN** 向 `/api/providers/:id` 发送 PUT 请求，携带有效的供应商数据
-- **THEN** 网关 SHALL 更新数据库中的供应商记录
-- **THEN** 网关 SHALL 返回更新后的供应商
-- **THEN** 更新 SHALL 支持修改 protocol 字段
-
-### Requirement: 删除供应商配置
-
-网关 SHALL 允许删除供应商配置。
-
-#### Scenario: 删除存在的供应商
-
-- **WHEN** 向 `/api/providers/:id` 发送 DELETE 请求，携带有效的供应商 ID
-- **THEN** 网关 SHALL 删除供应商记录
-- **THEN** 网关 SHALL 删除所有关联的模型（CASCADE）
-- **THEN** 网关 SHALL 返回状态码 204 (No Content)

openspec/changes/refactor-conversion-engine/specs/request-validation/spec.md

@@ -1,64 +0,0 @@
-## MODIFIED Requirements
-
-### Requirement: 验证 OpenAI 请求
-
-系统 SHALL 验证 OpenAI ChatCompletionRequest，验证逻辑位于 ProtocolAdapter 的 decodeRequest 内。
-
-#### Scenario: 必需字段验证
-
-- **WHEN** OpenAI Adapter 的 decodeRequest 解析请求
-- **THEN** SHALL 验证 model 字段不为空
-- **THEN** SHALL 验证 messages 字段不为空且至少有一条消息
-- **THEN** 验证失败 SHALL 返回 INVALID_INPUT 类型的 ConversionError
-
-#### Scenario: 参数范围验证
-
-- **WHEN** OpenAI Adapter 的 decodeRequest 解析参数
-- **THEN** SHALL 验证 temperature 范围在 [0, 2]
-- **THEN** SHALL 验证 max_tokens 大于 0
-- **THEN** SHALL 验证 top_p 范围在 (0, 1]
-
-#### Scenario: 消息内容验证
-
-- **WHEN** 验证 messages 字段
-- **THEN** SHALL 验证每条消息的 role 有效（system、developer、user、assistant、tool）
-- **THEN** SHALL 验证 content 不为空
-
-### Requirement: 验证 Anthropic 请求
-
-系统 SHALL 验证 Anthropic MessagesRequest，验证逻辑位于 ProtocolAdapter 的 decodeRequest 内。
-
-#### Scenario: 必需字段验证
-
-- **WHEN** Anthropic Adapter 的 decodeRequest 解析请求
-- **THEN** SHALL 验证 model 字段不为空
-- **THEN** SHALL 验证 messages 字段不为空且至少有一条消息
-- **THEN** SHALL 验证 max_tokens 大于 0（或使用默认值）
-
-#### Scenario: 参数范围验证
-
-- **WHEN** Anthropic Adapter 的 decodeRequest 解析参数
-- **THEN** SHALL 验证 temperature 范围在 [0, 1]
-- **THEN** SHALL 验证 top_p 范围在 (0, 1]
-
-#### Scenario: 消息内容验证
-
-- **WHEN** 验证 messages 字段
-- **THEN** SHALL 验证每条消息的 role 有效（user、assistant）
-- **THEN** SHALL 验证 content 数组不为空
-
-### Requirement: 返回友好的验证错误
-
-系统 SHALL 返回友好的验证错误响应。
-
-#### Scenario: 转换错误格式
-
-- **WHEN** decodeRequest 验证失败返回 ConversionError
-- **THEN** ProxyHandler SHALL 使用 clientAdapter.encodeError 编码错误响应
-- **THEN** 错误 SHALL 使用客户端协议的格式
-
-#### Scenario: 多字段错误
-
-- **WHEN** 多个字段验证失败
-- **THEN** ConversionError.details SHALL 包含所有验证错误
-- **THEN** 错误响应 SHALL 包含完整的验证错误信息

openspec/changes/refactor-conversion-engine/tasks.md

@@ -1,49 +0,0 @@
-## 1. 基础类型层 — Canonical Model 和核心类型定义
-
-- [x] 1.1 创建 `internal/conversion/errors.go`：定义 ConversionError 结构体（Code, Message, ClientProtocol, ProviderProtocol, InterfaceType, Details, Cause）和 ErrorCode 枚举（INVALID_INPUT, MISSING_REQUIRED_FIELD, INCOMPATIBLE_FEATURE, FIELD_MAPPING_FAILURE, TOOL_CALL_PARSE_ERROR, JSON_PARSE_ERROR, STREAM_STATE_ERROR, UTF8_DECODE_ERROR, PROTOCOL_CONSTRAINT_VIOLATION, ENCODING_FAILURE, INTERFACE_NOT_SUPPORTED），实现 error 接口
-- [x] 1.2 创建 `internal/conversion/interface.go`：定义 InterfaceType 枚举（CHAT, MODELS, MODEL_INFO, EMBEDDINGS, RERANK）
-- [x] 1.3 创建 `internal/conversion/provider.go`：定义 TargetProvider 结构体（BaseURL, APIKey, ModelName, AdapterConfig map[string]any）；编写测试
-- [x] 1.4 创建 `internal/conversion/canonical/types.go`：定义 CanonicalRequest（model, system, messages, tools, tool_choice, parameters, thinking, stream, user_id, output_format, parallel_tool_use）、CanonicalMessage（role 枚举: system/user/assistant/tool, content []ContentBlock）、ContentBlock（使用 type 字段的 discriminated union：text/tool_use/tool_result/thinking，ToolInput 使用 json.RawMessage）、CanonicalTool（name, description, input_schema）、ToolChoice 联合体（auto/none/any/tool+name）、RequestParameters（max_tokens, temperature, top_p, top_k, frequency_penalty, presence_penalty, stop_sequences）、ThinkingConfig（type: enabled/disabled/adaptive, budget_tokens, effort）、OutputFormat（json_object/json_schema+schema/text）、CanonicalResponse（id, model, content, stop_reason 枚举, usage）、CanonicalUsage（input_tokens, output_tokens, cache_read_tokens, cache_creation_tokens, reasoning_tokens）、SystemBlock（text）；编写构造和序列化测试
-- [x] 1.5 创建 `internal/conversion/canonical/stream.go`：定义 CanonicalStreamEvent 联合体（message_start, content_block_start, content_block_delta, content_block_stop, message_delta, message_stop, error, ping）及各事件的具体结构（MessageStartEvent 含 message{id,model,usage}、ContentBlockStartEvent 含 index 和 content_block、ContentBlockDeltaEvent 含 index 和 delta、ContentBlockStopEvent 含 index、MessageDeltaEvent 含 delta{stop_reason} 和 usage、MessageStopEvent、ErrorEvent、PingEvent），delta 联合体（text_delta, input_json_delta, thinking_delta），content_block 联合体（text, tool_use, thinking）；编写测试
-- [x] 1.6 创建 `internal/conversion/canonical/extended.go`：定义扩展层 Canonical Models（CanonicalModelList, CanonicalModel, CanonicalModelInfo, CanonicalEmbeddingRequest, CanonicalEmbeddingResponse, CanonicalRerankRequest, CanonicalRerankResponse）；编写测试
-
-## 2. 接口定义层 — Adapter、Stream、Middleware 接口
-
-- [x] 2.1 创建 `internal/conversion/adapter.go`：定义 ProtocolAdapter 接口（protocolName, protocolVersion, supportsPassthrough, detectInterfaceType, buildUrl, buildHeaders, supportsInterface, decodeRequest, encodeRequest, decodeResponse, encodeResponse, createStreamDecoder, createStreamEncoder, encodeError, 扩展层编解码方法：decodeModelsResponse/encodeModelsResponse/decodeModelInfoResponse/encodeModelInfoResponse/decodeEmbeddingRequest/encodeEmbeddingRequest/decodeEmbeddingResponse/encodeEmbeddingResponse/decodeRerankRequest/encodeRerankRequest/decodeRerankResponse/encodeRerankResponse），定义 AdapterRegistry 接口（register, get, listProtocols）和 memoryRegistry 实现（sync.RWMutex 保护的 map）；编写 Registry 注册/查询/重复注册测试
-- [x] 2.2 创建 `internal/conversion/stream.go`：定义 StreamDecoder 接口（processChunk(rawChunk []byte) []CanonicalStreamEvent, flush() []CanonicalStreamEvent）、StreamEncoder 接口（encodeEvent(event CanonicalStreamEvent) [][]byte, flush() [][]byte）、StreamConverter 接口（processChunk(rawChunk []byte) [][]byte, flush() [][]byte）、PassthroughStreamConverter 实现（直接传递原始字节）、CanonicalStreamConverter 实现（组合 StreamDecoder + MiddlewareChain + StreamEncoder，processChunk 内部调用 decoder → middleware → encoder 管道）；编写 PassthroughStreamConverter 测试
-- [x] 2.3 创建 `internal/conversion/middleware.go`：定义 ConversionMiddleware 接口（intercept(canonical, clientProtocol, providerProtocol, context) (CanonicalRequest, error) 和可选的 interceptStreamEvent(event, clientProtocol, providerProtocol, context) (CanonicalStreamEvent, error)）、ConversionContext 结构体（conversionId, interfaceType, timestamp, metadata）、MiddlewareChain 结构体（按注册顺序链式执行，任一返回错误则中断后续）；编写链式执行和中断测试
-
-## 3. 引擎层 — ConversionEngine 门面
-
-- [x] 3.1 创建 `internal/conversion/engine.go`：定义 HTTPRequestSpec（URL, Method string, Headers map[string]string, Body []byte）、HTTPResponseSpec（StatusCode int, Headers map[string]string, Body []byte）、ConversionEngine struct（registry, middlewareChain）；实现 registerAdapter、use、isPassthrough、convertHttpRequest（接口识别 → 透传判断 → clientAdapter.decode → middleware → providerAdapter.encode → providerAdapter.buildUrl + buildHeaders）、convertHttpResponse（透传判断 → providerAdapter.decodeResponse → clientAdapter.encodeResponse）、createStreamConverter（透传 → PassthroughStreamConverter，否则 → CanonicalStreamConverter）、内部 convertBody 分发（CHAT 走深度转换，扩展层走轻量映射，默认透传）；编写集成测试：使用 mock adapter 测试跨协议转换、同协议透传、未知接口透传
-
-## 4. OpenAI Adapter 实现
-
-- [x] 4.1 创建 `internal/conversion/openai/types.go`：对照 `docs/conversion_openai.md` 全新定义 OpenAI 线路格式类型（不沿用旧 `internal/protocol/openai/types.go`），包含完整字段（developer role, custom tools, reasoning_effort, reasoning_content, max_completion_tokens, parallel_tool_calls, response_format 的 json_schema 类型, stream_options, 废弃的 functions/function_call）；编写序列化测试
-- [x] 4.2 创建 `internal/conversion/openai/decoder.go`：实现 decodeRequest（对照 conversion_openai.md §4.1：decodeSystemPrompt 提取 system+developer 消息、decodeMessage 含 tool_calls/refusal/reasoning_content 解码、tool 消息 tool_call_id→tool_use_id、decodeTools 含 function+custom 类型、decodeToolChoice 含 required→any/allowed_tools 降级、decodeParameters 含 max_completion_tokens 优先、decodeOutputFormat、decodeThinking 含 reasoning_effort→ThinkingConfig、废弃字段 functions→tools 兼容）、decodeResponse（§5.2：content/refusal/reasoning_content/tool_calls 解码、finish_reason 映射表、usage 映射含 cached_tokens/reasoning_tokens）、扩展层 decode（decodeModelsResponse、decodeEmbeddingRequest/Response、decodeRerankRequest/Response）；编写完整测试覆盖每类消息和字段映射
-- [x] 4.3 创建 `internal/conversion/openai/encoder.go`：实现 encodeRequest（对照 conversion_openai.md §4.2：provider.model_name 覆盖、system 注入到 messages[0]、encodeMessage 含 tool_calls 编码到 message 顶层、角色交替合并、encodeTools 含 function 包装、encodeToolChoice 含 any→required、encodeParameters 含 max_completion_tokens、encodeOutputFormat、encodeThinking 含 disabled→"none"）、encodeResponse（§5.3：text→content、tool_use→tool_calls、thinking→reasoning_content、finish_reason 反向映射、usage 编码含 prompt_tokens_details）、扩展层 encode（encodeModelsResponse、encodeEmbeddingRequest/Response、encodeRerankRequest/Response）；编写完整测试
-- [x] 4.4 创建 `internal/conversion/openai/adapter.go`：实现 OpenAI ProtocolAdapter（protocolName→"openai"、supportsPassthrough→true、detectInterfaceType 根据正则匹配识别 /v1/chat/completions→CHAT、/v1/models→MODELS 等、buildHeaders 含 Authorization+Content-Type+OpenAI-Organization、buildUrl 按接口类型映射、supportsInterface 对 CHAT/MODELS/MODEL_INFO/EMBEDDINGS/RERANK 返回 true、encodeError 含 ErrorCode→OpenAI 错误类型映射），组合 decoder 和 encoder 方法；编写测试覆盖所有路径模式和边界情况
-- [x] 4.5 创建 `internal/conversion/openai/stream_decoder.go`：实现 OpenAIStreamDecoder（对照 conversion_openai.md §6.2-§6.3：processChunk 解析 SSE data 行，维护状态机 messageStarted/openBlocks/toolCallIdMap/toolCallNameMap/toolCallArguments/textBlockStarted/thinkingBlockStarted/utf8Remainder/accumulatedUsage，首个 chunk→MessageStartEvent，delta.content→text block 生命周期，delta.tool_calls→tool_use block 生命周期含索引映射和参数累积，delta.reasoning_content→thinking block（非标准），delta.refusal→text block，finish_reason→关闭所有 open blocks + MessageDeltaEvent + MessageStopEvent，usage chunk→MessageDeltaEvent，[DONE]→flush 关闭）；编写测试覆盖每种 delta 类型和边界情况（空 chunk、多 tool_calls、UTF-8 截断）
-- [x] 4.6 创建 `internal/conversion/openai/stream_encoder.go`：实现 OpenAIStreamEncoder（对照 conversion_openai.md §6.4：encodeEvent，ContentBlockStart 缓冲策略等待首次 ContentBlockDelta 合并输出，tool_use id/name 在首次 delta 时合并编码，text_delta 直接输出 data: {choices:[{delta:{content}}]}，input_json_delta 含 tool_calls 数组编码，thinking_delta 含 reasoning_content 字段，MessageStartEvent→{choices:[{delta:{role:"assistant"}}]}，MessageDeltaEvent→{choices:[{delta:{},finish_reason}]}，MessageStopEvent→[DONE]，PingEvent/ErrorEvent 丢弃，flush 输出缓冲区）；编写测试
-
-## 5. Anthropic Adapter 实现（与 Layer 4 并行）
-
-- [x] 5.1 创建 `internal/conversion/anthropic/types.go`：对照 `docs/conversion_anthropic.md` 全新定义 Anthropic 线路格式类型（不沿用旧 `internal/protocol/anthropic/types.go`），包含完整字段（thinking.type 含 adaptive、output_config.format/effort、disable_parallel_tool_use、metadata.user_id、redacted_thinking、pause_turn/refusal stop_reason、stop_details、container、cache_control）；编写序列化测试
-- [x] 5.2 创建 `internal/conversion/anthropic/decoder.go`：实现 decodeRequest（对照 conversion_anthropic.md §4.1：decodeSystem 从顶层 system 提取、decodeMessage 含 tool_result 从 user 消息拆分为独立 tool 角色消息、参数直接映射含 top_k、decodeThinking 含 enabled/disabled/adaptive 三种类型、decodeOutputFormat 仅支持 json_schema、公共字段提取含 metadata.user_id/disable_parallel_tool_use 反转/output_config.effort、协议特有字段 redacted_thinking 丢弃/cache_control 忽略）、decodeResponse（§5.2：text/tool_use/thinking 块解码、redacted_thinking 丢弃、stop_reason 映射含 pause_turn/refusal、usage 映射含 cache_read_input_tokens/cache_creation_input_tokens）、扩展层 decode（decodeModelsResponse 含 RFC3339→Unix 时间戳转换、decodeModelInfoResponse）；编写完整测试覆盖角色拆分、thinking 三种类型、时间戳转换
-- [x] 5.3 创建 `internal/conversion/anthropic/encoder.go`：实现 encodeRequest（对照 conversion_anthropic.md §4.2：provider.model_name 覆盖、system 注入为顶层字段、encodeMessages 含 tool→user 合并（优先合并到相邻 user 消息）、首消息 user 保证（自动注入空 user）、角色交替合并、encodeThinkingConfig 含 enabled/disabled/adaptive、encodeOutputFormat 含 json_object→空 schema 降级/text 丢弃、公共字段编码含 metadata.user_id/disable_parallel_tool_use 反转/output_config、参数编码含 max_tokens 必填/top_k 直接映射）、encodeResponse（§5.3：text/tool_use/thinking 块直接编码、stop_reason 映射含 content_filter→end_turn 降级、usage 编码含 cache_read_input_tokens/cache_creation_input_tokens）、扩展层 encode（encodeModelsResponse 含 Unix→RFC3339 转换和 has_more/first_id/last_id 字段、encodeModelInfoResponse）；编写完整测试覆盖角色合并、首消息注入、降级处理
-- [x] 5.4 创建 `internal/conversion/anthropic/adapter.go`：实现 Anthropic ProtocolAdapter（protocolName→"anthropic"、supportsPassthrough→true、detectInterfaceType 根据正则匹配识别 /v1/messages→CHAT、/v1/models→MODELS 等、buildHeaders 含 x-api-key + anthropic-version + anthropic-beta + Content-Type、buildUrl 按接口类型映射、supportsInterface 对 CHAT/MODELS/MODEL_INFO 返回 true 对 EMBEDDINGS/RERANK 返回 false、encodeError 返回 {type:"error",error:{type,message}}）；编写测试覆盖所有路径模式和边界情况
-- [x] 5.5 创建 `internal/conversion/anthropic/stream_decoder.go`：实现 AnthropicStreamDecoder（对照 conversion_anthropic.md §6.2-§6.3：解析命名 SSE 事件 event: message_start/data: {...}，1:1 映射到 CanonicalStreamEvent，维护状态 messageStarted/redactedBlocks/utf8Remainder/accumulatedUsage，redacted_thinking 检测后加入 redactedBlocks 并丢弃后续 delta/stop，citations_delta/signature_delta 直接丢弃，server_tool_use 等服务端工具块丢弃，UTF-8 跨 chunk 安全处理）；编写测试覆盖所有事件类型和 redacted_thinking 丢弃
-- [x] 5.6 创建 `internal/conversion/anthropic/stream_encoder.go`：实现 AnthropicStreamEncoder（对照 conversion_anthropic.md §6.4：直接映射无缓冲，每个 CanonicalStreamEvent 直接编码为对应的 Anthropic 命名 SSE 事件，格式 event: `<type>`\ndata: `<json>`\n\n，delta 编码 text_delta/input_json_delta/thinking_delta 直接映射）；编写测试
-
-## 6. 基础设施改造 — Provider、Handler、Domain
-
-- [x] 6.1 修改 `internal/domain/provider.go`：Provider 结构体新增 Protocol string 字段；修改 `internal/config/models.go`：GORM Provider 模型同步新增 Protocol 字段（gorm:"column:protocol;default:'openai'"）；修改 `internal/repository/` 中 toDomainProvider 和 toConfigProvider 转换函数同步 Protocol 字段；修改 `internal/handler/provider_handler.go`：CreateProvider 和 UpdateProvider 的请求结构体新增 Protocol 字段（可选，默认 "openai"），创建/更新 Provider 时赋值 Protocol 字段，List/Get 响应中包含 Protocol 字段；更新 `internal/service/service_test.go` 中所有创建测试 Provider 的地方补充 Protocol 字段；更新 `internal/handler/handler_test.go` 中 Provider CRUD 测试的请求体补充 Protocol 字段；创建数据库迁移文件 `backend/migrations/YYYYMMDDHHMMSS_add_provider_protocol.sql`：ALTER TABLE providers ADD COLUMN protocol TEXT DEFAULT 'openai'
-- [x] 6.2 重写 `internal/provider/client.go`：定义 HTTPRequestSpec 和 HTTPResponseSpec（或引用 conversion 包的定义），简化 ProviderClient 接口为 Send(ctx, HTTPRequestSpec) → (*HTTPResponseSpec, error) 和 SendStream(ctx, HTTPRequestSpec) → (<-chan StreamEvent, error)，移除所有旧协议硬编码依赖，Send 方法直接使用 http.NewRequest + spec.URL/Headers/Body，SendStream 保留现有 readStream goroutine 逻辑但输入改为 HTTPRequestSpec；重写 `provider/client_test.go`：删除所有基于旧协议类型的测试用例，基于 HTTPRequestSpec 重写成功/失败/流式测试用例，使用 httptest.Server 验证请求构建和响应解析
-- [x] 6.3 创建 `internal/handler/proxy_handler.go`：实现 ProxyHandler struct（依赖 ConversionEngine、ProviderClient、RoutingService、StatsService），实现 HandleProxy(w, r) 方法：从 URL 提取 clientProtocol（仅支持 `/{protocol}/v1/...` 前缀路由，不支持旧路由）、解析请求体 JSON、调用 RoutingService.Route(modelName) 获取路由结果（含 Provider.Protocol 作为 providerProtocol）、构建 TargetProvider、调用 engine.convertHttpRequest、调用 providerClient.Send/SendStream、调用 engine.convertHttpResponse、设置响应 Content-Type 和状态码、流式处理设置 text/event-stream 并用 StreamConverter 逐块转换写入、错误处理使用 clientAdapter.encodeError、异步调用 StatsService.Record；编写测试使用 httptest + mock engine/client/service
-- [x] 6.4 修改 `cmd/server/main.go`：创建 AdapterRegistry 并注册 OpenAI 和 Anthropic Adapter、创建 ConversionEngine（注入 registry）、创建 ProxyHandler（注入 engine + providerClient + routingService + statsService）、配置 Gin 路由：新增 `/{protocol}/v1/{path:*}` → ProxyHandler.HandleProxy，删除旧路由 `/v1/chat/completions` 和 `/v1/messages`，移除旧的 OpenAIHandler 和 AnthropicHandler 的路由注册，删除旧 Adapter 创建代码
-
-## 7. 清理和文档
-
-- [x] 7.1 删除旧代码：删除 `internal/protocol/openai/` 目录（types.go, adapter.go, adapter_test.go）、删除 `internal/protocol/anthropic/` 目录（types.go, converter.go, converter_test.go, stream_converter.go, stream_converter_test.go）、删除 `internal/handler/openai_handler.go` 和 `internal/handler/anthropic_handler.go`、删除 `internal/handler/handler_test.go` 中旧 OpenAI/Anthropic handler 测试用例和旧 `mockProviderClient`（基于旧协议类型的签名）、重写 `handler_test.go` 为 ProxyHandler 测试（基于新 ProviderClient 接口和 ConversionEngine mock）、删除 `internal/protocol/` 空目录、确认所有编译通过且无残留 import
-- [x] 7.2 更新 `README.md`：更新项目结构说明（新增 internal/conversion/、删除 internal/protocol/）、更新 API 接口说明（代理接口变更：`/{protocol}/v1/...`，移除旧路由 `/v1/chat/completions` 和 `/v1/messages`）、更新配置说明（Provider 新增 protocol 字段）
-- [x] 7.3 端到端测试：在 `backend/tests/integration/` 中新增 `conversion_test.go`，使用 httptest mock 上游服务器验证完整请求流：OpenAI→OpenAI 同协议透传、Anthropic→Anthropic 同协议透传、OpenAI→Anthropic 跨协议非流式、Anthropic→OpenAI 跨协议非流式、4 种方向的流式转换（含 tool_calls 和 thinking）、Models 接口跨协议转换、错误响应格式验证（各协议格式）、旧路由 `/v1/chat/completions` 和 `/v1/messages` 返回 404；复用 `tests/helpers.go` 中的测试数据库和 Provider/Model 创建辅助函数

openspec/specs/anthropic-protocol-proxy/spec.md

@@ -4,42 +4,47 @@
 
 ### Requirement: 支持 Anthropic Messages API 端点
 
-网关 SHALL 提供 Anthropic Messages API 端点 `POST /v1/messages` 供外部应用调用。
+网关 SHALL 提供 Anthropic Messages API 端点供外部应用调用。
 
 #### Scenario: 成功的非流式请求
 
-- **WHEN** 应用发送 POST 请求到 `/v1/messages`，携带有效的 Anthropic 请求格式（非流式）
-- **THEN** 网关 SHALL 将 Anthropic 请求转换为 OpenAI 格式
-- **THEN** 网关 SHALL 将转换后的请求转发到配置的供应商
-- **THEN** 网关 SHALL 将 OpenAI 响应转换回 Anthropic 格式
-- **THEN** 网关 SHALL 将转换后的响应返回给应用
+- **WHEN** 应用发送 POST 请求到 `/anthropic/v1/messages`，携带有效的 Anthropic 请求格式（非流式）
+- **THEN** 网关 SHALL 通过 ConversionEngine 将 Anthropic 请求解码为 Canonical 格式
+- **THEN** 网关 SHALL 将 Canonical 请求编码为目标供应商协议格式
+- **THEN** 网关 SHALL 将供应商的响应通过 ConversionEngine 转换为 Anthropic 格式返回给应用
 
 #### Scenario: 成功的流式请求
 
-- **WHEN** 应用发送 POST 请求到 `/v1/messages`，携带 `stream: true`
-- **THEN** 网关 SHALL 将 Anthropic 请求转换为 OpenAI 格式
-- **THEN** 网关 SHALL 将转换后的请求转发给供应商
-- **THEN** 网关 SHALL 将 OpenAI 流事件转换为 Anthropic 流事件
-- **THEN** 网关 SHALL 使用 SSE 格式将转换后的事件流式返回给应用
+- **WHEN** 应用发送 POST 请求到 `/anthropic/v1/messages`，携带 `stream: true`
+- **THEN** 网关 SHALL 通过 ConversionEngine 创建 StreamConverter
+- **THEN** 网关 SHALL 将上游协议的 SSE 流转换为 Anthropic 命名事件格式
+- **THEN** 网关 SHALL 使用 `event: <type>\ndata: <json>\n\n` 格式流式返回给应用
 
-**变更说明：** handler 通过 service 层调用，而非直接调用 config 和 provider 包。API 接口保持不变。
+#### Scenario: 同协议透传（Anthropic → Anthropic Provider）
 
-### Requirement: 将 Anthropic 请求转换为 OpenAI 格式
+- **WHEN** 客户端使用 Anthropic 协议且目标供应商也是 Anthropic 协议
+- **THEN** 网关 SHALL 跳过 Canonical 转换，仅重建认证 Header 后原样转发
+- **THEN** 请求和响应 Body SHALL 保持原样
 
-网关 SHALL 将 Anthropic Messages API 请求转换为 OpenAI Chat Completions API 格式。
+### Requirement: 双向协议转换
 
-#### Scenario: System 消息转换
+网关 SHALL 支持 Anthropic 协议与任意已注册协议间的双向转换。
 
-- **WHEN** Anthropic 请求包含 `system` 字段
-- **THEN** 网关 SHALL 将其转换为 `messages` 数组中 `role: "system"` 的消息
+#### Scenario: Anthropic 客户端 → OpenAI 供应商
 
-#### Scenario: Messages 转换
+- **WHEN** 客户端使用 Anthropic 协议且供应商使用 OpenAI 协议
+- **THEN** SHALL 将 Anthropic MessagesRequest 解码为 CanonicalRequest
+- **THEN** SHALL 将 CanonicalRequest 编码为 OpenAI ChatCompletionRequest
+- **THEN** SHALL 将 OpenAI ChatCompletionResponse 解码为 CanonicalResponse
+- **THEN** SHALL 将 CanonicalResponse 编码为 Anthropic MessagesResponse
 
-- **WHEN** Anthropic 请求包含 `messages` 数组
-- **THEN** 网关 SHALL 在转换后的 OpenAI 请求中保留这些消息
-- **THEN** 网关 SHALL 保留每条消息的 role 和 content
+#### Scenario: OpenAI 客户端 → Anthropic 供应<E4BE9B><E5BA94>
 
-**变更说明：** 协议转换逻辑保持不变，仅调用方式改为通过 service 层。
+- **WHEN** 客户端使用 OpenAI 协议且供应商使用 Anthropic 协议
+- **THEN** SHALL 将 OpenAI ChatCompletionRequest 解码为 CanonicalRequest
+- **THEN** SHALL 将 CanonicalRequest 编码为 Anthropic MessagesRequest
+- **THEN** SHALL 将 Anthropic MessagesResponse 解码为 CanonicalResponse
+- **THEN** SHALL 将 CanonicalResponse 编码为 OpenAI ChatCompletionResponse
 
 ## ADDED Requirements
 
@@ -49,9 +54,9 @@ Handler SHALL 通过 service 层处理业务逻辑。
 
 #### Scenario: 调用 routing service
 
-- **WHEN** handler 收到请求并转换为 OpenAI 格式
+- **WHEN** ProxyHandler 收到 Anthropic 协议请求
 - **THEN** SHALL 调用 RoutingService.Route() 获取路由结果
-- **THEN** SHALL 使用路由结果中的供应商信息
+- **THEN** SHALL 从路由结果获取 Provider（含 protocol 字段）
 
 #### Scenario: 调用 stats service
 
@@ -61,16 +66,22 @@ Handler SHALL 通过 service 层处理业务逻辑。
 
 ### Requirement: 使用结构化错误处理
 
-Handler SHALL 使用结构化错误处理。
+ProxyHandler SHALL 使用 ConversionError 和 Anthropic 的 encodeError 处理错误。
 
 #### Scenario: 协议转换错误
 
-- **WHEN** 协议转换失败
-- **THEN** SHALL 返回结构化错误响应
-- **THEN** SHALL 包含详细的错误信息
+- **WHEN** ConversionEngine 返回 ConversionError
+- **THEN** SHALL 使用 Anthropic 的 Adapter.encodeError 编码错误响应
+- **THEN** SHALL 使用 Anthropic 错误格式（`{type: "error", error: {type, message}}`）
 
 #### Scenario: 路由错误处理
 
 - **WHEN** RoutingService 返回错误
-- **THEN** SHALL 转换为对应的 AppError
-- **THEN** SHALL 返回统一的错误响应
+- **THEN** SHALL 转换为 ConversionError
+- **THEN** SHALL 使用 Anthropic 错误格式返回
+
+#### Scenario: 供应商错误处理
+
+- **WHEN** ProviderClient 返回错误
+- **THEN** SHALL 包装为 ConversionError
+- **THEN** SHALL 使用 Anthropic 错误格式返回

openspec/specs/conversion-engine/spec.md

@@ -1,3 +1,5 @@
+# Conversion Engine
+
 ## ADDED Requirements
 
 ### Requirement: 定义 CanonicalRequest 规范模型
@@ -264,7 +266,7 @@ ErrorCode SHALL 包含：INVALID_INPUT、MISSING_REQUIRED_FIELD、INCOMPATIBLE_F
 - **THEN** SHALL 使用对应扩展层 Canonical Model 做轻量字段映射
 - **THEN** 双方都不支持时 SHALL 走透传逻辑
 
-### Requirement: 定义 TargetProvider 结构体
+### Requirement: <EFBFBD><EFBFBD>义 TargetProvider 结构体
 
 系统 SHALL 定义 `TargetProvider` 结构体，包含 `base_url`、`api_key`、`model_name`、`adapter_config`。
 

openspec/specs/error-handling/spec.md

@@ -29,6 +29,12 @@
 - **THEN** SHALL 使用 ErrModelNotFound、ErrProviderNotFound 等预定义错误
 - **THEN** SHALL 设置 HTTP 状态码为 404
 
+#### Scenario: 转换错误响应
+
+- **WHEN** ConversionEngine 在协议转换过程中产生 ConversionError
+- **THEN** SHALL 使用客户端协议的 Adapter.encodeError 编码错误响应
+- **THEN** 错误响应 SHALL 使用客户端可理解的协议格式
+
 #### Scenario: 验证错误
 
 - **WHEN** 请求验证失败
@@ -120,3 +126,26 @@
 - **WHEN** repository 层发生错误
 - **THEN** SHALL 包装数据库错误
 - **THEN** SHALL 转换为应用错误
+
+## ADDED Requirements
+
+### Requirement: 定义 ConversionError 错误类型
+
+系统 SHALL 定义 ConversionError 结构体和 ErrorCode 枚举。
+
+#### Scenario: ConversionError 结构
+
+- **WHEN** 定义转换错误
+- **THEN** SHALL 包含 Code（ErrorCode 枚举）、Message 字段
+- **THEN** SHALL 可选包含 ClientProtocol、ProviderProtocol、InterfaceType、Details、Cause 字段
+
+#### Scenario: ErrorCode 枚举
+
+- **WHEN** 定义错误码
+- **THEN** SHALL 包含 INVALID_INPUT、MISSING_REQUIRED_FIELD、INCOMPATIBLE_FEATURE、FIELD_MAPPING_FAILURE、TOOL_CALL_PARSE_ERROR、JSON_PARSE_ERROR、STREAM_STATE_ERROR、UTF8_DECODE_ERROR、PROTOCOL_CONSTRAINT_VIOLATION、ENCODING_FAILURE、INTERFACE_NOT_SUPPORTED
+
+#### Scenario: 错误码到协议错误类型的映射
+
+- **WHEN** 使用 encodeError 编码错误
+- **THEN** ErrorCode SHALL 映射为各协议的错误类型字符串
+- **THEN** 例如 INVALID_INPUT → OpenAI "invalid_request_error"，Anthropic "invalid_request_error"

openspec/specs/layered-architecture/spec.md

@@ -4,28 +4,37 @@
 
 ### Requirement: 实现三层架构
 
-系统 SHALL 实现 handler → service → repository 三层架构。
+系统 SHALL 实现 handler → service → repository 三层架构，并在 handler 和 provider 之间新增 conversion 层。
 
 #### Scenario: Handler 层职责
 
 - **WHEN** 处理 HTTP 请求
-- **THEN** handler 层 SHALL 仅负责 HTTP 请求解析和响应
+- **THEN** handler 层 SHALL 仅负责 HTTP 请求解析、URL 路由和响应写入
+- **THEN** handler 层 SHALL 调用 ConversionEngine 处理协议转换
 - **THEN** handler 层 SHALL 调用 service 层处理业务逻辑
-- **THEN** handler 层 SHALL NOT 直接访问数据库
+- **THEN** handler 层 SHALL NOT 直接访问数据库或执行协议转换逻辑
+
+#### Scenario: Conversion 层职责
+
+- **WHEN** 处理协议转换
+- **THEN** conversion 层 SHALL 包含 Canonical Model 定义
+- **THEN** conversion 层 SHALL 包含各协议的 ProtocolAdapter 实现
+- **THEN** conversion 层 SHALL 包含 ConversionEngine 门面
+- **THEN** conversion 层 SHALL NOT 依赖 handler 或 service 层
 
 #### Scenario: Service 层职责
 
 - **WHEN** 处理业务逻辑
 - **THEN** service 层 SHALL 包含业务规则和验证
 - **THEN** service 层 SHALL 调用 repository 层访问数据
-- **THEN** service 层 SHALL 协调多个 repository 的操作
+- **THEN** service 层 SHALL NOT 包含协议转换逻辑
 
 #### Scenario: Repository 层职责
 
 - **WHEN** 访问数据
 - **THEN** repository 层 SHALL 仅负责数据访问
 - **THEN** repository 层 SHALL 封装数据库操作
-- **THEN** repository 层 SHALL NOT 包含业务逻辑
+- **THEN** repository 层 SHALL NOT 包含业务逻辑或协议转换逻辑
 
 ### Requirement: 定义核心接口
 
@@ -49,9 +58,17 @@
 
 - **WHEN** 定义 provider client 接口
 - **THEN** SHALL 定义 ProviderClient 接口
-- **THEN** SHALL 包含 SendRequest 和 SendStreamRequest 方法
+- **THEN** SHALL 包含 Send（非流式）和 SendStream（流式）方法
+- **THEN** SHALL 接受 HTTPRequestSpec 作为参数，不绑定特定协议
 - **THEN** SHALL 支持接口 Mock
 
+#### Scenario: Conversion 层接口定义
+
+- **WHEN** 定义 conversion 层接口
+- **THEN** SHALL 定义 ProtocolAdapter、StreamDecoder、StreamEncoder、StreamConverter、ConversionMiddleware 接口
+- **THEN** SHALL 定义 AdapterRegistry 用于 Adapter 注册和查询
+- **THEN** SHALL 定义 ConversionEngine 作为统一门面
+
 ### Requirement: 实现依赖注入
 
 系统 SHALL 使用手动依赖注入。
@@ -65,15 +82,24 @@
 #### Scenario: Service 注入
 
 - **WHEN** 初始化 handler
-- **THEN** SHALL 通过构造函数注入 service 依赖
+- **THEN** SHALL 通过构造函数注入 service 依赖、ConversionEngine、ProviderClient
 - **THEN** SHALL 使用接口类型而非具体类型
 
+#### Scenario: Conversion 组装
+
+- **WHEN** 应用启动
+- **THEN** SHALL 创建 AdapterRegistry 并注册所有 ProtocolAdapter
+- **THEN** SHALL 创建 ConversionEngine（注入 registry 和 middleware chain）
+- **THEN** SHALL 将 ConversionEngine 注入到 ProxyHandler
+
 #### Scenario: 主函数组装
 
 - **WHEN** 应用启动
 - **THEN** main.go SHALL 按顺序构造所有依赖
 - **THEN** SHALL 先构造基础设施（logger、database）
-- **THEN** SHALL 再构造 repository、service、handler
+- **THEN** SHALL 再构造 repository、service
+- **THEN** SHALL 再构造 conversion 层（registry → engine）
+- **THEN** SHALL 最后构造 handler
 
 ### Requirement: 定义 Domain 模型
 
@@ -84,6 +110,7 @@
 - **WHEN** 定义领域模型
 - **THEN** SHALL 在 internal/domain/ 包中定义
 - **THEN** SHALL 包含 Provider、Model、UsageStats 等模型
+- **THEN** Provider SHALL 包含 Protocol 字段
 - **THEN** SHALL 与数据库模型分离
 
 #### Scenario: Domain 模型使用

openspec/specs/openai-protocol-proxy/spec.md

@@ -4,22 +4,27 @@
 
 ### Requirement: 支持 OpenAI Chat Completions API 端点
 
-网关 SHALL 提供 OpenAI Chat Completions API 端点 `POST /v1/chat/completions` 供外部应用调用。
+网关 SHALL 提供 OpenAI Chat Completions API 端点供外部应用调用。
 
 #### Scenario: 成功的非流式请求
 
-- **WHEN** 应用发送 POST 请求到 `/v1/chat/completions`，携带有效的 OpenAI 请求格式（非流式）
-- **THEN** 网关 SHALL 将请求转发到配置的供应商
-- **THEN** 网关 SHALL 将供应商的响应以 OpenAI 格式返回给应用
+- **WHEN** 应用发送 POST 请求到 `/openai/v1/chat/completions`，携带有效的 OpenAI 请求格式（非流式）
+- **THEN** 网关 SHALL 通过 ConversionEngine 转换请求
+- **THEN** 网关 SHALL 将转换后的请求转发到配置的供应商
+- **THEN** 网关 SHALL 将供应商的响应通过 ConversionEngine 转换为 OpenAI 格式返回给应用
 
 #### Scenario: 成功的流式请求
 
-- **WHEN** 应用发送 POST 请求到 `/v1/chat/completions`，携带 `stream: true`
-- **THEN** 网关 SHALL 将请求转发到配置的供应商
-- **THEN** 网关 SHALL 使用 SSE 格式将响应流式返回给应用
+- **WHEN** 应用发送 POST 请求到 `/openai/v1/chat/completions`，携带 `stream: true`
+- **THEN** 网关 SHALL 通过 ConversionEngine 创建 StreamConverter
+- **THEN** 网关 SHALL 使用 SSE 格式将转换后的响应流式返回给应用
 - **THEN** 网关 SHALL 在流完成时发送 `data: [DONE]`
 
-**变更说明：** handler 通过 service 层调用，而非直接调用 config 和 provider 包。API 接口保持不变。
+#### Scenario: 同协议透传（OpenAI → OpenAI Provider）
+
+- **WHEN** 客户端使用 OpenAI 协议且目标供应商也是 OpenAI 协议
+- **THEN** 网关 SHALL 跳过 Canonical 转换，仅重建认证 Header 后原样转发
+- **THEN** 请求和响应 Body SHALL 保持原样
 
 ### Requirement: 根据模型名称路由请求
 
@@ -30,38 +35,32 @@
 - **WHEN** 请求包含存在于配置模型中的 `model` 字段
 - **AND** 该模型已启用
 - **THEN** 网关 SHALL 将请求路由到该模型关联的供应商
+- **THEN** 网关 SHALL 从供应商的 `protocol` 字段获取 providerProtocol
 
 #### Scenario: 模型未找到
 
 - **WHEN** 请求包含不存在于配置模型中的 `model` 字段
-- **THEN** 网关 SHALL 返回带有适当错误消息的错误响应
+- **THEN** 网关 SHALL 使用 OpenAI 格式返回错误响应
 
 #### Scenario: 模型已禁用
 
 - **WHEN** 请求包含已禁用模型的 `model` 字段
-- **THEN** 网关 SHALL 返回错误响应，指示模型不可用
-
-**变更说明：** 路由逻辑从 router 包迁移到 RoutingService，通过 service 层调用。API 接口保持不变。
+- **THEN** 网关 SHALL 使用 OpenAI 格式返回错误响应
 
 ### Requirement: 对 OpenAI 兼容供应商透明代理
 
-网关 SHALL 对 OpenAI 兼容供应商的请求和响应进行透明转发，不做修改。
+网关 SHALL 对 OpenAI 兼容供应商的请求和响应通过 ConversionEngine 进行转换处理。
 
-#### Scenario: 请求转发
+#### Scenario: 跨协议请求转发
 
-- **WHEN** 网关收到 OpenAI 协议请求
-- **AND** 目标供应商是 OpenAI 兼容的
-- **THEN** 网关 SHALL 将请求体原样转发给供应商
-- **THEN** 网关 SHALL 在 Authorization 头中设置供应商的 API Key
-- **THEN** 网关 SHALL 使用供应商的 base URL
+- **WHEN** 网关收到 OpenAI 协议请求且目标供应商使用不同协议
+- **THEN** 网关 SHALL 通过 ConversionEngine 将请求转换为目标协议格式
+- **THEN** 网关 SHALL 使用目标协议的 Adapter 构建 URL 和 Header
 
-#### Scenario: 响应转发
+#### Scenario: 扩展层接口代理
 
-- **WHEN** 供应商返回响应
-- **THEN** 网关 SHALL 将响应体原样返回给应用
-- **THEN** 网关 SHALL 保留所有响应头和状态码
-
-**变更说明：** provider client 通过接口注入到 handler，便于测试和替换实现。API 接口保持不变。
+- **WHEN** 网关收到 `/openai/v1/models` 等 GET 请求
+- **THEN** 网关 SHALL 通过 ConversionEngine 转换扩展层接口的响应格式
 
 ## ADDED Requirements
 
@@ -81,18 +80,40 @@ Handler SHALL 通过 service 层处理业务逻辑。
 - **THEN** SHALL 调用 StatsService.Record() 记录统计
 - **THEN** SHALL 异步记录统计（不阻塞响应）
 
+### Requirement: 使用 service 层处理请求
+
+Handler SHALL 通过 service 层处理业务逻辑。
+
+#### Scenario: 调用 routing service
+
+- **WHEN** ProxyHandler 收到请求
+- **THEN** SHALL 调用 RoutingService.Route() 获取路由结果
+- **THEN** SHALL 从路由结果获取 Provider（含 protocol 字段）
+
+#### Scenario: 调用 stats service
+
+- **WHEN** 请求成功完成
+- **THEN** SHALL 调用 StatsService.Record() 记录统计
+- **THEN** SHALL 异步记录统计（不阻塞响应）
+
 ### Requirement: 使用结构化错误处理
 
-Handler SHALL 使用结构化错误处理。
+ProxyHandler SHALL 使用 ConversionError 和协议对应的 encodeError 处理错误。
+
+#### Scenario: 转换错误
+
+- **WHEN** ConversionEngine 返回 ConversionError
+- **THEN** SHALL 使用 clientProtocol 的 Adapter.encodeError 编码错误响应
+- **THEN** SHALL 使用 OpenAI 错误格式（`{error: {message, type, code}}`）
 
 #### Scenario: 路由错误处理
 
 - **WHEN** RoutingService 返回错误
-- **THEN** SHALL 转换为对应的 AppError
-- **THEN** SHALL 返回统一的错误响应
+- **THEN** SHALL 转换为 ConversionError
+- **THEN** SHALL 使用 OpenAI 错误格式返回
 
 #### Scenario: 供应商错误处理
 
 - **WHEN** ProviderClient 返回错误
-- **THEN** SHALL 包装为 AppError
-- **THEN** SHALL 包含请求上下文信息
+- **THEN** SHALL 包装为 ConversionError
+- **THEN** SHALL 使用 OpenAI 错误格式返回

openspec/specs/protocol-adapter-anthropic/spec.md

@@ -1,3 +1,5 @@
+# Protocol Adapter - Anthropic
+
 ## ADDED Requirements
 
 ### Requirement: 实现 Anthropic ProtocolAdapter

openspec/specs/protocol-adapter-openai/spec.md

@@ -1,3 +1,5 @@
+# Protocol Adapter - OpenAI
+
 ## ADDED Requirements
 
 ### Requirement: 实现 OpenAI ProtocolAdapter
@@ -85,7 +87,7 @@
 - **WHEN** canonical.system 不为空
 - **THEN** SHALL 编码为 messages 数组头部的 role="system" 消息
 
-#### Scenario: Assistant 消息中 tool_calls 编码
+#### Scenario: Assistant <EFBFBD><EFBFBD>息中 tool_calls 编码
 
 - **WHEN** CanonicalMessage{role: "assistant"} 包含 tool_use 类型 ContentBlock
 - **THEN** SHALL 提取到 message.tool_calls 数组（{id, type: "function", function: {name, arguments}}）

openspec/specs/provider-management/spec.md

@@ -8,10 +8,11 @@
 
 #### Scenario: 使用有效数据创建供应商
 
-- **WHEN** 向 `/api/providers` 发送 POST 请求，携带有效的供应商数据（id, name, api_key, base_url）
+- **WHEN** 向 `/api/providers` 发送 POST 请求，携带有效的供应商数据（id, name, api_key, base_url, protocol）
 - **THEN** 网关 SHALL 在数据库中创建新的供应商记录
 - **THEN** 网关 SHALL 返回创建的供应商，状态码为 201
 - **THEN** 供应商 SHALL 默认启用
+- **THEN** protocol 字段 SHALL 默认为 "openai"
 
 #### Scenario: 使用重复 ID 创建供应商
 
@@ -34,7 +35,7 @@
 
 - **WHEN** 向 `/api/providers` 发送 GET 请求
 - **THEN** 网关 SHALL 返回所有供应商的列表
-- **THEN** 每个供应商 SHALL 包含 id, name, api_key（已掩码）, base_url, enabled, created_at, updated_at
+- **THEN** 每个供应商 SHALL 包含 id, name, api_key（已掩码）, base_url, protocol, enabled, created_at, updated_at
 - **THEN** api_key SHALL 被掩码（仅显示最后 4 个字符）
 
 **变更说明：** 数据访问从 config 包迁移到 ProviderRepository。API 接口保持不变。
@@ -47,6 +48,7 @@
 
 - **WHEN** 向 `/api/providers/:id` 发送 GET 请求，携带有效的供应商 ID
 - **THEN** 网关 SHALL 返回供应商详情
+- **THEN** SHALL 包含 protocol 字段
 - **THEN** api_key SHALL 被掩码
 
 #### Scenario: 获取不存在的供应商
@@ -65,7 +67,7 @@
 - **WHEN** 向 `/api/providers/:id` 发送 PUT 请求，携带有效的供应商数据
 - **THEN** 网关 SHALL 更新数据库中的供应商记录
 - **THEN** 网关 SHALL 返回更新后的供应商
-- **THEN** updated_at 时间戳 SHALL 被更新
+- **THEN** 更新 SHALL 支持修改 protocol 字段
 
 **变更说明：** 通过 ProviderService 和 ProviderRepository 实现。API 接口保持不变。
 

openspec/specs/request-validation/spec.md

@@ -20,43 +20,42 @@
 
 ### Requirement: 验证 OpenAI 请求
 
-系统 SHALL 验证 OpenAI ChatCompletionRequest。
+系统 SHALL 验证 OpenAI ChatCompletionRequest，验证逻辑位于 ProtocolAdapter 的 decodeRequest 内。
 
 #### Scenario: 必需字段验证
 
-- **WHEN** 收到 OpenAI 请求
+- **WHEN** OpenAI Adapter 的 decodeRequest 解析请求
 - **THEN** SHALL 验证 model 字段不为空
 - **THEN** SHALL 验证 messages 字段不为空且至少有一条消息
+- **THEN** 验证失败 SHALL 返回 INVALID_INPUT 类型的 ConversionError
 
 #### Scenario: 参数范围验证
 
-- **WHEN** 收到 OpenAI 请求
+- **WHEN** OpenAI Adapter 的 decodeRequest 解析参数
 - **THEN** SHALL 验证 temperature 范围在 [0, 2]
 - **THEN** SHALL 验证 max_tokens 大于 0
 - **THEN** SHALL 验证 top_p 范围在 (0, 1]
-- **THEN** SHALL 验证 frequency_penalty 范围在 [-2, 2]
-- **THEN** SHALL 验证 presence_penalty 范围在 [-2, 2]
 
 #### Scenario: 消息内容验证
 
 - **WHEN** 验证 messages 字段
-- **THEN** SHALL 验证每条消息的 role 有效（system、user、assistant、tool）
+- **THEN** SHALL 验证每条消息的 role 有效（system、developer、user、assistant、tool）
 - **THEN** SHALL 验证 content 不为空
 
 ### Requirement: 验证 Anthropic 请求
 
-系统 SHALL 验证 Anthropic MessagesRequest。
+系统 SHALL 验证 Anthropic MessagesRequest，验证逻辑位于 ProtocolAdapter 的 decodeRequest 内。
 
 #### Scenario: 必需字段验证
 
-- **WHEN** 收到 Anthropic 请求
+- **WHEN** Anthropic Adapter 的 decodeRequest 解析请求
 - **THEN** SHALL 验证 model 字段不为空
 - **THEN** SHALL 验证 messages 字段不为空且至少有一条消息
 - **THEN** SHALL 验证 max_tokens 大于 0（或使用默认值）
 
 #### Scenario: 参数范围验证
 
-- **WHEN** 收到 Anthropic 请求
+- **WHEN** Anthropic Adapter 的 decodeRequest 解析参数
 - **THEN** SHALL 验证 temperature 范围在 [0, 1]
 - **THEN** SHALL 验证 top_p 范围在 (0, 1]
 
@@ -93,26 +92,17 @@
 
 系统 SHALL 返回友好的验证错误响应。
 
-#### Scenario: 错误消息格式
+#### Scenario: 转换错误格式
 
-- **WHEN** 验证失败
-- **THEN** SHALL 返回 400 状态码
-- **THEN** SHALL 返回详细的错误消息
-- **THEN** SHALL 指示哪些字段验证失败
+- **WHEN** decodeRequest 验证失败返回 ConversionError
+- **THEN** ProxyHandler SHALL 使用 clientAdapter.encodeError 编码错误响应
+- **THEN** 错误 SHALL 使用客户端协议的格式
 
 #### Scenario: 多字段错误
 
 - **WHEN** 多个字段验证失败
-- **THEN** SHALL 返回所有验证错误
-- **THEN** SHALL 使用结构化格式（字段名 → 错误消息）
-
-#### Scenario: 国际化支持
-
-- **WHEN** 返回验证错误（未来）
-- **THEN** SHALL 支持错误消息国际化
-- **THEN** SHALL 使用错误码作为国际化 key
-
-注：当前版本使用中文错误消息。
+- **THEN** ConversionError.details SHALL 包含所有验证错误
+- **THEN** 错误响应 SHALL 包含完整的验证错误信息
 
 ### Requirement: 在 handler 中应用验证
 

openspec/specs/unified-proxy-handler/spec.md

@@ -1,3 +1,5 @@
+# Unified Proxy Handler
+
 ## ADDED Requirements
 
 ### Requirement: 实现统一代理 Handler
@@ -92,7 +94,7 @@ ProxyHandler SHALL 记录请求统计。
 
 ProxyHandler SHALL 支持 GET 请求的扩展层接口代理。
 
-#### Scenario: Models 接口代理
+#### Scenario: Models 接口<EFBFBD><EFBFBD>理
 
 - **WHEN** 收到 GET /{protocol}/v1/models 请求
 - **THEN** SHALL 执行路由和协议识别