runtime/llm/anthropic.go

@@ -0,0 +1,404 @@
+package llm
+
+import (
+	"bufio"
+	"bytes"
+	"context"
+	"encoding/json"
+	"fmt"
+	"io"
+	"net/http"
+	"strings"
+	"time"
+)
+
+// anthropic.go — 原生 Anthropic Messages API 客户端（N2 WP2）。
+// 与 OpenAI 路径共用 Message/Event 表面；组包与 BuildBody 的 Anthropic
+// 分支是同一个函数（消灭「离线/线上两张皮」，审计 N4 的根因）。
+
+// AnthropicVersion 是请求头 anthropic-version 的钉住值。
+const AnthropicVersion = "2023-06-01"
+
+// anthropicBody 组 Anthropic Messages 请求体（线上 Client 与离线 BuildBody
+// 共用）。tools 挂 input_schema；system 与最后一个 tool 挂 cache_control
+// （静态前缀的缓存断点）；assistant 的思考块在内容块最前且带签名往返。
+func anthropicBody(model string, messages []Message, tools []ToolDecl, opts StreamOptions) ([]byte, error) {
+	sys, msgs := anthropicMessages(messages)
+	body := map[string]any{
+		"model":    model,
+		"messages": msgs,
+		"stream":   true,
+		// Anthropic 要求 max_tokens。0 表示调用方不限制：用足够大的默认值，
+		// 避免 4096 把长回复截断。真正的上限由调用方 StreamOptions.MaxTokens 决定。
+		"max_tokens": opts.MaxTokens,
+	}
+	if body["max_tokens"] == 0 {
+		body["max_tokens"] = 128000
+	}
+	if sys != "" {
+		body["system"] = []map[string]any{{
+			"type":          "text",
+			"text":          sys,
+			"cache_control": map[string]any{"type": "ephemeral"},
+		}}
+	}
+	if budget := AnthropicThinkingBudget(opts.ThinkingLevel); budget > 0 {
+		body["thinking"] = map[string]any{"type": "enabled", "budget_tokens": budget}
+	}
+	if len(tools) > 0 {
+		body["tools"] = anthropicTools(tools)
+	}
+	return json.Marshal(body)
+}
+
+// anthropicTools 声明工具（name/description/input_schema）；最后一个挂
+// cache_control 作为静态工具前缀的断点。
+func anthropicTools(decls []ToolDecl) []map[string]any {
+	out := make([]map[string]any, 0, len(decls))
+	for i, t := range decls {
+		params := t.Parameters
+		if len(params) == 0 {
+			params = json.RawMessage(`{"type":"object","properties":{}}`)
+		}
+		item := map[string]any{
+			"name":         t.Name,
+			"description":  t.Description,
+			"input_schema": json.RawMessage(params),
+		}
+		if i == len(decls)-1 {
+			item["cache_control"] = map[string]any{"type": "ephemeral"}
+		}
+		out = append(out, item)
+	}
+	return out
+}
+
+// anthropicMessages 把通用转录映射成 Anthropic 消息：
+// system 收成字符串（由调用方决定挂法）；assistant 的思考块在最前，
+// 其后 text/image 与 tool_use；连续 RoleTool 折叠成一条 user 消息内的
+// tool_result 块组（Anthropic 不允许 tool 角色）。
+func anthropicMessages(messages []Message) (string, []map[string]any) {
+	var system strings.Builder
+	var out []map[string]any
+	for i := 0; i < len(messages); i++ {
+		m := messages[i]
+		switch m.Role {
+		case RoleSystem:
+			if system.Len() > 0 {
+				system.WriteString("\n")
+			}
+			system.WriteString(m.Content)
+		case RoleAssistant:
+			var content []map[string]any
+			for _, th := range m.Thinking {
+				if th.Redacted {
+					content = append(content, map[string]any{"type": "redacted_thinking", "data": th.Data})
+					continue
+				}
+				content = append(content, map[string]any{
+					"type": "thinking", "thinking": th.Thinking, "signature": th.Signature,
+				})
+			}
+			if m.Content != "" {
+				content = append(content, map[string]any{"type": "text", "text": m.Content})
+			}
+			for _, img := range m.Images {
+				content = append(content, map[string]any{
+					"type": "image",
+					"source": map[string]any{
+						"type":       "base64",
+						"media_type": firstNonEmpty(img.MIME, "image/png"),
+						"data":       img.Data,
+					},
+				})
+			}
+			for _, tc := range m.ToolCalls {
+				args := json.RawMessage(tc.Arguments)
+				if !json.Valid(args) {
+					args = json.RawMessage(`{}`)
+				}
+				content = append(content, map[string]any{
+					"type": "tool_use", "id": tc.ID, "name": tc.Name, "input": args,
+				})
+			}
+			if len(content) == 0 {
+				content = append(content, map[string]any{"type": "text", "text": ""})
+			}
+			out = append(out, map[string]any{"role": "assistant", "content": content})
+		case RoleTool:
+			// 连续 tool 结果折叠进同一条 user 消息。
+			var results []map[string]any
+			for ; i < len(messages) && messages[i].Role == RoleTool; i++ {
+				t := messages[i]
+				results = append(results, map[string]any{
+					"type":        "tool_result",
+					"tool_use_id": t.ToolCallID,
+					"content":     t.Content,
+					"is_error":    t.IsError,
+				})
+			}
+			i--
+			out = append(out, map[string]any{"role": "user", "content": results})
+		default:
+			content := []map[string]any{}
+			if m.Content != "" {
+				content = append(content, map[string]any{"type": "text", "text": m.Content})
+			}
+			for _, img := range m.Images {
+				content = append(content, map[string]any{
+					"type": "image",
+					"source": map[string]any{
+						"type":       "base64",
+						"media_type": firstNonEmpty(img.MIME, "image/png"),
+						"data":       img.Data,
+					},
+				})
+			}
+			out = append(out, map[string]any{"role": "user", "content": content})
+		}
+	}
+	return system.String(), out
+}
+
+// streamAnthropic 发 POST {base}/messages 并把 Anthropic SSE 事件族映射到
+// 通用 Event。与 OpenAI 客户端共享错误/中止语义。
+func (c *Client) streamAnthropic(ctx context.Context, model string, messages []Message, tools []ToolDecl, opts StreamOptions, out chan<- Event) {
+	base := firstNonEmpty(opts.APIBase, c.BaseURL)
+	key := firstNonEmpty(opts.APIKey, c.APIKey)
+	if base == "" || key == "" || model == "" {
+		emitError(out, "", false, "agentx llm client is missing base url, api key, or model")
+		return
+	}
+	if opts.MessageShape != "" {
+		messages = PrepareMessages(messages, opts.MessageShape)
+	}
+	raw, err := anthropicBody(model, messages, tools, opts)
+	if err != nil {
+		emitError(out, "", false, err.Error())
+		return
+	}
+	req, err := http.NewRequestWithContext(ctx, http.MethodPost, strings.TrimRight(base, "/")+"/messages", bytes.NewReader(raw))
+	if err != nil {
+		emitError(out, "", false, err.Error())
+		return
+	}
+	req.Header.Set("x-api-key", key)
+	req.Header.Set("anthropic-version", AnthropicVersion)
+	req.Header.Set("Content-Type", "application/json")
+	req.Header.Set("Accept", "text/event-stream")
+	hc := c.HTTPClient
+	if hc == nil {
+		hc = &http.Client{Timeout: 30 * time.Minute}
+	}
+	resp, err := hc.Do(req)
+	if err != nil {
+		if ctx.Err() != nil {
+			emitAbort(out, "", ctx.Err())
+			return
+		}
+		emitError(out, "", true, err.Error())
+		return
+	}
+	defer resp.Body.Close()
+	if resp.StatusCode >= 400 {
+		slurp, _ := io.ReadAll(io.LimitReader(resp.Body, 4096))
+		msg := fmt.Sprintf("provider status %d: %s", resp.StatusCode, strings.TrimSpace(string(slurp)))
+		ev := Event{
+			Kind:         KindError,
+			HTTPStatus:   resp.StatusCode,
+			Retryable:    StatusRetryable(resp.StatusCode) && !ContextOverflow(msg),
+			ErrorMessage: msg,
+			Err:          &StreamError{Message: msg, Retryable: StatusRetryable(resp.StatusCode) && !ContextOverflow(msg)},
+			Message:      Message{Role: RoleAssistant, StopReason: StopError},
+		}
+		if d, ok := ParseRetryAfter(resp.Header.Get("Retry-After"), time.Now()); ok {
+			ev.HasRetryAfter = true
+			ev.RetryAfter = d
+		}
+		out <- ev
+		return
+	}
+	parseAnthropicSSE(ctx, resp.Body, out)
+}
+
+// ---- Anthropic SSE 解析 ----
+
+type anthropicBlock struct {
+	typ       string
+	toolID    string
+	toolName  string
+	args      strings.Builder
+	thinking  strings.Builder
+	signature strings.Builder
+	redacted  string
+}
+
+// parseAnthropicSSE 读 event:/data: 帧对，产出文本/思考/工具增量与一次
+// 终结 message_end（或 error）。
+func parseAnthropicSSE(ctx context.Context, r io.Reader, out chan<- Event) {
+	blocks := map[int]*anthropicBlock{}
+	var content strings.Builder
+	var thinking []ThinkingBlock
+	var calls []ToolCall
+	var usage Usage
+	stopReason := StopEnd
+	finishReason := ""
+	sc := bufio.NewScanner(r)
+	sc.Buffer(make([]byte, 0, 64*1024), 4*1024*1024)
+	eventType := ""
+	for sc.Scan() {
+		line := sc.Text()
+		select {
+		case <-ctx.Done():
+			emitAbort(out, "", ctx.Err())
+			return
+		default:
+		}
+		switch {
+		case strings.HasPrefix(line, "event:"):
+			eventType = strings.TrimSpace(strings.TrimPrefix(line, "event:"))
+		case strings.HasPrefix(line, "data:"):
+			data := strings.TrimSpace(strings.TrimPrefix(line, "data:"))
+			if data == "" {
+				continue
+			}
+			var frame map[string]json.RawMessage
+			if err := json.Unmarshal([]byte(data), &frame); err != nil {
+				continue
+			}
+			switch eventType {
+			case "message_start":
+				var v struct {
+					Message struct {
+						Usage struct {
+							InputTokens      int `json:"input_tokens"`
+							CacheReadTokens  int `json:"cache_read_input_tokens"`
+							CacheWriteTokens int `json:"cache_creation_input_tokens"`
+						} `json:"usage"`
+					} `json:"message"`
+				}
+				if json.Unmarshal(frame["message"], &v.Message) == nil {
+					usage.PromptTokens = v.Message.Usage.InputTokens
+					usage.CacheReadTokens = v.Message.Usage.CacheReadTokens
+					usage.CacheWriteTokens = v.Message.Usage.CacheWriteTokens
+				}
+			case "content_block_start":
+				var v struct {
+					Index       int `json:"index"`
+					ContentBlock struct {
+						Type string `json:"type"`
+						ID   string `json:"id"`
+						Name string `json:"name"`
+						Data string `json:"data"`
+					} `json:"content_block"`
+				}
+				if json.Unmarshal(frame["index"], &v.Index) == nil {
+					_ = json.Unmarshal(frame["content_block"], &v.ContentBlock)
+					b := &anthropicBlock{typ: v.ContentBlock.Type, toolID: v.ContentBlock.ID, toolName: v.ContentBlock.Name, redacted: v.ContentBlock.Data}
+					blocks[v.Index] = b
+					if v.ContentBlock.Type == "tool_use" {
+						out <- Event{Kind: KindToolCallDelta, ToolIndex: v.Index, ToolID: v.ContentBlock.ID, ToolName: v.ContentBlock.Name}
+					}
+				}
+			case "content_block_delta":
+				var v struct {
+					Index int `json:"index"`
+					Delta struct {
+						Type      string `json:"type"`
+						Text      string `json:"text"`
+						Thinking  string `json:"thinking"`
+						Signature string `json:"signature"`
+						Partial   string `json:"partial_json"`
+					} `json:"delta"`
+				}
+				if json.Unmarshal(frame["index"], &v.Index) != nil {
+					continue
+				}
+				_ = json.Unmarshal(frame["delta"], &v.Delta)
+				b := blocks[v.Index]
+				switch v.Delta.Type {
+				case "text_delta":
+					content.WriteString(v.Delta.Text)
+					out <- Event{Kind: KindTextDelta, Text: v.Delta.Text}
+				case "thinking_delta":
+					if b != nil {
+						b.thinking.WriteString(v.Delta.Thinking)
+					}
+					out <- Event{Kind: KindReasoningDelta, ReasoningDelta: v.Delta.Thinking}
+				case "signature_delta":
+					if b != nil {
+						b.signature.WriteString(v.Delta.Signature)
+					}
+				case "input_json_delta":
+					if b != nil {
+						b.args.WriteString(v.Delta.Partial)
+					}
+					out <- Event{Kind: KindToolCallDelta, ToolIndex: v.Index, ArgumentsDelta: v.Delta.Partial}
+				}
+			case "content_block_stop":
+				var v struct {
+					Index int `json:"index"`
+				}
+				if json.Unmarshal(frame["index"], &v.Index) != nil {
+					continue
+				}
+				b := blocks[v.Index]
+				if b == nil {
+					continue
+				}
+				switch b.typ {
+				case "tool_use":
+					calls = append(calls, ToolCall{ID: b.toolID, Name: b.toolName, Arguments: b.args.String()})
+				case "thinking":
+					thinking = append(thinking, ThinkingBlock{Thinking: b.thinking.String(), Signature: b.signature.String()})
+				case "redacted_thinking":
+					thinking = append(thinking, ThinkingBlock{Redacted: true, Data: b.redacted})
+				}
+			case "message_delta":
+				var v struct {
+					Delta struct {
+						StopReason string `json:"stop_reason"`
+					} `json:"delta"`
+					Usage struct {
+						OutputTokens int `json:"output_tokens"`
+					} `json:"usage"`
+				}
+				_ = json.Unmarshal(frame["delta"], &v.Delta)
+				_ = json.Unmarshal(frame["usage"], &v.Usage)
+				finishReason = v.Delta.StopReason
+				usage.CompletionTokens = v.Usage.OutputTokens
+				if v.Delta.StopReason == "tool_use" {
+					stopReason = StopToolCalls
+				}
+			case "error":
+				msg := data
+				var v struct {
+					Error struct {
+						Type    string `json:"type"`
+						Message string `json:"message"`
+					} `json:"error"`
+				}
+				if json.Unmarshal(frame["error"], &v.Error) == nil && v.Error.Message != "" {
+					msg = v.Error.Type + ": " + v.Error.Message
+				}
+				out <- Event{Kind: KindError, ErrorMessage: msg, Err: &StreamError{Message: msg}, Message: Message{Role: RoleAssistant, StopReason: StopError}}
+				return
+			}
+		}
+	}
+	if err := sc.Err(); err != nil {
+		if ctx.Err() != nil {
+			emitAbort(out, "", ctx.Err())
+			return
+		}
+		emitError(out, "", true, "anthropic sse read: "+err.Error())
+		return
+	}
+	usage.TotalTokens = usage.PromptTokens + usage.CompletionTokens
+	out <- Event{
+		Kind:         KindMessageEnd,
+		Message:      Message{Role: RoleAssistant, Content: content.String(), ToolCalls: calls, Thinking: thinking, StopReason: stopReason},
+		Usage:        usage,
+		FinishReason: finishReason,
+	}
+}

runtime/llm/anthropic_test.go

@@ -0,0 +1,250 @@
+package llm
+
+import (
+	"context"
+	"encoding/json"
+	"io"
+	"net/http"
+	"net/http/httptest"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+)
+
+// ---- N2 WP2：Anthropic 原生客户端（验收① 的机器证据）----
+
+// fixtureServer 按调用序回放 testdata/anthropic 下的录制 SSE，并记录请求。
+type fixtureServer struct {
+	t        *testing.T
+	requests []recordedRequest
+	count    int
+}
+
+type recordedRequest struct {
+	path    string
+	apiKey  string
+	version string
+	body    map[string]any
+}
+
+func (f *fixtureServer) handler(w http.ResponseWriter, r *http.Request) {
+	f.count++
+	body, _ := io.ReadAll(r.Body)
+	var parsed map[string]any
+	_ = json.Unmarshal(body, &parsed)
+	f.requests = append(f.requests, recordedRequest{
+		path:    r.URL.Path,
+		apiKey:  r.Header.Get("x-api-key"),
+		version: r.Header.Get("anthropic-version"),
+		body:    parsed,
+	})
+	names := []string{"01-tool-use.sse", "02-thinking-text.sse", "03-text.sse"}
+	raw, err := os.ReadFile(filepath.Join("testdata", "anthropic", names[f.count-1]))
+	if err != nil {
+		f.t.Fatalf("fixture %d: %v", f.count, err)
+	}
+	w.Header().Set("Content-Type", "text/event-stream")
+	_, _ = w.Write(raw)
+}
+
+// anthropicPresetClient 经预设表构建客户端（验收的是「选 anthropic 预设」本身），
+// BaseURL 指到夹具服务器（预设只是结构体，测试内覆盖端点，不打真实网络）。
+func anthropicPresetClient(t *testing.T, base string) *Client {
+	t.Helper()
+	var preset Provider
+	for _, p := range Presets() {
+		if p.Name == "anthropic" {
+			preset = p
+		}
+	}
+	if preset.Name == "" {
+		t.Fatal("anthropic preset missing")
+	}
+	preset.BaseURL = base
+	preset.APIKey = "fixture-key"
+	c, err := ClientFor(preset)
+	if err != nil {
+		t.Fatalf("ClientFor: %v", err)
+	}
+	return c
+}
+
+func collect(t *testing.T, ch <-chan Event) (end Event, texts, reasoning []string) {
+	t.Helper()
+	for ev := range ch {
+		switch ev.Kind {
+		case KindTextDelta:
+			texts = append(texts, ev.Text)
+		case KindReasoningDelta:
+			reasoning = append(reasoning, ev.ReasoningDelta)
+		case KindMessageEnd:
+			end = ev
+		case KindError:
+			t.Fatalf("stream error: %s", ev.ErrorMessage)
+		}
+	}
+	return end, texts, reasoning
+}
+
+// TestAnthropicPresetToolRound 验收①：选 anthropic 预设，用录制 HTTP 夹具
+// 完成一轮完整工具调用（tools 请求 → tool_use → tool_result → 终文本），
+// 且 thinking 块带原签名在下一请求里往返。
+func TestAnthropicPresetToolRound(t *testing.T) {
+	fs := &fixtureServer{t: t}
+	srv := httptest.NewServer(http.HandlerFunc(fs.handler))
+	defer srv.Close()
+	c := anthropicPresetClient(t, srv.URL)
+	ctx := context.Background()
+
+	tools := []ToolDecl{{Name: "read_file", Description: "读文件", Parameters: json.RawMessage(`{"type":"object","properties":{"path":{"type":"string"}}}`)}}
+
+	// 第一回合：带 tools 请求 → tool_use。
+	end1, _, _ := collect(t, c.Stream(ctx, "claude-fixture", []Message{{Role: RoleUser, Content: "查一下 main.go"}}, tools, StreamOptions{}))
+	if end1.Message.StopReason != StopToolCalls || len(end1.Message.ToolCalls) != 1 {
+		t.Fatalf("resp1: %+v", end1.Message)
+	}
+	call := end1.Message.ToolCalls[0]
+	if call.ID != "toolu_01ABC" || call.Name != "read_file" || call.Arguments != `{"path":"main.go"}` {
+		t.Fatalf("tool call: %+v", call)
+	}
+
+	req1 := fs.requests[0]
+	if req1.path != "/messages" || req1.apiKey != "fixture-key" || req1.version != AnthropicVersion {
+		t.Fatalf("req1 headers: %+v", req1)
+	}
+	toolsBody, _ := req1.body["tools"].([]any)
+	if len(toolsBody) != 1 {
+		t.Fatalf("req1 tools: %v", req1.body["tools"])
+	}
+	tool0 := toolsBody[0].(map[string]any)
+	if tool0["name"] != "read_file" || tool0["input_schema"] == nil || tool0["cache_control"] == nil {
+		t.Fatalf("req1 tool decl: %v", tool0)
+	}
+
+	// 第二回合：assistant(tool_use) + tool_result → thinking(带签名) + 终文本。
+	msgs2 := []Message{
+		{Role: RoleUser, Content: "查一下 main.go"},
+		end1.Message,
+		{Role: RoleTool, ToolCallID: call.ID, Name: call.Name, Content: "1 package main\n..."},
+	}
+	end2, texts2, reasoning2 := collect(t, c.Stream(ctx, "claude-fixture", msgs2, tools, StreamOptions{}))
+	if end2.Message.StopReason != StopEnd || end2.Message.Content != "查到了，main.go 一共 42 行。" {
+		t.Fatalf("resp2: %+v", end2.Message)
+	}
+	if len(end2.Message.Thinking) != 1 || end2.Message.Thinking[0].Signature != "EqoBCkgIARABGAIiQA" {
+		t.Fatalf("resp2 thinking: %+v", end2.Message.Thinking)
+	}
+	if strings.Join(texts2, "") != end2.Message.Content || len(reasoning2) == 0 {
+		t.Fatalf("deltas: %v %v", texts2, reasoning2)
+	}
+
+	req2 := fs.requests[1]
+	msgsBody, _ := req2.body["messages"].([]any)
+	var toolResult map[string]any
+	for _, m := range msgsBody {
+		mm := m.(map[string]any)
+		if mm["role"] != "user" {
+			continue
+		}
+		for _, blk := range mm["content"].([]any) {
+			b := blk.(map[string]any)
+			if b["type"] == "tool_result" {
+				toolResult = b
+			}
+		}
+	}
+	if toolResult == nil || toolResult["tool_use_id"] != "toolu_01ABC" {
+		t.Fatalf("req2 tool_result: %v", toolResult)
+	}
+
+	// 第三回合：assistant(thinking+text) 回送 → 签名必须原样往返。
+	msgs3 := append(append([]Message{}, msgs2...), end2.Message, Message{Role: RoleUser, Content: "再说一遍"})
+	end3, _, _ := collect(t, c.Stream(ctx, "claude-fixture", msgs3, tools, StreamOptions{}))
+	if end3.Message.Content != "42 行。" {
+		t.Fatalf("resp3: %+v", end3.Message)
+	}
+
+	req3 := fs.requests[2]
+	var thinkingBlock map[string]any
+	for _, m := range req3.body["messages"].([]any) {
+		mm := m.(map[string]any)
+		if mm["role"] != "assistant" {
+			continue
+		}
+		for _, blk := range mm["content"].([]any) {
+			b := blk.(map[string]any)
+			if b["type"] == "thinking" {
+				thinkingBlock = b
+			}
+		}
+	}
+	if thinkingBlock == nil || thinkingBlock["signature"] != "EqoBCkgIARABGAIiQA" {
+		t.Fatalf("req3 thinking round-trip: %v", thinkingBlock)
+	}
+}
+
+// TestAnthropicBodyShape 组包单测：tool_use/tool_result/thinking 块形态、
+// system 与末位工具的 cache_control、max_tokens 默认。
+func TestAnthropicBodyShape(t *testing.T) {
+	msgs := []Message{
+		{Role: RoleSystem, Content: "sys"},
+		{Role: RoleUser, Content: "u"},
+		{Role: RoleAssistant, Thinking: []ThinkingBlock{{Thinking: "想", Signature: "sig"}}, ToolCalls: []ToolCall{{ID: "t1", Name: "read", Arguments: `{"a":1}`}}},
+		{Role: RoleTool, ToolCallID: "t1", Name: "read", Content: "r1"},
+		{Role: RoleTool, ToolCallID: "t2", Name: "grep", Content: "r2", IsError: true},
+	}
+	raw, err := anthropicBody("claude", msgs, []ToolDecl{{Name: "read", Description: "读"}}, StreamOptions{})
+	if err != nil {
+		t.Fatal(err)
+	}
+	var body map[string]any
+	if err := json.Unmarshal(raw, &body); err != nil {
+		t.Fatal(err)
+	}
+	if body["max_tokens"] != float64(128000) {
+		t.Fatalf("max_tokens 默认：%v", body["max_tokens"])
+	}
+	sys := body["system"].([]any)[0].(map[string]any)
+	if sys["cache_control"] == nil {
+		t.Fatal("system 应挂 cache_control")
+	}
+	out := body["messages"].([]any)
+	// assistant：思考块在最前，其后 tool_use。
+	assistant := out[1].(map[string]any)
+	blocks := assistant["content"].([]any)
+	if blocks[0].(map[string]any)["type"] != "thinking" || blocks[1].(map[string]any)["type"] != "tool_use" {
+		t.Fatalf("assistant 块序：%v", blocks)
+	}
+	// 连续两条 tool 结果折叠进同一条 user 消息。
+	folded := out[2].(map[string]any)
+	if folded["role"] != "user" || len(folded["content"].([]any)) != 2 {
+		t.Fatalf("tool_result 折叠：%v", folded)
+	}
+	second := folded["content"].([]any)[1].(map[string]any)
+	if second["type"] != "tool_result" || second["is_error"] != true {
+		t.Fatalf("tool_result is_error：%v", second)
+	}
+	// 末位工具挂 cache_control。
+	tool := body["tools"].([]any)[0].(map[string]any)
+	if tool["cache_control"] == nil {
+		t.Fatal("末位工具应挂 cache_control")
+	}
+}
+
+// TestAnthropicSSEErrorFrame 协议内 error 帧映射为 KindError。
+func TestAnthropicSSEErrorFrame(t *testing.T) {
+	raw := "event: error\ndata: {\"type\":\"error\",\"error\":{\"type\":\"overloaded_error\",\"message\":\"Overloaded\"}}\n\n"
+	ch := make(chan Event, 1)
+	go func() {
+		defer close(ch)
+		parseAnthropicSSE(context.Background(), strings.NewReader(raw), ch)
+	}()
+	var got []Event
+	for ev := range ch {
+		got = append(got, ev)
+	}
+	if len(got) != 1 || got[0].Kind != KindError || !strings.Contains(got[0].ErrorMessage, "Overloaded") {
+		t.Fatalf("%+v", got)
+	}
+}

runtime/llm/client.go

@@ -11,11 +11,15 @@ import (
 	"time"
 )
 
-// Client calls an OpenAI-compatible chat completions endpoint with streaming.
+// Client calls a chat completions endpoint with streaming.
+// 默认讲 OpenAI 兼容协议；Protocol 为 ProtocolAnthropic 时讲原生
+// Anthropic Messages（组包与 BuildBody 共用，见 anthropic.go）。
 type Client struct {
 	BaseURL    string
 	APIKey     string
 	HTTPClient *http.Client
+	// Protocol 空 = openai-chat（零值兼容所有直接构造点）。
+	Protocol Protocol
 }
 
 // Stream implements StreamFunc. Transport and provider failures are encoded as
@@ -24,6 +28,10 @@ func (c *Client) Stream(ctx context.Context, model string, messages []Message, t
 	out := make(chan Event)
 	go func() {
 		defer close(out)
+		if c.Protocol == ProtocolAnthropic {
+			c.streamAnthropic(ctx, model, messages, tools, opts, out)
+			return
+		}
 		c.run(ctx, model, messages, tools, opts, out)
 	}()
 	return out

runtime/llm/providers.go

@@ -140,7 +140,7 @@ type Provider struct {
 // Presets are OpenAI-compatible or native endpoints that only need an API key.
 func Presets() []Provider {
 	return []Provider{
-		{Name: "anthropic", Protocol: ProtocolOpenAIChat, BaseURL: "https://api.anthropic.com/v1", APIKeyEnv: "ANTHROPIC_API_KEY"},
+		{Name: "anthropic", Protocol: ProtocolAnthropic, BaseURL: "https://api.anthropic.com", APIKeyEnv: "ANTHROPIC_API_KEY"},
 		{Name: "openai", Protocol: ProtocolOpenAIChat, BaseURL: "https://api.openai.com/v1", APIKeyEnv: "OPENAI_API_KEY"},
 		{Name: "openai-responses", Protocol: ProtocolOpenAIResponses, BaseURL: "https://api.openai.com/v1", APIKeyEnv: "OPENAI_API_KEY"},
 		{Name: "google", Protocol: ProtocolOpenAIChat, BaseURL: "https://generativelanguage.googleapis.com/v1beta/openai", APIKeyEnv: "GEMINI_API_KEY"},
@@ -196,12 +196,14 @@ func Lookup(name string, custom []Provider) (Provider, bool) {
 	return Provider{}, false
 }
 
-// ClientFor 返回会说该协议的客户端。responses、azure 以及原生 Anthropic/Google
+// ClientFor 返回会说该协议的客户端。responses、azure 以及原生 Google
 // 协议当前没有接到客户端上，返回明确错误而不是发到错误的路径。
 func ClientFor(p Provider) (*Client, error) {
 	switch p.Protocol {
 	case ProtocolOpenAIChat:
 		return &Client{BaseURL: p.BaseURL, APIKey: p.APIKey}, nil
+	case ProtocolAnthropic:
+		return &Client{BaseURL: p.BaseURL, APIKey: p.APIKey, Protocol: ProtocolAnthropic}, nil
 	default:
 		return nil, fmt.Errorf("llm: protocol %q is not implemented", p.Protocol)
 	}
@@ -264,29 +266,10 @@ func BuildBody(protocol Protocol, model string, messages []Message, tools []Tool
 			body["reasoning"] = map[string]any{"effort": effort}
 		}
 		return json.Marshal(body)
-	// ProtocolAnthropic 与 ProtocolGoogle 只供离线组包。
-	// 当前 Client 只发 OpenAI chat completions，这两支没有接到客户端上。
-	// 原生协议是后续阶段的事；预设表里的 anthropic / google 已改走兼容端点。
+	// ProtocolGoogle 只供离线组包（客户端未接）；Anthropic 与线上客户端
+	// 共用 anthropicBody（同一组包，消灭离线/线上两张皮）。
 	case ProtocolAnthropic:
-		sys, msgs := anthropicMessages(messages)
-		body := map[string]any{
-			"model":      model,
-			"messages":   msgs,
-			"max_tokens": opts.MaxTokens,
-			"stream":     true,
-		}
-		// Anthropic 要求 max_tokens。0 表示调用方不限制：用足够大的默认值，
-		// 避免 4096 把长回复截断。真正的上限由调用方 StreamOptions.MaxTokens 决定。
-		if body["max_tokens"] == 0 {
-			body["max_tokens"] = 128000
-		}
-		if sys != "" {
-			body["system"] = sys
-		}
-		if budget := AnthropicThinkingBudget(opts.ThinkingLevel); budget > 0 {
-			body["thinking"] = map[string]any{"type": "enabled", "budget_tokens": budget}
-		}
-		return json.Marshal(body)
+		return anthropicBody(model, messages, tools, opts)
 	case ProtocolGoogle:
 		body := map[string]any{
 			"contents": googleContents(messages),
@@ -317,40 +300,6 @@ func responsesInput(messages []Message) []map[string]any {
 	return out
 }
 
-func anthropicMessages(messages []Message) (string, []map[string]any) {
-	var system strings.Builder
-	var out []map[string]any
-	for _, m := range messages {
-		if m.Role == RoleSystem {
-			if system.Len() > 0 {
-				system.WriteString("\n")
-			}
-			system.WriteString(m.Content)
-			continue
-		}
-		role := "user"
-		if m.Role == RoleAssistant {
-			role = "assistant"
-		}
-		content := []map[string]any{}
-		if m.Content != "" {
-			content = append(content, map[string]any{"type": "text", "text": m.Content})
-		}
-		for _, img := range m.Images {
-			content = append(content, map[string]any{
-				"type": "image",
-				"source": map[string]any{
-					"type":       "base64",
-					"media_type": firstNonEmpty(img.MIME, "image/png"),
-					"data":       img.Data,
-				},
-			})
-		}
-		out = append(out, map[string]any{"role": role, "content": content})
-	}
-	return system.String(), out
-}
-
 func googleContents(messages []Message) []map[string]any {
 	var out []map[string]any
 	for _, m := range messages {

runtime/llm/providers_test.go

@@ -69,19 +69,29 @@ func TestModelsJSONAndBodies(t *testing.T) {
 }
 
 func TestPresetsAndClientForChatCompletions(t *testing.T) {
-	want := map[string]string{
-		"anthropic": "https://api.anthropic.com/v1",
-		"google":    "https://generativelanguage.googleapis.com/v1beta/openai",
+	// google 仍走 OpenAI 兼容端点；anthropic 自 N2 起是原生 Messages 协议。
+	want := map[string]struct {
+		protocol Protocol
+		base     string
+	}{
+		"anthropic": {ProtocolAnthropic, "https://api.anthropic.com"},
+		"google":    {ProtocolOpenAIChat, "https://generativelanguage.googleapis.com/v1beta/openai"},
 	}
 	for _, p := range Presets() {
-		base, ok := want[p.Name]
+		w, ok := want[p.Name]
 		if !ok {
 			continue
 		}
-		if p.Protocol != ProtocolOpenAIChat || p.BaseURL != base {
+		if p.Protocol != w.protocol || p.BaseURL != w.base {
 			t.Fatalf("%s preset: %+v", p.Name, p)
 		}
 	}
+	// 原生 anthropic 有真实客户端（审计 N4 关闭）；responses/azure/google 仍 fail-closed。
+	if p, ok := Lookup("anthropic", nil); !ok {
+		t.Fatal("anthropic preset missing")
+	} else if c, err := ClientFor(p); err != nil || c.Protocol != ProtocolAnthropic {
+		t.Fatalf("anthropic ClientFor: %v %+v", err, c)
+	}
 	for _, name := range []string{"openai-responses", "azure"} {
 		p, ok := Lookup(name, nil)
 		if !ok {

runtime/llm/testdata/anthropic/01-tool-use.sse

@@ -0,0 +1,20 @@
+event: message_start
+data: {"type":"message_start","message":{"id":"msg_01","type":"message","role":"assistant","model":"claude-fixture","usage":{"input_tokens":88,"cache_read_input_tokens":0,"cache_creation_input_tokens":0}}}
+
+event: content_block_start
+data: {"type":"content_block_start","index":0,"content_block":{"type":"tool_use","id":"toolu_01ABC","name":"read_file"}}
+
+event: content_block_delta
+data: {"type":"content_block_delta","index":0,"delta":{"type":"input_json_delta","partial_json":"{\"path\":"}}
+
+event: content_block_delta
+data: {"type":"content_block_delta","index":0,"delta":{"type":"input_json_delta","partial_json":"\"main.go\"}"}}
+
+event: content_block_stop
+data: {"type":"content_block_stop","index":0}
+
+event: message_delta
+data: {"type":"message_delta","delta":{"stop_reason":"tool_use"},"usage":{"output_tokens":31}}
+
+event: message_stop
+data: {"type":"message_stop"}

runtime/llm/testdata/anthropic/02-thinking-text.sse

@@ -0,0 +1,29 @@
+event: message_start
+data: {"type":"message_start","message":{"id":"msg_02","type":"message","role":"assistant","model":"claude-fixture","usage":{"input_tokens":142,"cache_read_input_tokens":60,"cache_creation_input_tokens":0}}}
+
+event: content_block_start
+data: {"type":"content_block_start","index":0,"content_block":{"type":"thinking","thinking":""}}
+
+event: content_block_delta
+data: {"type":"content_block_delta","index":0,"delta":{"type":"thinking_delta","thinking":"用户问 main.go，工具已返回。"}}
+
+event: content_block_delta
+data: {"type":"content_block_delta","index":0,"delta":{"type":"signature_delta","signature":"EqoBCkgIARABGAIiQA"}}
+
+event: content_block_stop
+data: {"type":"content_block_stop","index":0}
+
+event: content_block_start
+data: {"type":"content_block_start","index":1,"content_block":{"type":"text","text":""}}
+
+event: content_block_delta
+data: {"type":"content_block_delta","index":1,"delta":{"type":"text_delta","text":"查到了，main.go 一共 42 行。"}}
+
+event: content_block_stop
+data: {"type":"content_block_stop","index":1}
+
+event: message_delta
+data: {"type":"message_delta","delta":{"stop_reason":"end_turn"},"usage":{"output_tokens":57}}
+
+event: message_stop
+data: {"type":"message_stop"}

runtime/llm/testdata/anthropic/03-text.sse

@@ -0,0 +1,17 @@
+event: message_start
+data: {"type":"message_start","message":{"id":"msg_03","type":"message","role":"assistant","model":"claude-fixture","usage":{"input_tokens":201}}}
+
+event: content_block_start
+data: {"type":"content_block_start","index":0,"content_block":{"type":"text","text":""}}
+
+event: content_block_delta
+data: {"type":"content_block_delta","index":0,"delta":{"type":"text_delta","text":"42 行。"}}
+
+event: content_block_stop
+data: {"type":"content_block_stop","index":0}
+
+event: message_delta
+data: {"type":"message_delta","delta":{"stop_reason":"end_turn"},"usage":{"output_tokens":9}}
+
+event: message_stop
+data: {"type":"message_stop"}

runtime/llm/types.go

@@ -52,8 +52,21 @@ type Message struct {
 	Images     []Image    `json:"images,omitempty"`
 	ToolCallID string     `json:"tool_call_id,omitempty"`
 	ToolCalls  []ToolCall `json:"tool_calls,omitempty"`
-	StopReason StopReason `json:"stop_reason,omitempty"`
-	IsError    bool       `json:"is_error,omitempty"`
+	// Thinking 是原生 Anthropic 路径的思考块（含签名），随 assistant 消息
+	// 在请求里原样往返（Anthropic 要求思考块在内容块最前且签名不失）。
+	// OpenAI 路径不写这个字段（reasoning 不落盘，见 sse.go 注释）。
+	Thinking   []ThinkingBlock `json:"thinking,omitempty"`
+	StopReason StopReason      `json:"stop_reason,omitempty"`
+	IsError    bool            `json:"is_error,omitempty"`
+}
+
+// ThinkingBlock 是一块 Anthropic 思考内容。Redacted 为真时 Data 是不透明
+// 密文（redacted_thinking），往返时原样回送。
+type ThinkingBlock struct {
+	Thinking  string `json:"thinking,omitempty"`
+	Signature string `json:"signature,omitempty"`
+	Redacted  bool   `json:"redacted,omitempty"`
+	Data      string `json:"data,omitempty"`
 }
 
 // ToolDecl is the JSON-schema declaration sent to the provider.