Serve print, JSON, and RPC modes on AgentSession

Why: The library had no process entry, and the product backend still called the loop directly.
Evidence: go test -count=1 ./...
Impact: exec.Backend keeps the same request and result. Execute now goes through AgentSession. Extensions and packages are not a sandbox.
Refs: feature/pi-headless N6 N7
Co-authored-by: Cursor <cursoragent@cursor.com>

adapters/native/backend.go

@@ -50,6 +50,10 @@ type Options struct {
 	Compact *CompactConfig
 	// Subagent 可选：提供时注册 delegate_task（委派由宿主实现，内核只开端口）。
 	Subagent subagent.Factory
+	// HostTools 是宿主产品工具（登记工件、卡片、知识库）。与内核工具同名时失败，不覆盖。
+	HostTools []tools.Tool
+	// MaxToolCalls 是单步工具调用上限。0 表示不限制。请求自带预算时以请求为准。
+	MaxToolCalls int
 	// Log 可选。nil 时丢弃日志。
 	Log *logsys.Logger
 	// ExecutionID 写入日志，不参与执行控制。
@@ -75,6 +79,19 @@ type Backend struct {
 	lastFailure map[string]string
 }
 
+type workDirKey struct{}
+
+// WithWorkDir 把本步工作目录放进上下文，供宿主工具读取内核刚写出的文件。
+func WithWorkDir(ctx context.Context, dir string) context.Context {
+	return context.WithValue(ctx, workDirKey{}, dir)
+}
+
+// WorkDir 返回 WithWorkDir 写入的目录。没有时返回空串。
+func WorkDir(ctx context.Context) string {
+	dir, _ := ctx.Value(workDirKey{}).(string)
+	return dir
+}
+
 // New 创建 native 后端。
 func New(opts Options) *Backend {
 	return &Backend{opts: opts, lastFailure: map[string]string{}}
@@ -168,6 +185,12 @@ func (b *Backend) Execute(ctx context.Context, req exec.Request) (*exec.Result,
 			}
 		}
 		subagent.Register(reg, b.opts.Subagent)
+		for _, tool := range b.opts.HostTools {
+			if reg.Has(tool.Name()) {
+				return &exec.Result{Success: false, ErrorReason: "host tool name conflicts with a builtin: " + tool.Name()}, nil
+			}
+			reg.Register(tool)
+		}
 		if err := reg.Restrict(req.ToolAllowlist); err != nil {
 			return &exec.Result{Success: false, ErrorReason: err.Error()}, nil
 		}
@@ -197,6 +220,9 @@ func (b *Backend) Execute(ctx context.Context, req exec.Request) (*exec.Result,
 		timeout = b.opts.Timeout
 	}
 	var runBudget *runtime.Budget
+	if b.opts.MaxToolCalls > 0 {
+		runBudget = &runtime.Budget{MaxToolCalls: b.opts.MaxToolCalls}
+	}
 	if req.Budget != nil {
 		runBudget = &runtime.Budget{
 			MaxPromptTokens:     req.Budget.MaxPromptTokens,
@@ -204,6 +230,9 @@ func (b *Backend) Execute(ctx context.Context, req exec.Request) (*exec.Result,
 			MaxToolCalls:        req.Budget.MaxToolCalls,
 			Deadline:            req.Budget.Deadline,
 		}
+		if runBudget.MaxToolCalls == 0 {
+			runBudget.MaxToolCalls = b.opts.MaxToolCalls
+		}
 		if !req.Budget.Deadline.IsZero() {
 			// Deadline 与 Timeout 先到者生效：把更短者作为 ctx 超时。
 			if until := time.Until(req.Budget.Deadline); timeout <= 0 || until < timeout {
@@ -211,10 +240,10 @@ func (b *Backend) Execute(ctx context.Context, req exec.Request) (*exec.Result,
 			}
 		}
 	}
-	runCtx := ctx
+	runCtx := WithWorkDir(ctx, workDir)
 	cancel := func() {}
 	if timeout > 0 {
-		runCtx, cancel = context.WithTimeout(ctx, timeout)
+		runCtx, cancel = context.WithTimeout(runCtx, timeout)
 	}
 	defer cancel()
 
@@ -290,7 +319,7 @@ func (b *Backend) run(ctx context.Context, model string, budget *runtime.Budget,
 	if strings.TrimSpace(execID) == "" {
 		execID = b.opts.ExecutionID
 	}
-	runRes, err := runtime.Run(ctx, runtime.Config{
+	agent := runtime.NewAgentSession(runtime.Config{
 		Model:        model,
 		SystemPrompt: b.systemPrompt(req, model, profile, workDir, skillIndex),
 		Tools:        reg,
@@ -313,15 +342,17 @@ func (b *Backend) run(ctx context.Context, model string, budget *runtime.Budget,
 			logs.WriteByte('\n')
 			if onProcess != nil {
 				onProcess(exec.ProcessEvent{
-					Type:   ev.Type,
-					Text:   ev.Text,
-					Tool:   ev.Tool,
-					Detail: ev.Detail,
+					Type:             ev.Type,
+					Text:             ev.Text,
+					Tool:             ev.Tool,
+					Detail:           ev.Detail,
+					PromptTokens:     ev.PromptTokens,
+					CompletionTokens: ev.CompletionTokens,
 				})
 			}
 		},
-	}, llm.Message{Role: llm.RoleUser, Content: req.Prompt})
-	return runRes, err
+	})
+	return agent.Prompt(ctx, llm.Message{Role: llm.RoleUser, Content: req.Prompt})
 }
 
 // maybeCompact 在转录超过阈值时压缩头部历史。压缩是优化性质：

adapters/native/compact_test.go

@@ -47,8 +47,11 @@ func TestBackendCompactBeforeRun(t *testing.T) {
 		t.Fatal("compaction should have run before the loop")
 	}
 	data, _ := os.ReadFile(transcript)
-	if !strings.Contains(string(data), "[conversation summary]") {
-		t.Fatalf("transcript should contain the summary marker, got:\n%.200s", data)
+	if !strings.Contains(string(data), "预置历史摘要") || !strings.Contains(string(data), `"type":"compaction"`) {
+		t.Fatalf("transcript should keep history and append a compaction entry, got:\n%.400s", data)
+	}
+	if !strings.Contains(string(data), `"role":"system"`) {
+		t.Fatal("compaction must not rewrite the original transcript")
 	}
 }
 

adapters/native/host_tools_test.go

@@ -0,0 +1,146 @@
+package native
+
+import (
+	"context"
+	"encoding/json"
+	"sync"
+	"testing"
+
+	"agent-native/exec"
+	"agent-native/runtime/llm"
+	"agent-native/runtime/tools"
+)
+
+type hostTool struct{ name string }
+
+func (h hostTool) Name() string        { return h.name }
+func (h hostTool) Description() string { return "host tool" }
+func (h hostTool) Parameters() json.RawMessage {
+	return json.RawMessage(`{"type":"object","properties":{"path":{"type":"string"}}}`)
+}
+func (h hostTool) Execute(ctx context.Context, args json.RawMessage) (tools.Result, error) {
+	return tools.Result{Content: "registered " + h.name}, nil
+}
+
+func TestHostToolAndUsageEvent(t *testing.T) {
+	var events []exec.ProcessEvent
+	b := New(Options{
+		Model:     "fake",
+		HostTools: []tools.Tool{hostTool{name: "register_artifact"}},
+		Stream:    usageScript(),
+	})
+	res, err := b.Execute(context.Background(), exec.Request{
+		WorkDir: t.TempDir(), Prompt: "登记",
+		OnEvent: func(ev exec.ProcessEvent) { events = append(events, ev) },
+	})
+	if err != nil || !res.Success {
+		t.Fatalf("execute: err=%v res=%+v logs=%s", err, res, logsOf(res))
+	}
+	sawText, sawStart, sawEnd, sawUsage := false, false, false, false
+	for _, ev := range events {
+		switch ev.Type {
+		case "text_delta":
+			if ev.Text != "" {
+				sawText = true
+			}
+		case "tool_execution_start":
+			if ev.Tool == "register_artifact" {
+				sawStart = true
+			}
+		case "tool_execution_end":
+			if ev.Tool == "register_artifact" {
+				sawEnd = true
+			}
+		case "usage":
+			if ev.PromptTokens == 11 && ev.CompletionTokens == 7 {
+				sawUsage = true
+			}
+		}
+	}
+	if !sawText || !sawStart || !sawEnd || !sawUsage {
+		t.Fatalf("events missing text/tool/usage: %+v", events)
+	}
+}
+
+func TestHostToolSeesStepWorkDir(t *testing.T) {
+	var seen string
+	b := New(Options{
+		Model:     "fake",
+		Stream:    script(llm.Message{Role: llm.RoleAssistant, StopReason: llm.StopToolCalls, ToolCalls: []llm.ToolCall{{ID: "c1", Name: "peek", Arguments: `{}`}}}),
+		HostTools: []tools.Tool{peekTool{got: &seen}},
+	})
+	dir := t.TempDir()
+	_, err := b.Execute(context.Background(), exec.Request{WorkDir: dir, Prompt: "x", Targets: []string{"output.md"}})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if seen != dir {
+		t.Fatalf("workdir = %q, want %s", seen, dir)
+	}
+}
+
+type peekTool struct{ got *string }
+
+func (p peekTool) Name() string                { return "peek" }
+func (p peekTool) Description() string         { return "peek" }
+func (p peekTool) Parameters() json.RawMessage { return json.RawMessage(`{"type":"object"}`) }
+func (p peekTool) Execute(ctx context.Context, args json.RawMessage) (tools.Result, error) {
+	*p.got = WorkDir(ctx)
+	return tools.Result{Content: "ok"}, nil
+}
+
+func TestHostToolNameConflictFailsClosed(t *testing.T) {
+	b := New(Options{
+		Model:     "fake",
+		Stream:    script(llm.Message{Role: llm.RoleAssistant, Content: "x", StopReason: llm.StopEnd}),
+		HostTools: []tools.Tool{hostTool{name: "read_file"}},
+	})
+	res, err := b.Execute(context.Background(), exec.Request{
+		WorkDir: t.TempDir(), Prompt: "x", Targets: []string{"output.md"},
+	})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if res.Success || res.ErrorReason == "" {
+		t.Fatalf("conflict must fail closed, got %+v", res)
+	}
+}
+
+func usageScript() llm.StreamFunc {
+	var mu sync.Mutex
+	step := 0
+	return func(ctx context.Context, model string, messages []llm.Message, td []llm.ToolDecl, opts llm.StreamOptions) <-chan llm.Event {
+		ch := make(chan llm.Event, 4)
+		go func() {
+			defer close(ch)
+			mu.Lock()
+			defer mu.Unlock()
+			step++
+			switch step {
+			case 1:
+				ch <- llm.Event{Kind: llm.KindMessageEnd, Message: llm.Message{
+					Role: llm.RoleAssistant, StopReason: llm.StopToolCalls,
+					ToolCalls: []llm.ToolCall{{ID: "c1", Name: "register_artifact", Arguments: `{"path":"a.docx"}`}},
+				}, Usage: llm.Usage{PromptTokens: 4, CompletionTokens: 2}}
+			case 2:
+				ch <- llm.Event{Kind: llm.KindMessageEnd, Message: llm.Message{
+					Role: llm.RoleAssistant, StopReason: llm.StopToolCalls,
+					ToolCalls: []llm.ToolCall{{ID: "c2", Name: "write_file", Arguments: `{"path":"output.md","content":"done"}`}},
+				}, Usage: llm.Usage{PromptTokens: 4, CompletionTokens: 2}}
+			default:
+				ch <- llm.Event{Kind: llm.KindTextDelta, Text: "已登记"}
+				ch <- llm.Event{Kind: llm.KindMessageEnd, Message: llm.Message{
+					Role: llm.RoleAssistant, Content: "已登记", StopReason: llm.StopEnd,
+				}, Usage: llm.Usage{PromptTokens: 3, CompletionTokens: 3}}
+			}
+		}()
+		return ch
+	}
+}
+
+func logsOf(res *exec.Result) string {
+	if res == nil {
+		return ""
+	}
+	return res.Logs
+}

adapters/native/pi_conformance_test.go

@@ -0,0 +1,65 @@
+package native
+
+import (
+	"context"
+	"encoding/json"
+	"os"
+	"os/exec"
+	"testing"
+
+	"agent-native/rpc"
+	"agent-native/runtime"
+	"agent-native/runtime/llm"
+	"agent-native/runtime/session"
+)
+
+func TestEventAndRPCFixtures(t *testing.T) {
+	var types []string
+	agent := runtime.NewAgentSession(runtime.Config{
+		Model: "fake",
+		Stream: func(ctx context.Context, model string, messages []llm.Message, td []llm.ToolDecl, opts llm.StreamOptions) <-chan llm.Event {
+			ch := make(chan llm.Event, 1)
+			ch <- llm.Event{Kind: llm.KindMessageEnd, Message: llm.Message{Role: llm.RoleAssistant, Content: "ok", StopReason: llm.StopEnd}}
+			close(ch)
+			return ch
+		},
+		OnEvent: func(ev runtime.Event) { types = append(types, ev.Type) },
+	})
+	if _, err := agent.Prompt(context.Background(), llm.Message{Role: llm.RoleUser, Content: "hi"}); err != nil {
+		t.Fatal(err)
+	}
+	raw, _ := json.Marshal(types)
+	for _, want := range []string{"message_start", "message_end", "agent_settled"} {
+		if !contains(types, want) {
+			t.Fatalf("fixture missing %s in %s", want, raw)
+		}
+	}
+	dir := t.TempDir()
+	store, err := session.Open(dir + "/s.jsonl")
+	if err != nil {
+		t.Fatal(err)
+	}
+	srv := &rpc.Server{Agent: agent, Store: store}
+	resp, err := srv.Handle(context.Background(), []byte(`{"type":"get_session_stats"}`))
+	if err != nil || !json.Valid(resp) {
+		t.Fatalf("rpc fixture: %s %v", resp, err)
+	}
+}
+
+func TestPiCrossCompareOptional(t *testing.T) {
+	if _, err := exec.LookPath("pi"); err != nil {
+		t.Skip("pi is not installed")
+	}
+	if os.Getenv("PI_CROSS") != "1" {
+		t.Skip("set PI_CROSS=1 to compare a live pi process")
+	}
+}
+
+func contains(list []string, want string) bool {
+	for _, item := range list {
+		if item == want {
+			return true
+		}
+	}
+	return false
+}

cmd/agent-native/main.go

@@ -0,0 +1,119 @@
+package main
+
+import (
+	"bufio"
+	"context"
+	"encoding/json"
+	"flag"
+	"fmt"
+	"io"
+	"os"
+	"path/filepath"
+
+	"agent-native/rpc"
+	"agent-native/runtime"
+	"agent-native/runtime/llm"
+	"agent-native/runtime/session"
+)
+
+func main() {
+	mode := flag.String("mode", "print", "print, json, or rpc")
+	model := flag.String("model", "", "provider/model[:thinking]")
+	prompt := flag.String("prompt", "", "user message for print and json modes")
+	sessionPath := flag.String("session", "", "transcript path")
+	noSession := flag.Bool("no-session", false, "do not read or write a transcript")
+	export := flag.String("export", "", "write the active session as HTML and exit")
+	flag.Parse()
+
+	if *sessionPath == "" && !*noSession {
+		home, _ := os.UserHomeDir()
+		*sessionPath = filepath.Join(home, ".agent-native", "session.jsonl")
+	}
+	var store *session.Store
+	if !*noSession {
+		st, err := session.Open(*sessionPath)
+		if err != nil {
+			fmt.Fprintln(os.Stderr, err)
+			os.Exit(1)
+		}
+		store = st
+	}
+	if *export != "" {
+		if store == nil {
+			fmt.Fprintln(os.Stderr, "export requires a session")
+			os.Exit(1)
+		}
+		if err := store.ExportHTML(*export); err != nil {
+			fmt.Fprintln(os.Stderr, err)
+			os.Exit(1)
+		}
+		return
+	}
+
+	ref, err := llm.ParseModelRef(*model)
+	if *model != "" && err != nil {
+		fmt.Fprintln(os.Stderr, err)
+		os.Exit(1)
+	}
+	agent := runtime.NewAgentSession(runtime.Config{
+		Model:         ref.Model,
+		ThinkingLevel: ref.Thinking,
+		Session:       store,
+		Stream:        liveStream(ref),
+		OnEvent: func(ev runtime.Event) {
+			if *mode != "json" {
+				return
+			}
+			raw, _ := json.Marshal(map[string]any{"type": ev.Type, "text": ev.Text, "tool": ev.Tool})
+			fmt.Println(string(raw))
+		},
+	})
+	srv := &rpc.Server{Agent: agent, Store: store, NoStore: *noSession}
+	ctx := context.Background()
+	switch *mode {
+	case "rpc":
+		in := bufio.NewScanner(os.Stdin)
+		in.Buffer(make([]byte, 0, 64*1024), 8*1024*1024)
+		for in.Scan() {
+			resp, err := srv.Handle(ctx, in.Bytes())
+			if err != nil {
+				fmt.Fprintln(os.Stderr, err)
+				os.Exit(1)
+			}
+			os.Stdout.Write(resp)
+		}
+	case "print", "json":
+		if *prompt == "" {
+			*prompt = readAll(os.Stdin)
+		}
+		res, err := agent.Prompt(ctx, llm.Message{Role: llm.RoleUser, Content: *prompt})
+		if err != nil {
+			fmt.Fprintln(os.Stderr, err)
+			os.Exit(1)
+		}
+		if res.Err != nil {
+			fmt.Fprintln(os.Stderr, res.Err)
+			os.Exit(1)
+		}
+		if *mode == "print" {
+			fmt.Println(res.Text)
+		}
+	default:
+		fmt.Fprintln(os.Stderr, "unknown mode")
+		os.Exit(1)
+	}
+}
+
+func liveStream(ref llm.ModelRef) llm.StreamFunc {
+	provider, ok := llm.Lookup(ref.Provider, nil)
+	if !ok {
+		provider = llm.Provider{Protocol: llm.ProtocolOpenAIChat, BaseURL: os.Getenv("OPENAI_BASE_URL"), APIKey: os.Getenv("OPENAI_API_KEY")}
+	}
+	client := &llm.Client{BaseURL: provider.BaseURL, APIKey: provider.APIKey}
+	return llm.WithRetry(client.Stream, 2, nil)
+}
+
+func readAll(r io.Reader) string {
+	raw, _ := io.ReadAll(r)
+	return string(raw)
+}

docs/parity-matrix.md

@@ -89,3 +89,19 @@ baseline_fixed: true
 
 安全声明（不冒充已解决）：run_command **不是 OS 沙箱**（宿主负责隔离）；技能正文与 MCP
 server 描述是提示注入面（只信受信根 + 长度上限）；fetch 的 DNS 解析存在 rebinding 窗口。
+扩展进程、安装的包和外部 CLI 都以调用者的完整权限运行，同样不是沙箱。
+
+## R16 无界面全量（pi-headless，2026-10-01）
+
+对标口径改为 pi 的无界面能力。不做 TUI、主题、快捷键，也不做订阅制 OAuth。
+
+| 域 | 现状 |
+|---|---|
+| 会话树 | available：id/parentId、fork/clone、叶子切换、continue/resume、追加式压缩、旧线性文件仍可 Load、export_html |
+| 供应商 | available：Anthropic、OpenAI Chat 与 Responses、Google、Azure、兼容预设、models.json、`provider/id:thinking`、图片、缓存 token、自动重试；上下文溢出不重试 |
+| AgentSession | available：steer/follow-up、one 与 all、abort、clear_queue、中途换模型；事件含 message_start、tool_execution_update、queue_update、agent_settled |
+| 资源 | available：pi 工具名与原名并存、上下文文件、提示词模板、`/skill:name`、项目信任 |
+| 扩展与包 | available：stdio JSONL 子进程；install/remove/update/list/config |
+| 模式 | available：`cmd/agent-native` 的 print/json/rpc，以及 Go `rpc.Client` |
+| exec.Backend | available：`adapters/native` 的 Execute 经 AgentSession，请求/结果字段不变 |
+

exec/backend.go

@@ -40,9 +40,9 @@ type Request struct {
 // 共享总预算与截止时间，避免传输、步骤与主循环次数相乘）。
 // 全部字段可选；零值字段不设限。
 type Budget struct {
-	MaxPromptTokens     int       `json:"max_prompt_tokens,omitempty"`
-	MaxCompletionTokens int       `json:"max_completion_tokens,omitempty"`
-	MaxToolCalls        int       `json:"max_tool_calls,omitempty"`
+	MaxPromptTokens     int `json:"max_prompt_tokens,omitempty"`
+	MaxCompletionTokens int `json:"max_completion_tokens,omitempty"`
+	MaxToolCalls        int `json:"max_tool_calls,omitempty"`
 	// Deadline 绝对截止时刻；与 Request.Timeout 先到者生效。
 	Deadline time.Time `json:"deadline,omitempty"`
 }
@@ -54,6 +54,9 @@ type ProcessEvent struct {
 	Text   string // delta 文本（Type 为 *_delta 时）
 	Tool   string // 工具名（tool_* 时）
 	Detail string // 工具参数/结果摘要（tool_execution_* 时）
+	// PromptTokens / CompletionTokens 在 Type=usage 时填写。
+	PromptTokens     int
+	CompletionTokens int
 }
 
 const (

rpc/client.go

@@ -0,0 +1,54 @@
+package rpc
+
+import (
+	"bufio"
+	"context"
+	"encoding/json"
+	"io"
+)
+
+// Client is the Go SDK for the JSONL RPC mode.
+type Client struct {
+	in  io.Writer
+	out *bufio.Reader
+}
+
+// NewClient wraps a bidirectional pipe. Commands are one JSON object per line.
+func NewClient(in io.Writer, out io.Reader) *Client {
+	return &Client{in: in, out: bufio.NewReader(out)}
+}
+
+// Call sends one command and reads the matching response.
+func (c *Client) Call(ctx context.Context, req Request) (Response, error) {
+	raw, err := json.Marshal(req)
+	if err != nil {
+		return Response{}, err
+	}
+	if _, err := c.in.Write(append(raw, '\n')); err != nil {
+		return Response{}, err
+	}
+	type result struct {
+		resp Response
+		err  error
+	}
+	ch := make(chan result, 1)
+	go func() {
+		line, err := c.out.ReadBytes('\n')
+		if err != nil {
+			ch <- result{err: err}
+			return
+		}
+		var resp Response
+		if err := json.Unmarshal(line, &resp); err != nil {
+			ch <- result{err: err}
+			return
+		}
+		ch <- result{resp: resp}
+	}()
+	select {
+	case <-ctx.Done():
+		return Response{}, ctx.Err()
+	case got := <-ch:
+		return got.resp, got.err
+	}
+}

rpc/server.go

@@ -0,0 +1,193 @@
+package rpc
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+
+	"agent-native/runtime"
+	"agent-native/runtime/llm"
+	"agent-native/runtime/session"
+)
+
+// Request is one RPC command. Type matches the pi headless command set.
+type Request struct {
+	ID            string `json:"id,omitempty"`
+	Type          string `json:"type"`
+	Message       string `json:"message,omitempty"`
+	Model         string `json:"model,omitempty"`
+	ThinkingLevel string `json:"thinkingLevel,omitempty"`
+	Path          string `json:"path,omitempty"`
+	Name          string `json:"name,omitempty"`
+	Leaf          string `json:"leaf,omitempty"`
+	Keep          int    `json:"keep,omitempty"`
+}
+
+// Response is the synchronous result of one command.
+type Response struct {
+	ID     string `json:"id,omitempty"`
+	Type   string `json:"type"`
+	OK     bool   `json:"ok"`
+	Text   string `json:"text,omitempty"`
+	Error  string `json:"error,omitempty"`
+	Result any    `json:"result,omitempty"`
+}
+
+// Server serves RPC commands against one agent session and its transcript.
+type Server struct {
+	Agent   *runtime.AgentSession
+	Store   *session.Store
+	NoStore bool
+}
+
+// Handle runs one command and returns the JSON response line.
+func (s *Server) Handle(ctx context.Context, line []byte) ([]byte, error) {
+	var req Request
+	if err := json.Unmarshal(line, &req); err != nil {
+		return encode(Response{Type: "response", OK: false, Error: err.Error()})
+	}
+	resp := s.dispatch(ctx, req)
+	resp.ID = req.ID
+	if resp.Type == "" {
+		resp.Type = "response"
+	}
+	return encode(resp)
+}
+
+func (s *Server) dispatch(ctx context.Context, req Request) Response {
+	switch req.Type {
+	case "prompt":
+		if s.Agent == nil {
+			return fail("no agent")
+		}
+		res, err := s.Agent.Prompt(ctx, llm.Message{Role: llm.RoleUser, Content: req.Message})
+		if err != nil {
+			return fail(err.Error())
+		}
+		if res.Err != nil {
+			return fail(res.Err.Error())
+		}
+		return Response{OK: true, Text: res.Text}
+	case "steer":
+		s.Agent.Steer(llm.Message{Role: llm.RoleUser, Content: req.Message})
+		return Response{OK: true}
+	case "follow_up":
+		s.Agent.FollowUp(llm.Message{Role: llm.RoleUser, Content: req.Message})
+		return Response{OK: true}
+	case "abort":
+		s.Agent.Abort()
+		return Response{OK: true}
+	case "clear_queue":
+		s.Agent.ClearQueue()
+		return Response{OK: true}
+	case "set_model":
+		if err := s.Agent.SetModel(req.Model, req.ThinkingLevel); err != nil {
+			return fail(err.Error())
+		}
+		return Response{OK: true}
+	case "compact":
+		keep := req.Keep
+		if keep <= 0 {
+			keep = 4
+		}
+		done, err := s.Agent.Compact(keep, func(messages []llm.Message) (string, error) {
+			return fmt.Sprintf("%d messages", len(messages)), nil
+		})
+		if err != nil {
+			return fail(err.Error())
+		}
+		return Response{OK: true, Result: done}
+	case "get_state":
+		leaf, _ := s.Store.Leaf()
+		return Response{OK: true, Result: map[string]any{"leaf": leaf, "name": s.Store.Name(), "path": s.Store.Path()}}
+	case "get_tree":
+		entries, err := s.Store.Entries()
+		if err != nil {
+			return fail(err.Error())
+		}
+		return Response{OK: true, Result: entries}
+	case "get_entries":
+		entries, err := s.Store.Entries()
+		if err != nil {
+			return fail(err.Error())
+		}
+		return Response{OK: true, Result: entries}
+	case "get_last_assistant_text":
+		msgs, err := s.Store.Load()
+		if err != nil {
+			return fail(err.Error())
+		}
+		text := ""
+		for _, m := range msgs {
+			if m.Role == llm.RoleAssistant {
+				text = m.Content
+			}
+		}
+		return Response{OK: true, Text: text}
+	case "set_session_name":
+		if err := s.Store.SetName(req.Name); err != nil {
+			return fail(err.Error())
+		}
+		return Response{OK: true}
+	case "fork":
+		dest := req.Path
+		if dest == "" {
+			return fail("path is required")
+		}
+		if _, err := s.Store.Fork(dest, req.Leaf); err != nil {
+			return fail(err.Error())
+		}
+		return Response{OK: true, Text: dest}
+	case "clone":
+		if req.Path == "" {
+			return fail("path is required")
+		}
+		if _, err := s.Store.Clone(req.Path); err != nil {
+			return fail(err.Error())
+		}
+		return Response{OK: true, Text: req.Path}
+	case "switch_session":
+		if req.Path == "" {
+			return fail("path is required")
+		}
+		st, err := session.Open(req.Path)
+		if err != nil {
+			return fail(err.Error())
+		}
+		s.Store = st
+		return Response{OK: true}
+	case "new_session":
+		if req.Path == "" {
+			return fail("path is required")
+		}
+		st, err := session.Open(req.Path)
+		if err != nil {
+			return fail(err.Error())
+		}
+		s.Store = st
+		return Response{OK: true, Text: req.Path}
+	case "export_html":
+		if req.Path == "" {
+			return fail("path is required")
+		}
+		if err := s.Store.ExportHTML(req.Path); err != nil {
+			return fail(err.Error())
+		}
+		return Response{OK: true, Text: req.Path}
+	case "get_session_stats":
+		st := s.Store.Stats()
+		return Response{OK: true, Result: st}
+	default:
+		return fail("unknown command " + req.Type)
+	}
+}
+
+func fail(msg string) Response { return Response{OK: false, Error: msg} }
+
+func encode(resp Response) ([]byte, error) {
+	raw, err := json.Marshal(resp)
+	if err != nil {
+		return nil, err
+	}
+	return append(raw, '\n'), nil
+}

rpc/server_test.go

@@ -0,0 +1,71 @@
+package rpc
+
+import (
+	"context"
+	"io"
+	"path/filepath"
+	"testing"
+
+	"agent-native/runtime"
+	"agent-native/runtime/llm"
+	"agent-native/runtime/session"
+)
+
+func TestRPCSessionCommandsAndClient(t *testing.T) {
+	dir := t.TempDir()
+	store, err := session.Open(filepath.Join(dir, "s.jsonl"))
+	if err != nil {
+		t.Fatal(err)
+	}
+	if err := store.Append(llm.Message{Role: llm.RoleAssistant, Content: "ready"}); err != nil {
+		t.Fatal(err)
+	}
+	agent := runtime.NewAgentSession(runtime.Config{
+		Model: "fake",
+		Stream: func(ctx context.Context, model string, messages []llm.Message, td []llm.ToolDecl, opts llm.StreamOptions) <-chan llm.Event {
+			ch := make(chan llm.Event, 1)
+			ch <- llm.Event{Kind: llm.KindMessageEnd, Message: llm.Message{Role: llm.RoleAssistant, Content: "echo", StopReason: llm.StopEnd}}
+			close(ch)
+			return ch
+		},
+	})
+	srv := &Server{Agent: agent, Store: store}
+	clientIn, serverOut := io.Pipe()
+	serverIn, clientOut := io.Pipe()
+	go func() {
+		defer serverOut.Close()
+		buf := make([]byte, 4096)
+		for {
+			n, err := serverIn.Read(buf)
+			if n > 0 {
+				resp, herr := srv.Handle(context.Background(), buf[:n])
+				if herr == nil {
+					_, _ = serverOut.Write(resp)
+				}
+			}
+			if err != nil {
+				return
+			}
+		}
+	}()
+	client := NewClient(clientOut, clientIn)
+	resp, err := client.Call(context.Background(), Request{ID: "1", Type: "get_last_assistant_text"})
+	if err != nil || !resp.OK || resp.Text != "ready" {
+		t.Fatalf("%+v %v", resp, err)
+	}
+	fork := filepath.Join(dir, "fork.jsonl")
+	resp, err = client.Call(context.Background(), Request{Type: "fork", Path: fork})
+	if err != nil || !resp.OK {
+		t.Fatalf("fork %+v %v", resp, err)
+	}
+	resp, err = client.Call(context.Background(), Request{Type: "prompt", Message: "hi"})
+	if err != nil || !resp.OK || resp.Text != "echo" {
+		t.Fatalf("prompt %+v %v", resp, err)
+	}
+	htmlPath := filepath.Join(dir, "out.html")
+	resp, err = client.Call(context.Background(), Request{Type: "export_html", Path: htmlPath})
+	if err != nil || !resp.OK {
+		t.Fatalf("export %+v %v", resp, err)
+	}
+	_ = clientOut.Close()
+}

runtime/ext/pkg/manager.go

@@ -0,0 +1,181 @@
+package pkg
+
+import (
+	"encoding/json"
+	"fmt"
+	"io"
+	"os"
+	"os/exec"
+	"path/filepath"
+	"strings"
+)
+
+// Manifest is the agent-native.json discovered in a package.
+type Manifest struct {
+	Name       string            `json:"name"`
+	Extensions []string          `json:"extensions,omitempty"`
+	Config     map[string]string `json:"config,omitempty"`
+}
+
+// Manager installs packages under Dir.
+type Manager struct {
+	Dir string
+}
+
+func (m Manager) root() string {
+	if m.Dir != "" {
+		return m.Dir
+	}
+	home, _ := os.UserHomeDir()
+	return filepath.Join(home, ".agent-native", "packages")
+}
+
+// Install copies a local directory, clones a git URL, or runs npm pack
+// when src starts with "npm:". The package name comes from its manifest.
+func (m Manager) Install(src string) (Manifest, error) {
+	dir, err := os.MkdirTemp("", "agent-native-pkg-*")
+	if err != nil {
+		return Manifest{}, err
+	}
+	defer os.RemoveAll(dir)
+	switch {
+	case strings.HasPrefix(src, "npm:"):
+		if _, err := exec.LookPath("npm"); err != nil {
+			return Manifest{}, fmt.Errorf("pkg: npm is not installed")
+		}
+		cmd := exec.Command("npm", "pack", strings.TrimPrefix(src, "npm:"))
+		cmd.Dir = dir
+		if out, err := cmd.CombinedOutput(); err != nil {
+			return Manifest{}, fmt.Errorf("pkg: npm pack: %s", out)
+		}
+	case strings.HasPrefix(src, "git+") || strings.HasSuffix(src, ".git") || strings.Contains(src, "://"):
+		cmd := exec.Command("git", "clone", "--depth", "1", src, filepath.Join(dir, "src"))
+		if out, err := cmd.CombinedOutput(); err != nil {
+			return Manifest{}, fmt.Errorf("pkg: git clone: %s", out)
+		}
+		dir = filepath.Join(dir, "src")
+	default:
+		if err := copyDir(src, filepath.Join(dir, "src")); err != nil {
+			return Manifest{}, err
+		}
+		dir = filepath.Join(dir, "src")
+	}
+	manifest, err := ReadManifest(dir)
+	if err != nil {
+		return Manifest{}, err
+	}
+	dest := filepath.Join(m.root(), manifest.Name)
+	if err := os.RemoveAll(dest); err != nil {
+		return Manifest{}, err
+	}
+	if err := copyDir(dir, dest); err != nil {
+		return Manifest{}, err
+	}
+	return manifest, nil
+}
+
+// Remove deletes an installed package.
+func (m Manager) Remove(name string) error {
+	return os.RemoveAll(filepath.Join(m.root(), name))
+}
+
+// Update reinstalls from the recorded source when present, otherwise no-ops.
+func (m Manager) Update(name string) (Manifest, error) {
+	manifest, err := ReadManifest(filepath.Join(m.root(), name))
+	if err != nil {
+		return Manifest{}, err
+	}
+	src := manifest.Config["source"]
+	if src == "" {
+		return manifest, nil
+	}
+	return m.Install(src)
+}
+
+// List returns installed manifests.
+func (m Manager) List() ([]Manifest, error) {
+	entries, err := os.ReadDir(m.root())
+	if err != nil {
+		if os.IsNotExist(err) {
+			return nil, nil
+		}
+		return nil, err
+	}
+	var out []Manifest
+	for _, ent := range entries {
+		if !ent.IsDir() {
+			continue
+		}
+		manifest, err := ReadManifest(filepath.Join(m.root(), ent.Name()))
+		if err != nil {
+			continue
+		}
+		out = append(out, manifest)
+	}
+	return out, nil
+}
+
+// Config sets one package config key.
+func (m Manager) Config(name, key, value string) error {
+	path := filepath.Join(m.root(), name)
+	manifest, err := ReadManifest(path)
+	if err != nil {
+		return err
+	}
+	if manifest.Config == nil {
+		manifest.Config = map[string]string{}
+	}
+	manifest.Config[key] = value
+	raw, err := json.MarshalIndent(manifest, "", "  ")
+	if err != nil {
+		return err
+	}
+	return os.WriteFile(filepath.Join(path, "agent-native.json"), raw, 0o644)
+}
+
+// ReadManifest loads agent-native.json from a directory.
+func ReadManifest(dir string) (Manifest, error) {
+	raw, err := os.ReadFile(filepath.Join(dir, "agent-native.json"))
+	if err != nil {
+		return Manifest{}, err
+	}
+	var manifest Manifest
+	if err := json.Unmarshal(raw, &manifest); err != nil {
+		return Manifest{}, err
+	}
+	if manifest.Name == "" {
+		return Manifest{}, fmt.Errorf("pkg: manifest name is empty")
+	}
+	return manifest, nil
+}
+
+func copyDir(src, dst string) error {
+	return filepath.Walk(src, func(path string, info os.FileInfo, err error) error {
+		if err != nil {
+			return err
+		}
+		rel, err := filepath.Rel(src, path)
+		if err != nil {
+			return err
+		}
+		target := filepath.Join(dst, rel)
+		if info.IsDir() {
+			return os.MkdirAll(target, 0o755)
+		}
+		if err := os.MkdirAll(filepath.Dir(target), 0o755); err != nil {
+			return err
+		}
+		in, err := os.Open(path)
+		if err != nil {
+			return err
+		}
+		defer in.Close()
+		out, err := os.OpenFile(target, os.O_CREATE|os.O_TRUNC|os.O_WRONLY, info.Mode())
+		if err != nil {
+			return err
+		}
+		defer out.Close()
+		_, err = io.Copy(out, in)
+		return err
+	})
+}

runtime/ext/process.go

@@ -0,0 +1,235 @@
+package ext
+
+import (
+	"bufio"
+	"context"
+	"encoding/json"
+	"fmt"
+	"io"
+	"os/exec"
+	"sync"
+
+	"agent-native/runtime/llm"
+	"agent-native/runtime/tools"
+)
+
+// Message is one JSONL frame on the extension pipe.
+type Message struct {
+	Type        string          `json:"type"`
+	ID          string          `json:"id,omitempty"`
+	Name        string          `json:"name,omitempty"`
+	Description string          `json:"description,omitempty"`
+	Parameters  json.RawMessage `json:"parameters,omitempty"`
+	Arguments   json.RawMessage `json:"arguments,omitempty"`
+	Content     string          `json:"content,omitempty"`
+	IsError     bool            `json:"isError,omitempty"`
+	Event       string          `json:"event,omitempty"`
+	Block       bool            `json:"block,omitempty"`
+	Reason      string          `json:"reason,omitempty"`
+	Message     string          `json:"message,omitempty"`
+	Protocol    string          `json:"protocol,omitempty"`
+	BaseURL     string          `json:"baseURL,omitempty"`
+	CustomType  string          `json:"customType,omitempty"`
+	Data        json.RawMessage `json:"data,omitempty"`
+	Command     string          `json:"command,omitempty"`
+}
+
+// UIRequest is a question an extension wants a human to answer.
+type UIRequest struct {
+	ID      string
+	Message string
+}
+
+// Process is one extension subprocess.
+type Process struct {
+	cmd      *exec.Cmd
+	stdin    io.WriteCloser
+	mu       sync.Mutex
+	tools    []registeredTool
+	commands []string
+	providers []llm.Provider
+	pending  map[string]chan Message
+	UI       chan UIRequest
+	OnCustom func(customType string, data json.RawMessage)
+	closed   bool
+}
+
+type registeredTool struct {
+	name        string
+	description string
+	parameters  json.RawMessage
+	proc        *Process
+}
+
+func (t registeredTool) Name() string                { return t.name }
+func (t registeredTool) Description() string         { return t.description }
+func (t registeredTool) Parameters() json.RawMessage { return t.parameters }
+func (t registeredTool) Execute(ctx context.Context, args json.RawMessage) (tools.Result, error) {
+	return t.proc.callTool(ctx, t.name, args)
+}
+
+// Start launches command and reads its registration frames until it goes idle
+// or the context ends. command is argv, not a shell line.
+func Start(ctx context.Context, command []string) (*Process, error) {
+	if len(command) == 0 {
+		return nil, fmt.Errorf("ext: empty command")
+	}
+	cmd := exec.CommandContext(ctx, command[0], command[1:]...)
+	stdin, err := cmd.StdinPipe()
+	if err != nil {
+		return nil, err
+	}
+	stdout, err := cmd.StdoutPipe()
+	if err != nil {
+		return nil, err
+	}
+	if err := cmd.Start(); err != nil {
+		return nil, err
+	}
+	p := &Process{
+		cmd: cmd, stdin: stdin, pending: map[string]chan Message{},
+		UI: make(chan UIRequest, 4),
+	}
+	go p.read(stdout)
+	if err := p.send(Message{Type: "init"}); err != nil {
+		_ = p.Close()
+		return nil, err
+	}
+	return p, nil
+}
+
+// Tools returns the tools the extension registered.
+func (p *Process) Tools() []tools.Tool {
+	p.mu.Lock()
+	defer p.mu.Unlock()
+	out := make([]tools.Tool, len(p.tools))
+	for i := range p.tools {
+		out[i] = p.tools[i]
+	}
+	return out
+}
+
+// Commands returns slash-command names.
+func (p *Process) Commands() []string {
+	p.mu.Lock()
+	defer p.mu.Unlock()
+	return append([]string(nil), p.commands...)
+}
+
+// Providers returns endpoints the extension registered.
+func (p *Process) Providers() []llm.Provider {
+	p.mu.Lock()
+	defer p.mu.Unlock()
+	return append([]llm.Provider(nil), p.providers...)
+}
+
+// Hook tells the extension about an event. A block result stops the host action.
+func (p *Process) Hook(ctx context.Context, event, name string, args json.RawMessage) (bool, string) {
+	id := name + ":" + event
+	ch := make(chan Message, 1)
+	p.mu.Lock()
+	p.pending[id] = ch
+	p.mu.Unlock()
+	if err := p.send(Message{Type: "event", ID: id, Event: event, Name: name, Arguments: args}); err != nil {
+		return false, ""
+	}
+	select {
+	case <-ctx.Done():
+		return false, ctx.Err().Error()
+	case msg := <-ch:
+		return msg.Block, msg.Reason
+	}
+}
+
+func (p *Process) callTool(ctx context.Context, name string, args json.RawMessage) (tools.Result, error) {
+	id := name
+	ch := make(chan Message, 1)
+	p.mu.Lock()
+	p.pending[id] = ch
+	p.mu.Unlock()
+	if err := p.send(Message{Type: "tool_call", ID: id, Name: name, Arguments: args}); err != nil {
+		return tools.Result{IsError: true, Content: err.Error()}, err
+	}
+	select {
+	case <-ctx.Done():
+		return tools.Result{IsError: true, Content: ctx.Err().Error()}, ctx.Err()
+	case msg := <-ch:
+		return tools.Result{Content: msg.Content, IsError: msg.IsError}, nil
+	}
+}
+
+func (p *Process) send(msg Message) error {
+	raw, err := json.Marshal(msg)
+	if err != nil {
+		return err
+	}
+	p.mu.Lock()
+	defer p.mu.Unlock()
+	if p.closed {
+		return fmt.Errorf("ext: closed")
+	}
+	_, err = p.stdin.Write(append(raw, '\n'))
+	return err
+}
+
+func (p *Process) read(r io.Reader) {
+	sc := bufio.NewScanner(r)
+	sc.Buffer(make([]byte, 0, 64*1024), 4*1024*1024)
+	for sc.Scan() {
+		var msg Message
+		if err := json.Unmarshal(sc.Bytes(), &msg); err != nil {
+			continue
+		}
+		p.dispatch(msg)
+	}
+}
+
+func (p *Process) dispatch(msg Message) {
+	switch msg.Type {
+	case "register_tool":
+		p.mu.Lock()
+		p.tools = append(p.tools, registeredTool{
+			name: msg.Name, description: msg.Description, parameters: msg.Parameters, proc: p,
+		})
+		p.mu.Unlock()
+	case "register_command":
+		p.mu.Lock()
+		p.commands = append(p.commands, msg.Name)
+		p.mu.Unlock()
+	case "register_provider":
+		p.mu.Lock()
+		p.providers = append(p.providers, llm.Provider{
+			Name: msg.Name, Protocol: llm.Protocol(msg.Protocol), BaseURL: msg.BaseURL,
+		})
+		p.mu.Unlock()
+	case "tool_result", "hook_result":
+		p.mu.Lock()
+		ch := p.pending[msg.ID]
+		delete(p.pending, msg.ID)
+		p.mu.Unlock()
+		if ch != nil {
+			ch <- msg
+		}
+	case "ui_request":
+		select {
+		case p.UI <- UIRequest{ID: msg.ID, Message: msg.Message}:
+		default:
+		}
+	case "custom_entry":
+		if p.OnCustom != nil {
+			p.OnCustom(msg.CustomType, msg.Data)
+		}
+	}
+}
+
+// Close stops the subprocess.
+func (p *Process) Close() error {
+	p.mu.Lock()
+	p.closed = true
+	_ = p.stdin.Close()
+	p.mu.Unlock()
+	if p.cmd.Process != nil {
+		_ = p.cmd.Process.Kill()
+	}
+	return p.cmd.Wait()
+}

runtime/ext/process_test.go

@@ -0,0 +1,100 @@
+package ext
+
+import (
+	"context"
+	"os"
+	"path/filepath"
+	"testing"
+	"time"
+
+	"agent-native/runtime/ext/pkg"
+)
+
+func TestExtensionRegistersAndHooks(t *testing.T) {
+	dir := t.TempDir()
+	script := filepath.Join(dir, "ext.sh")
+	body := `#!/bin/sh
+while IFS= read -r line; do
+  case "$line" in
+    *'"type":"init"'*)
+      echo '{"type":"register_tool","name":"ping","description":"p","parameters":{"type":"object"}}'
+      echo '{"type":"register_command","name":"hi"}'
+      echo '{"type":"register_provider","name":"local","protocol":"openai-chat","baseURL":"http://127.0.0.1:9"}'
+      echo '{"type":"ui_request","id":"u1","message":"choose"}'
+      ;;
+    *'"type":"tool_call"'*)
+      echo '{"type":"tool_result","id":"ping","content":"pong"}'
+      ;;
+    *'"type":"event"'*)
+      echo '{"type":"hook_result","id":"ping:tool_call","block":true,"reason":"no"}'
+      ;;
+  esac
+done
+`
+	if err := os.WriteFile(script, []byte(body), 0o755); err != nil {
+		t.Fatal(err)
+	}
+	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
+	defer cancel()
+	proc, err := Start(ctx, []string{"/bin/sh", script})
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer proc.Close()
+	deadline := time.Now().Add(2 * time.Second)
+	for len(proc.Tools()) == 0 && time.Now().Before(deadline) {
+		time.Sleep(20 * time.Millisecond)
+	}
+	tools := proc.Tools()
+	if len(tools) != 1 || tools[0].Name() != "ping" {
+		t.Fatalf("tools %+v", tools)
+	}
+	res, err := tools[0].Execute(ctx, []byte(`{}`))
+	if err != nil || res.Content != "pong" {
+		t.Fatalf("tool: %+v %v", res, err)
+	}
+	block, reason := proc.Hook(ctx, "tool_call", "ping", []byte(`{}`))
+	if !block || reason != "no" {
+		t.Fatalf("hook %v %s", block, reason)
+	}
+	if len(proc.Commands()) != 1 || len(proc.Providers()) != 1 {
+		t.Fatalf("commands %v providers %v", proc.Commands(), proc.Providers())
+	}
+	select {
+	case ui := <-proc.UI:
+		if ui.Message != "choose" {
+			t.Fatal(ui)
+		}
+	case <-time.After(time.Second):
+		t.Fatal("missing ui request")
+	}
+}
+
+func TestPackageInstallListConfigRemove(t *testing.T) {
+	src := t.TempDir()
+	if err := os.WriteFile(filepath.Join(src, "agent-native.json"), []byte(`{"name":"demo","extensions":["./ext.sh"]}`), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	mgr := pkg.Manager{Dir: t.TempDir()}
+	manifest, err := mgr.Install(src)
+	if err != nil || manifest.Name != "demo" {
+		t.Fatalf("%+v %v", manifest, err)
+	}
+	if err := mgr.Config("demo", "source", src); err != nil {
+		t.Fatal(err)
+	}
+	list, err := mgr.List()
+	if err != nil || len(list) != 1 || list[0].Config["source"] != src {
+		t.Fatalf("list %+v %v", list, err)
+	}
+	if _, err := mgr.Update("demo"); err != nil {
+		t.Fatal(err)
+	}
+	if err := mgr.Remove("demo"); err != nil {
+		t.Fatal(err)
+	}
+	list, _ = mgr.List()
+	if len(list) != 0 {
+		t.Fatalf("still installed: %+v", list)
+	}
+}

runtime/llm/client.go

@@ -53,6 +53,9 @@ func (c *Client) run(ctx context.Context, model string, messages []Message, tool
 	if opts.MaxTokens > 0 {
 		body["max_tokens"] = opts.MaxTokens
 	}
+	if effort := OpenAIReasoningEffort(opts.ThinkingLevel); effort != "" {
+		body["reasoning_effort"] = effort
+	}
 	if len(tools) > 0 {
 		body["tools"] = toProviderTools(tools)
 		body["tool_choice"] = "auto"
@@ -125,7 +128,7 @@ func toProviderMessages(messages []Message) []map[string]any {
 		default:
 			out = append(out, map[string]any{
 				"role":    string(m.Role),
-				"content": m.Content,
+				"content": messageContent(m),
 			})
 		}
 	}

runtime/llm/providers.go

@@ -0,0 +1,431 @@
+package llm
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"os"
+	"strings"
+	"time"
+)
+
+// Thinking levels, from off to max. Unknown values are rejected by NormalizeThinking.
+var ThinkingLevels = []string{"off", "minimal", "low", "medium", "high", "xhigh", "max"}
+
+// NormalizeThinking returns the canonical level. Empty stays empty.
+func NormalizeThinking(level string) (string, error) {
+	if level == "" {
+		return "", nil
+	}
+	level = strings.ToLower(strings.TrimSpace(level))
+	for _, known := range ThinkingLevels {
+		if level == known {
+			return level, nil
+		}
+	}
+	return "", fmt.Errorf("llm: unknown thinking level %q", level)
+}
+
+// OpenAIReasoningEffort maps a thinking level onto OpenAI reasoning_effort.
+// off and empty produce no parameter.
+func OpenAIReasoningEffort(level string) string {
+	switch level {
+	case "minimal":
+		return "minimal"
+	case "low":
+		return "low"
+	case "medium":
+		return "medium"
+	case "high", "xhigh", "max":
+		return "high"
+	default:
+		return ""
+	}
+}
+
+// AnthropicThinkingBudget maps a thinking level onto a token budget.
+// off and empty disable thinking.
+func AnthropicThinkingBudget(level string) int {
+	switch level {
+	case "minimal":
+		return 1024
+	case "low":
+		return 2048
+	case "medium":
+		return 4096
+	case "high":
+		return 8192
+	case "xhigh":
+		return 16384
+	case "max":
+		return 32768
+	default:
+		return 0
+	}
+}
+
+// GoogleThinkingBudget maps a thinking level onto Gemini thinkingBudget.
+func GoogleThinkingBudget(level string) int {
+	switch level {
+	case "minimal", "low":
+		return 1024
+	case "medium":
+		return 4096
+	case "high":
+		return 8192
+	case "xhigh", "max":
+		return 24576
+	default:
+		return 0
+	}
+}
+
+// ModelRef is provider/model with an optional :thinking suffix.
+type ModelRef struct {
+	Provider string
+	Model    string
+	Thinking string
+}
+
+// ParseModelRef accepts "model", "provider/model", and "provider/model:level".
+func ParseModelRef(spec string) (ModelRef, error) {
+	spec = strings.TrimSpace(spec)
+	if spec == "" {
+		return ModelRef{}, fmt.Errorf("llm: empty model ref")
+	}
+	thinking := ""
+	if i := strings.LastIndex(spec, ":"); i >= 0 {
+		thinking = spec[i+1:]
+		spec = spec[:i]
+		norm, err := NormalizeThinking(thinking)
+		if err != nil {
+			return ModelRef{}, err
+		}
+		thinking = norm
+	}
+	provider, model := "", spec
+	if i := strings.Index(spec, "/"); i >= 0 {
+		provider = spec[:i]
+		model = spec[i+1:]
+	}
+	if model == "" {
+		return ModelRef{}, fmt.Errorf("llm: model id missing in %q", spec)
+	}
+	return ModelRef{Provider: provider, Model: model, Thinking: thinking}, nil
+}
+
+// Protocol names the wire format a provider speaks.
+type Protocol string
+
+const (
+	ProtocolOpenAIChat      Protocol = "openai-chat"
+	ProtocolOpenAIResponses Protocol = "openai-responses"
+	ProtocolAnthropic       Protocol = "anthropic"
+	ProtocolGoogle          Protocol = "google"
+	ProtocolAzure           Protocol = "azure"
+)
+
+// Provider is one configured endpoint.
+type Provider struct {
+	Name      string   `json:"name"`
+	Protocol  Protocol `json:"protocol"`
+	BaseURL   string   `json:"baseURL"`
+	APIKeyEnv string   `json:"apiKeyEnv,omitempty"`
+	APIKey    string   `json:"-"`
+}
+
+// Presets are OpenAI-compatible or native endpoints that only need an API key.
+func Presets() []Provider {
+	return []Provider{
+		{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: ProtocolGoogle, BaseURL: "https://generativelanguage.googleapis.com/v1beta", APIKeyEnv: "GEMINI_API_KEY"},
+		{Name: "azure", Protocol: ProtocolAzure, BaseURL: "", APIKeyEnv: "AZURE_OPENAI_API_KEY"},
+		{Name: "deepseek", Protocol: ProtocolOpenAIChat, BaseURL: "https://api.deepseek.com/v1", APIKeyEnv: "DEEPSEEK_API_KEY"},
+		{Name: "minimax", Protocol: ProtocolOpenAIChat, BaseURL: "https://api.minimaxi.com/v1", APIKeyEnv: "MINIMAX_API_KEY"},
+		{Name: "openrouter", Protocol: ProtocolOpenAIChat, BaseURL: "https://openrouter.ai/api/v1", APIKeyEnv: "OPENROUTER_API_KEY"},
+		{Name: "groq", Protocol: ProtocolOpenAIChat, BaseURL: "https://api.groq.com/openai/v1", APIKeyEnv: "GROQ_API_KEY"},
+	}
+}
+
+// modelsFile is the on-disk shape of models.json.
+type modelsFile struct {
+	Providers []Provider `json:"providers"`
+}
+
+// LoadModelsJSON reads custom providers. Missing file yields an empty list.
+func LoadModelsJSON(path string) ([]Provider, error) {
+	raw, err := os.ReadFile(path)
+	if err != nil {
+		if os.IsNotExist(err) {
+			return nil, nil
+		}
+		return nil, err
+	}
+	var file modelsFile
+	if err := json.Unmarshal(raw, &file); err != nil {
+		return nil, err
+	}
+	for i := range file.Providers {
+		if file.Providers[i].APIKey == "" && file.Providers[i].APIKeyEnv != "" {
+			file.Providers[i].APIKey = os.Getenv(file.Providers[i].APIKeyEnv)
+		}
+	}
+	return file.Providers, nil
+}
+
+// Lookup finds a preset or a custom provider by name.
+func Lookup(name string, custom []Provider) (Provider, bool) {
+	for _, p := range custom {
+		if p.Name == name {
+			return p, true
+		}
+	}
+	for _, p := range Presets() {
+		if p.Name == name {
+			if p.APIKey == "" {
+				p.APIKey = os.Getenv(p.APIKeyEnv)
+			}
+			return p, true
+		}
+	}
+	return Provider{}, false
+}
+
+func messageContent(m Message) any {
+	if len(m.Images) == 0 {
+		return m.Content
+	}
+	parts := []map[string]any{}
+	if m.Content != "" {
+		parts = append(parts, map[string]any{"type": "text", "text": m.Content})
+	}
+	for _, img := range m.Images {
+		url := img.URL
+		if url == "" && img.Data != "" {
+			mime := img.MIME
+			if mime == "" {
+				mime = "image/png"
+			}
+			url = "data:" + mime + ";base64," + img.Data
+		}
+		parts = append(parts, map[string]any{
+			"type":      "image_url",
+			"image_url": map[string]any{"url": url},
+		})
+	}
+	return parts
+}
+
+// BuildBody returns the JSON body for one native protocol.
+func BuildBody(protocol Protocol, model string, messages []Message, tools []ToolDecl, opts StreamOptions) ([]byte, error) {
+	switch protocol {
+	case ProtocolOpenAIChat, ProtocolAzure:
+		body := map[string]any{
+			"model":    model,
+			"messages": toProviderMessages(messages),
+			"stream":   true,
+		}
+		if opts.MaxTokens > 0 {
+			body["max_tokens"] = opts.MaxTokens
+		}
+		if effort := OpenAIReasoningEffort(opts.ThinkingLevel); effort != "" {
+			body["reasoning_effort"] = effort
+		}
+		if len(tools) > 0 {
+			body["tools"] = toProviderTools(tools)
+		}
+		return json.Marshal(body)
+	case ProtocolOpenAIResponses:
+		body := map[string]any{
+			"model":  model,
+			"input":  responsesInput(messages),
+			"stream": true,
+		}
+		if opts.MaxTokens > 0 {
+			body["max_output_tokens"] = opts.MaxTokens
+		}
+		if effort := OpenAIReasoningEffort(opts.ThinkingLevel); effort != "" {
+			body["reasoning"] = map[string]any{"effort": effort}
+		}
+		return json.Marshal(body)
+	case ProtocolAnthropic:
+		sys, msgs := anthropicMessages(messages)
+		body := map[string]any{
+			"model":      model,
+			"messages":   msgs,
+			"max_tokens": opts.MaxTokens,
+			"stream":     true,
+		}
+		if body["max_tokens"] == 0 {
+			body["max_tokens"] = 4096
+		}
+		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)
+	case ProtocolGoogle:
+		body := map[string]any{
+			"contents": googleContents(messages),
+		}
+		if budget := GoogleThinkingBudget(opts.ThinkingLevel); budget > 0 {
+			body["generationConfig"] = map[string]any{
+				"thinkingConfig": map[string]any{"thinkingBudget": budget},
+			}
+		}
+		return json.Marshal(body)
+	default:
+		return nil, fmt.Errorf("llm: unknown protocol %q", protocol)
+	}
+}
+
+func responsesInput(messages []Message) []map[string]any {
+	out := make([]map[string]any, 0, len(messages))
+	for _, m := range messages {
+		role := string(m.Role)
+		if role == "" {
+			role = "user"
+		}
+		out = append(out, map[string]any{
+			"role":    role,
+			"content": messageContent(m),
+		})
+	}
+	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 {
+		if m.Role == RoleSystem {
+			continue
+		}
+		role := "user"
+		if m.Role == RoleAssistant {
+			role = "model"
+		}
+		parts := []map[string]any{}
+		if m.Content != "" {
+			parts = append(parts, map[string]any{"text": m.Content})
+		}
+		for _, img := range m.Images {
+			parts = append(parts, map[string]any{
+				"inlineData": map[string]any{
+					"mimeType": firstNonEmpty(img.MIME, "image/png"),
+					"data":     img.Data,
+				},
+			})
+		}
+		out = append(out, map[string]any{"role": role, "parts": parts})
+	}
+	return out
+}
+
+func firstNonEmpty(a, b string) string {
+	if a != "" {
+		return a
+	}
+	return b
+}
+
+// ContextOverflow reports provider errors that ask the caller to compact.
+func ContextOverflow(message string) bool {
+	m := strings.ToLower(message)
+	return strings.Contains(m, "context") && (strings.Contains(m, "overflow") || strings.Contains(m, "too long") || strings.Contains(m, "length"))
+}
+
+// WithRetry re-invokes base on a retryable terminal error, up to attempts
+// extra tries. onEvent receives auto_retry_start and auto_retry_end.
+// Context-overflow errors are returned immediately so the caller can compact.
+func WithRetry(base StreamFunc, attempts int, onEvent func(kind, detail string)) StreamFunc {
+	if attempts < 1 {
+		attempts = 1
+	}
+	return func(ctx context.Context, model string, messages []Message, tools []ToolDecl, opts StreamOptions) <-chan Event {
+		out := make(chan Event)
+		go func() {
+			defer close(out)
+			var last []Event
+			for try := 0; try <= attempts; try++ {
+				if try > 0 {
+					detail := fmt.Sprintf("attempt %d", try+1)
+					if onEvent != nil {
+						onEvent("auto_retry_start", detail)
+					}
+					delay := time.Duration(200*(1<<uint(try-1))) * time.Millisecond
+					select {
+					case <-ctx.Done():
+						emitAbort(out, "", ctx.Err())
+						return
+					case <-time.After(delay):
+					}
+				}
+				last = nil
+				for ev := range base(ctx, model, messages, tools, opts) {
+					last = append(last, ev)
+				}
+				overflow, retryable := false, false
+				for _, ev := range last {
+					if ev.Kind == KindError && ContextOverflow(ev.ErrorMessage) {
+						overflow = true
+					}
+					if ev.Kind == KindError && ev.Retryable {
+						retryable = true
+					}
+				}
+				if overflow || !retryable || try == attempts {
+					if try > 0 && onEvent != nil {
+						onEvent("auto_retry_end", fmt.Sprintf("attempt %d", try+1))
+					}
+					for _, ev := range last {
+						select {
+						case <-ctx.Done():
+							return
+						case out <- ev:
+						}
+					}
+					return
+				}
+			}
+		}()
+		return out
+	}
+}

runtime/llm/providers_test.go

@@ -0,0 +1,115 @@
+package llm
+
+import (
+	"context"
+	"encoding/json"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+)
+
+func TestParseModelRefAndThinking(t *testing.T) {
+	ref, err := ParseModelRef("anthropic/claude-sonnet:high")
+	if err != nil {
+		t.Fatal(err)
+	}
+	if ref.Provider != "anthropic" || ref.Model != "claude-sonnet" || ref.Thinking != "high" {
+		t.Fatalf("%+v", ref)
+	}
+	if _, err := ParseModelRef("openai/gpt:nope"); err == nil {
+		t.Fatal("unknown thinking level must fail")
+	}
+	if AnthropicThinkingBudget("high") != 8192 || OpenAIReasoningEffort("off") != "" {
+		t.Fatal("thinking map")
+	}
+	p, ok := Lookup("groq", nil)
+	if !ok || !strings.Contains(p.BaseURL, "groq") || p.Protocol != ProtocolOpenAIChat {
+		t.Fatalf("preset: %+v", p)
+	}
+}
+
+func TestModelsJSONAndBodies(t *testing.T) {
+	dir := t.TempDir()
+	path := filepath.Join(dir, "models.json")
+	raw := []byte(`{"providers":[{"name":"local","protocol":"openai-chat","baseURL":"http://127.0.0.1:9","apiKeyEnv":"UNUSED_KEY"}]}`)
+	if err := os.WriteFile(path, raw, 0o644); err != nil {
+		t.Fatal(err)
+	}
+	got, err := LoadModelsJSON(path)
+	if err != nil || len(got) != 1 || got[0].Name != "local" {
+		t.Fatalf("%+v %v", got, err)
+	}
+	msgs := []Message{{
+		Role: RoleUser, Content: "see",
+		Images: []Image{{MIME: "image/png", Data: "aaaa"}},
+	}}
+	body, err := BuildBody(ProtocolAnthropic, "claude", append([]Message{{Role: RoleSystem, Content: "sys"}}, msgs...), nil, StreamOptions{ThinkingLevel: "low"})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if !strings.Contains(string(body), "budget_tokens") || !strings.Contains(string(body), "base64") {
+		t.Fatalf("anthropic body: %s", body)
+	}
+	gbody, err := BuildBody(ProtocolGoogle, "gemini", msgs, nil, StreamOptions{ThinkingLevel: "max"})
+	if err != nil || !strings.Contains(string(gbody), "inlineData") || !strings.Contains(string(gbody), "thinkingBudget") {
+		t.Fatalf("google: %s %v", gbody, err)
+	}
+	rbody, err := BuildBody(ProtocolOpenAIResponses, "gpt", msgs, nil, StreamOptions{ThinkingLevel: "medium", MaxTokens: 20})
+	if err != nil || !strings.Contains(string(rbody), "max_output_tokens") || !strings.Contains(string(rbody), "image_url") {
+		t.Fatalf("responses: %s %v", rbody, err)
+	}
+	var decoded map[string]any
+	if err := json.Unmarshal(rbody, &decoded); err != nil {
+		t.Fatal(err)
+	}
+}
+
+func TestRetryThenSuccess(t *testing.T) {
+	n := 0
+	base := StreamFunc(func(ctx context.Context, model string, messages []Message, tools []ToolDecl, opts StreamOptions) <-chan Event {
+		out := make(chan Event, 2)
+		n++
+		if n == 1 {
+			out <- Event{Kind: KindError, Retryable: true, ErrorMessage: "429"}
+		} else {
+			out <- Event{Kind: KindTextDelta, Text: "ok"}
+			out <- Event{Kind: KindMessageEnd, Message: Message{Role: RoleAssistant, Content: "ok"}}
+		}
+		close(out)
+		return out
+	})
+	var events []string
+	wrapped := WithRetry(base, 2, func(kind, detail string) { events = append(events, kind) })
+	var texts []string
+	for ev := range wrapped(context.Background(), "m", nil, nil, StreamOptions{}) {
+		if ev.Kind == KindTextDelta {
+			texts = append(texts, ev.Text)
+		}
+	}
+	if len(texts) != 1 || texts[0] != "ok" {
+		t.Fatalf("texts %v calls %d", texts, n)
+	}
+	if len(events) == 0 || events[0] != "auto_retry_start" {
+		t.Fatalf("events %v", events)
+	}
+}
+
+func TestContextOverflowIsNotRetried(t *testing.T) {
+	if !ContextOverflow("maximum context length exceeded") {
+		t.Fatal("expected overflow")
+	}
+	n := 0
+	base := StreamFunc(func(ctx context.Context, model string, messages []Message, tools []ToolDecl, opts StreamOptions) <-chan Event {
+		out := make(chan Event, 1)
+		n++
+		out <- Event{Kind: KindError, Retryable: true, ErrorMessage: "context length overflow"}
+		close(out)
+		return out
+	})
+	for range WithRetry(base, 3, nil)(context.Background(), "m", nil, nil, StreamOptions{}) {
+	}
+	if n != 1 {
+		t.Fatalf("overflow must not retry, calls=%d", n)
+	}
+}

runtime/llm/sse.go

@@ -26,12 +26,19 @@ type sseChoice struct {
 
 type sseChunk struct {
 	Choices []sseChoice `json:"choices"`
-	Usage   *Usage      `json:"usage"`
+	Usage   *usageWire  `json:"usage"`
 	Error   *struct {
 		Message string `json:"message"`
 	} `json:"error"`
 }
 
+type usageWire struct {
+	Usage
+	Details struct {
+		Cached int `json:"cached_tokens"`
+	} `json:"prompt_tokens_details"`
+}
+
 // ParseSSE reads an OpenAI-compatible server-sent event body and emits
 // deltas plus one terminal message_end or error event.
 func ParseSSE(ctx context.Context, r io.Reader) <-chan Event {
@@ -77,7 +84,10 @@ func emitParsed(ctx context.Context, r io.Reader, out chan<- Event) {
 			return
 		}
 		if chunk.Usage != nil {
-			usage = *chunk.Usage
+			usage = chunk.Usage.Usage
+			if chunk.Usage.Details.Cached > 0 {
+				usage.CacheReadTokens = chunk.Usage.Details.Cached
+			}
 		}
 		if len(chunk.Choices) == 0 {
 			continue

runtime/llm/types.go

@@ -48,6 +48,7 @@ type Message struct {
 	Role       Role       `json:"role"`
 	Content    string     `json:"content,omitempty"`
 	Name       string     `json:"name,omitempty"`
+	Images     []Image    `json:"images,omitempty"`
 	ToolCallID string     `json:"tool_call_id,omitempty"`
 	ToolCalls  []ToolCall `json:"tool_calls,omitempty"`
 	StopReason StopReason `json:"stop_reason,omitempty"`
@@ -61,11 +62,20 @@ type ToolDecl struct {
 	Parameters  json.RawMessage `json:"parameters"`
 }
 
+// Image is one picture attached to a user message.
+type Image struct {
+	MIME string `json:"mime,omitempty"`
+	URL  string `json:"url,omitempty"`
+	Data string `json:"data,omitempty"`
+}
+
 // Usage counts tokens reported by the provider. Zero means the provider omitted usage.
 type Usage struct {
 	PromptTokens     int `json:"prompt_tokens"`
 	CompletionTokens int `json:"completion_tokens"`
 	TotalTokens      int `json:"total_tokens"`
+	CacheReadTokens  int `json:"cache_read_tokens,omitempty"`
+	CacheWriteTokens int `json:"cache_write_tokens,omitempty"`
 }
 
 // Event is one streamed update. Failures are KindError plus a final assistant
@@ -90,6 +100,9 @@ type StreamOptions struct {
 	MaxTokens int
 	APIBase   string
 	APIKey    string
+	// ThinkingLevel is off, minimal, low, medium, high, xhigh, or max.
+	// Empty leaves the provider default unchanged.
+	ThinkingLevel string
 }
 
 // StreamFunc is the provider boundary. It must return a channel that is closed

runtime/loop.go

@@ -29,6 +29,9 @@ type Event struct {
 	// Detail carries tool arguments/results or reasoning deltas for
 	// process-visualization consumers; log writers may ignore it.
 	Detail string
+	// PromptTokens / CompletionTokens 只在 type=usage 时有意义。
+	PromptTokens     int
+	CompletionTokens int
 }
 
 // Config drives one agent session.
@@ -43,6 +46,10 @@ type Config struct {
 	GetSteering   func() []llm.Message
 	GetFollowUp   func() []llm.Message
 	OnEvent       func(Event)
+	// ThinkingLevel is forwarded on each provider request.
+	ThinkingLevel string
+	// Resolve, when set, overrides Model and ThinkingLevel at the start of each turn.
+	Resolve func() (model, thinking string)
 	// Log 可选。nil 时丢弃日志，不改变循环结果。
 	Log *logsys.Logger
 	// ExecutionID 只用于日志字段，不参与控制流。
@@ -114,6 +121,7 @@ func Run(ctx context.Context, cfg Config, user llm.Message) (Result, error) {
 			return budgetStop(cfg, last, messages, usage, reason), nil
 		}
 		emit(cfg, Event{Type: "turn_start"})
+		emit(cfg, Event{Type: "message_start"})
 		end, errEv := consume(ctx, cfg, messages)
 		if errEv != nil {
 			cfg.logger().Error("runtime turn failed", errEv, "retryable", end.Retryable)
@@ -164,6 +172,8 @@ func Run(ctx context.Context, cfg Config, user llm.Message) (Result, error) {
 			continue
 		}
 		emit(cfg, Event{Type: "agent_end", Message: last})
+		emit(cfg, Event{Type: "agent_settled", Message: last})
+		emitUsage(cfg, usage)
 		return Result{Text: last.Content, Messages: messages, StopReason: last.StopReason, Usage: usage}, nil
 	}
 	errCap := errString("max turns exceeded")
@@ -218,9 +228,21 @@ func consume(ctx context.Context, cfg Config, messages []llm.Message) (llm.Event
 			decls = append(decls, llm.ToolDecl{Name: d.Name, Description: d.Description, Parameters: d.Parameters})
 		}
 	}
+	model := cfg.Model
+	opts := llm.StreamOptions{ThinkingLevel: cfg.ThinkingLevel}
+	if cfg.Resolve != nil {
+		if m, th := cfg.Resolve(); m != "" || th != "" {
+			if m != "" {
+				model = m
+			}
+			if th != "" {
+				opts.ThinkingLevel = th
+			}
+		}
+	}
 	var end llm.Event
 	sawEnd := false
-	for ev := range cfg.Stream(ctx, cfg.Model, messages, decls, llm.StreamOptions{}) {
+	for ev := range cfg.Stream(ctx, model, messages, decls, opts) {
 		switch ev.Kind {
 		case llm.KindTextDelta:
 			emit(cfg, Event{Type: "text_delta", Text: ev.Text})
@@ -267,7 +289,8 @@ func executeParallel(ctx context.Context, cfg Config, assistant llm.Message) []l
 }
 
 func runOne(ctx context.Context, cfg Config, call llm.ToolCall) llm.Message {
-	emit(cfg, Event{Type: "tool_execution_start", Tool: call.Name})
+	emit(cfg, Event{Type: "tool_execution_start", Tool: call.Name, Detail: call.Arguments})
+	emit(cfg, Event{Type: "tool_execution_update", Tool: call.Name, Detail: call.Arguments})
 	res := tools.Result{Content: "no tools configured", IsError: true}
 	if cfg.Tools != nil {
 		res = cfg.Tools.Execute(ctx, call.Name, json.RawMessage(call.Arguments))
@@ -296,6 +319,10 @@ func drain(fn func() []llm.Message) []llm.Message {
 	return fn()
 }
 
+func emitUsage(cfg Config, usage llm.Usage) {
+	emit(cfg, Event{Type: "usage", PromptTokens: usage.PromptTokens, CompletionTokens: usage.CompletionTokens})
+}
+
 func emit(cfg Config, ev Event) {
 	if cfg.OnEvent != nil {
 		cfg.OnEvent(ev)

runtime/resources/context.go

@@ -0,0 +1,37 @@
+package resources
+
+import (
+	"os"
+	"path/filepath"
+	"strings"
+)
+
+// ContextFiles walks from dir toward the filesystem root and concatenates
+// AGENTS.md and CLAUDE.md (nearest first). SYSTEM.md and APPEND_SYSTEM.md
+// are read from dir only.
+func ContextFiles(dir string) (string, error) {
+	var parts []string
+	cur := dir
+	for {
+		for _, name := range []string{"AGENTS.md", "CLAUDE.md"} {
+			raw, err := os.ReadFile(filepath.Join(cur, name))
+			if err != nil {
+				continue
+			}
+			parts = append(parts, "# "+name+" ("+cur+")\n"+string(raw))
+		}
+		parent := filepath.Dir(cur)
+		if parent == cur {
+			break
+		}
+		cur = parent
+	}
+	for _, name := range []string{"SYSTEM.md", "APPEND_SYSTEM.md"} {
+		raw, err := os.ReadFile(filepath.Join(dir, name))
+		if err != nil {
+			continue
+		}
+		parts = append(parts, "# "+name+"\n"+string(raw))
+	}
+	return strings.Join(parts, "\n\n"), nil
+}

runtime/resources/prompt.go

@@ -0,0 +1,44 @@
+package resources
+
+import (
+	"os"
+	"path/filepath"
+	"strings"
+)
+
+// Expand replaces {{var}} placeholders and /name includes.
+// An include reads <dir>/<name>.md. Unknown variables are left intact.
+func Expand(text, dir string, vars map[string]string) (string, error) {
+	out := text
+	for key, value := range vars {
+		out = strings.ReplaceAll(out, "{{"+key+"}}", value)
+	}
+	for {
+		i := strings.Index(out, "/")
+		if i < 0 {
+			break
+		}
+		rest := out[i+1:]
+		j := 0
+		for j < len(rest) && isName(rest[j]) {
+			j++
+		}
+		if j == 0 || (i > 0 && out[i-1] != ' ' && out[i-1] != '\n') {
+			break
+		}
+		name := rest[:j]
+		raw, err := os.ReadFile(filepath.Join(dir, name+".md"))
+		if err != nil {
+			break
+		}
+		out = out[:i] + string(raw) + out[i+1+j:]
+	}
+	for key, value := range vars {
+		out = strings.ReplaceAll(out, "{{"+key+"}}", value)
+	}
+	return out, nil
+}
+
+func isName(b byte) bool {
+	return b == '-' || b == '_' || (b >= 'a' && b <= 'z') || (b >= 'A' && b <= 'Z') || (b >= '0' && b <= '9')
+}

runtime/resources/resources_test.go

@@ -0,0 +1,72 @@
+package resources
+
+import (
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+
+	"agent-native/runtime/skills"
+	"agent-native/runtime/tools"
+)
+
+func TestContextAndPromptAndTrust(t *testing.T) {
+	root := t.TempDir()
+	proj := filepath.Join(root, "proj")
+	if err := os.MkdirAll(proj, 0o755); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.WriteFile(filepath.Join(root, "AGENTS.md"), []byte("root-agents"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.WriteFile(filepath.Join(proj, "SYSTEM.md"), []byte("sys"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	text, err := ContextFiles(proj)
+	if err != nil || !strings.Contains(text, "root-agents") || !strings.Contains(text, "sys") {
+		t.Fatalf("context: %s %v", text, err)
+	}
+	if err := os.WriteFile(filepath.Join(proj, "greet.md"), []byte("hello {{name}}"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	got, err := Expand("say /greet", proj, map[string]string{"name": "Ada"})
+	if err != nil || got != "say hello Ada" {
+		t.Fatalf("expand: %q %v", got, err)
+	}
+	if ProjectTrusted(TrustAsk, false, false) {
+		t.Fatal("ask is not trusted without approve")
+	}
+	if !ProjectTrusted(TrustNever, true, false) {
+		t.Fatal("approve overrides never")
+	}
+	if ProjectTrusted(TrustAlways, false, true) {
+		t.Fatal("no-approve wins")
+	}
+}
+
+func TestPiAliasesFilterAndSkillCommand(t *testing.T) {
+	reg := tools.WorkspaceTools(t.TempDir())
+	reg.RegisterPiAliases()
+	if !reg.Has("read") || !reg.Has("ls") || !reg.Has("read_file") {
+		t.Fatal("aliases missing")
+	}
+	if err := reg.Filter(tools.FilterOptions{Exclude: []string{"read"}, NoBuiltin: false}); err != nil {
+		t.Fatal(err)
+	}
+	if reg.Has("read") || !reg.Has("read_file") {
+		t.Fatal("exclude should drop only the alias")
+	}
+	dir := t.TempDir()
+	skillDir := filepath.Join(dir, "review")
+	if err := os.MkdirAll(skillDir, 0o755); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.WriteFile(filepath.Join(skillDir, "SKILL.md"), []byte("be thorough"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	store := skills.NewStore(dir)
+	out, err := store.ExpandCommands("use /skill:review now")
+	if err != nil || !strings.Contains(out, "be thorough") {
+		t.Fatalf("skill expand: %q %v", out, err)
+	}
+}

runtime/resources/settings.go

@@ -0,0 +1,102 @@
+package resources
+
+import (
+	"encoding/json"
+	"os"
+	"path/filepath"
+)
+
+// Trust is how a project directory is treated in non-interactive runs.
+type Trust string
+
+const (
+	TrustAsk    Trust = "ask"
+	TrustAlways Trust = "always"
+	TrustNever  Trust = "never"
+)
+
+// Settings is the global plus project configuration.
+type Settings struct {
+	Trust    Trust             `json:"trust,omitempty"`
+	Model    string            `json:"model,omitempty"`
+	Thinking string            `json:"thinking,omitempty"`
+	Tools    []string          `json:"tools,omitempty"`
+	Exclude  []string          `json:"exclude,omitempty"`
+	Vars     map[string]string `json:"vars,omitempty"`
+}
+
+// LoadSettings reads global (optional) then project settings.
+// Project keys override global keys. A missing file is an empty settings.
+func LoadSettings(globalPath, projectPath string) (Settings, error) {
+	global, err := readSettings(globalPath)
+	if err != nil {
+		return Settings{}, err
+	}
+	project, err := readSettings(projectPath)
+	if err != nil {
+		return Settings{}, err
+	}
+	return mergeSettings(global, project), nil
+}
+
+// ProjectTrusted reports whether project resources may be loaded.
+// ask and never refuse in non-interactive mode. approve forces trust.
+func ProjectTrusted(trust Trust, approve, noApprove bool) bool {
+	if noApprove {
+		return false
+	}
+	if approve {
+		return true
+	}
+	return trust == TrustAlways
+}
+
+func readSettings(path string) (Settings, error) {
+	if path == "" {
+		return Settings{}, nil
+	}
+	raw, err := os.ReadFile(path)
+	if err != nil {
+		if os.IsNotExist(err) {
+			return Settings{}, nil
+		}
+		return Settings{}, err
+	}
+	var s Settings
+	if err := json.Unmarshal(raw, &s); err != nil {
+		return Settings{}, err
+	}
+	return s, nil
+}
+
+func mergeSettings(base, over Settings) Settings {
+	if over.Trust != "" {
+		base.Trust = over.Trust
+	}
+	if over.Model != "" {
+		base.Model = over.Model
+	}
+	if over.Thinking != "" {
+		base.Thinking = over.Thinking
+	}
+	if over.Tools != nil {
+		base.Tools = over.Tools
+	}
+	if over.Exclude != nil {
+		base.Exclude = over.Exclude
+	}
+	if len(over.Vars) > 0 {
+		if base.Vars == nil {
+			base.Vars = map[string]string{}
+		}
+		for k, v := range over.Vars {
+			base.Vars[k] = v
+		}
+	}
+	return base
+}
+
+// DefaultPaths returns the global and project settings paths for a cwd.
+func DefaultPaths(home, cwd string) (string, string) {
+	return filepath.Join(home, ".agent-native", "settings.json"), filepath.Join(cwd, ".agent-native", "settings.json")
+}

runtime/session/catalog.go

@@ -0,0 +1,105 @@
+package session
+
+import (
+	"os"
+	"path/filepath"
+	"sort"
+	"strings"
+)
+
+// Info describes one session file in a directory.
+type Info struct {
+	ID      string
+	Path    string
+	Name    string
+	CWD     string
+	Leaf    string
+	Updated string
+}
+
+// List returns session files in dir, newest first.
+func List(dir string) ([]Info, error) {
+	ents, err := os.ReadDir(dir)
+	if err != nil {
+		if os.IsNotExist(err) {
+			return nil, nil
+		}
+		return nil, err
+	}
+	var out []Info
+	for _, ent := range ents {
+		name := ent.Name()
+		if ent.IsDir() || !strings.HasSuffix(name, ".jsonl") {
+			continue
+		}
+		path := filepath.Join(dir, name)
+		st, err := Open(path)
+		if err != nil {
+			continue
+		}
+		info, err := st.Info()
+		if err != nil {
+			continue
+		}
+		out = append(out, info)
+	}
+	sort.Slice(out, func(i, j int) bool { return out[i].Updated > out[j].Updated })
+	return out, nil
+}
+
+// Continue opens the newest session whose cwd matches. A second return
+// value of false means there is nothing to continue.
+func Continue(dir, cwd string) (*Store, bool, error) {
+	list, err := List(dir)
+	if err != nil {
+		return nil, false, err
+	}
+	for _, info := range list {
+		if cwd == "" || info.CWD == cwd {
+			st, err := Open(info.Path)
+			return st, err == nil, err
+		}
+	}
+	return nil, false, nil
+}
+
+// Resume opens a session by id (filename stem) or by name.
+func Resume(dir, idOrName string) (*Store, error) {
+	list, err := List(dir)
+	if err != nil {
+		return nil, err
+	}
+	for _, info := range list {
+		if info.ID == idOrName || info.Name == idOrName {
+			return Open(info.Path)
+		}
+	}
+	return nil, os.ErrNotExist
+}
+
+// SetCWD records the working directory used by Continue.
+func (s *Store) SetCWD(cwd string) error {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	return s.writeMetaKeyLocked("cwd", cwd)
+}
+
+// Info reads the catalog fields for this file.
+func (s *Store) Info() (Info, error) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	entries, err := s.readEntriesLocked()
+	if err != nil {
+		return Info{}, err
+	}
+	meta := s.readMetaLocked()
+	updated := ""
+	if len(entries) > 0 {
+		updated = entries[len(entries)-1].Timestamp
+	}
+	id := strings.TrimSuffix(filepath.Base(s.path), ".jsonl")
+	return Info{
+		ID: id, Path: s.path, Name: meta["name"], CWD: meta["cwd"],
+		Leaf: s.leafLocked(entries), Updated: updated,
+	}, nil
+}

runtime/session/compact.go

@@ -66,73 +66,53 @@ func (s *Store) ReadMeta() map[string]string {
 	return meta
 }
 
-// Compact 把转录改写为「系统消息 + 摘要消息 + 最近 keepLastN 条」。
-// 摘要由 summarize 生成（由适配层注入，通常是一次无工具模型调用）；
-// 改写原子落盘（临时文件 + rename）。
-// 返回是否发生了压缩（消息数不足/已是紧凑形态时返回 false 且不动文件）。
+// Compact appends a compaction entry. The file keeps every earlier record.
+// Load hides the summarized prefix behind the summary plus the last
+// keepLastN non-system messages. Returns false when there is nothing to fold.
 func (s *Store) Compact(keepLastN int, summarize func(messages []llm.Message) (string, error)) (bool, error) {
-	messages, err := s.Load()
+	s.mu.Lock()
+	entries, err := s.readEntriesLocked()
+	s.mu.Unlock()
 	if err != nil {
 		return false, err
 	}
-	if len(messages) <= keepLastN+1 { // 系统消息 + 尾部，无可压缩头部
+	leaf := ""
+	s.mu.Lock()
+	leaf = s.leafLocked(entries)
+	s.mu.Unlock()
+	path := pathTo(entries, leaf)
+	var messages []llm.Message
+	var msgEntries []Entry
+	for _, e := range path {
+		if e.Type == TypeMessage && e.Message != nil {
+			messages = append(messages, *e.Message)
+			msgEntries = append(msgEntries, e)
+		}
+	}
+	if len(messages) <= keepLastN+1 {
 		return false, nil
 	}
-	var system *llm.Message
 	rest := messages
+	restEntries := msgEntries
 	if len(rest) > 0 && rest[0].Role == llm.RoleSystem {
-		copyOf := rest[0]
-		system = &copyOf
 		rest = rest[1:]
+		restEntries = restEntries[1:]
 	}
 	tailStart := len(rest) - keepLastN
 	if tailStart <= 0 {
 		return false, nil
 	}
 	head := rest[:tailStart]
-	tail := rest[tailStart:]
 	summary, err := summarize(head)
 	if err != nil {
 		return false, err
 	}
-	summaryMsg := llm.Message{
-		Role:    llm.RoleUser,
-		Content: "[conversation summary]\n" + summary + "\n[recent messages follow]",
+	var tailIDs []string
+	for _, e := range restEntries[tailStart:] {
+		tailIDs = append(tailIDs, e.ID)
 	}
-	var rebuilt []llm.Message
-	if system != nil {
-		rebuilt = append(rebuilt, *system)
-	}
-	rebuilt = append(rebuilt, summaryMsg)
-	rebuilt = append(rebuilt, tail...)
-
 	s.mu.Lock()
 	defer s.mu.Unlock()
-	tmp := s.path + ".compact.tmp"
-	f, err := os.OpenFile(tmp, os.O_CREATE|os.O_TRUNC|os.O_WRONLY, 0o644)
-	if err != nil {
-		return false, err
-	}
-	w := bufio.NewWriter(f)
-	for _, m := range rebuilt {
-		raw, err := json.Marshal(m)
-		if err != nil {
-			_ = f.Close()
-			_ = os.Remove(tmp)
-			return false, err
-		}
-		w.Write(raw)
-		w.WriteByte('\n')
-	}
-	if err := w.Flush(); err != nil {
-		_ = f.Close()
-		_ = os.Remove(tmp)
-		return false, err
-	}
-	_ = f.Close()
-	if err := os.Rename(tmp, s.path); err != nil {
-		_ = os.Remove(tmp)
-		return false, err
-	}
-	return true, nil
+	_, err = s.appendEntryLocked(Entry{Type: TypeCompaction, Summary: summary, TailIDs: tailIDs})
+	return err == nil, err
 }

runtime/session/compact_test.go

@@ -114,9 +114,12 @@ func TestCompactRewritesTranscript(t *testing.T) {
 	if msgs[2].Content != "reply" || msgs[3].Content != "earlyj" || msgs[4].Content != "reply" {
 		t.Fatalf("tail order wrong: %+v", msgs[2:])
 	}
-	// 无 tmp 残留。
-	if _, err := os.Stat(s.Path() + ".compact.tmp"); !os.IsNotExist(err) {
-		t.Fatal("compact tmp file must not remain")
+	entries, err := s.Entries()
+	if err != nil {
+		t.Fatal(err)
+	}
+	if len(entries) <= 5 {
+		t.Fatalf("compaction must append, not rewrite; got %d entries", len(entries))
 	}
 }
 

runtime/session/html.go

@@ -0,0 +1,32 @@
+package session
+
+import (
+	"fmt"
+	"html"
+	"os"
+	"strings"
+)
+
+// ExportHTML writes the active branch as a standalone HTML document.
+func (s *Store) ExportHTML(path string) error {
+	msgs, err := s.Load()
+	if err != nil {
+		return err
+	}
+	var b strings.Builder
+	b.WriteString("<!DOCTYPE html><html><head><meta charset=\"utf-8\"><title>")
+	title := s.Name()
+	if title == "" {
+		title = "session"
+	}
+	b.WriteString(html.EscapeString(title))
+	b.WriteString("</title></head><body>\n")
+	for _, m := range msgs {
+		fmt.Fprintf(&b, "<article data-role=\"%s\"><h2>%s</h2><pre>%s</pre></article>\n",
+			html.EscapeString(string(m.Role)),
+			html.EscapeString(string(m.Role)),
+			html.EscapeString(m.Content))
+	}
+	b.WriteString("</body></html>\n")
+	return os.WriteFile(path, []byte(b.String()), 0o644)
+}

runtime/session/jsonl.go

@@ -1,17 +1,16 @@
 package session
 
 import (
-	"bufio"
-	"encoding/json"
 	"os"
 	"path/filepath"
-	"strings"
 	"sync"
 
 	"agent-native/runtime/llm"
 )
 
-// Store is an append-only JSONL transcript. Each line is one llm.Message.
+// Store is an append-only JSONL transcript. New sessions are a tree of
+// entries (id / parentId). Load still returns the model-visible message
+// path, and files written as one llm.Message per line remain readable.
 type Store struct {
 	path string
 	mu   sync.Mutex
@@ -31,56 +30,25 @@ func Open(path string) (*Store, error) {
 
 func (s *Store) Path() string { return s.path }
 
+// Append records one message on the active branch.
 func (s *Store) Append(msg llm.Message) error {
 	s.mu.Lock()
 	defer s.mu.Unlock()
-	raw, err := json.Marshal(msg)
-	if err != nil {
-		return err
-	}
-	f, err := os.OpenFile(s.path, os.O_APPEND|os.O_WRONLY, 0o644)
-	if err != nil {
-		return err
-	}
-	defer f.Close()
-	raw = append(raw, '\n')
-	_, err = f.Write(raw)
+	_, err := s.appendEntryLocked(Entry{Type: TypeMessage, Message: &msg})
 	return err
 }
 
-// Load 全量读取转录。崩溃残留容错：Append 恒以换行收尾，因此无换行结尾的
-// 尾部若不是完整 JSON 即为崩溃半行，丢弃而不是让整个会话不可用；
-// 行中间的坏行仍报错（那是文件损坏，不是崩溃残留）。
+// Load returns the model-visible messages on the active branch.
+// A compaction entry replaces earlier turns with its summary plus the
+// retained tail; the file itself is never rewritten.
+// Crash residue: a trailing line without a newline that is not valid JSON
+// is dropped. A bad line in the middle is still an error.
 func (s *Store) Load() ([]llm.Message, error) {
 	s.mu.Lock()
 	defer s.mu.Unlock()
-	raw, err := os.ReadFile(s.path)
+	entries, err := s.readEntriesLocked()
 	if err != nil {
 		return nil, err
 	}
-	text := string(raw)
-	if text != "" && !strings.HasSuffix(text, "\n") {
-		if idx := strings.LastIndexByte(text, '\n'); idx >= 0 {
-			if !json.Valid([]byte(text[idx+1:])) {
-				text = text[:idx+1]
-			}
-		} else if !json.Valid([]byte(text)) {
-			text = ""
-		}
-	}
-	var out []llm.Message
-	sc := bufio.NewScanner(strings.NewReader(text))
-	sc.Buffer(make([]byte, 0, 64*1024), 8*1024*1024)
-	for sc.Scan() {
-		line := sc.Bytes()
-		if len(line) == 0 {
-			continue
-		}
-		var msg llm.Message
-		if err := json.Unmarshal(line, &msg); err != nil {
-			return nil, err
-		}
-		out = append(out, msg)
-	}
-	return out, sc.Err()
+	return visibleMessages(entries, s.leafLocked(entries)), nil
 }

runtime/session/tree.go

@@ -0,0 +1,397 @@
+package session
+
+import (
+	"bufio"
+	"crypto/rand"
+	"encoding/hex"
+	"encoding/json"
+	"fmt"
+	"io"
+	"os"
+	"strings"
+	"time"
+
+	"agent-native/runtime/llm"
+)
+
+// EntryType is the kind of one JSONL record.
+type EntryType string
+
+const (
+	TypeHeader              EntryType = "header"
+	TypeMessage             EntryType = "message"
+	TypeModelChange         EntryType = "model_change"
+	TypeThinkingLevelChange EntryType = "thinking_level_change"
+	TypeCompaction          EntryType = "compaction"
+	TypeLabel               EntryType = "label"
+	TypeCustom              EntryType = "custom"
+	TypeSessionInfo         EntryType = "session_info"
+)
+
+// Entry is one node in the session tree. Files written before the tree
+// format (a bare llm.Message per line) are loaded as message entries.
+type Entry struct {
+	ID            string          `json:"id"`
+	ParentID      string          `json:"parentId,omitempty"`
+	Timestamp     string          `json:"timestamp"`
+	Type          EntryType       `json:"type"`
+	Message       *llm.Message    `json:"message,omitempty"`
+	Model         string          `json:"model,omitempty"`
+	ThinkingLevel string          `json:"thinkingLevel,omitempty"`
+	Label         string          `json:"label,omitempty"`
+	Name          string          `json:"name,omitempty"`
+	CustomType    string          `json:"customType,omitempty"`
+	Data          json.RawMessage `json:"data,omitempty"`
+	Summary       string          `json:"summary,omitempty"`
+	TailIDs       []string        `json:"tailIds,omitempty"`
+	CWD           string          `json:"cwd,omitempty"`
+}
+
+// Entries returns every record in file order, including branches that are
+// not on the active path.
+func (s *Store) Entries() ([]Entry, error) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	return s.readEntriesLocked()
+}
+
+// Leaf is the id of the current branch tip. Empty when the file has no entries.
+func (s *Store) Leaf() (string, error) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	entries, err := s.readEntriesLocked()
+	if err != nil {
+		return "", err
+	}
+	return s.leafLocked(entries), nil
+}
+
+// SetLeaf moves the active branch to an existing entry.
+func (s *Store) SetLeaf(id string) error {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	entries, err := s.readEntriesLocked()
+	if err != nil {
+		return err
+	}
+	if !hasID(entries, id) {
+		return fmt.Errorf("session: unknown entry %s", id)
+	}
+	return s.writeMetaKeyLocked("leaf", id)
+}
+
+// AppendEntry adds a non-message record (label, model change, custom, …)
+// as a child of the current leaf and moves the leaf to it.
+func (s *Store) AppendEntry(e Entry) (Entry, error) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	return s.appendEntryLocked(e)
+}
+
+// Fork copies the tree into a new file and sets that file's leaf to atID.
+// The source file is not modified. atID empty means the current leaf.
+func (s *Store) Fork(dest, atID string) (*Store, error) {
+	return s.copyTo(dest, atID, true)
+}
+
+// Clone copies the tree and keeps the same leaf.
+func (s *Store) Clone(dest string) (*Store, error) {
+	return s.copyTo(dest, "", false)
+}
+
+func (s *Store) copyTo(dest, atID string, moveLeaf bool) (*Store, error) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	entries, err := s.readEntriesLocked()
+	if err != nil {
+		return nil, err
+	}
+	if err := copyFile(s.path, dest); err != nil {
+		return nil, err
+	}
+	out, err := Open(dest)
+	if err != nil {
+		return nil, err
+	}
+	meta := s.readMetaLocked()
+	if meta == nil {
+		meta = map[string]string{}
+	}
+	leaf := s.leafLocked(entries)
+	if moveLeaf && atID != "" {
+		if !hasID(entries, atID) {
+			return nil, fmt.Errorf("session: unknown entry %s", atID)
+		}
+		leaf = atID
+	}
+	if leaf != "" {
+		meta["leaf"] = leaf
+	}
+	if err := out.SaveMeta(meta); err != nil {
+		return nil, err
+	}
+	return out, nil
+}
+
+// SetName records a human name for continue/resume.
+func (s *Store) SetName(name string) error {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	if err := s.writeMetaKeyLocked("name", name); err != nil {
+		return err
+	}
+	_, err := s.appendEntryLocked(Entry{Type: TypeSessionInfo, Name: name})
+	return err
+}
+
+// Name returns the session name, or empty.
+func (s *Store) Name() string {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	return s.readMetaLocked()["name"]
+}
+
+func (s *Store) appendEntryLocked(e Entry) (Entry, error) {
+	entries, err := s.readEntriesLocked()
+	if err != nil {
+		return Entry{}, err
+	}
+	if e.ID == "" {
+		e.ID = newID()
+	}
+	if e.Timestamp == "" {
+		e.Timestamp = time.Now().UTC().Format(time.RFC3339Nano)
+	}
+	if e.Type == "" {
+		e.Type = TypeCustom
+	}
+	if e.ParentID == "" {
+		e.ParentID = s.leafLocked(entries)
+	}
+	raw, err := json.Marshal(e)
+	if err != nil {
+		return Entry{}, err
+	}
+	f, err := os.OpenFile(s.path, os.O_APPEND|os.O_WRONLY, 0o644)
+	if err != nil {
+		return Entry{}, err
+	}
+	defer f.Close()
+	if _, err := f.Write(append(raw, '\n')); err != nil {
+		return Entry{}, err
+	}
+	if err := s.writeMetaKeyLocked("leaf", e.ID); err != nil {
+		return Entry{}, err
+	}
+	return e, nil
+}
+
+func (s *Store) readEntriesLocked() ([]Entry, error) {
+	raw, err := os.ReadFile(s.path)
+	if err != nil {
+		return nil, err
+	}
+	text := string(raw)
+	if text != "" && !strings.HasSuffix(text, "\n") {
+		if idx := strings.LastIndexByte(text, '\n'); idx >= 0 {
+			if !json.Valid([]byte(text[idx+1:])) {
+				text = text[:idx+1]
+			}
+		} else if !json.Valid([]byte(text)) {
+			text = ""
+		}
+	}
+	var out []Entry
+	sc := bufio.NewScanner(strings.NewReader(text))
+	sc.Buffer(make([]byte, 0, 64*1024), 8*1024*1024)
+	legacyParent := ""
+	n := 0
+	for sc.Scan() {
+		line := sc.Bytes()
+		if len(line) == 0 {
+			continue
+		}
+		var probe map[string]json.RawMessage
+		if err := json.Unmarshal(line, &probe); err != nil {
+			return nil, err
+		}
+		if _, ok := probe["type"]; ok {
+			var e Entry
+			if err := json.Unmarshal(line, &e); err != nil {
+				return nil, err
+			}
+			out = append(out, e)
+			legacyParent = e.ID
+			n++
+			continue
+		}
+		var msg llm.Message
+		if err := json.Unmarshal(line, &msg); err != nil {
+			return nil, err
+		}
+		id := fmt.Sprintf("legacy-%d", n)
+		out = append(out, Entry{
+			ID: id, ParentID: legacyParent, Type: TypeMessage, Message: &msg,
+		})
+		legacyParent = id
+		n++
+	}
+	return out, sc.Err()
+}
+
+func (s *Store) leafLocked(entries []Entry) string {
+	if leaf := s.readMetaLocked()["leaf"]; leaf != "" && hasID(entries, leaf) {
+		return leaf
+	}
+	if len(entries) == 0 {
+		return ""
+	}
+	return entries[len(entries)-1].ID
+}
+
+func visibleMessages(entries []Entry, leaf string) []llm.Message {
+	path := pathTo(entries, leaf)
+	var last *Entry
+	for i := range path {
+		if path[i].Type == TypeCompaction {
+			last = &path[i]
+		}
+	}
+	if last == nil {
+		var msgs []llm.Message
+		for _, e := range path {
+			if e.Type == TypeMessage && e.Message != nil {
+				msgs = append(msgs, *e.Message)
+			}
+		}
+		return msgs
+	}
+	byID := map[string]Entry{}
+	for _, e := range entries {
+		byID[e.ID] = e
+	}
+	var msgs []llm.Message
+	for _, e := range path {
+		if e.ID == last.ID {
+			break
+		}
+		if e.Type == TypeMessage && e.Message != nil && e.Message.Role == llm.RoleSystem {
+			msgs = append(msgs, *e.Message)
+		}
+	}
+	msgs = append(msgs, llm.Message{
+		Role:    llm.RoleUser,
+		Content: "[conversation summary]\n" + last.Summary + "\n[recent messages follow]",
+	})
+	for _, id := range last.TailIDs {
+		e, ok := byID[id]
+		if ok && e.Message != nil {
+			msgs = append(msgs, *e.Message)
+		}
+	}
+	after := false
+	for _, e := range path {
+		if e.ID == last.ID {
+			after = true
+			continue
+		}
+		if after && e.Type == TypeMessage && e.Message != nil {
+			msgs = append(msgs, *e.Message)
+		}
+	}
+	return msgs
+}
+
+func pathTo(entries []Entry, leaf string) []Entry {
+	byID := map[string]Entry{}
+	for _, e := range entries {
+		byID[e.ID] = e
+	}
+	var rev []Entry
+	seen := map[string]bool{}
+	for id := leaf; id != "" && !seen[id]; {
+		e, ok := byID[id]
+		if !ok {
+			break
+		}
+		seen[id] = true
+		rev = append(rev, e)
+		id = e.ParentID
+	}
+	for i, j := 0, len(rev)-1; i < j; i, j = i+1, j-1 {
+		rev[i], rev[j] = rev[j], rev[i]
+	}
+	return rev
+}
+
+func hasID(entries []Entry, id string) bool {
+	for _, e := range entries {
+		if e.ID == id {
+			return true
+		}
+	}
+	return false
+}
+
+func (s *Store) readMetaLocked() map[string]string {
+	raw, err := os.ReadFile(s.path + ".meta.json")
+	if err != nil {
+		return nil
+	}
+	var meta map[string]string
+	if err := json.Unmarshal(raw, &meta); err != nil {
+		return nil
+	}
+	return meta
+}
+
+func (s *Store) writeMetaKeyLocked(key, value string) error {
+	meta := s.readMetaLocked()
+	if meta == nil {
+		meta = map[string]string{}
+	}
+	meta[key] = value
+	raw, err := json.MarshalIndent(meta, "", "  ")
+	if err != nil {
+		return err
+	}
+	tmp := s.path + ".meta.json.tmp"
+	if err := os.WriteFile(tmp, raw, 0o644); err != nil {
+		return err
+	}
+	return os.Rename(tmp, s.path+".meta.json")
+}
+
+func newID() string {
+	var b [8]byte
+	if _, err := rand.Read(b[:]); err != nil {
+		return fmt.Sprintf("%d", time.Now().UnixNano())
+	}
+	return hex.EncodeToString(b[:])
+}
+
+func copyFile(src, dst string) error {
+	in, err := os.Open(src)
+	if err != nil {
+		return err
+	}
+	defer in.Close()
+	if err := os.MkdirAll(dirOf(dst), 0o755); err != nil {
+		return err
+	}
+	out, err := os.OpenFile(dst, os.O_CREATE|os.O_TRUNC|os.O_WRONLY, 0o644)
+	if err != nil {
+		return err
+	}
+	defer out.Close()
+	_, err = io.Copy(out, in)
+	return err
+}
+
+func dirOf(path string) string {
+	i := strings.LastIndex(path, "/")
+	if i < 0 {
+		return "."
+	}
+	return path[:i]
+}

runtime/session/tree_test.go

@@ -0,0 +1,128 @@
+package session
+
+import (
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+
+	"agent-native/runtime/llm"
+)
+
+func TestTreeForkCloneAndLeaf(t *testing.T) {
+	dir := t.TempDir()
+	s, err := Open(filepath.Join(dir, "a.jsonl"))
+	if err != nil {
+		t.Fatal(err)
+	}
+	if err := s.Append(llm.Message{Role: llm.RoleUser, Content: "one"}); err != nil {
+		t.Fatal(err)
+	}
+	if err := s.Append(llm.Message{Role: llm.RoleAssistant, Content: "two"}); err != nil {
+		t.Fatal(err)
+	}
+	entries, err := s.Entries()
+	if err != nil || len(entries) != 2 || entries[1].ParentID != entries[0].ID {
+		t.Fatalf("tree links: %+v %v", entries, err)
+	}
+	fork, err := s.Fork(filepath.Join(dir, "fork.jsonl"), entries[0].ID)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if err := fork.Append(llm.Message{Role: llm.RoleUser, Content: "branch"}); err != nil {
+		t.Fatal(err)
+	}
+	msgs, err := fork.Load()
+	if err != nil || len(msgs) != 2 || msgs[1].Content != "branch" {
+		t.Fatalf("fork path: %+v %v", msgs, err)
+	}
+	orig, _ := s.Load()
+	if len(orig) != 2 || orig[1].Content != "two" {
+		t.Fatalf("source branch changed: %+v", orig)
+	}
+	clone, err := s.Clone(filepath.Join(dir, "clone.jsonl"))
+	if err != nil {
+		t.Fatal(err)
+	}
+	cmsgs, _ := clone.Load()
+	if len(cmsgs) != 2 || cmsgs[1].Content != "two" {
+		t.Fatalf("clone: %+v", cmsgs)
+	}
+	if err := s.SetLeaf(entries[0].ID); err != nil {
+		t.Fatal(err)
+	}
+	switched, _ := s.Load()
+	if len(switched) != 1 || switched[0].Content != "one" {
+		t.Fatalf("set leaf: %+v", switched)
+	}
+}
+
+func TestLegacyTranscriptStillLoads(t *testing.T) {
+	dir := t.TempDir()
+	path := filepath.Join(dir, "old.jsonl")
+	body := "{\"role\":\"system\",\"content\":\"sys\"}\n{\"role\":\"user\",\"content\":\"hi\"}\n"
+	if err := os.WriteFile(path, []byte(body), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	s, err := Open(path)
+	if err != nil {
+		t.Fatal(err)
+	}
+	msgs, err := s.Load()
+	if err != nil || len(msgs) != 2 || msgs[1].Content != "hi" {
+		t.Fatalf("legacy: %+v %v", msgs, err)
+	}
+	if err := s.Append(llm.Message{Role: llm.RoleAssistant, Content: "next"}); err != nil {
+		t.Fatal(err)
+	}
+	msgs, err = s.Load()
+	if err != nil || len(msgs) != 3 || msgs[2].Content != "next" {
+		t.Fatalf("legacy append: %+v %v", msgs, err)
+	}
+}
+
+func TestCatalogContinueResumeAndHTML(t *testing.T) {
+	dir := t.TempDir()
+	s, err := Open(filepath.Join(dir, "sess.jsonl"))
+	if err != nil {
+		t.Fatal(err)
+	}
+	if err := s.SetCWD("/work"); err != nil {
+		t.Fatal(err)
+	}
+	if err := s.SetName("notes"); err != nil {
+		t.Fatal(err)
+	}
+	if err := s.Append(llm.Message{Role: llm.RoleUser, Content: "hello"}); err != nil {
+		t.Fatal(err)
+	}
+	got, ok, err := Continue(dir, "/work")
+	if err != nil || !ok || got.Path() != s.Path() {
+		t.Fatalf("continue: %v %v %v", got, ok, err)
+	}
+	resumed, err := Resume(dir, "notes")
+	if err != nil || resumed.Name() != "notes" {
+		t.Fatalf("resume: %v %v", resumed, err)
+	}
+	htmlPath := filepath.Join(dir, "out.html")
+	if err := s.ExportHTML(htmlPath); err != nil {
+		t.Fatal(err)
+	}
+	raw, _ := os.ReadFile(htmlPath)
+	if !strings.Contains(string(raw), "hello") || !strings.Contains(string(raw), "notes") {
+		t.Fatalf("html: %s", raw)
+	}
+	if _, err := s.AppendEntry(Entry{Type: TypeLabel, Label: "checkpoint"}); err != nil {
+		t.Fatal(err)
+	}
+	entries, _ := s.Entries()
+	var labels int
+	for _, e := range entries {
+		if e.Type == TypeLabel && e.Label == "checkpoint" {
+			labels++
+		}
+	}
+	if labels != 1 {
+		t.Fatalf("label missing: %+v", entries)
+	}
+}

runtime/session_agent.go

@@ -0,0 +1,200 @@
+package runtime
+
+import (
+	"context"
+	"sync"
+
+	"agent-native/runtime/llm"
+	"agent-native/runtime/session"
+)
+
+// Delivery selects how a queued steer or follow-up is handed to the loop.
+type Delivery string
+
+const (
+	DeliverOne Delivery = "one"
+	DeliverAll Delivery = "all"
+)
+
+// AgentSession is the headless session object: queues, abort, and
+// mid-run model changes around Run.
+type AgentSession struct {
+	cfg Config
+
+	mu         sync.Mutex
+	steer      []llm.Message
+	follow     []llm.Message
+	steerMode  Delivery
+	followMode Delivery
+	model      string
+	thinking   string
+	cancel     context.CancelFunc
+	runID      int
+}
+
+// NewAgentSession copies cfg. Queue modes default to one-at-a-time.
+func NewAgentSession(cfg Config) *AgentSession {
+	if cfg.Model == "" {
+		cfg.Model = "default"
+	}
+	return &AgentSession{
+		cfg:        cfg,
+		steerMode:  DeliverOne,
+		followMode: DeliverOne,
+		model:      cfg.Model,
+		thinking:   cfg.ThinkingLevel,
+	}
+}
+
+// SetQueueMode sets the delivery mode for steer and follow-up.
+func (a *AgentSession) SetQueueMode(steer, follow Delivery) {
+	a.mu.Lock()
+	defer a.mu.Unlock()
+	if steer != "" {
+		a.steerMode = steer
+	}
+	if follow != "" {
+		a.followMode = follow
+	}
+}
+
+// Steer queues a message to be delivered after the current tool batch.
+func (a *AgentSession) Steer(msg llm.Message) {
+	a.mu.Lock()
+	a.steer = append(a.steer, msg)
+	a.mu.Unlock()
+	a.emitQueue("steer")
+}
+
+// FollowUp queues a message for after the assistant stops.
+func (a *AgentSession) FollowUp(msg llm.Message) {
+	a.mu.Lock()
+	a.follow = append(a.follow, msg)
+	a.mu.Unlock()
+	a.emitQueue("follow_up")
+}
+
+// ClearQueue drops both queues.
+func (a *AgentSession) ClearQueue() {
+	a.mu.Lock()
+	a.steer = nil
+	a.follow = nil
+	a.mu.Unlock()
+	a.emitQueue("clear")
+}
+
+// Abort cancels the in-flight Prompt.
+func (a *AgentSession) Abort() {
+	a.mu.Lock()
+	cancel := a.cancel
+	a.mu.Unlock()
+	if cancel != nil {
+		cancel()
+	}
+}
+
+// SetModel changes the model and thinking level used on the next turn.
+// A session store, when present, records the change.
+func (a *AgentSession) SetModel(model, thinking string) error {
+	if thinking != "" {
+		norm, err := llm.NormalizeThinking(thinking)
+		if err != nil {
+			return err
+		}
+		thinking = norm
+	}
+	a.mu.Lock()
+	if model != "" {
+		a.model = model
+	}
+	if thinking != "" {
+		a.thinking = thinking
+	}
+	store := a.cfg.Session
+	a.mu.Unlock()
+	if store == nil {
+		return nil
+	}
+	if model != "" {
+		if _, err := store.AppendEntry(session.Entry{Type: session.TypeModelChange, Model: model}); err != nil {
+			return err
+		}
+	}
+	if thinking != "" {
+		if _, err := store.AppendEntry(session.Entry{Type: session.TypeThinkingLevelChange, ThinkingLevel: thinking}); err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
+// Prompt runs one user message until the agent settles.
+func (a *AgentSession) Prompt(ctx context.Context, user llm.Message) (Result, error) {
+	ctx, cancel := context.WithCancel(ctx)
+	a.mu.Lock()
+	a.runID++
+	id := a.runID
+	a.cancel = cancel
+	cfg := a.cfg
+	a.mu.Unlock()
+	defer func() {
+		cancel()
+		a.mu.Lock()
+		if a.runID == id {
+			a.cancel = nil
+		}
+		a.mu.Unlock()
+	}()
+	cfg.GetSteering = func() []llm.Message { return a.take("steer") }
+	cfg.GetFollowUp = func() []llm.Message { return a.take("follow") }
+	cfg.Resolve = func() (string, string) {
+		a.mu.Lock()
+		defer a.mu.Unlock()
+		return a.model, a.thinking
+	}
+	return Run(ctx, cfg, user)
+}
+
+// Compact appends a compaction entry and emits compaction events.
+func (a *AgentSession) Compact(keep int, summarize func([]llm.Message) (string, error)) (bool, error) {
+	if a.cfg.Session == nil {
+		return false, nil
+	}
+	emit(a.cfg, Event{Type: "compaction_start"})
+	done, err := a.cfg.Session.Compact(keep, summarize)
+	text := "noop"
+	if done {
+		text = "compacted"
+	}
+	if err != nil {
+		text = err.Error()
+	}
+	emit(a.cfg, Event{Type: "compaction_end", Text: text})
+	return done, err
+}
+
+func (a *AgentSession) take(kind string) []llm.Message {
+	a.mu.Lock()
+	defer a.mu.Unlock()
+	q := &a.steer
+	mode := a.steerMode
+	if kind == "follow" {
+		q = &a.follow
+		mode = a.followMode
+	}
+	if len(*q) == 0 {
+		return nil
+	}
+	if mode == DeliverAll {
+		out := append([]llm.Message(nil), (*q)...)
+		*q = nil
+		return out
+	}
+	out := []llm.Message{(*q)[0]}
+	*q = (*q)[1:]
+	return out
+}
+
+func (a *AgentSession) emitQueue(text string) {
+	emit(a.cfg, Event{Type: "queue_update", Text: text})
+}

runtime/session_agent_test.go

@@ -0,0 +1,88 @@
+package runtime
+
+import (
+	"context"
+	"sync"
+	"testing"
+
+	"agent-native/runtime/llm"
+)
+
+func TestAgentSessionSteerFollowAndAbort(t *testing.T) {
+	var mu sync.Mutex
+	step := 0
+	stream := llm.StreamFunc(func(ctx context.Context, model string, messages []llm.Message, td []llm.ToolDecl, opts llm.StreamOptions) <-chan llm.Event {
+		ch := make(chan llm.Event, 2)
+		go func() {
+			defer close(ch)
+			mu.Lock()
+			n := step
+			step++
+			mu.Unlock()
+			if opts.ThinkingLevel != "low" && n == 0 {
+				t.Errorf("thinking level not forwarded: %s", opts.ThinkingLevel)
+			}
+			if n == 0 {
+				ch <- llm.Event{Kind: llm.KindMessageEnd, Message: llm.Message{
+					Role: llm.RoleAssistant, StopReason: llm.StopToolCalls,
+					ToolCalls: []llm.ToolCall{{ID: "1", Name: "missing", Arguments: "{}"}},
+				}}
+				return
+			}
+			ch <- llm.Event{Kind: llm.KindMessageEnd, Message: llm.Message{
+				Role: llm.RoleAssistant, Content: "done", StopReason: llm.StopEnd,
+			}}
+		}()
+		return ch
+	})
+	var types []string
+	sess := NewAgentSession(Config{
+		Model:         "m",
+		ThinkingLevel: "low",
+		Stream:        stream,
+		MaxTurns:      6,
+		OnEvent: func(ev Event) {
+			types = append(types, ev.Type)
+		},
+	})
+	sess.SetQueueMode(DeliverOne, DeliverAll)
+	sess.Steer(llm.Message{Role: llm.RoleUser, Content: "steer-me"})
+	sess.FollowUp(llm.Message{Role: llm.RoleUser, Content: "later"})
+	res, err := sess.Prompt(context.Background(), llm.Message{Role: llm.RoleUser, Content: "go"})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if res.StopReason != llm.StopEnd {
+		t.Fatalf("stop %s err %v", res.StopReason, res.Err)
+	}
+	found := map[string]bool{}
+	for _, typ := range types {
+		found[typ] = true
+	}
+	for _, want := range []string{"message_start", "queue_update", "tool_execution_update", "agent_settled"} {
+		if !found[want] {
+			t.Fatalf("missing %s in %v", want, types)
+		}
+	}
+}
+
+func TestClearQueueAndSetModel(t *testing.T) {
+	sess := NewAgentSession(Config{Model: "old", Stream: func(ctx context.Context, model string, messages []llm.Message, td []llm.ToolDecl, opts llm.StreamOptions) <-chan llm.Event {
+		ch := make(chan llm.Event, 1)
+		ch <- llm.Event{Kind: llm.KindMessageEnd, Message: llm.Message{Role: llm.RoleAssistant, Content: model, StopReason: llm.StopEnd}}
+		close(ch)
+		return ch
+	}})
+	sess.Steer(llm.Message{Role: llm.RoleUser, Content: "drop"})
+	sess.ClearQueue()
+	if err := sess.SetModel("new-model", "high"); err != nil {
+		t.Fatal(err)
+	}
+	res, err := sess.Prompt(context.Background(), llm.Message{Role: llm.RoleUser, Content: "hi"})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if res.Text != "new-model" {
+		t.Fatalf("model switch did not apply: %q", res.Text)
+	}
+}

runtime/skills/skills.go

@@ -107,6 +107,35 @@ func (s *Store) Load(name string) (Skill, error) {
 	return Skill{}, fmt.Errorf("unknown skill %q (not installed in any configured root)", name)
 }
 
+// ExpandCommands replaces /skill:name with that skill's body.
+// An unknown name returns an error and leaves the text unchanged.
+func (s *Store) ExpandCommands(text string) (string, error) {
+	const marker = "/skill:"
+	var b strings.Builder
+	rest := text
+	for {
+		i := strings.Index(rest, marker)
+		if i < 0 {
+			b.WriteString(rest)
+			break
+		}
+		b.WriteString(rest[:i])
+		name := rest[i+len(marker):]
+		end := 0
+		for end < len(name) && name[end] != ' ' && name[end] != '\n' && name[end] != '\t' {
+			end++
+		}
+		skillName := name[:end]
+		skill, err := s.Load(skillName)
+		if err != nil {
+			return text, err
+		}
+		b.WriteString(skill.Body)
+		rest = name[end:]
+	}
+	return b.String(), nil
+}
+
 // parse 解析一份 SKILL.md：目录名权威，frontmatter 可选。
 func parse(dirName, path string, data []byte) (Skill, error) {
 	if !validSkillName(dirName) {

runtime/tools/alias.go

@@ -0,0 +1,83 @@
+package tools
+
+import (
+	"context"
+	"encoding/json"
+)
+
+// aliasTool exposes an existing tool under a pi-style name.
+type aliasTool struct {
+	name  string
+	inner Tool
+}
+
+func (a aliasTool) Name() string                 { return a.name }
+func (a aliasTool) Description() string          { return a.inner.Description() }
+func (a aliasTool) Parameters() json.RawMessage  { return a.inner.Parameters() }
+func (a aliasTool) Execute(ctx context.Context, args json.RawMessage) (Result, error) {
+	return a.inner.Execute(ctx, args)
+}
+
+// piAliases maps the headless pi tool name onto the name already registered.
+var piAliases = map[string]string{
+	"read":  "read_file",
+	"write": "write_file",
+	"edit":  "edit_file",
+	"bash":  "run_command",
+	"grep":  "grep",
+	"find":  "glob",
+	"ls":    "list_dir",
+}
+
+// RegisterPiAliases adds pi tool names beside the existing ones.
+// A name is skipped when the underlying tool is not registered.
+func (r *Registry) RegisterPiAliases() {
+	for alias, inner := range piAliases {
+		if alias == inner || r.Has(alias) || !r.Has(inner) {
+			continue
+		}
+		r.Register(aliasTool{name: alias, inner: r.tools[inner]})
+	}
+}
+
+// FilterOptions selects which tools stay visible.
+type FilterOptions struct {
+	Tools     []string
+	Exclude   []string
+	NoBuiltin bool
+}
+
+// Filter applies an allow list, then exclusions. NoBuiltin drops the
+// workspace tool names and their pi aliases.
+func (r *Registry) Filter(opts FilterOptions) error {
+	if len(opts.Tools) > 0 {
+		if err := r.Restrict(opts.Tools); err != nil {
+			return err
+		}
+	}
+	drop := map[string]bool{}
+	for _, name := range opts.Exclude {
+		drop[name] = true
+	}
+	if opts.NoBuiltin {
+		for _, name := range []string{"read_file", "write_file", "edit_file", "multi_edit", "list_dir", "grep", "glob", "run_command", "run_python"} {
+			drop[name] = true
+		}
+		for alias := range piAliases {
+			drop[alias] = true
+		}
+	}
+	if len(drop) == 0 {
+		return nil
+	}
+	var order []string
+	for _, name := range r.order {
+		if drop[name] {
+			delete(r.tools, name)
+			continue
+		}
+		order = append(order, name)
+	}
+	r.order = order
+	return nil
+}

runtime/tools/registry.go

@@ -45,6 +45,11 @@ func NewRegistry(root string) *Registry {
 
 func (r *Registry) Root() string { return r.root }
 
+func (r *Registry) Has(name string) bool {
+	_, ok := r.tools[name]
+	return ok
+}
+
 func (r *Registry) Register(t Tool) {
 	if _, ok := r.tools[t.Name()]; !ok {
 		r.order = append(r.order, t.Name())