feat(ai): session persist + continue HTTP/UI, checkpoints, queue

落地 docs/16 P1：会话落盘与真续跑（prior tool 轨迹）、incomplete「加预算继续」、
awaiting_confirm outcome、同参拒+第4/6轮检查点、侧栏忙时单槽排队。
Ask 真 token 流式记缺口（三协议 SSE 成本高）。

server/internal/ai/agent.go

@@ -59,6 +59,8 @@ type AgentRes struct {
 	IncompleteReason string      `json:"incompleteReason,omitempty"`
 	Resumable        bool        `json:"resumable,omitempty"`
 	StepsBudget      int         `json:"stepsBudget,omitempty"`
+	// Trajectory 含 system/user/assistant.tool_calls/tool 的完整对话（供落盘续跑；JSON 默认省略大字段可选）
+	Trajectory       []Message   `json:"trajectory,omitempty"`
 }
 
 // RunAgent 多轮：LLM tool_call → aitools.Runtime.Call → 回灌。
@@ -90,6 +92,11 @@ func runAgent(ctx context.Context, llm ToolChat, rt *aitools.Runtime, req AgentR
 	if runID == "" {
 		runID = NewRunID()
 	}
+	sessionID := strings.TrimSpace(req.SessionID)
+	if sessionID == "" {
+		sessionID = NewSessionID()
+		req.SessionID = sessionID
+	}
 	rt.RunID = runID
 	// Agent 路径禁止模型自行带确认令牌
 	rt.RejectCallerConfirm = true
@@ -119,13 +126,14 @@ func runAgent(ctx context.Context, llm ToolChat, rt *aitools.Runtime, req AgentR
 		}
 	}
 
-	baseMeta := func(steps []AgentStep, rounds int) AgentRes {
+	finish := func(steps []AgentStep, rounds int, msgs []Message) AgentRes {
 		return AgentRes{
 			Steps:       steps,
 			Rounds:      rounds,
 			RunID:       runID,
-			SessionID:   req.SessionID,
+			SessionID:   sessionID,
 			StepsBudget: maxRounds,
+			Trajectory:  append([]Message{}, msgs...),
 		}
 	}
 
@@ -134,19 +142,22 @@ func runAgent(ctx context.Context, llm ToolChat, rt *aitools.Runtime, req AgentR
 	rounds := 0
 	for round := 0; round < maxRounds; round++ {
 		if err := ctx.Err(); err != nil {
-			res := baseMeta(steps, rounds)
+			res := finish(steps, rounds, msgs)
 			res.Content = "已取消"
 			res.Outcome = OutcomeCancelled
 			res.Resumable = len(steps) > 0
-			emitEvent(emit, AgentEvent{Type: AgentEventError, Error: "cancelled", Steps: steps, Rounds: rounds, RunID: runID, Outcome: OutcomeCancelled})
+			emitEvent(emit, AgentEvent{Type: AgentEventError, Error: "cancelled", Steps: steps, Rounds: rounds, RunID: runID, SessionID: sessionID, Outcome: OutcomeCancelled})
 			return res, err
 		}
 		rounds = round + 1
+		if tip := CheckpointHint(rounds); tip != "" {
+			msgs = append(msgs, Message{Role: "system", Content: tip})
+		}
 		turn, err := llm.ChatWithTools(ctx, req.Model, msgs, tools)
 		if err != nil {
-			res := baseMeta(steps, rounds)
+			res := finish(steps, rounds, msgs)
 			res.Outcome = OutcomeFailed
-			emitEvent(emit, AgentEvent{Type: AgentEventError, Error: err.Error(), Steps: steps, Rounds: rounds, RunID: runID, Outcome: OutcomeFailed})
+			emitEvent(emit, AgentEvent{Type: AgentEventError, Error: err.Error(), Steps: steps, Rounds: rounds, RunID: runID, SessionID: sessionID, Outcome: OutcomeFailed})
 			return res, err
 		}
 		if len(turn.ToolCalls) == 0 {
@@ -157,13 +168,13 @@ func runAgent(ctx context.Context, llm ToolChat, rt *aitools.Runtime, req AgentR
 			for _, part := range chunkDelta(content, 36) {
 				emitEvent(emit, AgentEvent{Type: AgentEventDelta, Delta: part})
 			}
-			res := baseMeta(steps, rounds)
+			res := finish(steps, rounds, msgs)
 			res.Content = content
 			res.SQL = extractSQL(content)
 			res.Outcome = OutcomeCompleted
 			emitEvent(emit, AgentEvent{
 				Type: AgentEventDone, Content: res.Content, SQL: res.SQL, Steps: steps, Rounds: rounds,
-				RunID: runID, Outcome: OutcomeCompleted, StepsBudget: maxRounds,
+				RunID: runID, SessionID: sessionID, Outcome: OutcomeCompleted, StepsBudget: maxRounds,
 			})
 			return res, nil
 		}
@@ -217,13 +228,13 @@ func runAgent(ctx context.Context, llm ToolChat, rt *aitools.Runtime, req AgentR
 		}
 		if awaiting {
 			content := "需要确认写操作后才能继续。"
-			res := baseMeta(steps, rounds)
+			res := finish(steps, rounds, msgs)
 			res.Content = content
 			res.Outcome = OutcomeAwaitingConfirm
 			res.Resumable = true
 			emitEvent(emit, AgentEvent{
 				Type: AgentEventDone, Content: content, Steps: steps, Rounds: rounds,
-				RunID: runID, Outcome: OutcomeAwaitingConfirm, Resumable: true, StepsBudget: maxRounds,
+				RunID: runID, SessionID: sessionID, Outcome: OutcomeAwaitingConfirm, Resumable: true, StepsBudget: maxRounds,
 			})
 			return res, nil
 		}
@@ -234,14 +245,14 @@ func runAgent(ctx context.Context, llm ToolChat, rt *aitools.Runtime, req AgentR
 	for _, part := range chunkDelta(content, 36) {
 		emitEvent(emit, AgentEvent{Type: AgentEventDelta, Delta: part})
 	}
-	res := baseMeta(steps, rounds)
+	res := finish(steps, rounds, msgs)
 	res.Content = content
 	res.Outcome = OutcomeIncomplete
 	res.IncompleteReason = ReasonMaxSteps
 	res.Resumable = true
 	emitEvent(emit, AgentEvent{
 		Type: AgentEventIncomplete, Content: content, Steps: steps, Rounds: rounds,
-		RunID: runID, Outcome: OutcomeIncomplete, IncompleteReason: ReasonMaxSteps,
+		RunID: runID, SessionID: sessionID, Outcome: OutcomeIncomplete, IncompleteReason: ReasonMaxSteps,
 		Resumable: true, StepsBudget: maxRounds,
 	})
 	return res, nil

server/internal/ai/agent_stream.go

@@ -37,6 +37,7 @@ type AgentEvent struct {
 	Error            string            `json:"error,omitempty"`
 	Result           *AgentQueryResult `json:"result,omitempty"`
 	RunID            string            `json:"runId,omitempty"`
+	SessionID        string            `json:"sessionId,omitempty"`
 	Outcome          string            `json:"runOutcome,omitempty"`
 	IncompleteReason string            `json:"incompleteReason,omitempty"`
 	Resumable        bool              `json:"resumable,omitempty"`

server/internal/ai/harness.go

@@ -163,3 +163,15 @@ func DetectRepeatedToolCall(history []string, fp string, limit int) bool {
 	}
 	return n >= limit
 }
+
+// CheckpointHint 中段检查点（第 4 / 6 轮，1-based）；空串表示本轮无注入。
+func CheckpointHint(round1Based int) string {
+	switch round1Based {
+	case 4:
+		return "[检查点] 若已有足够信息请直接作答；避免重复相同参数的工具调用。"
+	case 6:
+		return "[检查点] 请尽快收尾给出结论，勿再扩大探索范围。"
+	default:
+		return ""
+	}
+}

server/internal/ai/session.go

@@ -1,33 +1,48 @@
 package ai
 
 import (
+	"crypto/rand"
+	"encoding/hex"
 	"encoding/json"
 	"fmt"
 	"os"
 	"path/filepath"
+	"sort"
+	"strings"
 	"time"
 )
 
-// Session 桌面本地会话（目标态落盘；HTTP 路由可后挂）。
+// Session 桌面本地会话（落盘 ~/.opsxdb/ai-sessions/<id>.json）。
 type Session struct {
-	ID        string         `json:"id"`
-	Title     string         `json:"title,omitempty"`
-	Status    string         `json:"status"` // idle|running|incomplete|completed|failed|cancelled
-	CreatedAt string         `json:"createdAt"`
-	UpdatedAt string         `json:"updatedAt"`
-	Messages  []Message      `json:"messages"`
-	Runs      []SessionRun   `json:"runs,omitempty"`
+	ID        string       `json:"id"`
+	Title     string       `json:"title,omitempty"`
+	Status    string       `json:"status"` // idle|running|incomplete|completed|failed|cancelled|awaiting_confirm
+	CreatedAt string       `json:"createdAt"`
+	UpdatedAt string       `json:"updatedAt"`
+	Messages  []Message    `json:"messages"`
+	Runs      []SessionRun `json:"runs,omitempty"`
 }
 
 // SessionRun 一次运行摘要。
 type SessionRun struct {
-	RunID             string `json:"runId"`
-	Outcome           string `json:"outcome"`
-	IncompleteReason  string `json:"incompleteReason,omitempty"`
-	Resumable         bool   `json:"resumable"`
-	StepsUsed         int    `json:"stepsUsed"`
-	StepsBudget       int    `json:"stepsBudget"`
-	FinishedAt        string `json:"finishedAt"`
+	RunID            string `json:"runId"`
+	Outcome          string `json:"outcome"`
+	IncompleteReason string `json:"incompleteReason,omitempty"`
+	Resumable        bool   `json:"resumable"`
+	StepsUsed        int    `json:"stepsUsed"`
+	StepsBudget      int    `json:"stepsBudget"`
+	FinishedAt       string `json:"finishedAt"`
+}
+
+// SessionSummary 列表项（不含 messages）。
+type SessionSummary struct {
+	ID        string `json:"id"`
+	Title     string `json:"title,omitempty"`
+	Status    string `json:"status"`
+	CreatedAt string `json:"createdAt"`
+	UpdatedAt string `json:"updatedAt"`
+	MsgCount  int    `json:"msgCount"`
+	LastRunID string `json:"lastRunId,omitempty"`
 }
 
 func sessionsDir() (string, error) {
@@ -44,9 +59,20 @@ func SessionPath(id string) (string, error) {
 	if err != nil {
 		return "", err
 	}
+	id = strings.TrimSpace(id)
+	if id == "" || strings.ContainsAny(id, `/\`) {
+		return "", fmt.Errorf("invalid session id")
+	}
 	return filepath.Join(dir, id+".json"), nil
 }
 
+// NewSessionID 生成 sess_… 短 id。
+func NewSessionID() string {
+	var b [8]byte
+	_, _ = rand.Read(b[:])
+	return "sess_" + hex.EncodeToString(b[:])
+}
+
 // SaveSession 脱敏后写入磁盘。
 func SaveSession(s Session) error {
 	if s.ID == "" {
@@ -87,3 +113,118 @@ func LoadSession(id string) (Session, error) {
 	}
 	return s, nil
 }
+
+// ListSessions 按 UpdatedAt 倒序；limit<=0 表示全部。
+func ListSessions(limit int) ([]SessionSummary, error) {
+	dir, err := sessionsDir()
+	if err != nil {
+		return nil, err
+	}
+	entries, err := os.ReadDir(dir)
+	if err != nil {
+		if os.IsNotExist(err) {
+			return []SessionSummary{}, nil
+		}
+		return nil, err
+	}
+	var out []SessionSummary
+	for _, e := range entries {
+		if e.IsDir() || !strings.HasSuffix(e.Name(), ".json") {
+			continue
+		}
+		id := strings.TrimSuffix(e.Name(), ".json")
+		s, err := LoadSession(id)
+		if err != nil {
+			continue
+		}
+		sum := SessionSummary{
+			ID: s.ID, Title: s.Title, Status: s.Status,
+			CreatedAt: s.CreatedAt, UpdatedAt: s.UpdatedAt,
+			MsgCount: len(s.Messages),
+		}
+		if n := len(s.Runs); n > 0 {
+			sum.LastRunID = s.Runs[n-1].RunID
+		}
+		out = append(out, sum)
+	}
+	sort.Slice(out, func(i, j int) bool { return out[i].UpdatedAt > out[j].UpdatedAt })
+	if limit > 0 && len(out) > limit {
+		out = out[:limit]
+	}
+	return out, nil
+}
+
+// DeleteSession 删除会话文件。
+func DeleteSession(id string) error {
+	path, err := SessionPath(id)
+	if err != nil {
+		return err
+	}
+	err = os.Remove(path)
+	if os.IsNotExist(err) {
+		return nil
+	}
+	return err
+}
+
+// OutcomeToSessionStatus 映射 RunOutcome → Session.Status。
+func OutcomeToSessionStatus(outcome string) string {
+	switch outcome {
+	case OutcomeIncomplete:
+		return "incomplete"
+	case OutcomeAwaitingConfirm:
+		return "awaiting_confirm"
+	case OutcomeFailed:
+		return "failed"
+	case OutcomeCancelled:
+		return "cancelled"
+	case OutcomeCompleted:
+		return "completed"
+	default:
+		if outcome == "" {
+			return "idle"
+		}
+		return outcome
+	}
+}
+
+// PersistAgentRun 把本趟轨迹与 outcome 写入会话（真·续跑用）。
+func PersistAgentRun(sessionID, title string, trajectory []Message, res AgentRes) (Session, error) {
+	sessionID = strings.TrimSpace(sessionID)
+	if sessionID == "" {
+		sessionID = NewSessionID()
+	}
+	s, err := LoadSession(sessionID)
+	if err != nil {
+		if !os.IsNotExist(err) {
+			return Session{}, err
+		}
+		s = Session{ID: sessionID, Status: "idle"}
+	}
+	if title != "" && s.Title == "" {
+		// 标题取首条 user 或短截断
+		t := strings.TrimSpace(title)
+		r := []rune(t)
+		if len(r) > 48 {
+			t = string(r[:48]) + "…"
+		}
+		s.Title = t
+	}
+	if len(trajectory) > 0 {
+		s.Messages = trajectory
+	}
+	s.Status = OutcomeToSessionStatus(res.Outcome)
+	s.Runs = append(s.Runs, SessionRun{
+		RunID:            res.RunID,
+		Outcome:          res.Outcome,
+		IncompleteReason: res.IncompleteReason,
+		Resumable:        res.Resumable,
+		StepsUsed:        res.Rounds,
+		StepsBudget:      res.StepsBudget,
+		FinishedAt:       time.Now().Format(time.RFC3339),
+	})
+	if err := SaveSession(s); err != nil {
+		return Session{}, err
+	}
+	return s, nil
+}

server/internal/ai/session_test.go

@@ -0,0 +1,175 @@
+package ai
+
+import (
+	"context"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+
+	"opsxdb/server/internal/aitools"
+	"opsxdb/server/internal/db"
+)
+
+func TestSessionPersistAndLoadTrajectory(t *testing.T) {
+	dir := t.TempDir()
+	t.Setenv("HOME", dir)
+	id := NewSessionID()
+	traj := []Message{
+		{Role: "system", Content: "sys"},
+		{Role: "user", Content: "list tables"},
+		{Role: "assistant", Content: "", ToolCalls: []ToolCall{{ID: "c1", Type: "function", Function: ToolCallFunc{Name: "opsxdb_list_tables", Arguments: "{}"}}}},
+		{Role: "tool", ToolCallID: "c1", Name: "opsxdb_list_tables", Content: `{"tables":["users"]}`},
+	}
+	res := AgentRes{RunID: "run_x", Outcome: OutcomeIncomplete, IncompleteReason: ReasonMaxSteps, Resumable: true, Rounds: 8, StepsBudget: 8, SessionID: id}
+	s, err := PersistAgentRun(id, "list tables", traj, res)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if s.Status != "incomplete" {
+		t.Fatalf("status=%s", s.Status)
+	}
+	loaded, err := LoadSession(id)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if len(loaded.Messages) != 4 {
+		t.Fatalf("msgs=%d", len(loaded.Messages))
+	}
+	if loaded.Messages[3].ToolCallID != "c1" {
+		t.Fatal("tool trajectory lost")
+	}
+	list, err := ListSessions(10)
+	if err != nil || len(list) < 1 {
+		t.Fatalf("list: %v %#v", err, list)
+	}
+	path := filepath.Join(dir, ".opsxdb", "ai-sessions", id+".json")
+	if _, err := os.Stat(path); err != nil {
+		t.Fatal(err)
+	}
+}
+
+func TestRunAgentContinueInjectsPrior(t *testing.T) {
+	rt := aitools.NewRuntime(aitools.Policy{Mode: aitools.ModeReadOnly, AllowedConnectionIDs: []string{"c1"}, DefaultLimit: 100, MaxLimit: 500})
+	rt.LoadConns = func() ([]aitools.SavedConnection, error) {
+		return []aitools.SavedConnection{{ID: "c1", Name: "d", Driver: "mysql"}}, nil
+	}
+	rt.Audit = func(aitools.AuditEntry) {}
+	prior := []Message{
+		{Role: "system", Content: "sys"},
+		{Role: "user", Content: "prev"},
+		{Role: "assistant", Content: "partial", ToolCalls: []ToolCall{{ID: "c1", Type: "function", Function: ToolCallFunc{Name: aitools.ToolListConnections, Arguments: "{}"}}}},
+		{Role: "tool", ToolCallID: "c1", Name: aitools.ToolListConnections, Content: `{"connections":[]}`},
+	}
+	llm := &scriptedLLM{turns: []ToolsTurn{{Content: "continued answer"}}}
+	res, err := RunAgent(context.Background(), llm, rt, AgentReq{
+		Prompt: "加预算继续", Continue: true, Prior: prior, MaxRounds: 4, SessionID: "sess_test",
+	})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if res.Outcome != OutcomeCompleted {
+		t.Fatalf("outcome=%s", res.Outcome)
+	}
+	if len(llm.seen) < 1 {
+		t.Fatal("no llm call")
+	}
+	sawTool := false
+	for _, m := range llm.seen[0] {
+		if m.Role == "tool" && m.ToolCallID == "c1" {
+			sawTool = true
+		}
+	}
+	if !sawTool {
+		t.Fatalf("prior tool not injected: %+v", llm.seen[0])
+	}
+	if res.SessionID == "" || len(res.Trajectory) < 4 {
+		t.Fatalf("session/traj: sid=%s traj=%d", res.SessionID, len(res.Trajectory))
+	}
+}
+
+func TestRunAgentAwaitingConfirmOutcome(t *testing.T) {
+	rt := aitools.NewRuntime(aitools.Policy{
+		Mode:                 aitools.ModeReadWrite,
+		AllowedConnectionIDs: []string{"c1"},
+		DefaultLimit:         100,
+		MaxLimit:             500,
+	})
+	rt.FindConn = func(id string) (*aitools.SavedConnection, error) {
+		return &aitools.SavedConnection{ID: id, Name: "d", Driver: "mysql", Host: "127.0.0.1", Port: 3306, Database: "demo"}, nil
+	}
+	rt.Audit = func(aitools.AuditEntry) {}
+	execCalls := 0
+	rt.ExecSQL = func(_ context.Context, _ db.ConnReq) db.ExecResult {
+		execCalls++
+		return db.ExecResult{RowCount: 1}
+	}
+	args := `{"connection_id":"c1","database":"demo","sql":"INSERT INTO users(name) VALUES('x')"}`
+	llm := &scriptedLLM{turns: []ToolsTurn{{
+		ToolCalls: []ToolCall{{ID: "w1", Type: "function", Function: ToolCallFunc{Name: aitools.ToolExecuteQuery, Arguments: args}}},
+	}}}
+	res, err := RunAgent(context.Background(), llm, rt, AgentReq{Prompt: "insert one", MaxRounds: 4})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if res.Outcome != OutcomeAwaitingConfirm {
+		t.Fatalf("outcome=%s want awaiting_confirm", res.Outcome)
+	}
+	if !res.Resumable {
+		t.Fatal("expected resumable")
+	}
+	if execCalls != 0 {
+		t.Fatal("must not execute write without confirm")
+	}
+	found := false
+	for _, s := range res.Steps {
+		if s.NeedConfirm {
+			found = true
+		}
+	}
+	if !found {
+		t.Fatalf("steps missing needConfirm: %+v", res.Steps)
+	}
+}
+
+func TestCheckpointHint(t *testing.T) {
+	if CheckpointHint(4) == "" || CheckpointHint(6) == "" {
+		t.Fatal("expected hints")
+	}
+	if CheckpointHint(1) != "" {
+		t.Fatal("no hint round 1")
+	}
+}
+
+func TestRunAgentInjectsCheckpoint(t *testing.T) {
+	rt := aitools.NewRuntime(aitools.Policy{Mode: aitools.ModeReadOnly, AllowedConnectionIDs: []string{"c1"}, DefaultLimit: 100, MaxLimit: 500})
+	rt.LoadConns = func() ([]aitools.SavedConnection, error) {
+		return []aitools.SavedConnection{{ID: "c1", Name: "d", Driver: "mysql"}}, nil
+	}
+	rt.Audit = func(aitools.AuditEntry) {}
+	var turns []ToolsTurn
+	for i := 0; i < 4; i++ {
+		turns = append(turns, ToolsTurn{ToolCalls: []ToolCall{{
+			ID: "c" + string(rune('a'+i)), Type: "function",
+			Function: ToolCallFunc{Name: aitools.ToolListConnections, Arguments: `{"n":` + string(rune('0'+i)) + `}`},
+		}}})
+	}
+	turns = append(turns, ToolsTurn{Content: "ok"})
+	llm := &scriptedLLM{turns: turns}
+	_, err := RunAgent(context.Background(), llm, rt, AgentReq{Prompt: "x", MaxRounds: 8})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if len(llm.seen) < 4 {
+		t.Fatalf("seen=%d", len(llm.seen))
+	}
+	found := false
+	for _, m := range llm.seen[3] {
+		if m.Role == "system" && strings.Contains(m.Content, "检查点") {
+			found = true
+		}
+	}
+	if !found {
+		t.Fatalf("checkpoint not in round4 msgs")
+	}
+}