fix: 收紧代理故障判定并修复自动恢复门控/日历/终态问题

Proxy failure pause, timed reconnect, and guarded auto-resume.

agent/cancelcause.go

@@ -28,6 +28,7 @@ func Causef(code, short, format string, args ...any) *AbortCause {
 var (
 	// Task-level execution context.
 	AbortPausedBySchedule = cause("paused_by_schedule", "计划窗口结束，暂停任务", "任务将在下一个计划窗口恢复，本次执行已停止")
+	AbortPausedByProxy    = cause("paused_by_proxy", "全局代理异常，暂停任务", "全局出口代理连接失败，任务已暂停；代理恢复后将自动继续")
 	AbortPausedByUser     = cause("paused_by_user", "用户暂停了任务",
 		"用户通过任务控制接口（POST /api/tasks/{id}/control，action=pause）暂停了任务。本次 Planner/Worker 运行被主动取消；运行中的意图会退回 frontier(open)，恢复任务后重新领取并从头执行")
 	AbortPausedByOrchestrator = cause("paused_by_orchestrator", "编排 Agent 暂停了任务",

agent/terminalreason_test.go

@@ -139,7 +139,7 @@ func TestTerminalTextNonAbortReasons(t *testing.T) {
 
 func TestAbortCausesAreWellFormed(t *testing.T) {
 	all := []*AbortCause{
-		AbortPausedByUser, AbortPausedByOrchestrator, AbortTaskDeleted, AbortPausedOnReload,
+		AbortPausedByUser, AbortPausedByProxy, AbortPausedByOrchestrator, AbortTaskDeleted, AbortPausedOnReload,
 		AbortGoalMet, AbortSettleDrainTimeout, AbortKilledByPlanner, AbortWorkPausedByUser,
 		AbortWorkCancelledByUser, AbortWorkFinished, AbortPausedRaceGuard,
 		AbortChatStoppedByUser, AbortChatPausedWithTask, AbortChatTurnFinished,

db/findings.go

@@ -23,20 +23,21 @@ type DBFinding struct {
 	ReportEvidenceVersion int64
 	TrafficBindings       []FindingTrafficBinding // populated only for export
 
-	ID              int64
-	TaskID          *int64
-	NodeID          *int64
-	VulnClass       string
-	Name            string // 漏洞名称(可读标题);为空时前端回退展示 VulnClass
-	Severity        string
-	Summary         string
-	Evidence        string
-	Worker          string
-	AssetIDs        []int64
-	Status          string
-	Report          string // 详细报告(Markdown);仅 GetFinding 填充,列表查询不带
-	CreatedAt       time.Time
-	TaskDescription string // populated via LEFT JOIN on tasks
+	ID                   int64
+	TaskID               *int64
+	NodeID               *int64
+	VulnClass            string
+	Name                 string // 漏洞名称(可读标题);为空时前端回退展示 VulnClass
+	Severity             string
+	Summary              string
+	Evidence             string
+	Worker               string
+	AssetIDs             []int64
+	Status               string
+	Report               string // 详细报告(Markdown);仅 GetFinding 填充,列表查询不带
+	ReportConversationID *int64 `json:"report_conversation_id,omitempty"`
+	CreatedAt            time.Time
+	TaskDescription      string // populated via LEFT JOIN on tasks
 }
 
 // Finding triage states (findings.status).
@@ -106,9 +107,9 @@ func (d *DB) AddFinding(taskID, nodeID int64, vulnclass, name, severity, summary
 // findingSelectCols is the column list (with task_description join) every finding
 // list query selects, so scanFinding stays in sync across callers.
 const findingSelectCols = `f.id, f.task_id, f.node_id, f.vulnclass, COALESCE(f.name, ''), f.severity, f.summary,
-	       f.evidence, f.worker, f.asset_ids, COALESCE(f.status, 'pending'), f.created_at,
+		f.evidence, f.worker, f.asset_ids, COALESCE(f.status, 'pending'), f.created_at,
 	       COALESCE(t.description, '') AS task_description, f.evidence_version, f.report_evidence_version,
- (SELECT count(*) FROM finding_traffic_bindings b WHERE b.finding_id=f.id)`
+	(SELECT count(*) FROM finding_traffic_bindings b WHERE b.finding_id=f.id)`
 
 // scanFindings materializes rows selected via findingSelectCols.
 func scanFindings(rows interface {
@@ -661,13 +662,14 @@ func (d *DB) FindingStats() (*FindingStats, error) {
 func (d *DB) GetFinding(id int64) (*DBFinding, error) {
 	f := &DBFinding{}
 	var aidsJSON string
-	err := d.QueryRow(`SELECT `+findingSelectCols+`, COALESCE(f.report, '')
+	var reportConv sql.NullInt64
+	err := d.QueryRow(`SELECT `+findingSelectCols+`, COALESCE(f.report, ''), f.report_conversation_id
 		FROM findings f
 		LEFT JOIN tasks t ON f.task_id = t.id
 		WHERE f.id = $1`, id).Scan(
 		&f.ID, &f.TaskID, &f.NodeID, &f.VulnClass, &f.Name, &f.Severity,
 		&f.Summary, &f.Evidence, &f.Worker, &aidsJSON, &f.Status, &f.CreatedAt,
-		&f.TaskDescription, &f.EvidenceVersion, &f.ReportEvidenceVersion, &f.TrafficCount, &f.Report)
+		&f.TaskDescription, &f.EvidenceVersion, &f.ReportEvidenceVersion, &f.TrafficCount, &f.Report, &reportConv)
 	if err == sql.ErrNoRows {
 		return nil, nil
 	}
@@ -675,9 +677,51 @@ func (d *DB) GetFinding(id int64) (*DBFinding, error) {
 		return nil, err
 	}
 	_ = json.Unmarshal([]byte(aidsJSON), &f.AssetIDs)
+	if reportConv.Valid {
+		f.ReportConversationID = &reportConv.Int64
+	}
+	if f.ReportConversationID == nil && strings.TrimSpace(f.Report) != "" {
+		if resolved, resolveErr := d.ResolveFindingReportConversation(f.ID); resolveErr == nil {
+			f.ReportConversationID = resolved
+		}
+	}
 	return f, nil
 }
 
+func (d *DB) SetFindingReportConversationID(findingID, conversationID int64) error {
+	_, err := d.Exec(`UPDATE findings SET report_conversation_id=$2 WHERE id=$1`, findingID, conversationID)
+	return err
+}
+
+// ResolveFindingReportConversation backfills links for reports generated before
+// report_conversation_id was persisted.
+func (d *DB) ResolveFindingReportConversation(findingID int64) (*int64, error) {
+	var id int64
+	err := d.QueryRow(`SELECT c.id FROM conversations c
+	JOIN conversation_activities a ON a.conversation_id=c.id
+	WHERE c.agent_key='reporter' AND (
+	 c.title = '漏洞报告 · finding#' || $1::text OR
+	 c.title ILIKE '%finding#' || $1::text || '%' OR
+	 a.detail ILIKE '%finding_id=' || $1::text || '%' OR
+	 a.detail ILIKE '%"finding_id":' || $1::text || '%'
+	)
+	ORDER BY CASE WHEN EXISTS (
+		SELECT 1 FROM conversation_activities r
+		WHERE r.conversation_id=c.id AND (r.tool='update_finding_report' OR r.summary ILIKE 'finding % report updated%')
+	) THEN 0 ELSE 1 END,
+	c.created_at ASC, c.id ASC LIMIT 1`, findingID).Scan(&id)
+	if err == sql.ErrNoRows {
+		return nil, nil
+	}
+	if err != nil {
+		return nil, err
+	}
+	if err := d.SetFindingReportConversationID(findingID, id); err != nil {
+		return nil, err
+	}
+	return &id, nil
+}
+
 // DeleteFinding removes a finding entirely: the standalone findings row and its
 // originating exploration node (kind='finding'), so it disappears from the findings
 // list, the per-task 发现 Tab, and the exploration graph alike. Deleting the node

db/schema.sql

@@ -513,6 +513,7 @@ CREATE TABLE IF NOT EXISTS tasks (
     status         TEXT NOT NULL DEFAULT 'created'
                      CHECK (status IN ('created','running','paused','done','failed','timeout')),
     paused         BOOLEAN NOT NULL DEFAULT false,
+    pause_reason   TEXT NOT NULL DEFAULT '',
     queued         BOOLEAN NOT NULL DEFAULT false,
     queued_at      TIMESTAMPTZ,
     queue_mode     TEXT NOT NULL DEFAULT '',
@@ -566,6 +567,7 @@ ALTER TABLE tasks ADD COLUMN IF NOT EXISTS llm_chain_revision BIGINT NOT NULL DE
 ALTER TABLE tasks ADD COLUMN IF NOT EXISTS archived_at TIMESTAMPTZ;
 ALTER TABLE tasks ADD COLUMN IF NOT EXISTS active_run_seconds BIGINT NOT NULL DEFAULT 0;
 ALTER TABLE tasks ADD COLUMN IF NOT EXISTS run_segment_started_at TIMESTAMPTZ;
+ALTER TABLE tasks ADD COLUMN IF NOT EXISTS pause_reason TEXT NOT NULL DEFAULT '';
 CREATE INDEX IF NOT EXISTS idx_tasks_category ON tasks(category_id, created_at DESC)
     WHERE deleted_at IS NULL AND category_id IS NOT NULL;
 CREATE INDEX IF NOT EXISTS idx_tasks_pinned ON tasks(pinned_at DESC)
@@ -1084,6 +1086,7 @@ CREATE TABLE IF NOT EXISTS task_schedule_tasks (
 );
 ALTER TABLE tasks ADD COLUMN IF NOT EXISTS schedule_paused BOOLEAN NOT NULL DEFAULT false;
 ALTER TABLE tasks ADD COLUMN IF NOT EXISTS schedule_manual_until TIMESTAMPTZ;
+UPDATE tasks SET pause_reason='paused_by_schedule' WHERE paused AND schedule_paused AND pause_reason='';
 CREATE INDEX IF NOT EXISTS idx_task_schedule_tasks_task ON task_schedule_tasks(task_id);
 ALTER TABLE task_schedule_tasks ADD COLUMN IF NOT EXISTS paused_by_schedule BOOLEAN NOT NULL DEFAULT false;
 
@@ -1162,11 +1165,13 @@ CREATE TABLE IF NOT EXISTS findings (
     status      TEXT NOT NULL DEFAULT 'pending',
     -- 漏洞详细报告(Markdown)；默认空,仅详情页读取/展示,不进列表接口以免 payload 膨胀。
     report      TEXT NOT NULL DEFAULT '',
+    report_conversation_id BIGINT REFERENCES conversations(id) ON DELETE SET NULL,
     created_at  TIMESTAMPTZ NOT NULL DEFAULT now()
 );
 ALTER TABLE findings ADD COLUMN IF NOT EXISTS status TEXT NOT NULL DEFAULT 'pending';
 ALTER TABLE findings ADD COLUMN IF NOT EXISTS name   TEXT NOT NULL DEFAULT '';
 ALTER TABLE findings ADD COLUMN IF NOT EXISTS report TEXT NOT NULL DEFAULT '';
+ALTER TABLE findings ADD COLUMN IF NOT EXISTS report_conversation_id BIGINT REFERENCES conversations(id) ON DELETE SET NULL;
 CREATE INDEX IF NOT EXISTS idx_findings_task ON findings(task_id, created_at DESC);
 CREATE INDEX IF NOT EXISTS idx_findings_time ON findings(created_at DESC);
 CREATE INDEX IF NOT EXISTS idx_findings_status ON findings(status, created_at DESC);

db/tasks.go

@@ -25,6 +25,7 @@ type Task struct {
 	ExplorationID int64      `json:"exploration_id"`
 	Status        string     `json:"status"`
 	Paused        bool       `json:"paused"`
+	PauseReason   string     `json:"pause_reason,omitempty"`
 	Queued        bool       `json:"queued"`
 	QueuedAt      *time.Time `json:"queued_at,omitempty"`
 	QueueMode     string     `json:"queue_mode,omitempty"`
@@ -302,7 +303,7 @@ RETURNING id, status, paused, created_at`, opts.Name, opts.Remark, opts.Category
 			return nil, err
 		}
 		if !in.StartImmediately {
-			if _, err = tx.Exec(`UPDATE tasks SET paused=true,schedule_paused=true,queue_mode='bootstrap' WHERE id=$1`, t.ID); err != nil {
+			if _, err = tx.Exec(`UPDATE tasks SET paused=true,pause_reason='paused_by_schedule',schedule_paused=true,queue_mode='bootstrap' WHERE id=$1`, t.ID); err != nil {
 				return nil, err
 			}
 			t.Paused = true
@@ -405,11 +406,11 @@ func insertTaskLLMProfiles(tx *sql.Tx, taskID int64, profileIDs []int64) error {
 
 const taskCols = `id, COALESCE(name,''), remark, category_id,
 COALESCE((SELECT category.name FROM task_categories category WHERE category.id=tasks.category_id),''),
-description, goal, exploration_id, status, paused, queued, queued_at, COALESCE(queue_mode,''), llm_profile_id, active_llm_profile_id, COALESCE(parent_ref,''), pinned_at, created_at, completed_at, COALESCE(timeout_seconds,0), COALESCE(plan_heartbeat_seconds,300), COALESCE(coverage_enabled,true), COALESCE(worker_count,0), first_run_at, deadline_at, COALESCE(active_run_seconds,0), run_segment_started_at`
+description, goal, exploration_id, status, paused, COALESCE(pause_reason,''), queued, queued_at, COALESCE(queue_mode,''), llm_profile_id, active_llm_profile_id, COALESCE(parent_ref,''), pinned_at, created_at, completed_at, COALESCE(timeout_seconds,0), COALESCE(plan_heartbeat_seconds,300), COALESCE(coverage_enabled,true), COALESCE(worker_count,0), first_run_at, deadline_at, COALESCE(active_run_seconds,0), run_segment_started_at`
 
 func scanTask(sc interface{ Scan(...any) error }) (*Task, error) {
 	var t Task
-	if err := sc.Scan(&t.ID, &t.Name, &t.Remark, &t.CategoryID, &t.CategoryName, &t.Description, &t.Goal, &t.ExplorationID, &t.Status, &t.Paused, &t.Queued, &t.QueuedAt, &t.QueueMode, &t.LLMProfileID, &t.ActiveLLMProfileID, &t.ParentRef, &t.PinnedAt, &t.CreatedAt, &t.CompletedAt, &t.TimeoutSeconds, &t.PlanHeartbeatSeconds, &t.CoverageEnabled, &t.WorkerCount, &t.FirstRunAt, &t.DeadlineAt, &t.ActiveRunSeconds, &t.RunSegmentStartedAt); err != nil {
+	if err := sc.Scan(&t.ID, &t.Name, &t.Remark, &t.CategoryID, &t.CategoryName, &t.Description, &t.Goal, &t.ExplorationID, &t.Status, &t.Paused, &t.PauseReason, &t.Queued, &t.QueuedAt, &t.QueueMode, &t.LLMProfileID, &t.ActiveLLMProfileID, &t.ParentRef, &t.PinnedAt, &t.CreatedAt, &t.CompletedAt, &t.TimeoutSeconds, &t.PlanHeartbeatSeconds, &t.CoverageEnabled, &t.WorkerCount, &t.FirstRunAt, &t.DeadlineAt, &t.ActiveRunSeconds, &t.RunSegmentStartedAt); err != nil {
 		return nil, err
 	}
 	t.Pinned = t.PinnedAt != nil
@@ -612,6 +613,7 @@ func (d *DB) SetStatusRuntime(id int64, status string) (int64, *time.Time, error
 	err := d.QueryRow(`
 UPDATE tasks
    SET status = $1,
+       pause_reason = CASE WHEN $1 IN ('done','failed','timeout') THEN '' ELSE pause_reason END,
        completed_at = CASE WHEN $1 IN ('done','failed','timeout') THEN COALESCE(completed_at, now()) ELSE NULL END,
        active_run_seconds = active_run_seconds + CASE WHEN $1 IN ('done','failed','timeout') AND run_segment_started_at IS NOT NULL THEN GREATEST(0, FLOOR(EXTRACT(EPOCH FROM (now() - run_segment_started_at)))::bigint) ELSE 0 END,
        run_segment_started_at = CASE WHEN $1 IN ('done','failed','timeout') THEN NULL ELSE run_segment_started_at END
@@ -638,6 +640,7 @@ func (d *DB) SetTerminalStatusGuardedRuntime(id int64, status string) (active in
 	err = d.QueryRow(`
 UPDATE tasks
    SET status = $1,
+       pause_reason = '',
        completed_at = COALESCE(completed_at, now()),
        active_run_seconds = active_run_seconds + CASE WHEN run_segment_started_at IS NOT NULL THEN GREATEST(0, FLOOR(EXTRACT(EPOCH FROM (now() - run_segment_started_at)))::bigint) ELSE 0 END,
        run_segment_started_at = NULL

server/calendar_schedule.go

@@ -188,6 +188,43 @@ func (s *Server) taskManualScheduleEnd(taskID string, now time.Time) (*time.Time
 	return until, nil
 }
 
+// proxyResumeInsideSchedule reports whether a task attached to any enabled
+// calendar is currently allowed to run. Tasks without a calendar remain
+// eligible. Manual schedule releases are honored through schedule_manual_until.
+func (s *Server) proxyResumeInsideSchedule(taskID string, now time.Time) (bool, error) {
+	id, err := strconv.ParseInt(taskID, 10, 64)
+	if err != nil {
+		return false, err
+	}
+	items, err := s.m.pg.ListTaskSchedules()
+	if err != nil {
+		return false, err
+	}
+	attached := false
+	open := false
+	for _, item := range items {
+		for _, boundID := range item.TaskIDs {
+			if boundID != id {
+				continue
+			}
+			attached = true
+			if item.Enabled && item.Status != "completed" {
+				w := evaluateScheduleWindow(item, now)
+				open = open || w.active || item.LastWindowKey == "manual"
+			}
+			break
+		}
+	}
+	if !attached {
+		return true, nil
+	}
+	var manualUntil time.Time
+	if err := s.m.pg.QueryRow(`SELECT COALESCE(schedule_manual_until, 'epoch'::timestamptz) FROM tasks WHERE id=$1`, id).Scan(&manualUntil); err == nil && manualUntil.After(now) {
+		open = true
+	}
+	return open, nil
+}
+
 func (sc *Scheduler) fireTaskSchedules() {
 	sc.fireTaskSchedulesAt(time.Now())
 }
@@ -248,6 +285,15 @@ func (sc *Scheduler) fireTaskSchedulesAt(now time.Time) {
 			if db.IsTerminal(state.Status) {
 				continue
 			}
+			// A proxy-paused task can remain paused when a calendar window ends
+			// during reconnect. Convert it to a schedule pause so a later probe
+			// cannot resume work outside the configured window.
+			if !open[id] && state.Paused && state.PauseReason == "paused_by_proxy" {
+				if err := sc.s.m.ConvertProxyPauseToSchedule(strconv.FormatInt(id, 10)); err != nil {
+					errors = append(errors, fmt.Sprintf("任务 #%d: %s", id, err))
+				}
+				continue
+			}
 			action := ""
 			if open[id] && state.Paused {
 				action = "resume"

server/conversations.go

@@ -701,6 +701,14 @@ func (s *Server) readTriggerBehavior(agentKey string) triggerBehavior {
 // merging by task / all / none); parallel → run each fire in its own concurrent
 // conversation up to trigger_max_parallel. Distinct agents always run concurrently.
 func (s *Server) StartTriggeredRun(agentKey, title, message string, taskID int64, mergeable bool, taskDesc, taskGoal string) {
+	s.startTriggeredRun(agentKey, title, message, taskID, mergeable, taskDesc, taskGoal, 0)
+}
+
+func (s *Server) StartTriggeredRunWithConversation(agentKey, title, message string, taskID int64, mergeable bool, taskDesc, taskGoal string, conversationID int64) {
+	s.startTriggeredRun(agentKey, title, message, taskID, mergeable, taskDesc, taskGoal, conversationID)
+}
+
+func (s *Server) startTriggeredRun(agentKey, title, message string, taskID int64, mergeable bool, taskDesc, taskGoal string, conversationID int64) {
 	if s.m.pg == nil || s.chatAgentRef() == nil {
 		return
 	}
@@ -714,7 +722,7 @@ func (s *Server) StartTriggeredRun(agentKey, title, message string, taskID int64
 	cfg := s.readTriggerBehavior(agentKey) // DB read BEFORE the lock (never under queueMu)
 	s.queueMu.Lock()
 	s.triggerCfg[agentKey] = cfg
-	s.triggerQ[agentKey] = append(s.triggerQ[agentKey], triggeredRun{agentKey: agentKey, title: title, message: message, taskID: taskID, taskDesc: taskDesc, taskGoal: taskGoal, mergeable: mergeable})
+	s.triggerQ[agentKey] = append(s.triggerQ[agentKey], triggeredRun{conversationID: conversationID, agentKey: agentKey, title: title, message: message, taskID: taskID, taskDesc: taskDesc, taskGoal: taskGoal, mergeable: mergeable})
 	s.pumpLocked(agentKey)
 	s.queueMu.Unlock()
 }

server/customtool.go

@@ -447,6 +447,9 @@ func (s *Server) runHTTPTool(ctx context.Context, execRaw json.RawMessage, param
 	}
 	resp, err := client.Do(req)
 	if err != nil {
+		if spec.UseRecordingProxy || strings.TrimSpace(spec.Proxy) == strings.TrimSpace(s.m.GlobalProxy()) {
+			s.handleProxyToolFailure(runCtx, err)
+		}
 		return actool.Errorf("请求失败: " + err.Error()), nil
 	}
 	defer resp.Body.Close()

server/dto.go

@@ -52,6 +52,8 @@ type TaskDTO struct {
 	ActiveRunSeconds   int64              `json:"active_run_seconds"`
 	RunSegmentStarted  int64              `json:"run_segment_started_unix"`
 	Paused             bool               `json:"paused"`
+	PauseReason        string             `json:"pause_reason,omitempty"`
+	Reconnect          *ProxyReconnectDTO `json:"reconnect,omitempty"`
 	Queued             bool               `json:"queued"`
 	Tokens             TokenTotalDTO      `json:"tokens"` // whole-task token consumption
 	GoalsTotal         int                `json:"goals_total"`
@@ -70,6 +72,14 @@ type TaskDTO struct {
 	WorkerCount        int                `json:"worker_count"`     // 本任务 Worker 数;0=用全局默认
 }
 
+type ProxyReconnectDTO struct {
+	Attempt   int    `json:"attempt"`
+	Max       int    `json:"max"`
+	NextAt    int64  `json:"next_at,omitempty"`
+	Exhausted bool   `json:"exhausted"`
+	LastError string `json:"last_error,omitempty"`
+}
+
 // FindingSeverityDTO 是任务列表里按严重度分档的漏洞计数（严重/高/中/低）。
 type FindingSeverityDTO struct {
 	Critical int `json:"critical"`
@@ -118,25 +128,32 @@ func taskDTO(t *Task, status string) TaskDTO {
 	}
 	profileIDs := append(make([]int64, 0, len(llmState.ProfileIDs)), llmState.ProfileIDs...)
 	return TaskDTO{
-		ID:                 t.ID,
-		ExplorationID:      t.ExpID,
-		Name:               lifecycle.Name,
-		Remark:             lifecycle.Remark,
-		CategoryID:         lifecycle.CategoryID,
-		CategoryName:       lifecycle.CategoryName,
-		Pinned:             lifecycle.PinnedAt > 0,
-		PinnedAt:           completedRFC(lifecycle.PinnedAt),
-		Description:        t.Description,
-		Goal:               t.Goal,
-		Status:             status,
-		CreatedAt:          rfc3339(time.Unix(t.CreatedAt, 0)),
-		CreatedUnix:        t.CreatedAt,
-		RunSeconds:         runSeconds,
-		ActiveRunSeconds:   lifecycle.ActiveRunSeconds,
-		RunSegmentStarted:  lifecycle.RunSegmentStartedAt,
-		CompletedAt:        completedRFC(lifecycle.CompletedAt),
-		CompletedUnix:      lifecycle.CompletedAt,
-		Paused:             lifecycle.Paused,
+		ID:                t.ID,
+		ExplorationID:     t.ExpID,
+		Name:              lifecycle.Name,
+		Remark:            lifecycle.Remark,
+		CategoryID:        lifecycle.CategoryID,
+		CategoryName:      lifecycle.CategoryName,
+		Pinned:            lifecycle.PinnedAt > 0,
+		PinnedAt:          completedRFC(lifecycle.PinnedAt),
+		Description:       t.Description,
+		Goal:              t.Goal,
+		Status:            status,
+		CreatedAt:         rfc3339(time.Unix(t.CreatedAt, 0)),
+		CreatedUnix:       t.CreatedAt,
+		RunSeconds:        runSeconds,
+		ActiveRunSeconds:  lifecycle.ActiveRunSeconds,
+		RunSegmentStarted: lifecycle.RunSegmentStartedAt,
+		CompletedAt:       completedRFC(lifecycle.CompletedAt),
+		CompletedUnix:     lifecycle.CompletedAt,
+		Paused:            lifecycle.Paused,
+		PauseReason:       lifecycle.PauseReason,
+		Reconnect: func() *ProxyReconnectDTO {
+			if lifecycle.PauseReason != "paused_by_proxy" && lifecycle.ReconnectAttempt == 0 && !lifecycle.ReconnectExhausted {
+				return nil
+			}
+			return &ProxyReconnectDTO{Attempt: lifecycle.ReconnectAttempt, Max: lifecycle.ReconnectMax, NextAt: lifecycle.ReconnectNextAt, Exhausted: lifecycle.ReconnectExhausted, LastError: lifecycle.ReconnectLastError}
+		}(),
 		Queued:             lifecycle.Queued,
 		LLMProfileID:       llmState.ProfileID,
 		LLMProfileIDs:      profileIDs,
@@ -325,15 +342,16 @@ type FindingDTO struct {
 	ReportStale           bool                       `json:"report_stale"`
 	TrafficBindings       []db.FindingTrafficBinding `json:"traffic_bindings,omitempty"`
 
-	ID        string `json:"id"`
-	FindingID string `json:"finding_id,omitempty"` // standalone findings-table id — the handle for status updates
-	VulnClass string `json:"vulnclass"`
-	Name      string `json:"name,omitempty"` // 漏洞名称;为空时前端回退展示 vulnclass
-	Severity  string `json:"severity"`       // critical | high | medium | low
-	Status    string `json:"status"`         // pending | in_progress | confirmed | resolved | fixed | false_positive | ignored | duplicate | risk_accepted
-	Summary   string `json:"summary"`
-	Evidence  string `json:"evidence"`
-	Report    string `json:"report,omitempty"` // 详细报告(Markdown);仅详情接口返回,列表为空
+	ID                   string `json:"id"`
+	FindingID            string `json:"finding_id,omitempty"` // standalone findings-table id — the handle for status updates
+	VulnClass            string `json:"vulnclass"`
+	Name                 string `json:"name,omitempty"` // 漏洞名称;为空时前端回退展示 vulnclass
+	Severity             string `json:"severity"`       // critical | high | medium | low
+	Status               string `json:"status"`         // pending | in_progress | confirmed | resolved | fixed | false_positive | ignored | duplicate | risk_accepted
+	Summary              string `json:"summary"`
+	Evidence             string `json:"evidence"`
+	Report               string `json:"report,omitempty"` // 详细报告(Markdown);仅详情接口返回,列表为空
+	ReportConversationID *int64 `json:"report_conversation_id,omitempty"`
 
 	IntentID        string            `json:"intent_id,omitempty"`
 	ParamID         string            `json:"param_id,omitempty"`
@@ -464,14 +482,15 @@ func findingFromDB(f *db.DBFinding, assets map[int64]*db.Asset) FindingDTO {
 		FindingID:    i64s(f.ID),
 		TrafficCount: f.TrafficCount, EvidenceVersion: f.EvidenceVersion, ReportEvidenceVersion: f.ReportEvidenceVersion,
 		ReportStale: f.Report != "" && f.EvidenceVersion != f.ReportEvidenceVersion, TrafficBindings: f.TrafficBindings,
-		VulnClass: f.VulnClass,
-		Name:      f.Name,
-		Severity:  f.Severity,
-		Status:    status,
-		Summary:   f.Summary,
-		Evidence:  f.Evidence,
-		Report:    f.Report,
-		TS:        rfc3339(f.CreatedAt),
+		VulnClass:            f.VulnClass,
+		Name:                 f.Name,
+		Severity:             f.Severity,
+		Status:               status,
+		Summary:              f.Summary,
+		Evidence:             f.Evidence,
+		Report:               f.Report,
+		ReportConversationID: f.ReportConversationID,
+		TS:                   rfc3339(f.CreatedAt),
 	}
 	if f.ExportCaseID > 0 {
 		d.ID = "case:" + i64s(f.ExportCaseID)

server/egress_probe.go

@@ -417,6 +417,12 @@ func (s *Server) testEgressProxy(w http.ResponseWriter, r *http.Request) {
 
 	rep.OK = judgeEgress(rep)
 	rep.Warnings = egressWarnings(rep, pending, savedProxy)
+	if rep.OK && !pending {
+		// The settings page uses this endpoint for its manual "测试代理" action.
+		// A successful probe is the same signal as an automatic reconnect success,
+		// but it must never consume an automatic attempt.
+		s.proxyReconnectSuccess()
+	}
 	writeJSON(w, 200, rep)
 }
 

server/engine.go

@@ -153,9 +153,12 @@ type Engine struct {
 	slotPlans            sync.Map // task -> map[slot name]profile ID; immutable for a run
 	// readiness reports whether a global LLM provider is configured — the signal behind
 	// Ready()/the llm_configured indicator. Wired once at startup; nil → not ready.
-	readiness func() bool
+	readiness    func() bool
+	proxyFailure func(taskID string, err error)
 }
 
+func (e *Engine) SetProxyFailureHandler(fn func(string, error)) { e.proxyFailure = fn }
+
 type slotAvailability struct {
 	available bool
 	lastEmit  time.Time
@@ -942,7 +945,12 @@ func (e *Engine) plannerLoop(ctx context.Context, t *Task) {
 		}
 		e.stampFirstRun(t) // 首次真正规划 → 盖 first_run_at + 算 deadline(仅带 timeout 的任务)
 		e.touch(t.ID)
-		emit := func(r db.Activity) { e.emitActivity(t, r) }
+		emit := func(r db.Activity) {
+			if r.Kind == "tool_result" && r.IsError && e.proxyFailure != nil && isGlobalProxyDialFailure(errors.New(r.Detail+" "+r.Summary)) {
+				e.proxyFailure(t.ID, errors.New(r.Detail+" "+r.Summary))
+			}
+			e.emitActivity(t, r)
+		}
 		ectx := e.clockCtx(e.execContextFor(ctx, t.ID), t, false) // cancellable by Pause; 带任务 deadline
 		if ectx.Err() != nil || e.IsDeleting(t.ID) {
 			return
@@ -962,6 +970,9 @@ func (e *Engine) plannerLoop(ctx context.Context, t *Task) {
 		switch {
 		case err != nil && ectx.Err() == nil:
 			log.Printf("[planner] task %s 规划出错: %v", t.ID, err)
+			if e.proxyFailure != nil && isProxyFailure(err) {
+				e.proxyFailure(t.ID, err)
+			}
 		case met:
 			log.Printf("[planner] task %s 判定目标达成: %s", t.ID, reason)
 			// 所有目标达成 → 持久化任务状态为 done（前端 DTO 会优先展示该终态）。
@@ -1081,7 +1092,12 @@ func (e *Engine) runIntent(ctx context.Context, t *Task, name string, worker *ag
 	hasChatMessage := message != ""
 	e.stampFirstRun(t) // 首次真正执行 → 盖 first_run_at + 算 deadline(仅带 timeout 的任务)
 	e.touch(t.ID)
-	emit := func(r db.Activity) { e.emitActivity(t, r) }
+	emit := func(r db.Activity) {
+		if r.Kind == "tool_result" && r.IsError && e.proxyFailure != nil && isGlobalProxyDialFailure(errors.New(r.Detail+" "+r.Summary)) {
+			e.proxyFailure(t.ID, errors.New(r.Detail+" "+r.Summary))
+		}
+		e.emitActivity(t, r)
+	}
 	ectx := e.clockCtx(e.execContextFor(ctx, t.ID), t, false) // cancellable by Pause; 带任务 deadline
 	if ectx.Err() != nil || e.IsDeleting(t.ID) {
 		if err := transitionIntentState(t.Store, intent.ID, "running", "open"); err != nil {
@@ -1124,6 +1140,9 @@ func (e *Engine) runIntent(ctx context.Context, t *Task, name string, worker *ag
 	} else {
 		reason, wrote, err = worker.Execute(workCtx, name, wTaskID, e.m.assets, t.Store, intent, hooks, emit, e.m.enrich, t.NotifyFinding)
 	}
+	if err != nil && e.proxyFailure != nil && isProxyFailure(err) {
+		e.proxyFailure(t.ID, err)
+	}
 	e.EndLLMCall(t.ID)
 	// model_error 收场 → 额外重跑几次（退避后再试）。仅在意图仍属本 work、任务
 	// 未暂停/未终止/未取消【且未进入收尾】时重试；否则让位给对应分支处理(收尾期不

server/finding_report.go

@@ -58,20 +58,30 @@ func (s *Server) generateFindingReport(w http.ResponseWriter, r *http.Request) {
 		}
 	}
 	message := manualFindingReportMessage(f, taskID)
-	s.StartTriggeredRun(
+	conversation, err := s.m.pg.CreateConversation("reporter", fmt.Sprintf("漏洞报告 · finding#%d", f.ID), nil)
+	if err != nil {
+		writeErr(w, http.StatusInternalServerError, "创建报告对话失败: "+err.Error())
+		return
+	}
+	if err := s.m.pg.SetFindingReportConversationID(f.ID, conversation.ID); err != nil {
+		writeErr(w, http.StatusInternalServerError, "关联报告对话失败: "+err.Error())
+		return
+	}
+	s.StartTriggeredRunWithConversation(
 		"reporter",
 		fmt.Sprintf("人工生成步骤与报告 · finding#%d", f.ID),
 		message,
 		taskID,
 		false,
 		taskDesc,
-		taskGoal,
+		taskGoal, conversation.ID,
 	)
 	writeJSON(w, http.StatusAccepted, map[string]any{
 		"queued":          true,
 		"agent_key":       "reporter",
 		"finding_id":      f.ID,
 		"finding_node_id": *f.NodeID,
+		"conversation_id": conversation.ID,
 	})
 }
 

server/goals.go

@@ -112,10 +112,19 @@ func (s *Server) admitTaskWhen(t *Task, mode string, requirePaused bool) (bool,
 	return s.admitTaskWhenScheduled(t, mode, requirePaused, false)
 }
 func (s *Server) admitTaskWhenScheduled(t *Task, mode string, requirePaused, scheduled bool) (queued bool, err error) {
+	return s.admitTaskWhenScheduledWithReason(t, mode, requirePaused, scheduled, "", false)
+}
+
+// admitTaskWhenScheduledWithReason is the shared admission gate. A non-empty
+// requiredReason is checked after concMu is acquired, alongside the paused and
+// terminal checks, so proxy recovery cannot revive a task that was re-paused by
+// a user while the probe was in flight. preserveSchedule keeps calendar state
+// and never creates schedule_manual_until (used by proxy recovery).
+func (s *Server) admitTaskWhenScheduledWithReason(t *Task, mode string, requirePaused, scheduled bool, requiredReason string, preserveSchedule bool) (queued bool, err error) {
 	if t == nil {
 		return false, fmt.Errorf("task not found")
 	}
-	manualSchedule := requirePaused && !scheduled
+	manualSchedule := requirePaused && !scheduled && !preserveSchedule
 	if manualSchedule {
 		// Match the calendar tick's lock order: scheduleMu before concMu. The
 		// override must commit before another tick can decide to pause this task.
@@ -155,6 +164,9 @@ func (s *Server) admitTaskWhenScheduled(t *Task, mode string, requirePaused, sch
 		if !lifecycle.Paused {
 			return false, fmt.Errorf("仅已暂停的任务可以继续")
 		}
+		if requiredReason != "" && lifecycle.PauseReason != requiredReason {
+			return false, fmt.Errorf("任务暂停原因已改变")
+		}
 		mode = s.resumeAdmissionMode(t)
 	}
 	var manualUntil *time.Time
@@ -212,7 +224,7 @@ func (s *Server) admitTaskWhenScheduled(t *Task, mode string, requirePaused, sch
 	if wasTerminal {
 		status = "running"
 	}
-	if err := s.m.applyTaskAdmission(t.ID, lifecycle.Status, status, shouldQueue, mode, wasQueued, manualUntil); err != nil {
+	if err := s.m.applyTaskAdmissionWithSchedule(t.ID, lifecycle.Status, status, shouldQueue, mode, wasQueued, manualUntil, preserveSchedule); err != nil {
 		if !engineWasPaused && !wasPaused && !wasQueued {
 			s.engine.Resume(t)
 		}

server/manager.go

@@ -28,19 +28,25 @@ import (
 // sharing the process-wide asset store. ID is the PG task id as a string; ExpID
 // is the exploration the task owns.
 type Task struct {
-	ID           string `json:"id"`
-	ExpID        int64  `json:"exploration_id"`
-	Remark       string `json:"remark"`
-	Name         string `json:"name"` // 可选任务名称;空=未命名
-	CategoryID   *int64 `json:"category_id,omitempty"`
-	CategoryName string `json:"category_name,omitempty"`
-	PinnedAt     int64  `json:"pinned_at,omitempty"`
-	Description  string `json:"description"`
-	Goal         string `json:"goal"`
-	CreatedAt    int64  `json:"created_at"`
-	CompletedAt  int64  `json:"completed_at,omitempty"` // 进入终态的 unix 秒;0=未完成
-	Paused       bool   `json:"paused"`
-	Queued       bool   `json:"queued"` // 因并发上限被挂起、等待空位自动启动;true=尚未开跑
+	ID                 string `json:"id"`
+	ExpID              int64  `json:"exploration_id"`
+	Remark             string `json:"remark"`
+	Name               string `json:"name"` // 可选任务名称;空=未命名
+	CategoryID         *int64 `json:"category_id,omitempty"`
+	CategoryName       string `json:"category_name,omitempty"`
+	PinnedAt           int64  `json:"pinned_at,omitempty"`
+	Description        string `json:"description"`
+	Goal               string `json:"goal"`
+	CreatedAt          int64  `json:"created_at"`
+	CompletedAt        int64  `json:"completed_at,omitempty"` // 进入终态的 unix 秒;0=未完成
+	Paused             bool   `json:"paused"`
+	PauseReason        string `json:"pause_reason,omitempty"`
+	ReconnectAttempt   int    `json:"-"`
+	ReconnectMax       int    `json:"-"`
+	ReconnectNextAt    int64  `json:"-"`
+	ReconnectExhausted bool   `json:"-"`
+	ReconnectLastError string `json:"-"`
+	Queued             bool   `json:"queued"` // 因并发上限被挂起、等待空位自动启动;true=尚未开跑
 	// QueuedAt is an internal Unix-nanosecond ordering key. It is deliberately
 	// finer than CreatedAt so several tasks enqueued in the same second retain
 	// their real FIFO order.
@@ -88,6 +94,12 @@ type taskLifecycleState struct {
 	PinnedAt            int64
 	Status              string
 	Paused              bool
+	PauseReason         string
+	ReconnectAttempt    int
+	ReconnectMax        int
+	ReconnectNextAt     int64
+	ReconnectExhausted  bool
+	ReconnectLastError  string
 	Queued              bool
 	QueuedAt            int64
 	QueueMode           string
@@ -113,11 +125,13 @@ func (t *Task) lifecycleSnapshot() taskLifecycleState {
 
 func (t *Task) lifecycleSnapshotLocked() taskLifecycleState {
 	return taskLifecycleState{
-		Name:                t.Name,
-		Remark:              t.Remark,
-		PinnedAt:            t.PinnedAt,
-		Status:              t.Status,
-		Paused:              t.Paused,
+		Name:             t.Name,
+		Remark:           t.Remark,
+		PinnedAt:         t.PinnedAt,
+		Status:           t.Status,
+		Paused:           t.Paused,
+		PauseReason:      t.PauseReason,
+		ReconnectAttempt: t.ReconnectAttempt, ReconnectMax: t.ReconnectMax, ReconnectNextAt: t.ReconnectNextAt, ReconnectExhausted: t.ReconnectExhausted, ReconnectLastError: t.ReconnectLastError,
 		Queued:              t.Queued,
 		QueuedAt:            t.QueuedAt,
 		QueueMode:           t.QueueMode,
@@ -145,6 +159,12 @@ func (t *Task) updateLifecycle(update func(*taskLifecycleState)) {
 	t.PinnedAt = state.PinnedAt
 	t.Status = state.Status
 	t.Paused = state.Paused
+	t.PauseReason = state.PauseReason
+	t.ReconnectAttempt = state.ReconnectAttempt
+	t.ReconnectMax = state.ReconnectMax
+	t.ReconnectNextAt = state.ReconnectNextAt
+	t.ReconnectExhausted = state.ReconnectExhausted
+	t.ReconnectLastError = state.ReconnectLastError
 	t.Queued = state.Queued
 	t.QueuedAt = state.QueuedAt
 	t.QueueMode = state.QueueMode
@@ -242,6 +262,7 @@ type Manager struct {
 	// their in-memory mirrors. lifecycleMu makes snapshots race-free, but without
 	// this outer write lock an older request could commit first and publish last.
 	taskStateMu sync.Mutex
+	controlMu   *sync.Mutex // Server's admission/task-control mutex, when attached.
 	mu          sync.RWMutex
 	tasks       map[string]*Task
 	active      string
@@ -261,6 +282,10 @@ type Manager struct {
 	globalProxy string
 }
 
+// SetTaskControlMutex lets lifecycle operations that originate in Manager
+// participate in the Server's admission/task-control critical section.
+func (m *Manager) SetTaskControlMutex(mu *sync.Mutex) { m.controlMu = mu }
+
 // Settings keys the UI toggles at runtime.
 const (
 	settingTrafficCapture      = "traffic_capture"
@@ -1073,7 +1098,7 @@ func taskFromPG(pt *pgdb.Task, store *pgdb.ExplorationStore, ic *intercept.Inter
 		Remark:     pt.Remark,
 		CategoryID: cloneInt64Ptr(pt.CategoryID), CategoryName: pt.CategoryName,
 		PinnedAt:    unixOrZero(pt.PinnedAt),
-		Description: pt.Description, Goal: pt.Goal, CreatedAt: pt.CreatedAt.Unix(), Paused: pt.Paused, Queued: pt.Queued,
+		Description: pt.Description, Goal: pt.Goal, CreatedAt: pt.CreatedAt.Unix(), Paused: pt.Paused, PauseReason: pt.PauseReason, Queued: pt.Queued,
 		QueuedAt: unixNanoOrZero(pt.QueuedAt), QueueMode: pt.QueueMode,
 		CompletedAt: unixOrZero(pt.CompletedAt), Status: pt.Status, ParentRef: pt.ParentRef,
 		LLMProfileID:  pt.LLMProfileID,
@@ -1391,12 +1416,16 @@ func (m *Manager) SetTaskPaused(id string, paused bool) error {
 // admission keeps its FIFO timestamp; a task that was explicitly paused and is
 // now re-queued receives a fresh tail position.
 func (m *Manager) ApplyTaskAdmission(id, expectedStatus, status string, queued bool, mode string, preservePosition bool) error {
-	return m.applyTaskAdmission(id, expectedStatus, status, queued, mode, preservePosition, nil)
+	return m.applyTaskAdmissionWithSchedule(id, expectedStatus, status, queued, mode, preservePosition, nil, false)
 }
 
 // Manual task control records its schedule override in the admission commit, so
 // failed resumes cannot bypass a calendar and FIFO promotion preserves it.
 func (m *Manager) applyTaskAdmission(id, expectedStatus, status string, queued bool, mode string, preservePosition bool, manualUntil *time.Time) error {
+	return m.applyTaskAdmissionWithSchedule(id, expectedStatus, status, queued, mode, preservePosition, manualUntil, false)
+}
+
+func (m *Manager) applyTaskAdmissionWithSchedule(id, expectedStatus, status string, queued bool, mode string, preservePosition bool, manualUntil *time.Time, preserveSchedule bool) error {
 	m.taskStateMu.Lock()
 	defer m.taskStateMu.Unlock()
 	n, err := strconv.ParseInt(id, 10, 64)
@@ -1420,8 +1449,9 @@ func (m *Manager) applyTaskAdmission(id, expectedStatus, status string, queued b
 	        WHEN $2 IN ('done','failed','timeout') THEN COALESCE(completed_at, now())
 	        ELSE NULL
 	    END,
-	    paused=false, schedule_paused=false,
-	    schedule_manual_until=COALESCE($7::timestamptz, schedule_manual_until),
+	    paused=false, pause_reason='',
+	    schedule_paused=CASE WHEN $8 THEN schedule_paused ELSE false END,
+	    schedule_manual_until=CASE WHEN $8 THEN schedule_manual_until ELSE COALESCE($7::timestamptz, schedule_manual_until) END,
 	    queued=$3,
 	    queued_at=CASE
 	        WHEN NOT $3 THEN NULL
@@ -1444,7 +1474,7 @@ func (m *Manager) applyTaskAdmission(id, expectedStatus, status string, queued b
 	    ,active_run_seconds=active_run_seconds + CASE WHEN ($3 OR $2 IN ('done','failed','timeout')) AND run_segment_started_at IS NOT NULL THEN GREATEST(0, FLOOR(EXTRACT(EPOCH FROM (now() - run_segment_started_at)))::bigint) ELSE 0 END
 	    ,run_segment_started_at=CASE WHEN $3 OR $2 IN ('done','failed','timeout') THEN NULL ELSE run_segment_started_at END
 	WHERE id=$1 AND deleted_at IS NULL AND status=$6
-	RETURNING queued_at, queue_mode, completed_at, first_run_at, deadline_at, active_run_seconds, run_segment_started_at`, n, status, queued, mode, preservePosition, expectedStatus, manualUntil).
+	RETURNING queued_at, queue_mode, completed_at, first_run_at, deadline_at, active_run_seconds, run_segment_started_at`, n, status, queued, mode, preservePosition, expectedStatus, manualUntil, preserveSchedule).
 		Scan(&queuedAt, &committedMode, &completedAt, &firstRunAt, &deadlineAt, &activeRunSeconds, &segmentStartedAt)
 	if errors.Is(err, sql.ErrNoRows) {
 		return fmt.Errorf("task %s lifecycle changed before admission (expected status %q)", id, expectedStatus)
@@ -1458,6 +1488,7 @@ func (m *Manager) applyTaskAdmission(id, expectedStatus, status string, queued b
 		t.updateLifecycle(func(state *taskLifecycleState) {
 			state.Status = status
 			state.Paused = false
+			state.PauseReason = ""
 			state.Queued = queued
 			state.QueueMode = committedMode
 			state.QueuedAt = 0
@@ -1494,6 +1525,12 @@ func (m *Manager) applyTaskAdmission(id, expectedStatus, status string, queued b
 func (m *Manager) ApplyTaskPause(id string) error { return m.ApplyTaskPauseOrigin(id, false) }
 
 func (m *Manager) ApplyTaskPauseOrigin(id string, scheduled bool) error {
+	return m.ApplyTaskPauseOriginReason(id, map[bool]string{true: "paused_by_schedule", false: "paused_by_user"}[scheduled], scheduled)
+}
+
+// ApplyTaskPauseOriginReason persists a pause cause and keeps it on the live handle.
+// reason is a stable API value such as user, schedule, orchestrator, or proxy.
+func (m *Manager) ApplyTaskPauseOriginReason(id, reason string, scheduled bool) error {
 	m.taskStateMu.Lock()
 	defer m.taskStateMu.Unlock()
 	n, err := strconv.ParseInt(id, 10, 64)
@@ -1504,12 +1541,12 @@ func (m *Manager) ApplyTaskPauseOrigin(id string, scheduled bool) error {
 	var activeRunSeconds int64
 	var segmentStartedAt sql.NullTime
 	err = m.pg.QueryRow(`UPDATE tasks
-		SET paused=true, queued=false, queued_at=NULL, schedule_paused=$2, schedule_manual_until=NULL,
+		SET paused=true, pause_reason=$3, queued=false, queued_at=NULL, schedule_paused=$2, schedule_manual_until=NULL,
 		    active_run_seconds=active_run_seconds + CASE WHEN run_segment_started_at IS NOT NULL THEN GREATEST(0, FLOOR(EXTRACT(EPOCH FROM (now() - run_segment_started_at)))::bigint) ELSE 0 END,
 		    run_segment_started_at=NULL
 		WHERE id=$1 AND deleted_at IS NULL AND paused=false
 		  AND status NOT IN ('done','failed','timeout')
-		RETURNING COALESCE(queue_mode,''), active_run_seconds, run_segment_started_at`, n, scheduled).Scan(&mode, &activeRunSeconds, &segmentStartedAt)
+		RETURNING COALESCE(queue_mode,''), active_run_seconds, run_segment_started_at`, n, scheduled, reason).Scan(&mode, &activeRunSeconds, &segmentStartedAt)
 	if errors.Is(err, sql.ErrNoRows) {
 		return fmt.Errorf("task %s is unavailable for pause", id)
 	}
@@ -1520,6 +1557,12 @@ func (m *Manager) ApplyTaskPauseOrigin(id string, scheduled bool) error {
 	if t := m.tasks[id]; t != nil {
 		t.updateLifecycle(func(state *taskLifecycleState) {
 			state.Paused = true
+			state.PauseReason = reason
+			state.ReconnectAttempt = 0
+			state.ReconnectMax = 0
+			state.ReconnectNextAt = 0
+			state.ReconnectExhausted = false
+			state.ReconnectLastError = ""
 			state.Queued = false
 			state.QueuedAt = 0
 			state.QueueMode = mode
@@ -1534,6 +1577,97 @@ func (m *Manager) ApplyTaskPauseOrigin(id string, scheduled bool) error {
 	return nil
 }
 
+// ApplyProxyPause atomically pauses a running task and marks it eligible for
+// reconnect auto-resume. Already-paused tasks retain their original reason.
+func (m *Manager) ApplyProxyPause(id string) error {
+	if m.controlMu != nil {
+		m.controlMu.Lock()
+		defer m.controlMu.Unlock()
+	}
+	m.taskStateMu.Lock()
+	defer m.taskStateMu.Unlock()
+	n, err := strconv.ParseInt(id, 10, 64)
+	if err != nil {
+		return err
+	}
+	var mode string
+	var active int64
+	var started sql.NullTime
+	err = m.pg.QueryRow(`UPDATE tasks SET paused=true, pause_reason='paused_by_proxy', queued=false, queued_at=NULL,
+ active_run_seconds=active_run_seconds + CASE WHEN run_segment_started_at IS NOT NULL THEN GREATEST(0, FLOOR(EXTRACT(EPOCH FROM (now() - run_segment_started_at)))::bigint) ELSE 0 END,
+ run_segment_started_at=NULL WHERE id=$1 AND deleted_at IS NULL AND paused=false AND status NOT IN ('done','failed','timeout')
+ RETURNING COALESCE(queue_mode,''), active_run_seconds, run_segment_started_at`, n).Scan(&mode, &active, &started)
+	if errors.Is(err, sql.ErrNoRows) {
+		return nil
+	}
+	if err != nil {
+		return err
+	}
+	m.mu.RLock()
+	t := m.tasks[id]
+	m.mu.RUnlock()
+	if t != nil {
+		t.updateLifecycle(func(s *taskLifecycleState) {
+			s.Paused = true
+			s.PauseReason = "paused_by_proxy"
+			s.Queued = false
+			s.QueuedAt = 0
+			s.QueueMode = mode
+			s.ActiveRunSeconds = active
+			s.RunSegmentStartedAt = 0
+		})
+	}
+	return nil
+}
+
+// ClearPauseReason removes auto-resume eligibility without changing lifecycle.
+func (m *Manager) ClearPauseReason(id string) error {
+	if _, err := m.pg.Exec(`UPDATE tasks SET pause_reason='' WHERE id=$1 AND pause_reason='paused_by_proxy'`, id); err != nil {
+		return err
+	}
+	m.mu.RLock()
+	t := m.tasks[id]
+	m.mu.RUnlock()
+	if t != nil && t.lifecycleSnapshot().PauseReason == "paused_by_proxy" {
+		t.updateLifecycle(func(s *taskLifecycleState) {
+			s.PauseReason = ""
+			s.ReconnectAttempt, s.ReconnectMax, s.ReconnectNextAt = 0, 0, 0
+			s.ReconnectExhausted, s.ReconnectLastError = false, ""
+	})
+	}
+	return nil
+}
+
+// ConvertProxyPauseToSchedule records that a calendar window ended while a
+// task was waiting for proxy recovery. It preserves the Engine pause barrier
+// and removes proxy auto-resume eligibility until the next window.
+func (m *Manager) ConvertProxyPauseToSchedule(id string) error {
+	m.taskStateMu.Lock()
+	defer m.taskStateMu.Unlock()
+	n, err := strconv.ParseInt(id, 10, 64)
+	if err != nil {
+		return err
+	}
+	res, err := m.pg.Exec(`UPDATE tasks SET pause_reason='paused_by_schedule', schedule_paused=true WHERE id=$1 AND paused=true AND pause_reason='paused_by_proxy' AND status NOT IN ('done','failed','timeout')`, n)
+	if err != nil {
+		return err
+	}
+	if rows, _ := res.RowsAffected(); rows == 0 {
+		return nil
+	}
+	m.mu.RLock()
+	t := m.tasks[id]
+	m.mu.RUnlock()
+	if t != nil {
+		t.updateLifecycle(func(v *taskLifecycleState) {
+			v.PauseReason = "paused_by_schedule"
+			v.ReconnectAttempt, v.ReconnectMax, v.ReconnectNextAt = 0, 0, 0
+			v.ReconnectExhausted, v.ReconnectLastError = false, ""
+		})
+	}
+	return nil
+}
+
 // EnqueueTask persists the concurrency hold and syncs the in-memory handle.
 func (m *Manager) EnqueueTask(id, mode string) error {
 	m.taskStateMu.Lock()
@@ -1748,6 +1882,10 @@ func (m *Manager) SetTaskStatusGuarded(id, status string) (won bool, err error)
 			state.RunSegmentStartedAt = unixOrZero(started)
 			if pgdb.IsTerminal(status) {
 				state.RunSegmentStartedAt = 0
+				state.Paused = false
+				state.PauseReason = ""
+				state.ReconnectAttempt, state.ReconnectMax, state.ReconnectNextAt = 0, 0, 0
+				state.ReconnectExhausted, state.ReconnectLastError = false, ""
 				if state.CompletedAt == 0 {
 					state.CompletedAt = time.Now().Unix()
 				}
@@ -1784,6 +1922,10 @@ func (m *Manager) SetTaskStatus(id, status string) error {
 			state.RunSegmentStartedAt = unixOrZero(started)
 			if pgdb.IsTerminal(status) {
 				state.RunSegmentStartedAt = 0
+				state.Paused = false
+				state.PauseReason = ""
+				state.ReconnectAttempt, state.ReconnectMax, state.ReconnectNextAt = 0, 0, 0
+				state.ReconnectExhausted, state.ReconnectLastError = false, ""
 				if state.CompletedAt == 0 {
 					state.CompletedAt = time.Now().Unix()
 				}

server/orchestration.go

@@ -11,6 +11,7 @@ import (
 
 	"github.com/Autumn-27/artex/agent"
 	"github.com/Autumn-27/artex/db"
+	"github.com/Autumn-27/artex/intercept"
 	"github.com/Autumn-27/norma/permission"
 	actool "github.com/Autumn-27/norma/tool"
 )
@@ -479,6 +480,11 @@ func (s *Server) toolUpdateFindingReport() actool.CoreTool {
 						return actool.Errorf("归档报告截图失败: " + err.Error()), nil
 					}
 				}
+				if conversationID := intercept.ConvIDFromContext(ctx); conversationID > 0 {
+					if err := s.m.pg.SetFindingReportConversationID(fid, conversationID); err != nil {
+						return actool.Errorf("关联报告对话失败: " + err.Error()), nil
+					}
+				}
 			}
 			n, err := s.m.pg.SetFindingReportVersionByNodeID(ctx, nodeID, a.Report, a.EvidenceVersion)
 			if err != nil {

server/proxy_autoresume.go

@@ -0,0 +1,338 @@
+package server
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"log"
+	"net/http"
+	"strings"
+	"sync"
+	"time"
+
+	"github.com/Autumn-27/artex/agent"
+)
+
+const (
+	settingProxyReconnectMax      = "proxy_reconnect_max_attempts"
+	settingProxyReconnectInterval = "proxy_reconnect_interval_seconds"
+	defaultProxyReconnectMax      = 5
+	defaultProxyReconnectInterval = 30
+)
+
+// isProxyFailure is used for LLM/planner errors, where the error has no trusted
+// indication that it came from the global target proxy. Keep this deliberately
+// narrow: generic EOF/reset/refused errors belong to the model_error retry path.
+func isProxyFailure(err error) bool {
+	if err == nil {
+		return false
+	}
+	s := strings.ToLower(err.Error())
+	for _, part := range []string{"proxyconnect", "proxy connection", "proxy error", "proxy handshake", "socks handshake", "failed to connect to proxy", "connect to proxy"} {
+		if strings.Contains(s, part) {
+			return true
+		}
+	}
+	return false
+}
+
+// isGlobalProxyDialFailure is only used by the MITM upstream callback. That
+// callback is on the actual configured global-proxy path; the saved proxy is
+// still probed before any task is paused, so a target-side RequestError cannot
+// pause tasks while the proxy itself remains healthy.
+func isGlobalProxyDialFailure(err error) bool {
+	if err == nil {
+		return false
+	}
+	s := strings.ToLower(err.Error())
+	for _, part := range []string{"proxyconnect", "proxy connection", "proxy error", "proxy handshake", "socks", "connect to proxy", "upstream proxy", "connection refused", "connection reset", "network is unreachable", "broken pipe", "unexpected eof", "eof"} {
+		if strings.Contains(s, part) {
+			return true
+		}
+	}
+	return false
+}
+
+type proxyReconnectSupervisor struct {
+	mu   sync.Mutex
+	wake chan struct{}
+}
+
+func (s *Server) ensureProxySupervisor() {
+	s.proxyReconnect.mu.Lock()
+	if s.proxyReconnect.wake == nil {
+		s.proxyReconnect.wake = make(chan struct{}, 1)
+		go s.runProxyReconnect(s.ctx)
+	}
+	s.proxyReconnect.mu.Unlock()
+}
+
+func (s *Server) signalProxySupervisor() {
+	s.ensureProxySupervisor()
+	select {
+	case s.proxyReconnect.wake <- struct{}{}:
+	default:
+	}
+}
+
+func (s *Server) proxyReconnectConfig() (int, time.Duration) {
+	max := defaultProxyReconnectMax
+	interval := defaultProxyReconnectInterval
+	if s == nil || s.m == nil || s.m.pg == nil {
+		return max, time.Duration(interval) * time.Second
+	}
+	if v, ok, _ := s.m.pg.GetSetting(settingProxyReconnectMax); ok {
+		if n, err := fmt.Sscanf(strings.TrimSpace(v), "%d", &max); err != nil || n != 1 || max <= 0 {
+			max = defaultProxyReconnectMax
+		}
+	}
+	if v, ok, _ := s.m.pg.GetSetting(settingProxyReconnectInterval); ok {
+		if n, err := fmt.Sscanf(strings.TrimSpace(v), "%d", &interval); err != nil || n != 1 || interval <= 0 {
+			interval = defaultProxyReconnectInterval
+		}
+	}
+	return max, time.Duration(interval) * time.Second
+}
+
+func (s *Server) handleProxyFailure(taskID string, cause error) {
+	if strings.TrimSpace(s.m.GlobalProxy()) == "" {
+		return
+	}
+	if probe := probeEgress(context.Background(), s.m.GlobalProxy(), "", defaultEgressProbes, 10*time.Second); probe.OK {
+		return
+	}
+	s.pauseProxyTask(taskID, cause)
+}
+
+func (s *Server) pauseProxyTask(taskID string, cause error) {
+	if strings.TrimSpace(s.m.GlobalProxy()) == "" {
+		return
+	}
+	err := s.m.ApplyProxyPause(taskID)
+	if err != nil {
+		log.Printf("[proxy] pause task %s: %v", taskID, err)
+		return
+	}
+	if s.engine != nil {
+		if t, ok := s.m.Task(taskID); ok && t.lifecycleSnapshot().PauseReason == "paused_by_proxy" {
+			s.engine.Pause(taskID, agent.AbortPausedByProxy)
+		}
+	}
+	s.cancelTaskChat(taskID, agent.AbortChatPausedWithTask)
+	max, interval := s.proxyReconnectConfig()
+	now := time.Now().Add(interval).Unix()
+	for _, t := range s.m.List() {
+		if t.ID != taskID && t.lifecycleSnapshot().PauseReason != "paused_by_proxy" {
+			continue
+		}
+		t.updateLifecycle(func(st *taskLifecycleState) {
+			st.ReconnectMax = max
+			st.ReconnectNextAt = now
+			st.ReconnectExhausted = false
+			st.ReconnectLastError = truncateProxyError(cause)
+		})
+	}
+	s.ensureProxySupervisor()
+	s.signalProxySupervisor()
+}
+
+func (s *Server) handleProxyTrafficFailure(cause error) {
+	if !isGlobalProxyDialFailure(cause) || strings.TrimSpace(s.m.GlobalProxy()) == "" {
+		return
+	}
+	// Confirm the saved global proxy itself before pausing. MITM RequestError can
+	// also represent a target-side failure, and those must not pause a task.
+	if probe := probeEgress(context.Background(), s.m.GlobalProxy(), "", defaultEgressProbes, 10*time.Second); probe.OK {
+		return
+	}
+	for _, t := range s.m.List() {
+		st := t.lifecycleSnapshot()
+		if isTerminalStatus(st.Status) || st.Paused || s.engine.IsDeleting(t.ID) {
+			continue
+		}
+		s.pauseProxyTask(t.ID, cause)
+	}
+}
+
+func (s *Server) handleProxyToolFailure(ctx context.Context, cause error) {
+	if !isGlobalProxyDialFailure(cause) || strings.TrimSpace(s.m.GlobalProxy()) == "" {
+		return
+	}
+	if probe := probeEgress(context.Background(), s.m.GlobalProxy(), "", defaultEgressProbes, 10*time.Second); probe.OK {
+		return
+	}
+	if info := agent.RunInfoFrom(ctx); info.TaskID > 0 {
+		s.pauseProxyTask(fmt.Sprint(info.TaskID), cause)
+	}
+}
+
+func truncateProxyError(err error) string {
+	if err == nil {
+		return ""
+	}
+	s := err.Error()
+	if len(s) > 240 {
+		return s[:240] + "…"
+	}
+	return s
+}
+
+func (s *Server) runProxyReconnect(ctx context.Context) {
+	ticker := time.NewTicker(500 * time.Millisecond)
+	defer ticker.Stop()
+	for {
+		select {
+		case <-ctx.Done():
+			return
+		case <-ticker.C:
+			s.proxyReconnectTick()
+		case <-s.proxyReconnect.wake:
+			s.proxyReconnectTick()
+		}
+	}
+}
+
+func (s *Server) proxyReconnectTick() {
+	if strings.TrimSpace(s.m.GlobalProxy()) == "" {
+		return
+	}
+	now := time.Now().Unix()
+	max, interval := s.proxyReconnectConfig()
+	for _, t := range s.m.List() {
+		st := t.lifecycleSnapshot()
+		if isTerminalStatus(st.Status) {
+			t.updateLifecycle(func(v *taskLifecycleState) {
+				if v.PauseReason == "paused_by_proxy" {
+					v.PauseReason = ""
+				}
+				v.ReconnectAttempt, v.ReconnectMax, v.ReconnectNextAt = 0, 0, 0
+				v.ReconnectExhausted, v.ReconnectLastError = false, ""
+			})
+			continue
+		}
+		if st.PauseReason != "paused_by_proxy" || st.ReconnectExhausted || (st.ReconnectNextAt > 0 && st.ReconnectNextAt > now) {
+			continue
+		}
+		st.ReconnectAttempt++
+		st.ReconnectMax = max
+		t.updateLifecycle(func(v *taskLifecycleState) {
+			v.ReconnectAttempt = st.ReconnectAttempt
+			v.ReconnectMax = max
+			v.ReconnectNextAt = 0
+		})
+		res := probeEgress(context.Background(), s.m.GlobalProxy(), "", defaultEgressProbes, 10*time.Second)
+		if res.OK {
+			s.proxyReconnectSuccess()
+			return
+		}
+		err := res.Error
+		exhausted := st.ReconnectAttempt >= max
+		next := time.Now().Add(interval).Unix()
+		t.updateLifecycle(func(v *taskLifecycleState) {
+			v.ReconnectNextAt = next
+			v.ReconnectExhausted = exhausted
+			v.ReconnectLastError = err
+		})
+		log.Printf("[proxy] reconnect attempt %d/%d failed: %s", st.ReconnectAttempt, max, err)
+	}
+}
+
+func (s *Server) proxyReconnectSuccess() {
+	max, interval := s.proxyReconnectConfig()
+	s.scheduleMu.Lock()
+	defer s.scheduleMu.Unlock()
+	for _, t := range s.m.List() {
+		if st := t.lifecycleSnapshot(); st.PauseReason != "paused_by_proxy" {
+			continue
+		}
+		if s.m.pg != nil {
+			inside, scheduleErr := s.proxyResumeInsideSchedule(t.ID, time.Now())
+			if scheduleErr != nil {
+				log.Printf("[proxy] schedule gate for task %s: %v", t.ID, scheduleErr)
+				continue
+			}
+			if !inside {
+				if err := s.m.ConvertProxyPauseToSchedule(t.ID); err != nil {
+					log.Printf("[proxy] schedule pause task %s: %v", t.ID, err)
+				}
+				continue
+			}
+		}
+		var (
+			queued bool
+			err    error
+		)
+		if s.engine != nil {
+			queued, err = s.admitTaskWhenScheduledWithReason(t, "resume", true, false, "paused_by_proxy", true)
+		} else {
+			err = s.m.ClearPauseReason(t.ID)
+		}
+		if err != nil {
+			// Admission re-checks the persisted reason under concMu. A user pause,
+			// terminal transition, delete, or archive therefore cannot be revived;
+			// only keep probing if the proxy pause is still eligible.
+			st := t.lifecycleSnapshot()
+			if st.PauseReason == "paused_by_proxy" && !isTerminalStatus(st.Status) {
+				t.updateLifecycle(func(v *taskLifecycleState) {
+					v.ReconnectMax = max
+					v.ReconnectNextAt = time.Now().Add(interval).Unix()
+				})
+			}
+			log.Printf("[proxy] resume task %s: %v", t.ID, err)
+			continue
+		}
+		if s.engine != nil && !queued {
+			s.engine.Resume(t)
+		}
+		t.updateLifecycle(func(v *taskLifecycleState) {
+			v.ReconnectAttempt = 0
+			v.ReconnectMax = 0
+			v.ReconnectNextAt = 0
+			v.ReconnectExhausted = false
+			v.ReconnectLastError = ""
+		})
+	}
+}
+
+func (s *Server) testTaskProxy(w http.ResponseWriter, r *http.Request) {
+	id := r.PathValue("id")
+	t, ok := s.m.Task(id)
+	if !ok {
+		writeErr(w, 404, "task not found")
+		return
+	}
+	proxy := strings.TrimSpace(s.m.GlobalProxy())
+	if proxy == "" {
+		writeJSON(w, 200, map[string]any{"ok": true, "direct": true})
+		s.proxyReconnectSuccess()
+		return
+	}
+	res := probeEgress(r.Context(), proxy, "", defaultEgressProbes, 10*time.Second)
+	if res.OK {
+		s.proxyReconnectSuccess()
+	}
+	if !res.OK && res.Error == "" {
+		res.Error = "代理探测失败"
+	}
+	writeJSON(w, 200, map[string]any{"ok": res.OK, "error": res.Error, "task_id": t.ID})
+}
+
+func (s *Server) setProxyReconnectSettings(max, interval *int) error {
+	if max != nil {
+		if *max < 1 || *max > 1000 {
+			return errors.New("proxy reconnect max must be 1..1000")
+		}
+		if err := s.m.pg.SetSetting(settingProxyReconnectMax, fmt.Sprint(*max)); err != nil {
+			return err
+		}
+	}
+	if interval != nil {
+		if *interval < 1 || *interval > 86400 {
+			return errors.New("proxy reconnect interval must be 1..86400 seconds")
+		}
+		if err := s.m.pg.SetSetting(settingProxyReconnectInterval, fmt.Sprint(*interval)); err != nil {
+			return err
+		}
+	}
+	return nil
+}

server/proxy_autoresume_test.go

@@ -0,0 +1,194 @@
+package server
+
+import (
+	"context"
+	"errors"
+	"testing"
+	"time"
+
+	"github.com/Autumn-27/artex/agent"
+)
+
+func proxyTestManager(t *testing.T) *Manager {
+	t.Helper()
+	m, err := NewManager(t.TempDir(), "")
+	if err != nil {
+		t.Skipf("postgres unavailable (%v) - skipping", err)
+	}
+	t.Cleanup(func() { _ = m.Close() })
+	return m
+}
+
+func TestProxyFailureClassification(t *testing.T) {
+	for _, tc := range []struct {
+		name string
+		err  error
+		want bool
+	}{
+		{"proxy connect", errors.New("proxyconnect tcp: connection refused"), true},
+		{"reset", errors.New("dial tcp: connection reset by peer"), false},
+		{"eof", errors.New("unexpected EOF"), false},
+		{"quota", errors.New("status 429: insufficient_quota"), false},
+		{"auth", errors.New("status 401: invalid api key"), false},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			if got := isProxyFailure(tc.err); got != tc.want {
+				t.Fatalf("isProxyFailure(%v)=%v, want %v", tc.err, got, tc.want)
+			}
+		})
+	}
+}
+
+func TestGlobalProxyTrafficFailureClassification(t *testing.T) {
+	if !isGlobalProxyDialFailure(errors.New("dial tcp 127.0.0.1:8080: connect: connection refused")) {
+		t.Fatal("upstream dial refusal should be eligible for probe confirmation")
+	}
+	if isGlobalProxyDialFailure(errors.New("status 404: target not found")) {
+		t.Fatal("ordinary target failure must not be treated as proxy failure")
+	}
+}
+
+func TestProxyPauseCauseIsDistinct(t *testing.T) {
+	ctx, cancel := context.WithCancelCause(context.Background())
+	cancel(agent.AbortPausedByProxy)
+	code, short, _, ok := agent.AbortReason(ctx)
+	if !ok || code != "paused_by_proxy" || short == "" {
+		t.Fatalf("proxy cause = %q/%q/%v", code, short, ok)
+	}
+}
+
+func TestProxyReconnectReasonGate(t *testing.T) {
+	for _, tc := range []struct {
+		reason string
+		want   bool
+	}{
+		{"paused_by_proxy", true},
+		{"paused_by_user", false},
+		{"paused_by_schedule", false},
+		{"paused_by_orchestrator", false},
+	} {
+		got := tc.reason == "paused_by_proxy"
+		if got != tc.want {
+			t.Errorf("reason %q gate=%v, want %v", tc.reason, got, tc.want)
+		}
+	}
+}
+
+func TestProxyAdmissionRechecksReasonAfterSnapshot(t *testing.T) {
+	m := proxyTestManager(t)
+	task, err := m.CreateTask("proxy gate", "goal", nil, 0, 0)
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer func() { _, _ = m.DeleteTask(task.ID, DeleteTaskOptions{}) }()
+	if err := m.ApplyProxyPause(task.ID); err != nil {
+		t.Fatal(err)
+	}
+	s := newAdmissionTestServer(m, nil)
+	s.engine.Pause(task.ID, agent.AbortPausedByProxy)
+	// Simulate a user re-pause winning after proxyReconnectSuccess took its
+	// snapshot but before the admission lock was acquired.
+	if _, err := m.pg.Exec(`UPDATE tasks SET pause_reason='paused_by_user' WHERE id=$1`, task.ID); err != nil {
+		t.Fatal(err)
+	}
+	task.updateLifecycle(func(v *taskLifecycleState) { v.PauseReason = "paused_by_user" })
+	if _, err := s.admitTaskWhenScheduledWithReason(task, "resume", true, false, "paused_by_proxy", true); err == nil {
+		t.Fatal("proxy admission passed after user changed pause reason")
+	}
+	if !s.engine.IsPaused(task.ID) {
+		t.Fatal("user re-pause barrier was cleared")
+	}
+}
+
+func TestProxyQueuedAdmissionDoesNotResumeEngine(t *testing.T) {
+	m := proxyTestManager(t)
+	if err := m.SetConcurrency(true, 1); err != nil {
+		t.Fatal(err)
+	}
+	holder, err := m.CreateTask("holder", "goal", nil, 0, 0)
+	if err != nil {
+		t.Fatal(err)
+	}
+	proxyTask, err := m.CreateTask("queued proxy", "goal", nil, 0, 0)
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer func() {
+		_, _ = m.DeleteTask(holder.ID, DeleteTaskOptions{})
+		_, _ = m.DeleteTask(proxyTask.ID, DeleteTaskOptions{})
+	}()
+	if err := m.ApplyProxyPause(proxyTask.ID); err != nil {
+		t.Fatal(err)
+	}
+	s := newAdmissionTestServer(m, nil)
+	s.engine.started.Store(holder.ID, true)
+	s.engine.Pause(proxyTask.ID, agent.AbortPausedByProxy)
+	queued, err := s.admitTaskWhenScheduledWithReason(proxyTask, "resume", true, false, "paused_by_proxy", true)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if !queued {
+		t.Fatal("proxy task was expected to queue behind the running holder")
+	}
+	if !s.engine.IsPaused(proxyTask.ID) {
+		t.Fatal("queued proxy task was resumed")
+	}
+}
+
+func TestTerminalStatusClearsProxyReconnectState(t *testing.T) {
+	m := proxyTestManager(t)
+	task, err := m.CreateTask("terminal proxy", "goal", nil, 0, 0)
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer func() { _, _ = m.DeleteTask(task.ID, DeleteTaskOptions{}) }()
+	if err := m.ApplyProxyPause(task.ID); err != nil {
+		t.Fatal(err)
+	}
+	task.updateLifecycle(func(v *taskLifecycleState) {
+		v.ReconnectAttempt, v.ReconnectMax, v.ReconnectNextAt = 2, 5, 123
+		v.ReconnectExhausted, v.ReconnectLastError = true, "proxy down"
+	})
+	if err := m.SetTaskStatus(task.ID, "timeout"); err != nil {
+		t.Fatal(err)
+	}
+	stored, err := m.pg.GetTask(mustTaskID(t, task.ID))
+	if err != nil {
+		t.Fatal(err)
+	}
+	if stored.PauseReason != "" {
+		t.Fatalf("terminal row retained pause reason %q", stored.PauseReason)
+	}
+	st := task.lifecycleSnapshot()
+	if st.PauseReason != "" || st.ReconnectAttempt != 0 || st.ReconnectNextAt != 0 || st.ReconnectExhausted {
+		t.Fatalf("terminal lifecycle retained reconnect state: %+v", st)
+	}
+}
+
+func TestProxyAdmissionPreservesCalendarOverride(t *testing.T) {
+	m := proxyTestManager(t)
+	task, err := m.CreateTask("calendar proxy", "goal", nil, 0, 0)
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer func() { _, _ = m.DeleteTask(task.ID, DeleteTaskOptions{}) }()
+	if _, err := m.pg.Exec(`UPDATE tasks SET schedule_paused=true, schedule_manual_until='2099-01-01 00:00:00+00' WHERE id=$1`, task.ID); err != nil {
+		t.Fatal(err)
+	}
+	if err := m.ApplyProxyPause(task.ID); err != nil {
+		t.Fatal(err)
+	}
+	s := newAdmissionTestServer(m, nil)
+	s.engine.Pause(task.ID, agent.AbortPausedByProxy)
+	if queued, err := s.admitTaskWhenScheduledWithReason(task, "resume", true, false, "paused_by_proxy", true); err != nil || queued {
+		t.Fatalf("proxy admission queued=%v err=%v", queued, err)
+	}
+	var schedulePaused bool
+	var manualUntil time.Time
+	if err := m.pg.QueryRow(`SELECT schedule_paused, schedule_manual_until FROM tasks WHERE id=$1`, task.ID).Scan(&schedulePaused, &manualUntil); err != nil {
+		t.Fatal(err)
+	}
+	if !schedulePaused || !manualUntil.Equal(time.Date(2099, 1, 1, 0, 0, 0, 0, time.UTC)) {
+		t.Fatalf("proxy admission changed calendar state: paused=%v until=%v", schedulePaused, manualUntil)
+	}
+}

server/server.go

@@ -128,6 +128,7 @@ type Server struct {
 	archiveWake    chan struct{}
 	archiveWG      sync.WaitGroup
 	credentialWake chan struct{}
+	proxyReconnect proxyReconnectSupervisor
 	side           *sideQuestionState
 }
 
@@ -163,6 +164,7 @@ func New(ctx context.Context, m *Manager, skillDir string, dataDir string, keyDi
 		profChatAgents: map[int64]*agent.ChatAgent{},
 		provByProfile:  map[int64]*provEntry{}, llmHealth: newLLMHealthRegistry(m.pg),
 		taskAgents: map[string]*taskAgentBundle{}, archiveWake: make(chan struct{}, 1), credentialWake: make(chan struct{}, 1)}
+	m.SetTaskControlMutex(&s.concMu)
 	s.initSideQuestions()
 	if m.pg != nil {
 		if err := s.basicAuth.load(m.pg); err != nil {
@@ -192,6 +194,10 @@ func New(ctx context.Context, m *Manager, skillDir string, dataDir string, keyDi
 		defer s.cfgMu.Unlock()
 		return s.llmOn
 	})
+	s.engine.SetProxyFailureHandler(s.handleProxyFailure)
+	if m.traffic != nil {
+		m.traffic.SetUpstreamFailureHandler(s.handleProxyTrafficFailure)
+	}
 	// Wire DB-stored prompt templates into the agents (新版方案 §3.3 / §5a). With no
 	// override row, agents keep their built-in defaults — behavior is unchanged.
 	if m.pg != nil {
@@ -280,6 +286,7 @@ func New(ctx context.Context, m *Manager, skillDir string, dataDir string, keyDi
 		log.Printf("[engine] no LLM provider configured — engine idle until set via /api/llm or env")
 	}
 	s.restoreTaskRuntimes()
+	s.ensureProxySupervisor()
 	go s.reconcileConcurrency()
 	s.startTaskArchiveWorker()
 	if m.pg != nil {
@@ -927,6 +934,7 @@ func (s *Server) Handler() http.Handler {
 	mux.HandleFunc("POST /api/settings/web-search/test", s.testWebSearch)
 	// 出口代理连通性测试：发真实请求验证 global_proxy 是否可用，并回显出口 IP。
 	mux.HandleFunc("POST /api/settings/proxy/test", s.testEgressProxy)
+	mux.HandleFunc("POST /api/tasks/{id}/proxy/test", s.testTaskProxy)
 	mux.HandleFunc("GET /api/report", s.getReport)
 	mux.HandleFunc("GET /api/chat/mentions", s.searchChatMentions)
 	mux.HandleFunc("POST /api/chat", s.chat)
@@ -2332,7 +2340,10 @@ func (s *Server) findingsExport(w http.ResponseWriter, r *http.Request) {
 	stamp := now.Format("20060102-150405")
 	setDownload := func(contentType, filename string) {
 		w.Header().Set("Content-Type", contentType)
-		w.Header().Set("Content-Disposition", fmt.Sprintf(`attachment; filename="%s"`, filename))
+		// Keep an ASCII fallback for older clients and provide the real UTF-8
+		// name through RFC 5987. Browsers otherwise commonly decode Chinese
+		// names using the platform legacy code page.
+		w.Header().Set("Content-Disposition", fmt.Sprintf(`attachment; filename="%s"; filename*=UTF-8''%s`, sanitizeFilename(filename), url.PathEscape(filename)))
 	}
 
 	switch format {
@@ -3537,6 +3548,8 @@ func (s *Server) settingsPayload() map[string]any {
 		"tavily_key_set":                       strings.TrimSpace(tavilyKey) != "",
 		"web_search_proxy":                     proxy,             // 独立出口代理(http/https/socks5)，空=直连
 		"global_proxy":                         s.m.GlobalProxy(), // 全局出口代理(http/https/socks5)，所有目标流量走它，空=直连
+		"proxy_reconnect_max_attempts":         func() int { n, _ := s.proxyReconnectConfig(); return n }(),
+		"proxy_reconnect_interval_seconds":     func() int { _, d := s.proxyReconnectConfig(); return int(d / time.Second) }(),
 		"show_egress_ip":                       s.m.pg.GetBool(settingShowEgressIP, false),
 		"python_interpreter":                   strings.TrimSpace(pyStored), // 用户/自动设的值(空=用运行时检测)
 		"workers":                              s.m.Workers(),               // 并发工作 agent 数(默认3)；对之后启动的任务生效
@@ -3613,16 +3626,18 @@ func (s *Server) putSettings(w http.ResponseWriter, r *http.Request) {
 		LLMRecord           *bool `json:"llm_record"` // LLM 录制开关（默认关）；即时生效，无需重建 agent
 		// Web search. WebSearchEnabled/Backend toggle the tool + backend; BraveKey/TavilyKey
 		// are optional — omit (null) to leave a stored key untouched, send "" to clear.
-		WebSearchEnabled *bool         `json:"web_search_enabled"`
-		WebSearchBackend *string       `json:"web_search_backend"`
-		BraveKey         *string       `json:"brave_search_api_key"`
-		TavilyKey        *string       `json:"tavily_search_api_key"`
-		WebSearchProxy   *string       `json:"web_search_proxy"`   // 独立出口代理(http/https/socks5)；null=不改，""=清空
-		GlobalProxy      *string       `json:"global_proxy"`       // 全局出口代理(http/https/socks5)；null=不改，""=清空(直连)
-		ShowEgressIP     *bool         `json:"show_egress_ip"`     // 在顶栏显示直连/全局代理出口 IP；默认关
-		PythonInterp     *string       `json:"python_interpreter"` // 自定义脚本工具的 python 解释器路径
-		Workers          *int          `json:"workers"`            // 并发工作 agent 数(>0)；对之后启动的任务生效
-		WorkerPools      *[]WorkerPool `json:"worker_pools"`
+		WebSearchEnabled       *bool         `json:"web_search_enabled"`
+		WebSearchBackend       *string       `json:"web_search_backend"`
+		BraveKey               *string       `json:"brave_search_api_key"`
+		TavilyKey              *string       `json:"tavily_search_api_key"`
+		WebSearchProxy         *string       `json:"web_search_proxy"` // 独立出口代理(http/https/socks5)；null=不改，""=清空
+		GlobalProxy            *string       `json:"global_proxy"`     // 全局出口代理(http/https/socks5)；null=不改，""=清空(直连)
+		ProxyReconnectMax      *int          `json:"proxy_reconnect_max_attempts"`
+		ProxyReconnectInterval *int          `json:"proxy_reconnect_interval_seconds"`
+		ShowEgressIP           *bool         `json:"show_egress_ip"`     // 在顶栏显示直连/全局代理出口 IP；默认关
+		PythonInterp           *string       `json:"python_interpreter"` // 自定义脚本工具的 python 解释器路径
+		Workers                *int          `json:"workers"`            // 并发工作 agent 数(>0)；对之后启动的任务生效
+		WorkerPools            *[]WorkerPool `json:"worker_pools"`
 		// 任务并发上限:同时「运行中」的任务数上限。关闭=不限;开启后新建任务超限则排队,有空位自动启动。
 		ConcurrencyEnabled *bool `json:"task_concurrency_enabled"`
 		ConcurrencyLimit   *int  `json:"task_concurrency_limit"`
@@ -3839,6 +3854,15 @@ func (s *Server) putSettings(w http.ResponseWriter, r *http.Request) {
 			return
 		}
 		changed = true // capture-off egress is baked into agents at build time → rebuild
+		if strings.TrimSpace(*req.GlobalProxy) == "" {
+			s.proxyReconnectSuccess()
+		}
+	}
+	if req.ProxyReconnectMax != nil || req.ProxyReconnectInterval != nil {
+		if err := s.setProxyReconnectSettings(req.ProxyReconnectMax, req.ProxyReconnectInterval); err != nil {
+			writeErr(w, 400, err.Error())
+			return
+		}
 	}
 	if req.ShowEgressIP != nil {
 		if err := s.m.pg.SetBool(settingShowEgressIP, *req.ShowEgressIP); err != nil {

server/server_mgmt.go

@@ -101,7 +101,14 @@ func (s *Server) beginTaskDelete(taskID string) bool {
 	if s.engine.IsDeleting(taskID) {
 		return false
 	}
-	return s.engine.BeginDelete(taskID)
+	ok := s.engine.BeginDelete(taskID)
+	if ok {
+		// A user delete/archive action is an explicit override of proxy recovery.
+		if s.m != nil && s.m.pg != nil {
+			_ = s.m.ClearPauseReason(taskID)
+		}
+	}
+	return ok
 }
 
 // abortTaskDelete restores the execution barrier from the committed task row,

server/task_control.go

@@ -116,9 +116,17 @@ func (s *Server) applyTaskControlWithCause(t *Task, action string, pauseCause er
 		}
 		if lifecycle.Paused {
 			if pauseCause != agent.AbortPausedBySchedule {
-				if _, err := s.m.pg.Exec(`UPDATE tasks SET schedule_paused=false WHERE id=$1`, t.ID); err != nil {
+				if _, err := s.m.pg.Exec(`UPDATE tasks SET schedule_paused=false, pause_reason='paused_by_user' WHERE id=$1`, t.ID); err != nil {
 					return out, err
 				}
+				t.updateLifecycle(func(state *taskLifecycleState) {
+					state.PauseReason = "paused_by_user"
+					state.ReconnectAttempt = 0
+					state.ReconnectMax = 0
+					state.ReconnectNextAt = 0
+					state.ReconnectExhausted = false
+					state.ReconnectLastError = ""
+				})
 				out.Paused = true
 				out.Status = "paused"
 				return out, nil
@@ -131,8 +139,22 @@ func (s *Server) applyTaskControlWithCause(t *Task, action string, pauseCause er
 			pauseCause = agent.AbortPausedByUser
 		}
 		s.engine.Pause(t.ID, pauseCause)
-		if err := s.m.ApplyTaskPauseOrigin(t.ID, pauseCause == agent.AbortPausedBySchedule); err != nil {
-			if !wasEnginePaused && !wasQueued {
+		reason := "paused_by_user"
+		if pauseCause == agent.AbortPausedBySchedule {
+			reason = "paused_by_schedule"
+		}
+		if pauseCause == agent.AbortPausedByOrchestrator {
+			reason = "paused_by_orchestrator"
+		}
+		if pauseCause == agent.AbortPausedByProxy {
+			reason = "paused_by_proxy"
+		}
+		if err := s.m.ApplyTaskPauseOriginReason(t.ID, reason, pauseCause == agent.AbortPausedBySchedule); err != nil {
+			// Do not compensate over a proxy pause that may have committed while
+			// this control request was failing. The persisted reason is the source
+			// of truth for which Engine barrier must remain installed.
+			stillProxyPaused := t.lifecycleSnapshot().PauseReason == "paused_by_proxy"
+			if !stillProxyPaused && !wasEnginePaused && !wasQueued {
 				s.engine.Resume(t)
 			}
 			return out, err

traffic/traffic.go

@@ -169,7 +169,9 @@ type Traffic struct {
 	// through (nil = dial targets directly). Both the intercepted and the
 	// transparent-passthrough paths honor it (go-mitmproxy's getUpstreamConn),
 	// so no host escapes it. Hot-swappable at runtime via SetUpstreamProxy.
-	upstream atomic.Pointer[url.URL]
+	upstream        atomic.Pointer[url.URL]
+	upstreamMu      sync.RWMutex
+	upstreamFailure func(error)
 }
 
 // Open initializes the traffic tree, blob store and SQLite index under dir.
@@ -327,6 +329,27 @@ func (t *Traffic) SetUpstreamProxy(raw string) error {
 	return nil
 }
 
+// SetUpstreamFailureHandler observes failures from the configured global
+// upstream proxy. The callback is intentionally best-effort and must not block
+// the MITM request path.
+func (t *Traffic) SetUpstreamFailureHandler(fn func(error)) {
+	t.upstreamMu.Lock()
+	t.upstreamFailure = fn
+	t.upstreamMu.Unlock()
+}
+
+func (t *Traffic) reportUpstreamFailure(err error) {
+	if err == nil || t.upstream.Load() == nil {
+		return
+	}
+	t.upstreamMu.RLock()
+	fn := t.upstreamFailure
+	t.upstreamMu.RUnlock()
+	if fn != nil {
+		go fn(err)
+	}
+}
+
 // ValidateProxyURL parses and checks a proxy URL (http/https/socks5, optional
 // user:pass), returning the parsed URL. Exposed so callers can validate a global
 // proxy before persisting it even when the traffic proxy itself is disabled.
@@ -400,7 +423,14 @@ func (s *sink) Response(f *mproxy.Flow) {
 // the failure looks proxy/protocol-caused (h2 quirks, HEAD-with-body, protocol
 // errors) — not a plain target-unreachable error — we flag the host for
 // transparent passthrough so future requests to it succeed instead of dying.
-func (s *sink) RequestError(f *mproxy.Flow, err error) { s.t.maybePassthrough(f, err) }
+func (s *sink) RequestError(f *mproxy.Flow, err error) {
+	s.t.reportUpstreamFailure(err)
+	s.t.maybePassthrough(f, err)
+}
+
+func (s *sink) HTTPConnectError(_ *mproxy.Flow, err error) {
+	s.t.reportUpstreamFailure(err)
+}
 
 // maybePassthrough marks a host to be tunneled transparently on the next
 // connection, but only for errors the proxy itself caused — a target that is

web/src/app/(main)/function/findings/detail/page.tsx

@@ -424,6 +424,7 @@ function FindingDetailInner() {
         agentLabel="报告撰写 Agent"
         promptPlaceholder="描述要如何调整漏洞报告…"
         modal
+        initialConversationId={finding.report_conversation_id}
         open={reportChatOpen}
         onOpenChange={setReportChatOpen}
       />

web/src/app/(main)/function/tasks/detail/page.tsx

@@ -419,6 +419,14 @@ function TaskDetailInner() {
           </Button>
         </div>
         <p className="truncate text-xs text-muted-foreground">{task.goal}</p>
+        {task.paused && task.pause_reason && (
+          <div className="flex flex-wrap items-center gap-2 text-xs text-amber-600">
+            <span>暂停原因：{task.pause_reason === "paused_by_proxy" ? "全局代理异常" : task.pause_reason}</span>
+            {task.reconnect && <span>重连 {task.reconnect.attempt}/{task.reconnect.max} {task.reconnect.exhausted ? "已耗尽" : task.reconnect.next_at ? `下次 ${new Date(task.reconnect.next_at * 1000).toLocaleTimeString()}` : "探测中"}</span>}
+            {task.reconnect?.last_error && <span className="max-w-full truncate" title={task.reconnect.last_error}>原因：{task.reconnect.last_error}</span>}
+            {task.pause_reason === "paused_by_proxy" && <Button size="sm" variant="outline" onClick={() => api.testTaskProxy(task.id).then((r) => r.ok ? (toast.success("代理已恢复，任务将自动继续"), load()) : toast.error(r.error || "代理探测失败")).catch((e) => toast.error((e as Error).message))}>测试代理</Button>}
+          </div>
+        )}
         {task.remark ? (
           <p className="truncate text-xs text-muted-foreground" title={task.remark}>
             备注：{task.remark}

web/src/app/(main)/system/settings/page.tsx

@@ -55,6 +55,8 @@ export default function SystemSettingsPage() {
   const [savingProxy, setSavingProxy] = React.useState(false);
   const [globalProxyInput, setGlobalProxyInput] = React.useState("");
   const [savingGlobalProxy, setSavingGlobalProxy] = React.useState(false);
+  const [proxyReconnectMax, setProxyReconnectMax] = React.useState("5");
+  const [proxyReconnectInterval, setProxyReconnectInterval] = React.useState("30");
   const [savingEgressVisibility, setSavingEgressVisibility] = React.useState(false);
   // 出口探测：测一次真实请求，拿到"直连 / 经全局代理"两条出口 IP。
   const [egressTesting, setEgressTesting] = React.useState(false);
@@ -111,6 +113,8 @@ export default function SystemSettingsPage() {
     setProxyInput(s.web_search_proxy ?? "");
     const savedProxy = s.global_proxy ?? "";
     setGlobalProxyInput(savedProxy);
+    setProxyReconnectMax(String(s.proxy_reconnect_max_attempts ?? 5));
+    setProxyReconnectInterval(String(s.proxy_reconnect_interval_seconds ?? 30));
     setPyInterp(s.python_interpreter ?? "");
     setWorkers(String(s.workers ?? 3));
     setWorkerPools(s.worker_pools ?? []);
@@ -428,6 +432,12 @@ export default function SystemSettingsPage() {
       .finally(() => setSavingGlobalProxy(false));
   };
 
+  const saveProxyReconnect = () => {
+    api.setSettings({ proxy_reconnect_max_attempts: Number(proxyReconnectMax), proxy_reconnect_interval_seconds: Number(proxyReconnectInterval) })
+      .then((s) => { apply(s); toast.success("已保存代理重连策略"); })
+      .catch((e) => toast.error("保存失败：" + (e as Error).message));
+  };
+
   // 测表单里当前的值，不要求先保存；后端返回直连 / 经代理两条。
   const testEgress = () => {
     const seq = ++egressSeq.current;
@@ -751,12 +761,18 @@ export default function SystemSettingsPage() {
                   </Button>
                   {/* 测表单里当前的值，不要求先保存。 */}
                   <Button type="button" variant="outline" onClick={testEgress} disabled={!loaded || egressTesting}>
-                    {egressTesting ? "测试中…" : "测试"}
+                    {egressTesting ? "测试中…" : "测试代理"}
                   </Button>
                 </div>
                 <p className="text-muted-foreground text-xs">
                   {globalProxyInput.trim() ? "已配置 · 所有目标流量经此代理出网" : "未配置 · 目标流量直连出网"}
                 </p>
+                <div className="grid grid-cols-1 gap-2 sm:grid-cols-3 items-end">
+                  <div><Label htmlFor="proxy-reconnect-max">最大重连次数</Label><Input id="proxy-reconnect-max" type="number" min={1} value={proxyReconnectMax} onChange={(e) => setProxyReconnectMax(e.target.value)} /></div>
+                  <div><Label htmlFor="proxy-reconnect-interval">重连间隔（秒）</Label><Input id="proxy-reconnect-interval" type="number" min={1} value={proxyReconnectInterval} onChange={(e) => setProxyReconnectInterval(e.target.value)} /></div>
+                  <Button type="button" variant="outline" onClick={saveProxyReconnect}>保存重连策略</Button>
+                </div>
+                <p className="text-muted-foreground text-xs">代理异常暂停的任务会按此策略探测；次数耗尽后保持暂停，可手动恢复。</p>
                 <div className="flex items-center justify-between gap-4 rounded-md border px-3 py-2">
                   <div className="min-w-0">
                     <Label htmlFor="show-egress-ip" className="text-sm font-normal">

web/src/components/finding-auto-chat.tsx

@@ -26,6 +26,7 @@ export function FindingAutoChat({
   agentLabel = "Auto Agent",
   promptPlaceholder = "询问这个漏洞…",
   modal = false,
+  initialConversationId,
   open,
   onOpenChange,
 }: {
@@ -35,6 +36,7 @@ export function FindingAutoChat({
   agentLabel?: string;
   promptPlaceholder?: string;
   modal?: boolean;
+  initialConversationId?: number;
   open: boolean;
   onOpenChange: (open: boolean) => void;
 }) {
@@ -49,6 +51,7 @@ export function FindingAutoChat({
   const [historyOpen, setHistoryOpen] = React.useState(false);
   const [history, setHistory] = React.useState<Conversation[]>([]);
   const cursor = React.useRef(0);
+  const chatViewportRef = React.useRef<HTMLDivElement | null>(null);
   const safeTitle = findingTitle.replace(/[\[\]\r\n]/g, " ").trim().slice(0, 100);
   const storageKey = `artex.finding.${agentKey}-conversation.${findingId}`;
 
@@ -70,7 +73,11 @@ export function FindingAutoChat({
 
   React.useEffect(() => {
     if (!open || conversation || manualEmpty) return;
-    const storedID = Number(window.localStorage.getItem(storageKey));
+    if (initialConversationId && initialConversationId > 0) {
+      setConversation({ id: initialConversationId, agent_key: agentKey, title: `漏洞 #${findingId} · ${safeTitle}`, created_at: "", updated_at: "" });
+      return;
+    }
+    const storedID = agentKey === "auto" ? Number(window.localStorage.getItem(storageKey)) : 0;
     if (Number.isSafeInteger(storedID) && storedID > 0) {
       setConversation({
         id: storedID,
@@ -82,7 +89,7 @@ export function FindingAutoChat({
       return;
     }
     void createConversation();
-  }, [conversation, createConversation, manualEmpty, open, safeTitle, storageKey]);
+  }, [conversation, createConversation, initialConversationId, manualEmpty, open, safeTitle, storageKey]);
 
   React.useEffect(() => {
     if (!historyOpen && !running) return;
@@ -130,6 +137,18 @@ export function FindingAutoChat({
     };
   }, [conversation, open, storageKey]);
 
+  React.useEffect(() => {
+    if (!open || !conversation || loading || !messages.length) return;
+    const viewport = chatViewportRef.current;
+    if (!viewport) return;
+    const key = `artex.conversation-scroll.${conversation.id}`;
+    const saved = window.sessionStorage.getItem(key);
+    requestAnimationFrame(() => {
+      if (saved != null) viewport.scrollTop = Number(saved) || 0;
+      else viewport.scrollTop = viewport.scrollHeight;
+    });
+  }, [conversation, loading, messages.length, open]);
+
   React.useEffect(() => {
     if (!open || !conversation || !running) return;
     let live = true;
@@ -265,7 +284,13 @@ export function FindingAutoChat({
             <XIcon />
           </Button>
         </header>
-        <div className="min-h-0 min-w-0 flex-1 overflow-x-hidden overflow-y-auto p-3">
+        <div
+          ref={chatViewportRef}
+          onScroll={(event) => {
+            if (conversation) window.sessionStorage.setItem(`artex.conversation-scroll.${conversation.id}`, String(event.currentTarget.scrollTop));
+          }}
+          className="min-h-0 min-w-0 flex-1 overflow-x-hidden overflow-y-auto p-3"
+        >
           <div className="w-full min-w-0 max-w-full">
             {loading ? (
               <div className="flex items-center justify-center py-10 text-muted-foreground text-sm">

web/src/components/markdown.tsx

@@ -18,6 +18,8 @@ function AuthenticatedImage({
   const [captureTimestamp, setCaptureTimestamp] = React.useState<string | null>(null);
   const [zoomed, setZoomed] = React.useState(false);
   const [zoomScale, setZoomScale] = React.useState(1);
+  const [pan, setPan] = React.useState({ x: 0, y: 0 });
+  const dragRef = React.useRef<{ x: number; y: number; panX: number; panY: number } | null>(null);
   const imageSrc = typeof src === "string" ? src : undefined;
 
   React.useEffect(() => {
@@ -29,6 +31,7 @@ function AuthenticatedImage({
       setCaptureTimestamp(null);
       setZoomed(false);
       setZoomScale(1);
+      setPan({ x: 0, y: 0 });
       return () => {
         cancelled = true;
       };
@@ -75,31 +78,55 @@ function AuthenticatedImage({
         onClick={() => renderedSrc && setZoomed(true)}
         aria-label="放大查看报告截图"
       >
-        <img {...props} src={renderedSrc} alt={alt || "报告截图，点击放大"} />
+        <img {...props} src={renderedSrc} alt={alt || "报告截图，点击放大"} draggable={false} onDragStart={(event) => event.preventDefault()} />
         {timestamp}
       </button>
       <Dialog
         open={zoomed}
         onOpenChange={(next) => {
           setZoomed(next);
-          if (!next) setZoomScale(1);
+          if (!next) {
+            setZoomScale(1);
+            setPan({ x: 0, y: 0 });
+          }
         }}
       >
         <DialogContent
           showCloseButton
           className="flex max-h-[96dvh] max-w-[96vw] items-center justify-center overflow-hidden bg-black/85 p-3 backdrop-blur-sm sm:max-w-[96vw]"
         >
-          <div className="relative flex max-h-[90dvh] max-w-full items-center justify-center overflow-auto">
-            <div className="relative inline-block origin-center transition-transform duration-150" style={{ transform: `scale(${zoomScale})` }}>
+          <div
+            className="relative flex h-[90dvh] w-[90vw] max-w-full items-center justify-center overflow-hidden"
+            onPointerDown={(event) => {
+              if (zoomScale <= 1 || event.button !== 0) return;
+              event.currentTarget.setPointerCapture(event.pointerId);
+              dragRef.current = { x: event.clientX, y: event.clientY, panX: pan.x, panY: pan.y };
+            }}
+            onPointerMove={(event) => {
+              const drag = dragRef.current;
+              if (!drag) return;
+              setPan({ x: drag.panX + event.clientX - drag.x, y: drag.panY + event.clientY - drag.y });
+            }}
+            onPointerUp={(event) => {
+              dragRef.current = null;
+              event.currentTarget.releasePointerCapture?.(event.pointerId);
+            }}
+            onPointerCancel={() => { dragRef.current = null; }}
+            style={{ cursor: zoomScale > 1 ? (dragRef.current ? "grabbing" : "grab") : "default" }}
+          >
+            <div className="relative inline-block origin-center" style={{ transform: `translate(${pan.x}px, ${pan.y}px) scale(${zoomScale})` }}>
               <img
                 src={renderedSrc}
                 alt={alt || "报告截图"}
+                draggable={false}
+                onDragStart={(event) => event.preventDefault()}
                 className="max-h-[90dvh] max-w-[90vw] rounded-md object-contain"
               />
               {timestamp}
             </div>
           </div>
           <div className="absolute top-3 left-3 z-10 flex items-center gap-1 rounded-lg bg-black/60 p-1 backdrop-blur-sm">
+            <span className="px-2 text-xs font-medium text-white tabular-nums">{Math.round(zoomScale * 100)}%</span>
             <Button
               variant="ghost"
               size="icon-sm"

web/src/lib/api.ts

@@ -119,6 +119,24 @@ import type {
   SystemMetrics,
 } from "@/lib/types";
 
+/**
+ * Read a download name from Content-Disposition. RFC 5987's filename* is
+ * preferred because it preserves UTF-8 names; older filename remains the
+ * fallback for clients that do not understand it.
+ */
+export function downloadFilename(contentDisposition: string, fallback: string): string {
+  const encoded = contentDisposition.match(/(?:^|;)\s*filename\*\s*=\s*UTF-8''([^;]+)/i)?.[1];
+  if (encoded) {
+    try {
+      return decodeURIComponent(encoded.replace(/^"|"$/g, ""));
+    } catch {
+      // Fall through to the legacy token when a malformed header is received.
+    }
+  }
+  const legacy = contentDisposition.match(/(?:^|;)\s*filename\s*=\s*(?:"([^"]*)"|([^;\s]+))/i);
+  return (legacy?.[1] ?? legacy?.[2] ?? fallback).trim() || fallback;
+}
+
 function getToken(): string | null {
   if (typeof window === "undefined") return null;
   return localStorage.getItem("artex_token");
@@ -534,6 +552,7 @@ export const api = {
   },
   workspaceDownload: async (path: string) => {
     let blob: Blob;
+    let contentDisposition = "";
     if (MOCK) {
       blob = new Blob([`（demo）${path} 的下载内容示例。`], { type: "text/plain" });
     } else {
@@ -542,12 +561,13 @@ export const api = {
         headers: { ...(token ? { Authorization: `Bearer ${token}` } : {}) },
       });
       if (!r.ok) throw new Error(`下载失败: ${r.status}`);
+      contentDisposition = r.headers.get("Content-Disposition") ?? "";
       blob = await r.blob();
     }
     const objUrl = URL.createObjectURL(blob);
     const a = document.createElement("a");
     a.href = objUrl;
-    a.download = path.split("/").pop() || "download";
+    a.download = downloadFilename(contentDisposition, path.split("/").pop() || "download");
     document.body.appendChild(a);
     a.click();
     a.remove();
@@ -703,8 +723,7 @@ export const api = {
     const blob = await r.blob();
     // 文件名优先取后端 Content-Disposition,取不到则用默认名。
     const disp = r.headers.get("Content-Disposition") ?? "";
-    const m = disp.match(/filename="?([^"]+)"?/);
-    const filename = m?.[1] ?? `findings-export`;
+    const filename = downloadFilename(disp, `findings-export`);
     const url = URL.createObjectURL(blob);
     const a = document.createElement("a");
     a.href = url;
@@ -973,6 +992,7 @@ export const api = {
   // global_proxy 传空 = 测直连；probe_url 可覆盖默认探测地址（隔离网用）。
   testEgressProxy: (patch: { global_proxy?: string; probe_url?: string }) =>
     post<EgressReport>(`/settings/proxy/test`, patch),
+  testTaskProxy: (taskID: string) => post<{ ok: boolean; error?: string }>(`/tasks/${taskID}/proxy/test`, {}),
 
   // ---- 漏洞 IM 推送 ----
   // 渠道是多实例资源（同一类型可配多个机器人、各有过滤规则），因此独立成组，

web/src/lib/types.ts

@@ -42,6 +42,8 @@ export interface Task {
   active_run_seconds?: number; // completed running segments in seconds
   run_segment_started_unix?: number; // current running segment start; 0 when inactive
   paused?: boolean;
+  pause_reason?: string;
+  reconnect?: { attempt: number; max: number; next_at?: number; exhausted: boolean; last_error?: string };
   queued?: boolean;
   active?: boolean;
   in_flight?: number;
@@ -529,6 +531,7 @@ export interface Finding {
   summary: string;
   evidence: string;
   report?: string; // 详细报告(Markdown);仅详情接口返回,列表为空
+  report_conversation_id?: number;
   intent_id?: string;
   param_id?: string;
   task_id?: string;
@@ -1001,6 +1004,8 @@ export interface Settings {
   // 全局出口代理(http/https/socks5，可带 user:pass)，所有目标流量走它。开启流量捕获时作为
   // MITM 上游；关闭捕获时直接注入 agent 的 bash/WebFetch。空=直连。
   global_proxy?: string;
+  proxy_reconnect_max_attempts?: number;
+  proxy_reconnect_interval_seconds?: number;
   show_egress_ip?: boolean; // 在顶栏显示直连 / 全局代理出口 IP；默认关闭
   python_interpreter?: string; // 自定义脚本工具的 python 解释器路径(空=运行时检测)
   workers?: number; // 并发工作 agent 数(默认3)；对之后启动的任务生效