feat(issues): show handoff note in execution-log trigger text

An assignment-triggered run that carried a handoff note showed the
generic "Initial run" label. Surface the note inline (truncated, like
comment triggers show their text) so the row reads as the handoff.

taskToResponse now populates handoff_note for all callers (dropping the
now-redundant explicit set in ClaimTaskByRuntime); the field is added to
the AgentTask type + zod schema (optional, additive — old clients ignore
it via the loose schema, new clients fall back to "Initial run").

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

packages/core/api/schemas.ts

@@ -477,6 +477,7 @@ const AgentTaskResponseSchema = z.object({
   attempt: z.number().optional(),
   trigger_comment_id: z.string().optional(),
   trigger_summary: z.string().optional(),
+  handoff_note: z.string().optional(),
   kind: z.string().optional(),
   work_dir: z.string().optional(),
   relative_work_dir: z.string().optional(),

packages/core/realtime/use-realtime-sync.ts

@@ -517,11 +517,15 @@ export function useRealtimeSync(
         // open composer's chips (e.g. an agent finishing its run becomes
         // triggerable again mid-typing).
         qc.invalidateQueries({ queryKey: issueKeys.commentTriggerPreviewAll() });
-        // Issue-trigger previews (assign/status/create/batch) are queue-
-        // dependent the same way — the status source's pending dedup means a
-        // task finishing changes "will this start a run", so refresh any open
-        // picker/modal preview (MUL-3375).
-        qc.invalidateQueries({ queryKey: issueKeys.issueTriggerPreviewAll() });
+        // Issue-trigger previews (assign/status/create/batch) are deliberately
+        // NOT invalidated here. Unlike comment triggers, the assign source
+        // (create / assignee change) cancels existing tasks before enqueuing, so
+        // a task event can never change its verdict; only the status source's
+        // pending dedup could, and that preview is advisory — the write path
+        // re-evaluates authoritatively, so a rare stale label is harmless.
+        // Refetching every mounted preview on every workspace task event caused
+        // visible flicker, so the preview now refetches only on input change
+        // (signature), mirroring its query design (MUL-3375).
       },
     };
 

packages/core/types/agent.ts

@@ -196,6 +196,13 @@ export interface AgentTask {
    * or deleted.
    */
   trigger_summary?: string;
+  /**
+   * Handoff instruction the assigner attached when starting this run (MUL-3375).
+   * Present only on assignment-triggered runs that carried a note; the execution
+   * log shows it inline as the trigger reason. Absent (legacy / no note) falls
+   * back to the generic "initial run" label.
+   */
+  handoff_note?: string;
   /**
    * Server-computed source discriminator used by the activity row to label
    * tasks that have no linked issue (so e.g. quick-create tasks render

packages/views/issues/actions/__tests__/use-issue-actions.test.tsx

@@ -150,6 +150,24 @@ describe("useIssueActions", () => {
     expect(mockUpdateMutate).not.toHaveBeenCalled();
   });
 
+  it("assigning an agent to a backlog issue applies directly — backlog never starts a run", () => {
+    const backlogIssue = { ...mockIssue, status: "backlog" } as Issue;
+    const { result } = renderHook(() => useIssueActions(backlogIssue), { wrapper });
+
+    act(() => {
+      result.current.updateField({
+        assignee_type: "agent",
+        assignee_id: "agent-1",
+      });
+    });
+
+    expect(mockUpdateMutate).toHaveBeenCalledWith(
+      { id: "issue-1", assignee_type: "agent", assignee_id: "agent-1" },
+      expect.any(Object),
+    );
+    expect(mockOpenModal).not.toHaveBeenCalled();
+  });
+
   it("assigning a member applies directly without the run-confirm modal", () => {
     const { result } = renderHook(() => useIssueActions(mockIssue), { wrapper });
 

packages/views/issues/actions/use-issue-actions.ts

@@ -57,6 +57,7 @@ export function useIssueActions(issue: Issue | null): UseIssueActionsResult {
   const issueId = issue?.id ?? null;
   const issueIdentifier = issue?.identifier ?? null;
   const issueProjectId = issue?.project_id ?? null;
+  const issueStatus = issue?.status ?? null;
 
   const updateField = useCallback(
     (updates: Partial<UpdateIssueRequest>) => {
@@ -66,9 +67,15 @@ export function useIssueActions(issue: Issue | null): UseIssueActionsResult {
       // which applies the change itself — the four entry points share this one
       // backend-driven flow instead of guessing (MUL-3375). Every other field
       // change (status, priority, member assign, unassign) applies directly.
+      //
+      // Backlog is the parking lot: assigning a backlog issue never starts a run
+      // (server/internal/service/issue_trigger.go), so the modal would only show
+      // an empty "won't start" box with a single Apply button. Apply directly,
+      // matching the batch backlog short-circuit in BatchActionToolbar.
       if (
         (updates.assignee_type === "agent" || updates.assignee_type === "squad") &&
-        updates.assignee_id
+        updates.assignee_id &&
+        issueStatus !== "backlog"
       ) {
         openModal("issue-run-confirm", {
           issueIds: [issueId],
@@ -90,7 +97,7 @@ export function useIssueActions(issue: Issue | null): UseIssueActionsResult {
         },
       );
     },
-    [issueId, updateIssue, openModal, t],
+    [issueId, issueStatus, updateIssue, openModal, t],
   );
 
   const togglePin = useCallback(() => {

packages/views/issues/components/batch-action-toolbar.tsx

@@ -100,13 +100,23 @@ export function BatchActionToolbar({
 
   const handleBatchAssignee = (updates: Partial<UpdateIssueRequest>) => {
     if ((updates.assignee_type === "agent" || updates.assignee_type === "squad") && updates.assignee_id) {
-      openModal("issue-run-confirm", {
-        issueIds: ids,
-        mode: "assign",
-        assigneeType: updates.assignee_type,
-        assigneeId: updates.assignee_id,
-      });
-      return;
+      // Backlog never starts a run on assign (parking lot), so if every selected
+      // issue is in backlog the confirm modal would only render an empty "won't
+      // start" box — apply directly, matching handleBatchStatus's backlog short-
+      // circuit. A mixed selection still routes through the modal: the non-backlog
+      // issues will trigger and need confirmation. An empty intersection (selected
+      // ids not in `issues`) falls through to the modal — safer than skipping.
+      const selected = issues.filter((i) => selectedIds.has(i.id));
+      const allBacklog = selected.length > 0 && selected.every((i) => i.status === "backlog");
+      if (!allBacklog) {
+        openModal("issue-run-confirm", {
+          issueIds: ids,
+          mode: "assign",
+          assigneeType: updates.assignee_type,
+          assigneeId: updates.assignee_id,
+        });
+        return;
+      }
     }
     void handleBatchUpdate(updates);
   };

packages/views/issues/components/execution-log-section.tsx

@@ -224,6 +224,12 @@ function useTriggerText(task: AgentTask): string {
   }
   if (task.autopilot_run_id) return t(($) => $.execution_log.trigger_autopilot);
   if (task.trigger_comment_id) return t(($) => $.execution_log.trigger_comment);
+  // Assignment-triggered run that carried a handoff note: show the note inline
+  // (truncated by TriggerText) the way comment triggers show their text, so the
+  // row reads as the handoff instead of the generic "initial run".
+  if (task.handoff_note) {
+    return retryPrefix + t(($) => $.execution_log.trigger_handoff_prefix) + stripMentionMarkdown(task.handoff_note);
+  }
   return t(($) => $.execution_log.trigger_initial);
 }
 

packages/views/issues/hooks/use-issue-trigger-preview.ts

@@ -1,7 +1,7 @@
 "use client";
 
 import { useMemo } from "react";
-import { useQuery } from "@tanstack/react-query";
+import { keepPreviousData, useQuery } from "@tanstack/react-query";
 import { api } from "@multica/core/api";
 import { issueKeys } from "@multica/core/issues/queries";
 import type { IssueAssigneeType, IssueStatus, IssueTriggerPreviewItem } from "@multica/core/types";
@@ -41,8 +41,19 @@ function previewSignature(params: UseIssueTriggerPreviewParams): string {
 
 /** Reads the unified backend predicate via POST /api/issues/preview-trigger so
  *  the four entry points never re-implement "will this start a run" (MUL-3375).
- *  The answer is queue-dependent (status-source pending dedup), so it is never
- *  cached stale: staleTime 0, and WS task events invalidate issueTriggerPreviewAll. */
+ *
+ *  The verdict changes only with the inputs (assignee / status), so the query
+ *  refetches solely on signature change — it is deliberately NOT invalidated by
+ *  WS task events. The assign source (create / assignee change) cancels existing
+ *  tasks before enqueuing, so its verdict can't shift from a task event at all;
+ *  the status source's pending dedup could, but the preview is advisory and the
+ *  write path re-evaluates authoritatively, so a rare stale status label is
+ *  harmless — far better than refetching every mounted preview on every
+ *  workspace task event (the source of the visible flicker, MUL-3375).
+ *
+ *  Mirrors the comment-trigger preview's data handling: keepPreviousData so an
+ *  input switch swaps the answer in place instead of collapsing, and only the
+ *  very first load (no prior data) counts as loading. */
 export function useIssueTriggerPreview(
   params: UseIssueTriggerPreviewParams,
 ): UseIssueTriggerPreviewResult {
@@ -67,13 +78,18 @@ export function useIssueTriggerPreview(
     enabled,
     retry: false,
     staleTime: 0,
+    // Keep the prior verdict visible while a new signature (assignee/status
+    // switch) refetches, so the hint swaps in place rather than collapsing.
+    placeholderData: keepPreviousData,
   });
 
   const triggers = previewQuery.data?.triggers ?? EMPTY;
   return {
     triggers,
     totalCount: previewQuery.data?.total_count ?? 0,
-    isLoading: enabled && previewQuery.isFetching,
+    // Only the first load (no prior data) is "loading"; a background/placeholder
+    // refetch is not, so reveal animations gated on this never collapse mid-fetch.
+    isLoading: enabled && previewQuery.isLoading,
     handoffSupported: triggers.length > 0 && triggers.every((t) => t.handoff_supported === true),
   };
 }

packages/views/locales/en/issues.json

@@ -371,6 +371,7 @@
     "trigger_autopilot": "Autopilot run",
     "trigger_comment": "Comment trigger",
     "trigger_initial": "Initial run",
+    "trigger_handoff_prefix": "Handoff: ",
     "status_queued": "Queued",
     "status_dispatched": "Starting",
     "status_waiting_local_directory": "Waiting for local directory",

packages/views/locales/en/modals.json

@@ -53,6 +53,7 @@
   "run_confirm": {
     "title_assign": "Assign and start?",
     "title_status": "Start working now?",
+    "checking": "Checking…",
     "will_start_named": "Once assigned, {{name}} will start working on this issue right away.",
     "will_start_named_squad": "Once assigned, {{name}}'s leader will review it and start assigning the work.",
     "will_start": "This will start work on {{count}} issues right away.",

packages/views/locales/ja/issues.json

@@ -356,6 +356,7 @@
     "trigger_autopilot": "オートパイロット実行",
     "trigger_comment": "コメントトリガー",
     "trigger_initial": "初回実行",
+    "trigger_handoff_prefix": "引き継ぎ：",
     "status_queued": "待機中",
     "status_dispatched": "開始中",
     "status_waiting_local_directory": "ローカルディレクトリ待機中",

packages/views/locales/ja/modals.json

@@ -53,6 +53,7 @@
   "run_confirm": {
     "title_assign": "割り当てて開始しますか？",
     "title_status": "今すぐ作業を開始しますか？",
+    "checking": "確認中…",
     "will_start_named": "割り当てると、{{name}} がすぐにこのイシューの作業を開始します。",
     "will_start_named_squad": "割り当てると、{{name}} のリーダーが内容を評価し、タスクの割り振りを開始します。",
     "will_start": "{{count}} 件のイシューの作業をすぐに開始します。",

packages/views/locales/ko/issues.json

@@ -370,6 +370,7 @@
     "trigger_autopilot": "오토파일럿 실행",
     "trigger_comment": "댓글 트리거",
     "trigger_initial": "초기 실행",
+    "trigger_handoff_prefix": "인계: ",
     "status_queued": "대기 중",
     "status_dispatched": "시작 중",
     "status_waiting_local_directory": "로컬 디렉터리 대기 중",

packages/views/locales/ko/modals.json

@@ -53,6 +53,7 @@
   "run_confirm": {
     "title_assign": "할당하고 시작할까요?",
     "title_status": "지금 작업을 시작할까요?",
+    "checking": "확인 중…",
     "will_start_named": "할당하면 {{name}}이(가) 이 이슈 작업을 바로 시작합니다.",
     "will_start_named_squad": "할당하면 {{name}}의 리더가 내용을 평가한 뒤 작업 배분을 시작합니다.",
     "will_start": "{{count}}개 이슈의 작업을 바로 시작합니다.",

packages/views/locales/zh-Hans/issues.json

@@ -361,6 +361,7 @@
     "trigger_autopilot": "自动化运行",
     "trigger_comment": "评论触发",
     "trigger_initial": "首次运行",
+    "trigger_handoff_prefix": "交接：",
     "status_queued": "排队中",
     "status_dispatched": "启动中",
     "status_waiting_local_directory": "等待本地目录释放",

packages/views/locales/zh-Hans/modals.json

@@ -53,6 +53,7 @@
   "run_confirm": {
     "title_assign": "指派并开始处理？",
     "title_status": "现在开始处理？",
+    "checking": "检查中…",
     "will_start_named": "指派后，{{name}} 会立即开始工作。",
     "will_start_named_squad": "指派后，{{name}} 的队长会评估并开始安排任务。",
     "will_start": "将立即开始处理 {{count}} 个 issue。",

packages/views/modals/run-confirm.tsx

@@ -12,6 +12,7 @@ import {
 } from "@multica/ui/components/ui/dialog";
 import { Button } from "@multica/ui/components/ui/button";
 import { Textarea } from "@multica/ui/components/ui/textarea";
+import { Spinner } from "@multica/ui/components/ui/spinner";
 import type { IssueAssigneeType, IssueStatus, UpdateIssueRequest } from "@multica/core/types";
 import { useUpdateIssue, useBatchUpdateIssues } from "@multica/core/issues/mutations";
 import { useActorName } from "@multica/core/workspace/hooks";
@@ -69,7 +70,11 @@ export function RunConfirmModal({
   const mode = d.mode ?? "assign";
 
   const [note, setNote] = useState("");
-  const [submitting, setSubmitting] = useState(false);
+  // Which footer action is in flight, so only the clicked button shows a
+  // spinner (the request runs an agent on the server for note assigns, so it is
+  // not instant — the disabled-only state read as frozen).
+  const [pendingAction, setPendingAction] = useState<"go" | "suppress" | null>(null);
+  const submitting = pendingAction !== null;
 
   const updateIssue = useUpdateIssue();
   const batchUpdate = useBatchUpdateIssues();
@@ -82,6 +87,7 @@ export function RunConfirmModal({
     enabled: issueIds.length > 0,
   });
 
+  const loading = preview.isLoading;
   const willStart = preview.totalCount > 0;
   const canNote = mode === "assign" && willStart;
   // Soft gate: an old runtime can't render the note. Disable the box but let
@@ -98,7 +104,7 @@ export function RunConfirmModal({
 
   const submit = async (suppressRun: boolean) => {
     if (issueIds.length === 0 || submitting) return;
-    setSubmitting(true);
+    setPendingAction(suppressRun ? "suppress" : "go");
     const payload = applyTo({
       ...(suppressRun ? { suppress_run: true } : {}),
       ...(!suppressRun && canNote && !noteDisabled && note.trim()
@@ -114,7 +120,7 @@ export function RunConfirmModal({
       onClose();
     } catch (err) {
       toast.error(err instanceof Error && err.message ? err.message : t(($) => $.run_confirm.toast_failed));
-      setSubmitting(false);
+      setPendingAction(null);
     }
   };
 
@@ -152,10 +158,24 @@ export function RunConfirmModal({
           <DialogTitle>
             {mode === "assign" ? t(($) => $.run_confirm.title_assign) : t(($) => $.run_confirm.title_status)}
           </DialogTitle>
-          <DialogDescription>{headline}</DialogDescription>
+          <DialogDescription>
+            {loading ? (
+              <span className="flex items-center gap-1.5 text-muted-foreground">
+                <Spinner className="size-3.5" />
+                {t(($) => $.run_confirm.checking)}
+              </span>
+            ) : (
+              headline
+            )}
+          </DialogDescription>
         </DialogHeader>
 
-        {canNote ? (
+        {/* Assign mode keeps the note box mounted while the preview is in flight
+            (disabled), so the dialog opens at its resolved height instead of
+            growing when the predicate lands. Parked (no run) is the only case
+            without a note, and it can't be a Backlog assign (those skip this
+            modal), so it is rare. */}
+        {mode === "assign" && (loading || canNote) ? (
           <div className="grid gap-1.5">
             <label className="text-sm font-medium" htmlFor="handoff-note">
               {t(($) => $.run_confirm.note_label)}
@@ -164,30 +184,34 @@ export function RunConfirmModal({
               id="handoff-note"
               value={note}
               maxLength={MAX_HANDOFF_NOTE}
-              disabled={noteDisabled || submitting}
+              disabled={loading || noteDisabled || submitting}
               placeholder={t(($) => $.run_confirm.note_placeholder)}
               onChange={(e) => setNote(e.target.value)}
               rows={3}
             />
-            {noteDisabled ? (
+            {!loading && noteDisabled ? (
               <p className="text-xs text-muted-foreground">{t(($) => $.run_confirm.note_unsupported)}</p>
             ) : null}
           </div>
         ) : null}
 
         <DialogFooter>
-          {willStart ? (
+          {loading ? (
+            <Button type="button" disabled>
+              <Spinner className="size-4" />
+            </Button>
+          ) : willStart ? (
             <>
               <Button type="button" variant="outline" disabled={submitting} onClick={() => submit(true)}>
-                {t(($) => $.run_confirm.dont_start)}
+                {pendingAction === "suppress" ? <Spinner className="size-4" /> : t(($) => $.run_confirm.dont_start)}
               </Button>
               <Button type="button" disabled={submitting} onClick={() => submit(false)}>
-                {t(($) => $.run_confirm.start)}
+                {pendingAction === "go" ? <Spinner className="size-4" /> : t(($) => $.run_confirm.start)}
               </Button>
             </>
           ) : (
             <Button type="button" disabled={submitting} onClick={() => submit(false)}>
-              {t(($) => $.run_confirm.apply)}
+              {pendingAction === "go" ? <Spinner className="size-4" /> : t(($) => $.run_confirm.apply)}
             </Button>
           )}
         </DialogFooter>

server/internal/handler/agent.go

@@ -379,6 +379,10 @@ func taskToResponse(t db.AgentTaskQueue, workspaceID string) AgentTaskResponse {
 	if t.WorkDir.Valid {
 		workDir = t.WorkDir.String
 	}
+	handoffNote := ""
+	if t.HandoffNote.Valid {
+		handoffNote = t.HandoffNote.String
+	}
 	return AgentTaskResponse{
 		ID:               uuidToString(t.ID),
 		AgentID:          uuidToString(t.AgentID),
@@ -399,6 +403,7 @@ func taskToResponse(t db.AgentTaskQueue, workspaceID string) AgentTaskResponse {
 		CreatedAt:        timestampToString(t.CreatedAt),
 		TriggerCommentID: uuidToPtr(t.TriggerCommentID),
 		TriggerSummary:   textToPtr(t.TriggerSummary),
+		HandoffNote:      handoffNote,
 		WorkDir:          workDir,
 		RelativeWorkDir:  relativeWorkDir(workDir, workspaceID, uuidToString(t.ID)),
 		// Surface task source so the UI can distinguish issue-linked tasks

server/internal/handler/daemon.go

@@ -1588,12 +1588,9 @@ func (h *Handler) ClaimTaskByRuntime(w http.ResponseWriter, r *http.Request) {
 		}
 	}
 
-	// Handoff note (MUL-3375): a first-class instruction set when the issue was
-	// assigned/promoted. Surfaced to the daemon so its prompt + issue_context.md
-	// can render the assignment-handoff branch. Empty for all other task kinds.
-	if task.HandoffNote.Valid {
-		resp.HandoffNote = task.HandoffNote.String
-	}
+	// Handoff note (MUL-3375) is populated by taskToResponse (the shared mapper
+	// resp came from above), so the daemon's prompt + issue_context.md render the
+	// assignment-handoff branch. Empty for all other task kinds.
 
 	// Quick-create task: no issue / chat / autopilot link — workspace and
 	// prompt come from the task's context JSONB. Resolve workspace from