fix(issues): sync header agent chip with execution log via shared query

The header live chip derived its active-task state from the workspace-wide
agent-task-snapshot, while the right-panel Execution log read the per-issue
task list. Two queries, two endpoints, two independent refetches: the heavier
workspace snapshot lands later than the per-issue list, so the log could show
a running task while the header chip had not started yet.

Point the chip at the same `issueKeys.tasks(issueId)` cache the Execution log
uses (identical query options). Both surfaces now observe one cache entry and
update atomically. Drop the now-redundant workspace-id lookup and client-side
issue_id filter, since the endpoint is already issue-scoped.

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

packages/views/issues/components/issue-agent-header-chip.test.tsx

@@ -6,17 +6,13 @@ import type { AgentTask } from "@multica/core/types";
 import { renderWithI18n } from "../../test/i18n";
 
 const mockState = vi.hoisted(() => ({
-  snapshot: [] as unknown[],
+  tasks: [] as unknown[],
   taskMessagesOptions: vi.fn(),
   // Captures the props the chip passes to PopoverTrigger so a test can assert
   // the card is wired to open on hover, not only on click.
   triggerProps: undefined as Record<string, unknown> | undefined,
 }));
 
-vi.mock("@multica/core/hooks", () => ({
-  useWorkspaceId: () => "ws-1",
-}));
-
 vi.mock("@multica/core/workspace/hooks", () => ({
   useActorName: () => ({
     getActorName: (_type: string, id: string) =>
@@ -75,8 +71,9 @@ vi.mock("@tanstack/react-query", async () => {
   return {
     ...actual,
     useQuery: (opts: { queryKey?: readonly unknown[] }) => {
-      if (opts.queryKey?.[2] === "agent-task-snapshot") {
-        return { data: mockState.snapshot };
+      // Per-issue task list: issueKeys.tasks(issueId) === ["issues","tasks",id]
+      if (opts.queryKey?.[0] === "issues" && opts.queryKey?.[1] === "tasks") {
+        return { data: mockState.tasks };
       }
       return { data: undefined };
     },
@@ -106,13 +103,13 @@ function makeTask(overrides: Partial<AgentTask>): AgentTask {
 beforeEach(() => {
   cleanup();
   vi.clearAllMocks();
-  mockState.snapshot = [];
+  mockState.tasks = [];
   mockState.triggerProps = undefined;
 });
 
 describe("IssueAgentHeaderChip", () => {
   it("shows the active agent name without event count or elapsed time", () => {
-    mockState.snapshot = [makeTask({})];
+    mockState.tasks = [makeTask({})];
 
     renderWithI18n(<IssueAgentHeaderChip issueId="issue-1" />);
 
@@ -126,7 +123,7 @@ describe("IssueAgentHeaderChip", () => {
   });
 
   it("keeps the header popover card with active task rows", () => {
-    mockState.snapshot = [makeTask({ id: "task-running" })];
+    mockState.tasks = [makeTask({ id: "task-running" })];
 
     renderWithI18n(<IssueAgentHeaderChip issueId="issue-1" />);
 
@@ -138,7 +135,7 @@ describe("IssueAgentHeaderChip", () => {
   });
 
   it("opens the activity card on hover, not only on click", () => {
-    mockState.snapshot = [makeTask({})];
+    mockState.tasks = [makeTask({})];
 
     renderWithI18n(<IssueAgentHeaderChip issueId="issue-1" />);
 
@@ -152,7 +149,7 @@ describe("IssueAgentHeaderChip", () => {
   });
 
   it("uses the concise multi-agent working label", () => {
-    mockState.snapshot = [
+    mockState.tasks = [
       makeTask({ id: "task-1", agent_id: "agent-1" }),
       makeTask({ id: "task-2", agent_id: "agent-2" }),
     ];
@@ -168,7 +165,7 @@ describe("IssueAgentHeaderChip", () => {
   });
 
   it("uses the requested Chinese single-agent copy", () => {
-    mockState.snapshot = [makeTask({})];
+    mockState.tasks = [makeTask({})];
 
     renderWithI18n(<IssueAgentHeaderChip issueId="issue-1" />, {
       locale: "zh-Hans",
@@ -177,14 +174,20 @@ describe("IssueAgentHeaderChip", () => {
     expect(screen.getByText("Walt 在工作")).toBeInTheDocument();
   });
 
-  it("does not render for inactive or unrelated tasks", () => {
-    mockState.snapshot = [
+  it("does not render when the issue has only terminal tasks", () => {
+    // The list is issue-scoped by the endpoint, so the chip's only job is to
+    // ignore terminal statuses (those are the execution log's story).
+    mockState.tasks = [
       makeTask({
         id: "task-done",
         status: "completed",
         completed_at: "2026-06-08T08:05:00Z",
       }),
-      makeTask({ id: "task-other", issue_id: "issue-2" }),
+      makeTask({
+        id: "task-cancelled",
+        status: "cancelled",
+        completed_at: "2026-06-08T08:06:00Z",
+      }),
     ];
 
     renderWithI18n(<IssueAgentHeaderChip issueId="issue-1" />);

packages/views/issues/components/issue-agent-header-chip.tsx

@@ -7,10 +7,10 @@ import {
   PopoverContent,
   PopoverTrigger,
 } from "@multica/ui/components/ui/popover";
-import { useWorkspaceId } from "@multica/core/hooks";
 import { useActorName } from "@multica/core/workspace/hooks";
 import { cn } from "@multica/ui/lib/utils";
-import { agentTaskSnapshotOptions } from "@multica/core/agents";
+import { api } from "@multica/core/api";
+import { issueKeys } from "@multica/core/issues/queries";
 import type { AgentTask } from "@multica/core/types";
 import { AgentAvatarStack } from "../../agents/components/agent-avatar-stack";
 import { ActiveTaskRow } from "./execution-log-section";
@@ -21,10 +21,12 @@ import { useT } from "../../i18n";
 // signal stays in one fixed place and never competes with future sticky
 // banners in the content column. Replaces the in-body sticky live card.
 //
-// Derives state from the workspace-wide agent task snapshot filtered by
-// issue id — the same single source of truth that powers the board-card /
-// list-row IssueAgentActivityIndicator, so the chip is always consistent
-// with those surfaces and costs zero extra network.
+// Reads the same per-issue task list as the right-panel Execution log
+// (shared `issueKeys.tasks(issueId)` cache), so the header chip and the log
+// always agree on what is active. Both surfaces derive from one query, which
+// removes the race where the old workspace-wide agent-task-snapshot refetched
+// slower than this per-issue list and left the chip lagging behind the log's
+// "agent is working".
 //
 // Collapsed display stays intentionally shallow:
 //   - one running agent  → avatar + "{name} is working"
@@ -44,14 +46,20 @@ interface IssueAgentHeaderChipProps {
 export const IssueAgentHeaderChip = memo(function IssueAgentHeaderChip({
   issueId,
 }: IssueAgentHeaderChipProps) {
-  const wsId = useWorkspaceId();
-  const { data: snapshot = [] } = useQuery(agentTaskSnapshotOptions(wsId));
+  // Same query options as ExecutionLogSection so both observe one cache entry.
+  const { data: tasks = [] } = useQuery({
+    queryKey: issueKeys.tasks(issueId),
+    queryFn: () => api.listTasksByIssue(issueId),
+    staleTime: 30_000,
+    refetchOnWindowFocus: true,
+  });
 
   const { running, queued } = useMemo(() => {
     const running: AgentTask[] = [];
     const queued: AgentTask[] = [];
-    for (const task of snapshot) {
-      if (task.issue_id !== issueId) continue;
+    // The list is already issue-scoped by the endpoint, so only the status
+    // split matters here.
+    for (const task of tasks) {
       if (task.status === "running") running.push(task);
       else if (
         task.status === "queued" ||
@@ -64,7 +72,7 @@ export const IssueAgentHeaderChip = memo(function IssueAgentHeaderChip({
       // Terminal statuses are the execution log's story, not the live chip's.
     }
     return { running, queued };
-  }, [snapshot, issueId]);
+  }, [tasks]);
 
   // No active work → render nothing.
   if (running.length === 0 && queued.length === 0) return null;