feat(chat): answer the CLI's menus from the chat; Stop ends Thinking… (#74)

- A menu the CLI asks with (AskUserQuestion — one question or several,
  multiSelect —, a permission or trust prompt, Codex's approval, a
  first-start login menu) is read off the terminal under the chat
  (`parsePaneMenu`/`readRunMenu`) and shown in the prompt box's place as a
  card of buttons. A pick presses its digit (picks it, or ticks its box),
  Claude's unnumbered Submit row is reached with the arrows and Enter, and
  Cancel presses Esc. "Type something" opens a field in the card: its text
  goes in by the arrows, Ctrl+U over anything already typed there, a
  bracketed paste and Enter. That row presses no key — a second press of
  its digit typed the digit into the CLI's field ("44444…"). Keys verified
  on Claude Code 2.1.296 and Codex 0.161. AskUserQuestion fires no
  Notification, so a menu on screen also counts as waiting (live line, tab
  badge). Client only: no new route.
- #74: Esc before the model wrote anything puts the prompt back in Claude
  Code's input box and writes no "[Request interrupted by user]" line, so
  the transcript ended on the prompt and the chat read "Thinking…" for
  15 min. The chat's Stop now ends the live line until a newer transcript
  line arrives (matched by line, not by clock).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

client/src/components/conversation/ConversationView.tsx

@@ -97,6 +97,8 @@ interface ConversationViewProps {
   waiting?: boolean;
   /** Shows the pane's terminal, where that question is answered. */
   onOpenTerminal?: () => void;
+  /** Bumped by the chat's Stop: the live line ends with the turn on screen. */
+  stopped?: number;
   /** The newest reply's prompt size (input + cache) and model: how full the
    *  context is. */
   onContext?: (tokens: number, model: string | null) => void;
@@ -113,6 +115,7 @@ export function ConversationView({
   follow = 0,
   waiting = false,
   onOpenTerminal,
+  stopped = 0,
   onContext,
   onNotStarted,
 }: ConversationViewProps) {
@@ -558,7 +561,11 @@ export function ConversationView({
                 {t("conversation.notStartedBody")}
               </p>
               {onOpenTerminal && (
-                <button type="button" onClick={onOpenTerminal} className="btn-secondary mt-3 text-xs">
+                <button
+                  type="button"
+                  onClick={onOpenTerminal}
+                  className="btn-secondary mt-3 text-xs"
+                >
                   {t("conversation.openTerminal")}
                 </button>
               )}
@@ -578,6 +585,7 @@ export function ConversationView({
               live={live}
               waiting={waiting}
               onOpenTerminal={onOpenTerminal}
+              stopped={stopped}
             />
           )}
         </div>

client/src/components/conversation/MessageList.tsx

@@ -156,6 +156,9 @@ interface MessageListProps {
   waiting?: boolean;
   /** Shows the pane's terminal, where that question is answered. */
   onOpenTerminal?: () => void;
+  /** Bumped by the chat's Stop: the turn on screen is over, even when the
+   *  transcript never says so. */
+  stopped?: number;
 }
 
 /** Build a map from tool_use id → tool_result for matching */
@@ -611,15 +614,24 @@ export function MessageList({
   live = false,
   waiting = false,
   onOpenTerminal,
+  stopped = 0,
 }: MessageListProps) {
   const { t } = useTranslation("sessions");
   const toolResultMap = useMemo(() => buildToolResultMap(messages), [messages]);
+  const keys = useMemo(() => messageKeys(messages), [messages]);
+  const lastKey = keys[keys.length - 1] ?? null;
+  // The newest line when Stop was pressed. Esc before the model wrote anything
+  // puts the prompt back in Claude Code's input box and writes no
+  // "[Request interrupted…]" line (issue #74), so the transcript still ends on
+  // the prompt: no live line until a newer one arrives.
+  const [stop, setStop] = useState<{ n: number; on: string | null }>({ n: stopped, on: null });
+  if (stop.n !== stopped) setStop({ n: stopped, on: lastKey });
+  const halted = stop.on !== null && stop.on === lastKey;
   const pending = useMemo(
-    () => (live ? workingOn(messages, toolResultMap) : null),
-    [live, messages, toolResultMap]
+    () => (live && !halted ? workingOn(messages, toolResultMap) : null),
+    [live, halted, messages, toolResultMap]
   );
   const stats = useMemo(() => (pending ? turnStats(messages) : null), [pending, messages]);
-  const keys = useMemo(() => messageKeys(messages), [messages]);
   const items = useMemo(() => chatItems(messages, keys), [messages, keys]);
 
   if (loading) {

client/src/components/conversation/__tests__/MessageList.sender.test.tsx

@@ -182,6 +182,29 @@ describe("MessageList — sender attribution", () => {
     expect(screen.queryByTestId("chat-running-step")).toBeNull();
   });
 
+  // Issue #74: Esc before the model wrote anything takes the prompt back into
+  // Claude Code's input box and writes NO "[Request interrupted…]" line — the
+  // transcript still ends on the prompt, which read as "Thinking…" for 15 min.
+  it("drops Thinking… once Stop was pressed, until the transcript moves on", () => {
+    const prompt = msg({
+      sender: "user",
+      timestamp: new Date(Date.now() - 14_000).toISOString(),
+      content: [{ type: "text", text: "Phần đề xuất trong báo cáo" }],
+    });
+    const { rerender } = render(<MessageList loading={false} messages={[prompt]} live />);
+    expect(screen.getByTestId("chat-running-step")).toHaveTextContent("Thinking");
+    rerender(<MessageList loading={false} messages={[prompt]} live stopped={1} />);
+    expect(screen.queryByTestId("chat-running-step")).toBeNull();
+    // The next prompt is a new turn: its Thinking… shows.
+    const next = msg({
+      sender: "user",
+      timestamp: new Date().toISOString(),
+      content: [{ type: "text", text: "again" }],
+    });
+    rerender(<MessageList loading={false} messages={[prompt, next]} live stopped={1} />);
+    expect(screen.getByTestId("chat-running-step")).toHaveTextContent("Thinking");
+  });
+
   it("counts a turn's output once per API call, since its prompt", () => {
     const at = "2026-06-26T08:19:00.000Z";
     const line = (cache: number, out: number) =>