fix: v2.7.5 — the chat shows each AskUserQuestion with the user's answer

- Chat: an AskUserQuestion is its own card (AskCard in MessageList.tsx), not
  part of the collapsed work line — each question with the option picked or
  the text typed, "declined" when refused. The transcript route passes the
  answers from Claude Code's toolUseResult (askAnswers, beside diff).
- Chat: the activity panel keeps the user's open/close per session for the
  tab, and opens by default only once the session has an output.
- Workspace home: the sent prompt's draft is dropped even though opening the
  session unmounts the home.
- Terminal: Ctrl+V with text on the clipboard pastes the text; only an empty
  clipboard still sends ^V.
- AGENTS.md / CLAUDE.md: run only the tests a change touches while iterating,
  the full suite once before committing.

AGENTS.md

@@ -25,6 +25,7 @@ make while you work here as Codex.
 ## Validation expectations
 - Backend changes: run `npm run test:server` when possible.
 - Frontend changes: run `npm run test:client` when possible.
+- Full suites take 5–6 minutes, so while iterating run only what the change touches: `npm run test:server -- server/__tests__/<name>.test.js`, and in `client/` `NODE_ENV=test npx vitest related <changed files> --run` (or `vitest run <test file>`). Run the full suite once before committing, and always when the change touches something every screen uses (`Layout`, `Sidebar`, `StatusBar`, `tailwind.config.js`, global CSS, shared hooks/lib) — the per-screen snapshots catch that.
 - If any check is skipped, report it explicitly.
 
 ## Safety expectations

CLAUDE.md

@@ -1,6 +1,6 @@
 # Claude Code Working Guide
 
-_Last synced: v2.7.4 (2026-10-12) — Changes hides a staged change undone on disk (#72); chat says when the session has not started (#73); a member's missing folder is made as them, bypass offered only once accepted._ A change that makes any statement in this file, `AGENTS.md`, or a README wrong — a moved path, a new or changed extension point, a retired rule — updates that file in the same change-set and bumps this marker. **Bump = replace the version/date and a ≤ 1-line summary; never append a changelog here** — details belong in the commit message and `docs/`, because this whole file is loaded into every session.
+_Last synced: v2.7.5 (2026-10-13) — the chat shows each AskUserQuestion with the user's answer; while iterating, run only the tests a change touches._ A change that makes any statement in this file, `AGENTS.md`, or a README wrong — a moved path, a new or changed extension point, a retired rule — updates that file in the same change-set and bumps this marker. **Bump = replace the version/date and a ≤ 1-line summary; never append a changelog here** — details belong in the commit message and `docs/`, because this whole file is loaded into every session.
 
 ## Project mission
 - Maintain a reliable local-first dashboard for Claude Code session monitoring.
@@ -96,6 +96,7 @@ See `docs/LANES.md` for full guide: stage reporting, custom templates, lane acti
 ## Testing and verification policy
 - Backend changes: run `npm run test:server` before finishing.
 - Frontend changes: run `npm run test:client` when relevant. This includes per-screen render snapshots (`client/src/pages/__tests__/screens.snapshot.test.tsx`). If a UI change is intentional, review the snapshot diff and regenerate baselines with `cd client && npx vitest run -u`; never blindly update snapshots to make tests pass.
+- Full suites take 5–6 minutes — while iterating run only what the change touches (`npm run test:server -- server/__tests__/<name>.test.js`; in `client/` `NODE_ENV=test npx vitest related <changed files> --run`), then the full suite once before committing, and always when touching something every screen uses (`Layout`, `Sidebar`, `StatusBar`, `tailwind.config.js`, global CSS, shared hooks/lib).
 - If you cannot run a verification step, state exactly what was not run and why.
 
 ## Change guidelines by area

client/src/components/conversation/MessageList.tsx

@@ -15,7 +15,9 @@
  * files ends with Claude Desktop's "Edited N files" card: every file with the
  * lines added and removed (the tool results' `diff`); a click opens the file
  * when the caller can (`onOpenFile`), and a row (or the card's header, for
- * every file in it) drags into a prompt box as an `@path` attachment. A thinking block that carries text is
+ * every file in it) drags into a prompt box as an `@path` attachment. An
+ * AskUserQuestion is its own card (`AskCard`), not part of the collapsed work
+ * line: each question with the answer the user picked or typed. A thinking block that carries text is
  * the model's narration to the user (Claude Code ≥ 2.1.280 stores real
  * thinking empty) and reads as prose, like Claude Desktop shows it; a live
  * chat (`live`) ends, until its turn is over, with what the model is doing
@@ -95,6 +97,8 @@ import {
   Check,
   Hand,
   SquareTerminal,
+  MessageCircleQuestion,
+  CornerDownRight,
 } from "lucide-react";
 import type { TranscriptMessage, TranscriptContent, TranscriptSender } from "../../lib/types";
 
@@ -461,6 +465,7 @@ type Item =
   | { kind: "user"; key: string; msg: TranscriptMessage }
   | { kind: "text"; key: string; msg: TranscriptMessage; text: string }
   | { kind: "work"; key: string; blocks: TranscriptContent[] }
+  | { kind: "ask"; key: string; block: TranscriptContent }
   | { kind: "edits"; key: string; blocks: TranscriptContent[] }
   | { kind: "done"; key: string; ms: number; output: number };
 
@@ -522,7 +527,10 @@ export function chatItems(messages: TranscriptMessage[], keys: string[]): Item[]
     msg.content.forEach((block, b) => {
       const last = items[items.length - 1];
       if (block.type === "tool_use" && EDIT_TOOLS.has(block.name ?? "")) edits.push(block);
-      if (block.type === "tool_use") {
+      // A question to the user is a turn of the conversation, not hidden work.
+      if (block.type === "tool_use" && block.name === "AskUserQuestion") {
+        items.push({ kind: "ask", key: `${key}:${b}`, block });
+      } else if (block.type === "tool_use") {
         if (last?.kind === "work") last.blocks.push(block);
         else items.push({ kind: "work", key: `${key}:${b}`, blocks: [block] });
       } else if ((block.type === "text" || block.type === "thinking") && block.text?.trim()) {
@@ -669,6 +677,14 @@ export function MessageList({
               return (
                 <WorkGroup key={item.key} blocks={item.blocks} toolResultMap={toolResultMap} />
               );
+            case "ask":
+              return (
+                <AskCard
+                  key={item.key}
+                  block={item.block}
+                  result={item.block.id ? (toolResultMap.get(item.block.id) ?? null) : null}
+                />
+              );
             case "edits":
               // The card closes a finished turn; one still working has none yet.
               if (pending && item === items[items.length - 1]) return null;
@@ -935,6 +951,50 @@ function RunningStep({
 }
 
 /** Under a turn that ended: how long it took and the output it made. */
+/** An AskUserQuestion: each question with the answer the user picked or
+ *  typed (the result's `answers`), "declined" when they refused it, no answer
+ *  yet while the CLI still asks. */
+function AskCard({
+  block,
+  result,
+}: {
+  block: TranscriptContent;
+  result: TranscriptContent | null;
+}) {
+  const { t } = useTranslation("sessions");
+  const input = block.input as { questions?: unknown } | undefined;
+  const questions = (Array.isArray(input?.questions) ? input.questions : [])
+    .map((q) => (q && typeof q === "object" ? (q as { question?: unknown }).question : null))
+    .filter((q): q is string => typeof q === "string");
+  return (
+    <div
+      data-testid="chat-ask"
+      className="space-y-2 rounded-xl border border-border px-3 py-2 text-sm"
+    >
+      {questions.map((q, i) => {
+        const answer = result?.answers?.[q];
+        return (
+          <div key={i}>
+            <div className="flex items-start gap-1.5 text-fg-secondary">
+              <MessageCircleQuestion className="mt-0.5 h-3.5 w-3.5 flex-shrink-0" />
+              <span>{q}</span>
+            </div>
+            {answer && (
+              <div className="ml-5 mt-0.5 flex items-start gap-1.5 font-medium text-fg-primary">
+                <CornerDownRight className="mt-0.5 h-3.5 w-3.5 flex-shrink-0 text-fg-muted" />
+                <span className="whitespace-pre-wrap">{answer}</span>
+              </div>
+            )}
+          </div>
+        );
+      })}
+      {result?.is_error && (
+        <div className="text-xs text-fg-muted">{t("conversation.askDeclined")}</div>
+      )}
+    </div>
+  );
+}
+
 function TurnDone({ ms, output }: { ms: number; output: number }) {
   const { t } = useTranslation("sessions");
   return (

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

@@ -0,0 +1,85 @@
+/**
+ * @file MessageList.ask.test.tsx
+ * @description An AskUserQuestion call is not folded into the collapsed work
+ * line: the chat shows it as its own card — each question with the answer the
+ * user picked or typed, "declined" when they refused it, nothing yet while the
+ * CLI still waits.
+ * @author Nguyễn Ngọc Trí Vĩ <vinnt@smartgift.vn>
+ */
+import { describe, it, expect } from "vitest";
+import { render, screen, within } from "@testing-library/react";
+import { MessageList } from "../MessageList";
+import type { TranscriptMessage } from "../../../lib/types";
+
+const at = "2026-10-13T08:00:00.000Z";
+const user = (text: string): TranscriptMessage =>
+  ({
+    type: "user",
+    sender: "user",
+    timestamp: at,
+    content: [{ type: "text", text }],
+  }) as TranscriptMessage;
+const ask = (id: string, questions: string[]): TranscriptMessage =>
+  ({
+    type: "assistant",
+    sender: "assistant",
+    timestamp: at,
+    content: [
+      { type: "tool_use", id: `${id}r`, name: "Read", input: { file_path: "/w/a.ts" } },
+      {
+        type: "tool_use",
+        id,
+        name: "AskUserQuestion",
+        input: {
+          questions: questions.map((question) => ({
+            question,
+            header: "Pick",
+            options: [{ label: "Yes" }, { label: "No" }],
+          })),
+        },
+      },
+    ],
+  }) as TranscriptMessage;
+const result = (id: string, extra: Record<string, unknown>) =>
+  ({
+    type: "user",
+    sender: "tool",
+    timestamp: at,
+    content: [{ type: "tool_result", id, output: "x", ...extra }],
+  }) as TranscriptMessage;
+
+describe("an AskUserQuestion card", () => {
+  it("shows each question with the user's answer, outside the work line", () => {
+    render(
+      <MessageList
+        loading={false}
+        messages={[
+          user("go"),
+          ask("q1", ["Commit?", "Push?"]),
+          result("q1", { answers: { "Commit?": "Yes", "Push?": "only after review" } }),
+        ]}
+      />
+    );
+    const card = screen.getByTestId("chat-ask");
+    expect(within(card).getByText("Commit?")).toBeTruthy();
+    expect(within(card).getByText("Yes")).toBeTruthy();
+    expect(within(card).getByText("only after review")).toBeTruthy();
+    expect(within(screen.getByTestId("chat-tools")).queryByText(/Commit\?/)).toBeNull();
+  });
+
+  it("says declined when refused, and shows no answer while still asked", () => {
+    const { rerender } = render(
+      <MessageList loading={false} messages={[user("go"), ask("q1", ["Commit?"])]} />
+    );
+    const card = screen.getByTestId("chat-ask");
+    expect(within(card).getByText("Commit?")).toBeTruthy();
+    expect(within(card).queryByText(/declined/i)).toBeNull();
+    rerender(
+      <MessageList
+        loading={false}
+        messages={[user("go"), ask("q1", ["Commit?"]), result("q1", { is_error: true })]}
+      />
+    );
+    expect(within(screen.getByTestId("chat-ask")).getByText(/declined/i)).toBeTruthy();
+  });
+});

client/src/components/lanes/WorkspaceHome.tsx

@@ -277,6 +277,13 @@ function HomeBody({
         ...(claude && inWorktree && project?.root ? { worktree: true } : {}),
       });
       setPrompt("");
+      // Here, not only through the effect: the navigate below unmounts this
+      // home in the same batch, so the effect never sees the empty prompt.
+      try {
+        localStorage.removeItem(draftKey);
+      } catch {
+        /* storage unavailable: nothing was kept */
+      }
       images.forEach((i) => i.url && URL.revokeObjectURL(i.url));
       setImages([]);
       navigate(`/workspace?lane=${lane.id}&run=${encodeURIComponent(runId)}`);

client/src/components/lanes/__tests__/WorkspaceHome.test.tsx

@@ -14,7 +14,7 @@
 import { beforeEach, describe, expect, it, vi } from "vitest";
 import { fireEvent, render, screen, waitFor, within } from "@testing-library/react";
 import userEvent from "@testing-library/user-event";
-import { MemoryRouter, Route, Routes, useLocation } from "react-router-dom";
+import { MemoryRouter, Route, Routes, useLocation, useSearchParams } from "react-router-dom";
 import type { Lane, Session } from "../../../lib/types";
 import type { LaneSourcesView, RunHandle } from "../../../lib/api";
 
@@ -241,6 +241,27 @@ describe("WorkspaceHome", () => {
     await waitFor(() => expect(localStorage.getItem("agenthub.workspace.draft.1")).toBeNull());
   });
 
+  it("drops the sent prompt's draft even though opening the session unmounts the home", async () => {
+    // As the Workspace page: `&run=` replaces the home with the session view.
+    function HomeUntilRun() {
+      const [params] = useSearchParams();
+      return params.get("run") ? null : <WorkspaceHome lane={lane()} onMenuAction={() => undefined} />;
+    }
+    const user = userEvent.setup();
+    render(
+      <MemoryRouter initialEntries={["/workspace?lane=1"]}>
+        <Routes>
+          <Route path="*" element={<HomeUntilRun />} />
+        </Routes>
+      </MemoryRouter>
+    );
+    await screen.findByTestId("workspace-recent-s-live");
+    await user.type(screen.getByLabelText("How can I help you today?"), "sent once");
+    await user.click(screen.getByTestId("workspace-home-start"));
+    await waitFor(() => expect(screen.queryByTestId("workspace-home-start")).toBeNull());
+    expect(localStorage.getItem("agenthub.workspace.draft.1")).toBeNull();
+  });
+
   it("keeps AI and Model (with effort) in the box, Proxy/Permission/Worktree under it", async () => {
     renderHome();
     await screen.findByTestId("workspace-recent-s-live");

client/src/components/run/PaneChat.tsx

@@ -20,8 +20,10 @@
  * (`answerRunMenu`).
  * It also has Claude Desktop's activity panel (`ChatOutline`: progress,
  * outputs, what the session used), open at first in a pane spanning the
- * grid's full width or height (`panelOpen`) and collapsed into a file button
- * with the output count in a quarter-sized one; it sits beside the
+ * grid's full width or height (`panelOpen`) once the session has an output,
+ * else collapsed into a file button with the output count; the user's own
+ * open/close is kept per session for the tab (sessionStorage), so the chat
+ * leaving for the terminal and back keeps it; it sits beside the
  * conversation when the pane is wide and over it when narrow (a CSS container
  * query on the pane, not the window), dragged wider or narrower by its
  * splitter there (a fraction of the pane, remembered per browser). A file
@@ -92,6 +94,11 @@ const MODES: [PermissionMode, string][] = [
 const isSettable = (m: unknown): m is PermissionMode => MODES.some(([v]) => v === m);
 const SIDE_KEY = "agenthub.chat.panelWidth";
 const clampSide = (f: number) => Math.round(Math.min(0.7, Math.max(0.2, f)) * 100) / 100;
+const outlineKey = (sessionId: string) => `agenthub.chat.panel.${sessionId}`;
+const readOutlineChoice = (sessionId: string) => {
+  const v = sessionStorage.getItem(outlineKey(sessionId));
+  return v === null ? null : v === "1";
+};
 const storedSide = () => {
   const f = Number(localStorage.getItem(SIDE_KEY));
   return f ? clampSide(f) : null;
@@ -328,11 +335,20 @@ export function PaneChat({
   const [model, setModel] = useState(run.model ?? "");
   const [effort, setEffort] = useState<EffortLevel>(run.effort ?? "");
   const pickFiles = useRef<(() => void) | null>(null);
-  const [outlineOpen, setOutlineOpen] = useState(panelOpen);
+  // The user's own open/close, per session for this tab: it outlives the chat
+  // leaving for the terminal (an unmount). Unset, the default rules.
+  const [outlineChoice, setOutlineChoice] = useState(() => readOutlineChoice(sessionId));
   // A split or a close changes the pane's shape: its default comes back.
-  useEffect(() => setOutlineOpen(panelOpen), [panelOpen]);
+  const shape = useRef(panelOpen);
+  useEffect(() => {
+    if (shape.current === panelOpen) return;
+    shape.current = panelOpen;
+    sessionStorage.removeItem(outlineKey(sessionId));
+    setOutlineChoice(null);
+  }, [panelOpen, sessionId]);
   const openOutline = (open: boolean) => {
-    setOutlineOpen(open);
+    sessionStorage.setItem(outlineKey(sessionId), open ? "1" : "0");
+    setOutlineChoice(open);
     setPreview(null);
   };
   const [preview, setPreview] = useState<string | null>(null);
@@ -355,6 +371,8 @@ export function PaneChat({
   }, []);
   const [list, setList] = useState<OutlineList | null>(null);
   const activity = useSessionOutline(sessionId);
+  // Open by default only in a spanning pane, and only once there is an output.
+  const outlineOpen = outlineChoice ?? (panelOpen && activity.outputs.length > 0);
   const scope: PreviewScope = { laneId, laneCwd: laneCwd || cwd, cwd, sessionId };
   const text = promptWithImages(prompt, images);
 

client/src/components/run/TerminalView.tsx

@@ -40,7 +40,15 @@
  * here takes the lead back (`resize` with `drive: true`).
  * @author Nguyễn Ngọc Trí Vĩ <vinnt@smartgift.vn>
  */
-import { useEffect, useId, useMemo, useRef, useState, type ClipboardEvent, type DragEvent } from "react";
+import {
+  useEffect,
+  useId,
+  useMemo,
+  useRef,
+  useState,
+  type ClipboardEvent,
+  type DragEvent,
+} from "react";
 import { useTranslation } from "react-i18next";
 import type { Terminal } from "@xterm/xterm";
 import type { CanvasAddon } from "@xterm/addon-canvas";
@@ -200,13 +208,14 @@ export function TerminalView({ runId, wsBaseUrl, onExit, autoFocus = true }: Ter
 
   // A pasted image (a screenshot) goes the way of a drop: xterm only reads
   // text/plain and would paste nothing. Capture phase, because xterm's own
-  // paste listener stops the event. A Ctrl+V with no file in it is the pane's
-  // ^V again (vim's block select, readline's quoted insert), not a text paste.
+  // paste listener stops the event. Text is xterm's to paste, Ctrl+V's too;
+  // only a Ctrl+V with nothing on the clipboard is the pane's ^V again
+  // (vim's block select, readline's quoted insert).
   const onPaste = (e: ClipboardEvent) => {
     const ctrlV = ctrlVRef.current;
     ctrlVRef.current = false;
     const files = Array.from(e.clipboardData?.files ?? []);
-    if (files.length === 0 && !ctrlV) return;
+    if (files.length === 0 && (!ctrlV || e.clipboardData?.getData("text/plain"))) return;
     e.preventDefault();
     e.stopPropagation();
     if (files.length > 0) void attachFiles(files);
@@ -354,7 +363,8 @@ export function TerminalView({ runId, wsBaseUrl, onExit, autoFocus = true }: Ter
         }
         // Ctrl+V: xterm would send ^V, and claude would then read the
         // SERVER's clipboard, never this browser's. Let the browser paste
-        // instead (onPaste); a browser that fires none still sends ^V.
+        // instead (onPaste: a file's path, or the text); a browser that fires
+        // no paste, or one with nothing on the clipboard, still sends ^V.
         if (ev.ctrlKey && !ev.altKey && !ev.metaKey && !ev.shiftKey && key === "v") {
           ctrlVRef.current = true;
           setTimeout(() => {

client/src/components/run/__tests__/PaneChat.test.tsx

@@ -13,7 +13,7 @@
  * @author Nguyễn Ngọc Trí Vĩ <vinnt@smartgift.vn>
  */
 
-import { act, fireEvent, render, screen, waitFor } from "@testing-library/react";
+import { act, fireEvent, render, screen, waitFor, within } from "@testing-library/react";
 import { MemoryRouter } from "react-router-dom";
 import { beforeEach, describe, expect, it, vi } from "vitest";
 import { PaneChat, contextWindow } from "../PaneChat";
@@ -148,6 +148,7 @@ beforeEach(() => {
   vi.mocked(submitToRun).mockClear();
   // A prompt typed in one test is a saved draft in the next (`lib/chatDraft`).
   localStorage.removeItem("agenthub.chat.draft.s1");
+  sessionStorage.clear();
 });
 
 describe("PaneChat — the + menu", () => {
@@ -313,27 +314,63 @@ describe("PaneChat — the activity panel", () => {
     expect(screen.getByTestId("chat-outline")).toBeTruthy();
   });
 
-  it("starts open in a pane spanning the grid, and follows the pane when it is split", () => {
+  it("starts open in a pane spanning the grid, and follows the pane when it is split", async () => {
     const pane = (panelOpen: boolean) => (
       <MemoryRouter>
         <PaneChat run={RUN} sessionId="s1" cwd="/w" panelOpen={panelOpen} />
       </MemoryRouter>
     );
     const { rerender } = render(pane(true));
-    expect(screen.getByTestId("chat-outline")).toBeTruthy();
+    await screen.findByTestId("chat-outline");
     rerender(pane(false));
     expect(screen.queryByTestId("chat-outline")).toBeNull();
     expect(screen.getByTestId("chat-outline-show")).toBeTruthy();
   });
 
-  it("resizes from its splitter and remembers the width", () => {
+  it("starts closed while the session has no output, even in a spanning pane", async () => {
+    vi.mocked(api.sessions.outline).mockResolvedValueOnce({
+      progress: [],
+      outputs: [],
+      files: [],
+      uploads: [],
+      memory: [],
+      skills: [],
+      mcp: [],
+    } as never);
+    render(
+      <MemoryRouter>
+        <PaneChat run={RUN} sessionId="s1" cwd="/w" panelOpen />
+      </MemoryRouter>
+    );
+    await waitFor(() => expect(api.sessions.outline).toHaveBeenCalled());
+    expect(screen.queryByTestId("chat-outline")).toBeNull();
+    expect(screen.getByTestId("chat-outline-show").textContent).toBe("0");
+  });
+
+  it("keeps the user's open or close across the chat leaving for the terminal", async () => {
+    const pane = () => (
+      <MemoryRouter>
+        <PaneChat run={RUN} sessionId="s1" cwd="/w" panelOpen />
+      </MemoryRouter>
+    );
+    const first = render(pane());
+    fireEvent.click(
+      within(await screen.findByTestId("chat-outline")).getByRole("button", { name: "Hide side panel" })
+    );
+    first.unmount();
+    render(pane());
+    await waitFor(() => expect(screen.getByTestId("chat-outline-show").textContent).toBe("2"));
+    expect(screen.queryByTestId("chat-outline")).toBeNull();
+  });
+
+  it("resizes from its splitter and remembers the width", async () => {
     localStorage.removeItem("agenthub.chat.panelWidth");
     render(
       <MemoryRouter>
         <PaneChat run={RUN} sessionId="s1" cwd="/w" panelOpen />
       </MemoryRouter>
     );
-    const splitter = screen.getByRole("separator", { name: "Resize side panel" });
+    const splitter = await screen.findByRole("separator", { name: "Resize side panel" });
     expect(splitter.getAttribute("aria-valuenow")).toBe("70");
     fireEvent.keyDown(splitter, { key: "ArrowLeft" });
     expect(localStorage.getItem("agenthub.chat.panelWidth")).toBe("0.35");

client/src/components/run/__tests__/TerminalView.test.tsx

@@ -527,18 +527,30 @@ describe("TerminalView", () => {
   });
 
   // Ctrl+V is how claude pastes an image; xterm would send ^V, and claude then
-  // reads the SERVER's clipboard, not the browser's.
-  it("lets Ctrl+V reach the browser's paste, and still sends ^V when it holds no file", () => {
+  // reads the SERVER's clipboard, not the browser's. Text on the clipboard is
+  // pasted like any browser paste: sending ^V for it pasted nothing.
+  it("lets Ctrl+V reach the browser's paste: text is pasted, ^V only for an empty clipboard", () => {
+    vi.useFakeTimers();
     inputMock.mockClear();
-    pasteMock.mockClear();
     render(<TerminalView runId="r1" wsBaseUrl="ws://x" />);
     const handler = keyHandlerMock.mock.calls[0]![0]! as (e: KeyboardEvent) => boolean;
     expect(handler(new KeyboardEvent("keydown", { key: "v", ctrlKey: true }))).toBe(false);
+    // Not cancelled: xterm's own paste listener takes the text.
+    expect(
+      fireEvent.paste(screen.getByTestId("terminal-view"), {
+        clipboardData: { files: [], getData: () => "some text" },
+      })
+    ).toBe(true);
+    vi.advanceTimersByTime(1);
+    expect(inputMock).not.toHaveBeenCalled();
+    handler(new KeyboardEvent("keydown", { key: "v", ctrlKey: true }));
     fireEvent.paste(screen.getByTestId("terminal-view"), {
-      clipboardData: { files: [], getData: () => "some text" },
+      clipboardData: { files: [], getData: () => "" },
     });
+    vi.advanceTimersByTime(1);
+    expect(inputMock).toHaveBeenCalledTimes(1);
     expect(inputMock).toHaveBeenCalledWith("\x16", true);
-    expect(pasteMock).not.toHaveBeenCalled();
+    vi.useRealTimers();
   });
 
   it("sends ^V on Ctrl+V when the browser fires no paste at all", () => {

client/src/i18n/locales/en/sessions.json

@@ -209,6 +209,7 @@
     "moreSteps_one": "+{{count}} more",
     "moreSteps_other": "+{{count}} more",
     "quiet": "no new activity for {{time}}",
+    "askDeclined": "Declined — no answer given",
     "turnDone": "Done in {{time}}"
   },
   "overview": {

client/src/i18n/locales/vi/sessions.json

@@ -209,6 +209,7 @@
     "moreSteps_one": "+{{count}} việc nữa",
     "moreSteps_other": "+{{count}} việc nữa",
     "quiet": "không có hoạt động mới trong {{time}}",
+    "askDeclined": "Đã từ chối — không trả lời",
     "turnDone": "Xong trong {{time}}"
   },
   "overview": {

client/src/lib/types.ts

@@ -3213,6 +3213,9 @@ export interface TranscriptContent {
   /** Present on a Write/Edit/MultiEdit "tool_result": the lines it added and
    *  removed (from Claude Code's `structuredPatch`). */
   diff?: { added: number; removed: number };
+  /** Present on an AskUserQuestion "tool_result": question text → the option
+   *  the user picked, or what they typed. */
+  answers?: Record<string, string>;
 }
 
 /** Who actually sent a transcript message. A JSONL `type:"user"` line can be the

client/src/pages/__tests__/__snapshots__/screens.snapshot.test.tsx.snap

@@ -7178,7 +7178,7 @@ exports[`screen snapshots > Settings — about 1`] = `
                 <span
                   class="text-status-warning/90"
                 >
-                  UI build v2.7.4 (rebuild client to match)
+                  UI build v2.7.5 (rebuild client to match)
                 </span>
               </p>
             </div>

openapi.yaml

@@ -3,7 +3,7 @@
 openapi: 3.0.3
 info:
   title: Agent Dashboard for Claude Code API
-  version: 2.7.4
+  version: 2.7.5
   description: HTTP API for real-time Claude Code session monitoring, agent lifecycle tracking, analytics, pricing, hooks ingestion, and workflow intelligence.
   contact:
     name: Nguyễn Ngọc Trí Vĩ

package-lock.json

@@ -1,12 +1,12 @@
 {
   "name": "agenthub",
-  "version": "2.7.4",
+  "version": "2.7.5",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "agenthub",
-      "version": "2.7.4",
+      "version": "2.7.5",
       "hasInstallScript": true,
       "license": "UNLICENSED",
       "dependencies": {

package.json

@@ -1,6 +1,6 @@
 {
   "name": "agenthub",
-  "version": "2.7.4",
+  "version": "2.7.5",
   "bin": {
     "agenthub": "bin/agenthub.js"
   },

server/__tests__/transcript-sender.test.js

@@ -18,7 +18,7 @@ const fs = require("node:fs");
 const TEST_DB = path.join(os.tmpdir(), `transcript-sender-${Date.now()}-${process.pid}.db`);
 process.env.DASHBOARD_DB_PATH = TEST_DB;
 
-const { classifyTranscriptSender, patchCounts } = require("../routes/sessions");
+const { classifyTranscriptSender, patchCounts, askAnswers } = require("../routes/sessions");
 const { db } = require("../db");
 
 after(() => {
@@ -168,3 +168,22 @@ describe("patchCounts — the +/- lines of a Write/Edit result", () => {
     assert.equal(patchCounts(undefined), null);
   });
 });
+
+describe("askAnswers — what the user picked or typed for an AskUserQuestion", () => {
+  it("returns each question's answer", () => {
+    const r = {
+      questions: [{ question: "Commit?", options: [{ label: "Yes" }, { label: "No" }] }],
+      answers: { "Commit?": "Yes", "Notes?": "typed text" },
+      annotations: {},
+    };
+    assert.deepEqual(askAnswers(r), { "Commit?": "Yes", "Notes?": "typed text" });
+  });
+
+  it("drops a non-string answer and answers null for a result that is no question", () => {
+    assert.deepEqual(askAnswers({ answers: { a: "x", b: 3 } }), { a: "x" });
+    assert.equal(askAnswers({ answers: {} }), null);
+    assert.equal(askAnswers({ answers: ["x"] }), null);
+    assert.equal(askAnswers({ stdout: "ok" }), null);
+    assert.equal(askAnswers("text"), null);
+  });
+});

server/routes/sessions.js

@@ -1029,6 +1029,7 @@ router.get("/:id/transcript", async (req, res) => {
             } else if (block.type === "tool_result") {
               // A line carries one tool result; an edit's +/- counts ride on it.
               const diff = patchCounts(entry.toolUseResult);
+              const answers = askAnswers(entry.toolUseResult);
               content.push({
                 type: "tool_result",
                 id: block.tool_use_id || null,
@@ -1040,6 +1041,7 @@ router.get("/:id/transcript", async (req, res) => {
                 ),
                 is_error: !!block.is_error,
                 ...(diff ? { diff } : {}),
+                ...(answers ? { answers } : {}),
               });
             }
           }
@@ -1330,6 +1332,16 @@ function patchCounts(r) {
   return null;
 }
 
+/** An AskUserQuestion result's answers (`toolUseResult.answers`): question
+ *  text → the option picked, or what the user typed. Null for any other result. */
+function askAnswers(r) {
+  if (!r || typeof r !== "object" || !r.answers || typeof r.answers !== "object") return null;
+  if (Array.isArray(r.answers)) return null;
+  const out = {};
+  for (const [q, a] of Object.entries(r.answers)) if (typeof a === "string") out[q] = a;
+  return Object.keys(out).length ? out : null;
+}
+
 // A conversation turn — a prompt, an answer — is shown whole: it is bounded by
 // the model's output and what the user typed. Only tool payloads and thinking
 // (the parts that can run to megabytes) keep the 10 KB cap. This one guards
@@ -1353,3 +1365,4 @@ module.exports.sessionStats = sessionStats;
 // Exported for unit tests — sender attribution is correctness-critical.
 module.exports.classifyTranscriptSender = classifyTranscriptSender;
 module.exports.patchCounts = patchCounts;
+module.exports.askAnswers = askAnswers;