fix(desktop): keep clarify answerable across reconnect/hydration + tool-progress off (#69795)

* fix(desktop): keep clarify lifecycle when tool progress is off

* fix(desktop): render clarify prompt from the request event

Re-authored onto the current use-message-stream/gateway-event.ts (the
original patched the pre-split use-message-stream.ts). When the tool.start
row that normally mounts the inline clarify UI is missed (stream reconnect
/ hydration race), upsert a stable pending clarify tool row from
clarify.request itself so the prompt stays answerable; a real
tool.start/complete with the same request id merges rather than duplicates.

Co-authored-by: 정수환 <centerid@naver.com>

* chore(contributors): map centerid@naver.com -> lidises

Attribution mapping for the salvaged #47544 commit.

* fix(desktop): correlate clarify rows by question so hydration can't duplicate

The hydrated row (from clarify.request's request_id) and the real tool.start
row (the model's tool_call_id) have different ids, so id-only matching appended
a second clarify card in the normal path (caught by the BLOCKING_CLARIFY e2e:
'question' resolved to 2 elements). Add 'question' to the tool match-value keys
so a clarify upsert merges into the existing pending clarify row regardless of
id (same request<->args correlation ClarifyToolPending already uses); when no
row exists yet (reconnect/hydration) it still creates one.

---------

Co-authored-by: 정수환 <centerid@naver.com>

apps/desktop/src/app/session/hooks/use-message-stream/clarify-hydration.test.tsx

@@ -0,0 +1,116 @@
+import { QueryClient } from '@tanstack/react-query'
+import { act, cleanup, render, waitFor } from '@testing-library/react'
+import { type MutableRefObject, useEffect, useRef } from 'react'
+import { afterEach, beforeEach, describe, expect, it, vi } from 'vitest'
+
+import type { ClientSessionState } from '@/app/types'
+import { createClientSessionState } from '@/lib/chat-runtime'
+import { clearClarifyRequest } from '@/store/clarify'
+import type { RpcEvent } from '@/types/hermes'
+
+import { useMessageStream } from './index'
+
+// A `clarify.request` must leave an answerable inline row even when the
+// `tool.start` that normally mounts it was missed (stream reconnect /
+// hydration race). Without it the sidebar says "needs input" but the
+// transcript has nowhere to render the choices, so the agent blocks forever.
+
+const SID = 'session-1'
+
+let handleEvent: ((event: RpcEvent) => void) | null = null
+let stateRef: MutableRefObject<Map<string, ClientSessionState>> | null = null
+
+function Harness() {
+  const activeSessionIdRef = useRef<string | null>(SID)
+  const sessionStateByRuntimeIdRef = useRef(new Map<string, ClientSessionState>())
+  const queryClientRef = useRef(new QueryClient())
+
+  const stream = useMessageStream({
+    activeSessionIdRef,
+    hydrateFromStoredSession: vi.fn(async () => undefined),
+    queryClient: queryClientRef.current,
+    refreshHermesConfig: vi.fn(async () => undefined),
+    refreshSessions: vi.fn(async () => undefined),
+    sessionStateByRuntimeIdRef,
+    updateSessionState: (sessionId, updater) => {
+      const current = sessionStateByRuntimeIdRef.current.get(sessionId) ?? createClientSessionState()
+      const next = updater(current)
+      sessionStateByRuntimeIdRef.current.set(sessionId, next)
+
+      return next
+    }
+  })
+
+  useEffect(() => {
+    handleEvent = stream.handleGatewayEvent
+    stateRef = sessionStateByRuntimeIdRef
+  }, [stream.handleGatewayEvent])
+
+  return null
+}
+
+async function mountStream() {
+  render(<Harness />)
+  await waitFor(() => expect(handleEvent).not.toBeNull())
+}
+
+const clarifyRequest = (payload: Record<string, unknown>) =>
+  act(() => handleEvent!({ payload, session_id: SID, type: 'clarify.request' }))
+
+const toolStart = (payload: Record<string, unknown>) =>
+  act(() => handleEvent!({ payload, session_id: SID, type: 'tool.start' }))
+
+function clarifyParts() {
+  const messages = stateRef?.current.get(SID)?.messages ?? []
+
+  return messages.flatMap(m => m.parts).filter(p => p.type === 'tool-call' && p.toolName === 'clarify')
+}
+
+describe('clarify.request stream hydration', () => {
+  beforeEach(() => {
+    handleEvent = null
+    stateRef = null
+    clearClarifyRequest()
+  })
+
+  afterEach(() => {
+    cleanup()
+    clearClarifyRequest()
+    vi.restoreAllMocks()
+  })
+
+  it('mounts an answerable clarify row when the tool.start row was missed', async () => {
+    await mountStream()
+
+    clarifyRequest({ choices: ['yes', 'no'], question: 'Ship it?', request_id: 'req-1' })
+
+    const parts = clarifyParts()
+    expect(parts).toHaveLength(1)
+    expect(parts[0].type === 'tool-call' && parts[0].toolCallId).toBe('req-1')
+    expect(parts[0].type === 'tool-call' && parts[0].args).toMatchObject({
+      choices: ['yes', 'no'],
+      question: 'Ship it?'
+    })
+  })
+
+  it('merges with the real tool.start row even though its id differs from the request id', async () => {
+    await mountStream()
+
+    // Reality: tool.start carries the model's tool_call_id, clarify.request a
+    // separately-generated request_id. They must still collapse to ONE card
+    // (correlated by question), not two.
+    toolStart({ args: { choices: ['a'], question: 'Pick' }, name: 'clarify', tool_id: 'call-abc' })
+    clarifyRequest({ choices: ['a'], question: 'Pick', request_id: 'req-2' })
+
+    expect(clarifyParts()).toHaveLength(1)
+  })
+
+  it('does not duplicate when clarify.request arrives before the tool.start row', async () => {
+    await mountStream()
+
+    clarifyRequest({ choices: ['a'], question: 'Pick', request_id: 'req-3' })
+    toolStart({ args: { choices: ['a'], question: 'Pick' }, name: 'clarify', tool_id: 'call-xyz' })
+
+    expect(clarifyParts()).toHaveLength(1)
+  })
+})

apps/desktop/src/app/session/hooks/use-message-stream/gateway-event.ts

@@ -679,19 +679,30 @@ export function useGatewayEventHandler(deps: GatewayEventDeps) {
         const question = typeof payload?.question === 'string' ? payload.question : ''
 
         if (requestId && question) {
+          const choices = Array.isArray(payload?.choices) ? payload!.choices!.filter(c => typeof c === 'string') : null
+
           setClarifyRequest({
             requestId,
             question,
-            choices: Array.isArray(payload?.choices) ? payload!.choices!.filter(c => typeof c === 'string') : null,
+            choices,
             sessionId: sessionId ?? null
           })
 
-          // The transcript only renders the active session, so a background
-          // clarify is otherwise invisible (the row just keeps spinning like
-          // it's working). Flag the session so the sidebar shows a persistent
-          // "needs input" indicator on its row — works for the active session
-          // too, and survives alt-tab / window blur (unlike a toast).
           if (sessionId) {
+            // `clarify.request` is the blocking event the Python side waits on,
+            // while the inline UI normally mounts from the earlier `tool.start`
+            // row. If that row was missed (stream reconnect / hydration race) the
+            // sidebar still says "needs input" but there is nowhere to render the
+            // choices. Upsert a stable pending clarify tool row from the request
+            // itself so the prompt stays answerable; a real tool.start/complete
+            // with the same request id merges rather than duplicates.
+            upsertToolCall(sessionId, { args: { choices, question }, name: 'clarify', tool_id: requestId }, 'running')
+
+            // The transcript only renders the active session, so a background
+            // clarify is otherwise invisible (the row just keeps spinning like
+            // it's working). Flag the session so the sidebar shows a persistent
+            // "needs input" indicator on its row — works for the active session
+            // too, and survives alt-tab / window blur (unlike a toast).
             updateSessionState(sessionId, state => ({ ...state, needsInput: true }))
           }
 

apps/desktop/src/lib/chat-messages.ts

@@ -356,7 +356,11 @@ function collectToolMatchValues(query: string, context: string, preview: string)
 
 function toolPayloadMatchValues(payload: GatewayEventPayload | undefined): string[] {
   const payloadArgs = liveToolArgs(payload)
-  const query = firstStringField(payloadArgs, ['search_term', 'query'])
+  // `question` is clarify's identifying arg: a synthetic row hydrated from
+  // `clarify.request` (a fresh request id) must correlate with the `tool.start`
+  // row (the model's tool_call_id) so the two ids don't produce a duplicate
+  // clarify card — same correlation ClarifyToolPending uses for request↔args.
+  const query = firstStringField(payloadArgs, ['search_term', 'query', 'question'])
   const context = typeof payload?.context === 'string' ? payload.context.trim() : ''
   const preview = typeof payload?.preview === 'string' ? payload.preview.trim() : ''
 
@@ -369,7 +373,7 @@ function toolPartMatchValues(part: ChatMessagePart): string[] {
   }
 
   const args = part.args as Record<string, unknown>
-  const query = firstStringField(args, ['search_term', 'query'])
+  const query = firstStringField(args, ['search_term', 'query', 'question'])
   const context = typeof args.context === 'string' ? args.context.trim() : ''
   const preview = typeof args.preview === 'string' ? args.preview.trim() : ''
 

contributors/emails/centerid@naver.com

@@ -0,0 +1,2 @@
+lidises
+# PR salvage of #47544

tests/test_tui_gateway_server.py

@@ -865,6 +865,46 @@ def test_tui_tool_output_risk_event_exposes_metadata_without_raw_output(monkeypa
     assert "result" not in events[0][2]
 
 
+def test_tui_clarify_lifecycle_events_emit_when_tool_progress_off(monkeypatch):
+    events: list[tuple[str, str, dict]] = []
+    monkeypatch.setattr(
+        server, "_emit", lambda event_type, sid, payload: events.append((event_type, sid, payload))
+    )
+    monkeypatch.setitem(
+        server._sessions,
+        "clarify-off-test",
+        {"tool_progress_mode": "off", "tool_started_at": {}},
+    )
+
+    args = {"question": "Pick one", "choices": ["A", "B"]}
+    result = '{"question":"Pick one","choices_offered":["A","B"],"user_response":"A"}'
+
+    server._on_tool_start("clarify-off-test", "tool-clarify", "clarify", args)
+    server._on_tool_complete("clarify-off-test", "tool-clarify", "clarify", args, result)
+
+    assert [event[0] for event in events] == ["tool.start", "tool.complete"]
+    assert events[0][2]["name"] == "clarify"
+    assert events[0][2]["tool_id"] == "tool-clarify"
+    assert events[1][2]["result"]["user_response"] == "A"
+
+
+def test_tui_non_interactive_tool_lifecycle_stays_hidden_when_tool_progress_off(monkeypatch):
+    events: list[tuple[str, str, dict]] = []
+    monkeypatch.setattr(
+        server, "_emit", lambda event_type, sid, payload: events.append((event_type, sid, payload))
+    )
+    monkeypatch.setitem(
+        server._sessions,
+        "terminal-off-test",
+        {"tool_progress_mode": "off", "tool_started_at": {}},
+    )
+
+    server._on_tool_start("terminal-off-test", "tool-1", "terminal", {"command": "pwd"})
+    server._on_tool_complete("terminal-off-test", "tool-1", "terminal", {"command": "pwd"}, "done")
+
+    assert events == []
+
+
 def test_dispatch_rejects_non_object_request():
     resp = server.dispatch([])
 

tui_gateway/server.py

@@ -3299,6 +3299,14 @@ def _tool_progress_enabled(sid: str) -> bool:
     return _session_tool_progress_mode(sid) != "off"
 
 
+def _tool_lifecycle_required_for_ui(name: str) -> bool:
+    """Return True for tool events that are interactive UI, not optional chrome."""
+    # Desktop renders the clarify choices/question from the tool-call part, then
+    # wires request_id from clarify.request. If tool progress is off, suppressing
+    # clarify's lifecycle events leaves only the sidebar attention dot visible.
+    return name == "clarify"
+
+
 def _restart_slash_worker(sid: str, session: dict):
     worker = session.get("slash_worker")
     if worker:
@@ -4236,7 +4244,7 @@ def _on_tool_start(sid: str, tool_call_id: str, name: str, args: dict):
         except Exception:
             pass
         session.setdefault("tool_started_at", {})[tool_call_id] = time.time()
-    if _tool_progress_enabled(sid):
+    if _tool_progress_enabled(sid) or _tool_lifecycle_required_for_ui(name):
         payload = {
             "tool_id": tool_call_id,
             "name": name,
@@ -4294,7 +4302,7 @@ def _on_tool_complete(sid: str, tool_call_id: str, name: str, args: dict, result
             payload["inline_diff"] = "\n".join(rendered)
     except Exception:
         pass
-    if _tool_progress_enabled(sid) or payload.get("inline_diff"):
+    if _tool_progress_enabled(sid) or payload.get("inline_diff") or _tool_lifecycle_required_for_ui(name):
         _emit("tool.complete", sid, payload)