Preserve completion gating and handle explicit idle stream status safely

src/agent_runtime/completion.py

@@ -309,11 +309,6 @@ def with_completion_gate(func):
                                 has_final = False
                             answer += str(data.get('delta') or '')
                         answer_events.append(data)
-                        if data.get('thinking') is True and 'delta' in data:
-                            # Stream reasoning deltas live; only the answer
-                            # prose is held for the terminal evidence review.
-                            streamed_answer_events.add(id(data))
-                            yield _event(data)
                         continue
                     yield chunk
             if provider_error and not answer_events and not metrics_events:

static/js/chat.js

@@ -6149,7 +6149,12 @@ import { invalidateSettings } from './appConfig.js';
         cache: 'no-store',
       });
       if (!_getForegroundStreamState() || _backgroundStreams.has(sid)) return;
-      if (res.status !== 404) return;
+      if (res.status !== 404 && !res.ok) return;
+      let _probe = null;
+      try { _probe = await res.json(); } catch (_) { /* An invalid response is not an idle stream. */ }
+      const _noActiveStream = res.status === 404 || (res.ok && _probe &&
+        typeof _probe.status === 'string' && _probe.status !== 'streaming');
+      if (!_noActiveStream) return;
 
       console.warn('[stream-watchdog] Local stream was stale and server has no active stream. Unlocking composer.');
       if (active.abortCtrl && !active.abortCtrl.signal.aborted) {