fix(gateway): defer background review notifications until after main reply

Background review notifications ("💾 Skill created", "💾 Memory updated")
could race ahead of the main assistant reply in chat, making it look like
the agent stopped after creating a skill.

Gate bg-review notifications behind a threading.Event + pending queue.
Register a release callback on the adapter's _post_delivery_callbacks dict
so base.py's finally block fires it after the main response is delivered.

The queued-message path in _run_agent pops and calls the callback directly
to prevent double-fire.

Co-authored-by: Hermes Agent <hermes@nousresearch.com>
Closes #10541

tests/gateway/test_run_progress_topics.py

@@ -1,5 +1,6 @@
 """Tests for topic-aware gateway progress updates."""
 
+import asyncio
 import importlib
 import sys
 import time
@@ -415,6 +416,21 @@ class QueuedCommentaryAgent:
         }
 
 
+class BackgroundReviewAgent:
+    def __init__(self, **kwargs):
+        self.background_review_callback = kwargs.get("background_review_callback")
+        self.tools = []
+
+    def run_conversation(self, message, conversation_history=None, task_id=None):
+        if self.background_review_callback:
+            self.background_review_callback("💾 Skill 'prospect-scanner' created.")
+        return {
+            "final_response": "done",
+            "messages": [],
+            "api_calls": 1,
+        }
+
+
 class VerboseAgent:
     """Agent that emits a tool call with args whose JSON exceeds 200 chars."""
     LONG_CODE = "x" * 300
@@ -668,6 +684,66 @@ async def test_run_agent_queued_message_does_not_treat_commentary_as_final(monke
     assert "final response 1" in sent_texts
 
 
+@pytest.mark.asyncio
+async def test_run_agent_defers_background_review_notification_until_release(monkeypatch, tmp_path):
+    adapter, result = await _run_with_agent(
+        monkeypatch,
+        tmp_path,
+        BackgroundReviewAgent,
+        session_id="sess-bg-review-order",
+        config_data={"display": {"interim_assistant_messages": True}},
+    )
+
+    assert result["final_response"] == "done"
+    assert adapter.sent == []
+
+
+@pytest.mark.asyncio
+async def test_base_processing_releases_post_delivery_callback_after_main_send():
+    """Post-delivery callbacks on the adapter fire after the main response."""
+    adapter = ProgressCaptureAdapter()
+
+    async def _handler(event):
+        return "done"
+
+    adapter.set_message_handler(_handler)
+
+    released = []
+
+    def _post_delivery_cb():
+        released.append(True)
+        adapter.sent.append(
+            {
+                "chat_id": "bg-review",
+                "content": "💾 Skill 'prospect-scanner' created.",
+                "reply_to": None,
+                "metadata": None,
+            }
+        )
+
+    source = SessionSource(
+        platform=Platform.TELEGRAM,
+        chat_id="-1001",
+        chat_type="group",
+        thread_id="17585",
+    )
+    event = MessageEvent(
+        text="hello",
+        message_type=MessageType.TEXT,
+        source=source,
+        message_id="msg-1",
+    )
+    session_key = "agent:main:telegram:group:-1001:17585"
+    adapter._active_sessions[session_key] = asyncio.Event()
+    adapter._post_delivery_callbacks[session_key] = _post_delivery_cb
+
+    await adapter._process_message_background(event, session_key)
+
+    sent_texts = [call["content"] for call in adapter.sent]
+    assert sent_texts == ["done", "💾 Skill 'prospect-scanner' created."]
+    assert released == [True]
+
+
 @pytest.mark.asyncio
 async def test_verbose_mode_does_not_truncate_args_by_default(monkeypatch, tmp_path):
     """Verbose mode with default tool_preview_length (0) should NOT truncate args.