fix(tui): preserve session when switching personality

Previously, /personality in the TUI called _reset_session_agent() which
destroyed the agent, cleared conversation history, and effectively started
a new session. This made personality switching disruptive — users lost
their entire conversation context.

Now /personality updates the agent's ephemeral_system_prompt in-place and
injects a pivot marker into the conversation history. The marker tells
the model to adopt the new persona from that point forward, which is
necessary because LLMs tend to pattern-match their prior responses and
continue the established tone without an explicit signal.

Changes:
- tui_gateway/server.py: Rewrite _apply_personality_to_session to update
  the agent in-place instead of resetting. Inject a user-role pivot
  marker so the model actually switches style mid-conversation.
- ui-tui/src/app/slash/commands/session.ts: Update help text (no longer
  mentions history reset).
- tests/test_tui_gateway_server.py: Update test to verify history is
  preserved, pivot marker is injected, and ephemeral prompt is set.

tests/test_tui_gateway_server.py

@@ -1559,13 +1559,15 @@ def test_config_set_personality_rejects_unknown_name(monkeypatch):
     assert "Unknown personality" in resp["error"]["message"]
 
 
-def test_config_set_personality_resets_history_and_returns_info(monkeypatch):
+def test_config_set_personality_preserves_history_and_returns_info(monkeypatch):
+    agent = types.SimpleNamespace(
+        ephemeral_system_prompt=None, _cached_system_prompt="old"
+    )
     session = _session(
-        agent=types.SimpleNamespace(),
+        agent=agent,
         history=[{"role": "user", "text": "hi"}],
         history_version=4,
     )
-    new_agent = types.SimpleNamespace(model="x")
     emits = []
 
     server._sessions["sid"] = session
@@ -1574,13 +1576,9 @@ def test_config_set_personality_resets_history_and_returns_info(monkeypatch):
         "_available_personalities",
         lambda cfg=None: {"helpful": "You are helpful."},
     )
-    monkeypatch.setattr(
-        server, "_make_agent", lambda sid, key, session_id=None: new_agent
-    )
     monkeypatch.setattr(
         server, "_session_info", lambda agent: {"model": getattr(agent, "model", "?")}
     )
-    monkeypatch.setattr(server, "_restart_slash_worker", lambda session: None)
     monkeypatch.setattr(server, "_emit", lambda *args: emits.append(args))
     monkeypatch.setattr(server, "_write_config_key", lambda path, value: None)
 
@@ -1592,11 +1590,19 @@ def test_config_set_personality_resets_history_and_returns_info(monkeypatch):
         }
     )
 
-    assert resp["result"]["history_reset"] is True
+    assert resp["result"]["history_reset"] is False
-    assert resp["result"]["info"] == {"model": "x"}
+    assert resp["result"]["info"] == {"model": "?"}
-    assert session["history"] == []
+    # History is preserved with a pivot marker appended
+    assert len(session["history"]) == 2
+    assert session["history"][0] == {"role": "user", "text": "hi"}
+    assert session["history"][1]["role"] == "user"
+    assert "personality" in session["history"][1]["content"].lower()
+    assert "You are helpful." in session["history"][1]["content"]
     assert session["history_version"] == 5
-    assert ("session.info", "sid", {"model": "x"}) in emits
+    # Agent's system prompt was updated in-place; cached prompt untouched
+    assert agent.ephemeral_system_prompt == "You are helpful."
+    assert agent._cached_system_prompt == "old"
+    assert ("session.info", "sid", {"model": "?"}) in emits
 
 
 def test_session_compress_uses_compress_helper(monkeypatch):

tui_gateway/server.py

@@ -1680,21 +1680,46 @@ def _validate_personality(value: str, cfg: dict | None = None) -> tuple[str, str
 def _apply_personality_to_session(
     sid: str, session: dict, new_prompt: str
 ) -> tuple[bool, dict | None]:
+    """Apply a personality change to an existing session without resetting history.
+
+    Updates the agent's ephemeral system prompt in-place so the new personality
+    takes effect on the next turn.  The cached base system prompt is left intact
+    (ephemeral_system_prompt is appended at API-call time, not baked into the
+    cache), which preserves prompt-cache hits.
+
+    Also injects a system-role marker into the conversation history so the model
+    knows to pivot its style from this point forward (without this, LLMs tend to
+    continue the tone established by earlier messages in the transcript).
+
+    Returns (history_reset, info) — history_reset is always False since we
+    preserve the conversation.
+    """
     if not session:
         return False, None
 
-    try:
+    agent = session.get("agent")
-        info = _reset_session_agent(sid, session)
+    if agent:
-        return True, info
+        agent.ephemeral_system_prompt = new_prompt or None
-    except Exception:
+        # Inject a pivot marker into history so the model sees the change point.
-        if session.get("agent"):
+        # This prevents it from pattern-matching its prior style.
-            agent = session["agent"]
+        if new_prompt:
-            agent.ephemeral_system_prompt = new_prompt or None
+            marker = (
-            agent._cached_system_prompt = None
+                "[System: The user has changed the assistant's personality. "
-            info = _session_info(agent)
+                "From this point forward, adopt the following persona and respond "
-            _emit("session.info", sid, info)
+                f"accordingly: {new_prompt}]"
-            return False, info
+            )
-        return False, None
+        else:
+            marker = (
+                "[System: The user has cleared the personality overlay. "
+                "From this point forward, respond in your normal default style.]"
+            )
+        with session["history_lock"]:
+            session["history"].append({"role": "user", "content": marker})
+            session["history_version"] = int(session.get("history_version", 0)) + 1
+        info = _session_info(agent)
+        _emit("session.info", sid, info)
+        return False, info
+    return False, None
 
 
 def _cfg_max_turns(cfg: dict, default: int) -> int:

ui-tui/src/app/slash/commands/session.ts

@@ -109,7 +109,7 @@ export const sessionCommands: SlashCommand[] = [
   },
 
   {
-    help: 'switch or reset personality (history reset on set)',
+    help: 'switch personality for this session',
     name: 'personality',
     run: (arg, ctx) => {
       if (!arg) {