refactor: use mark_dirty() method instead of direct attribute assignment

agent.py

@@ -98,7 +98,7 @@ def _append_tool_result(
             "content": content,
         }
     )
-    session.history_dirty = True
+    session.mark_dirty()
 
 
 def _turn_state(user_msg: Optional[Any], session: Session, memory_service: Any) -> dict:
@@ -155,7 +155,7 @@ def _turn_state(user_msg: Optional[Any], session: Session, memory_service: Any)
             state["experience_update_start_index"] = len(session.message_history)
 
         session.message_history.append({"role": "user", "content": user_msg})
-        session.history_dirty = True
+        session.mark_dirty()
         log_agent(session.user_id, "IN", user_msg, Fore.CYAN)
 
         if _is_gui_interaction_request(user_msg):
@@ -316,7 +316,7 @@ async def run_and_store(name=func_name, arguments=args, call_id=tc_id):
         results = await asyncio.gather(*tasks)
         for r in results:
             session.message_history.append(r)
-        session.history_dirty = True
+        session.mark_dirty()
 
     if session.pending_confirmations:
         if signal_handler:
@@ -582,7 +582,7 @@ def _append_assistant_round(session: Session, round_result: dict) -> bool:
         msg_dict["tool_calls"] = valid_tool_calls
 
     session.message_history.append(msg_dict)
-    session.history_dirty = True
+    session.mark_dirty()
     return bool(valid_tool_calls)
 
 
@@ -616,7 +616,7 @@ async def _finalize_or_request_more_work(
                 "If `validation_pytest` is missing, run at least one `pytest` command via `run_bash`."
             )
             session.message_history.append({"role": "user", "content": guidance})
-            session.history_dirty = True
+            session.mark_dirty()
             return None
 
     if turn_state["experience_update_active"]:
@@ -630,7 +630,7 @@ async def _finalize_or_request_more_work(
                 "Then provide confirmation that the experience was actually stored."
             )
             session.message_history.append({"role": "user", "content": guidance})
-            session.history_dirty = True
+            session.mark_dirty()
             return None
 
     if memory_service and user_msg:
@@ -777,7 +777,7 @@ async def resolve_confirmations(
                     "content": result,
                 }
             )
-    session.history_dirty = True
+    session.mark_dirty()
 
     # After resolving, run the agent turn again to process results
     return await run_agent_turn(None, session, signal_handler=signal_handler)
@@ -814,7 +814,7 @@ async def deny_confirmations(session: Session, deny_all: bool = True) -> None:
                     "content": "Action denied by user.",
                 }
             )
-    session.history_dirty = True
+    session.mark_dirty()
 
 
 def main():

bot.py

@@ -529,7 +529,7 @@ async def cancel(update: Update, context: ContextTypes.DEFAULT_TYPE):
                     "content": "Action denied by user.",
                 }
             )
-        session.history_dirty = True
+        session.mark_dirty()
         count = len(session.pending_confirmations)
         session.pending_confirmations = []
         session_manager.save(user_id)

cli.py

@@ -152,7 +152,7 @@ async def cli_signal_handler(msg: str):
                             "content": result
                         })
 
-                    session.history_dirty = True
+                    session.mark_dirty()
                     session_manager.save(user_id)
                     is_resume = True # Continue the same turn
                     continue
@@ -180,7 +180,7 @@ async def cli_signal_handler(msg: str):
                             "name": e.action,
                             "content": "Action denied by user."
                         })
-                    session.history_dirty = True
+                    session.mark_dirty()
                     session_manager.save(user_id)
                     is_resume = True
                     continue

commit_message.txt

@@ -0,0 +1,3 @@
+refactor: use mark_dirty() method instead of direct attribute assignment
+
+Replaces all direct assignments to `session.history_dirty = True` with the encapsulated `session.mark_dirty()` method call to properly enforce the performance semantics documented in the Session dataclass.

desktop/api_bridge.py

@@ -633,7 +633,7 @@ async def handle_command(request: web.Request) -> web.Response:
                         })
                     count = len(session.pending_confirmations)
                     session.pending_confirmations = []
-                    session.history_dirty = True
+                    session.mark_dirty()
                     session_manager.save(user_id)
                     result = f"✅ {count} pending action(s) cancelled."
                 else:

prompt_builder.py

@@ -255,7 +255,7 @@ async def _compact_history(session: Session, router: LLMRouter) -> None:
         new_history.extend(last_messages)
 
         session.message_history = new_history
-        session.history_dirty = True
+        session.mark_dirty()
         log_agent(
             session.user_id, "COMPACT", "History successfully compacted.", Fore.GREEN
         )
@@ -397,7 +397,7 @@ def _inject_system_directive(session: Session, directive: str) -> None:
     if directive not in content:
         session.message_history[0]["content"] = content + "\n\n" + directive
         # System prompt content changed; sanitize output may differ.
-        session.history_dirty = True
+        session.mark_dirty()
 
 
 def _extract_memory_lines(results: Any, limit: int = 6) -> List[str]:
@@ -547,7 +547,7 @@ def _sync_basic_facts_into_system_prompt(
     )
     if updated != content:
         session.message_history[0]["content"] = updated
-        session.history_dirty = True
+        session.mark_dirty()
 
 
 def _build_memory_context(
@@ -874,7 +874,7 @@ def _merge_memory_context_into_system_prompt(
 
     if session.message_history[0].get("role") != "system":
         session.message_history.insert(0, get_initial_messages()[0])
-        session.history_dirty = True
+        session.mark_dirty()
 
     base_content = str(session.message_history[0].get("content") or "")
 
@@ -890,7 +890,7 @@ def _merge_memory_context_into_system_prompt(
     )
     if merged != str(session.message_history[0].get("content") or ""):
         session.message_history[0]["content"] = merged
-        session.history_dirty = True
+        session.mark_dirty()
 
 
 def _normalize_single_system_message(session: Session) -> None:
@@ -951,7 +951,7 @@ def _normalize_single_system_message(session: Session) -> None:
 
     if changed:
         session.message_history = normalized
-        session.history_dirty = True
+        session.mark_dirty()
 
 
 def _extract_tool_path(args: dict) -> str: