fix: remove redundant message_history prepend causing 2x token doubling

Both get_response (token counting) and _fetch_response (actual API call)
were prepending self.message_history to every request. But cli.py already
passes the full conversation history via history_context as conversation_input
to Runner.run, which threads it through as original_input to these methods.

Result: every historical message was sent TWICE in every API call, doubling
the effective context size. After auto-compact cleared message_history, the
duplication between runner-accumulated generated_items and message_history
rebuilt the doubled context within a single Runner.run invocation (after
just 3-4 tool calls), explaining why n_tokens never dropped post-compact.

Fix: remove the prepend loops. The runner's input parameter already contains
the full conversation (original_input + generated_items), so converted_messages
is built from input alone. message_history continues to serve its role as
cross-turn persistence (populated via add_to_message_history, consumed by
cli.py as history_context for the next Runner.run call).

Expected effect: halved token counts in normal operation; post-compact first
call starts at ~system_prompt + 1 user message and grows linearly with tool
calls rather than doubling.

src/cai/sdk/agents/models/openai_chatcompletions.py

@@ -544,20 +544,12 @@ class OpenAIChatCompletionsModel(Model):
             | {"base_url": str(self._get_client().base_url)},
             disabled=tracing.is_disabled(),
         ) as span_generation:
-            # Prepare the messages for consistent token counting
-            # IMPORTANT: Include existing message history for context
+            # Prepare the messages for consistent token counting.
+            # History is already included in `input` via cli.py's history_context mechanism
+            # (history_context = agent.model.message_history is passed as conversation_input
+            # to Runner.run, which then passes it as original_input to get_response).
+            # Prepending message_history here would double-count every message.
             converted_messages = []
-            
-            # First, add all existing messages from history
-            if self.message_history:
-                for msg in self.message_history:
-                    msg_copy = msg.copy()  # Use copy to avoid modifying original
-                    # Remove any existing cache_control to avoid exceeding the 4-block limit
-                    if "cache_control" in msg_copy:
-                        del msg_copy["cache_control"]
-                    converted_messages.append(msg_copy)
-            
-            # Then convert and add the new input
             new_messages = self._converter.items_to_messages(input, model_instance=self)
             converted_messages.extend(new_messages)
             
@@ -2545,19 +2537,12 @@ class OpenAIChatCompletionsModel(Model):
         # start by re-fetching self.is_ollama
         self.is_ollama = os.getenv("OLLAMA") is not None and os.getenv("OLLAMA").lower() == "true"
 
-        # IMPORTANT: Include existing message history for context
+        # Build the message list from `input` only.
+        # History is already included in `input` via cli.py's history_context mechanism:
+        # cli.py passes history_context (= message_history) as part of conversation_input
+        # to Runner.run, which passes it as original_input through to _fetch_response.
+        # Prepending message_history again would send every historical message twice.
         converted_messages = []
-        
-        # First, add all existing messages from history
-        if self.message_history:
-            for msg in self.message_history:
-                msg_copy = msg.copy()  # Use copy to avoid modifying original
-                # Remove any existing cache_control to avoid exceeding the 4-block limit
-                if "cache_control" in msg_copy:
-                    del msg_copy["cache_control"]
-                converted_messages.append(msg_copy)
-        
-        # Then convert and add the new input
         new_messages = self._converter.items_to_messages(input, model_instance=self)
         converted_messages.extend(new_messages)