Use system prompt prefix for commit messages

I've been using Qwen3 with reasoning disabled via a /no_think in the
system prompt prefix. I found that the commit message generation was
ignoring this setting. This change updates the commit message generation
loop to incorporate that setting if defined.

aider/repo.py

@@ -334,24 +334,32 @@ class GitRepo:
         content += diffs
 
         system_content = self.commit_prompt or prompts.commit_system
+
         language_instruction = ""
         if user_language:
             language_instruction = f"\n- Is written in {user_language}."
         system_content = system_content.format(language_instruction=language_instruction)
 
-        messages = [
-            dict(role="system", content=system_content),
-            dict(role="user", content=content),
-        ]
-
         commit_message = None
         for model in self.models:
             spinner_text = f"Generating commit message with {model.name}"
             with WaitingSpinner(spinner_text):
+                if model.system_prompt_prefix:
+                    current_system_content = model.system_prompt_prefix + "\n" + system_content
+                else:
+                    current_system_content = system_content
+
+                messages = [
+                    dict(role="system", content=current_system_content),
+                    dict(role="user", content=content),
+                ]
+
                 num_tokens = model.token_count(messages)
                 max_tokens = model.info.get("max_input_tokens") or 0
+
                 if max_tokens and num_tokens > max_tokens:
                     continue
+
                 commit_message = model.simple_send_with_retries(messages)
                 if commit_message:
                     break  # Found a model that could generate the message

tests/basic/test_repo.py

@@ -683,3 +683,34 @@ class TestRepo(unittest.TestCase):
             # Verify the commit was actually made
             latest_commit_msg = raw_repo.head.commit.message
             self.assertEqual(latest_commit_msg.strip(), "Should succeed")
+
+    @patch("aider.models.Model.simple_send_with_retries")
+    def test_get_commit_message_uses_system_prompt_prefix(self, mock_send):
+        """
+        Verify that GitRepo.get_commit_message() prepends the model.system_prompt_prefix
+        to the system prompt sent to the LLM.
+        """
+        mock_send.return_value = "good commit message"
+
+        prefix = "MY-CUSTOM-PREFIX"
+        model = Model("gpt-3.5-turbo")
+        model.system_prompt_prefix = prefix
+
+        with GitTemporaryDirectory():
+            repo = GitRepo(InputOutput(), None, None, models=[model])
+
+            # Call the function under test
+            repo.get_commit_message("dummy diff", "dummy context")
+
+            # Ensure the LLM was invoked once
+            mock_send.assert_called_once()
+
+            # Grab the system message sent to the model
+            messages = mock_send.call_args[0][0]
+            system_msg_content = messages[0]["content"]
+
+            # Verify the prefix is at the start of the system message
+            self.assertTrue(
+                system_msg_content.startswith(prefix),
+                "system_prompt_prefix should be prepended to the system prompt",
+            )