fix(server): treat developer messages as system prompts

.github/workflows/test.yml

@@ -59,6 +59,7 @@ jobs:
         python -m pytest tests/test_bon_candidate_filtering.py -v --tb=short
         python -m pytest tests/test_compact_plugin.py -v --tb=short
         python -m pytest tests/test_completions_endpoint.py -v --tb=short
+        python -m pytest tests/test_developer_role.py -v --tb=short
         python -m pytest tests/test_entropy_decoding_sampling.py -v --tb=short
         python -m pytest tests/test_mars_agent_routing.py -v --tb=short
         python -m pytest tests/test_mars_aggregation_sampling.py -v --tb=short

optillm/server.py

@@ -677,7 +677,8 @@ def parse_conversation(messages):
         if role == 'assistant' and not text_content:
             continue
 
-        if role == 'system':
+        # 'developer' is the OpenAI replacement for 'system' on o-series and newer models
+        if role in ('system', 'developer'):
             system_prompt, optillm_approach = extract_optillm_approach(text_content)
         elif role == 'user':
             if not optillm_approach:

tests/test_developer_role.py

@@ -0,0 +1,106 @@
+"""
+Offline unit tests for `developer` role messages.
+
+OpenAI clients send `developer` messages in place of `system` messages for
+o-series and newer models. optillm used to drop them when it split a request
+into system prompt and conversation, so the instructions never reached the
+approach. Uses the Flask test client with a mocked approach (no network, no model).
+"""
+
+import sys
+import os
+sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+
+import unittest
+
+from optillm import server
+
+
+CAPTURED = {}
+
+
+def fake_chat_with_mcts(system_prompt, initial_query, client, model,
+                        num_simulations, exploration_weight, depth,
+                        request_config=None, request_id=None):
+    CAPTURED['system_prompt'] = system_prompt
+    CAPTURED['initial_query'] = initial_query
+    return "ok", 1
+
+
+class TestParseConversationDeveloperRole(unittest.TestCase):
+
+    def test_developer_message_becomes_system_prompt(self):
+        messages = [
+            {"role": "developer", "content": "Answer in French."},
+            {"role": "user", "content": "Hello"},
+        ]
+        system_prompt, initial_query, approach = server.parse_conversation(messages)
+        self.assertEqual(system_prompt, "Answer in French.")
+        self.assertEqual(initial_query, "User: Hello")
+        self.assertIsNone(approach)
+
+    def test_developer_message_with_content_parts(self):
+        messages = [
+            {"role": "developer", "content": [{"type": "text", "text": "Answer in French."}]},
+            {"role": "user", "content": "Hello"},
+        ]
+        system_prompt, _, _ = server.parse_conversation(messages)
+        self.assertEqual(system_prompt, "Answer in French.")
+
+    def test_developer_message_can_select_approach(self):
+        messages = [
+            {"role": "developer", "content": "<optillm_approach>re2</optillm_approach> Be brief."},
+            {"role": "user", "content": "Hello"},
+        ]
+        system_prompt, _, approach = server.parse_conversation(messages)
+        self.assertEqual(system_prompt, "Be brief.")
+        self.assertEqual(approach, "re2")
+
+    def test_system_message_unchanged(self):
+        messages = [
+            {"role": "system", "content": "Answer in French."},
+            {"role": "user", "content": "Hello"},
+        ]
+        system_prompt, initial_query, _ = server.parse_conversation(messages)
+        self.assertEqual(system_prompt, "Answer in French.")
+        self.assertEqual(initial_query, "User: Hello")
+
+
+class TestDeveloperRoleReachesApproach(unittest.TestCase):
+
+    def setUp(self):
+        CAPTURED.clear()
+        self._originals = {
+            'get_config': server.get_config,
+            'chat_with_mcts': server.chat_with_mcts,
+            'config': dict(server.server_config),
+        }
+        server.get_config = lambda: (object(), 'optillm')
+        server.chat_with_mcts = fake_chat_with_mcts
+        server.server_config['optillm_api_key'] = ''
+        server.app.config['TESTING'] = True
+        self.client = server.app.test_client()
+
+    def tearDown(self):
+        server.get_config = self._originals['get_config']
+        server.chat_with_mcts = self._originals['chat_with_mcts']
+        server.server_config.clear()
+        server.server_config.update(self._originals['config'])
+
+    def test_developer_message_is_passed_to_approach(self):
+        resp = self.client.post(
+            '/v1/chat/completions',
+            json={"model": "gpt-4o-mini",
+                  "messages": [
+                      {"role": "developer", "content": "Answer in French."},
+                      {"role": "user", "content": "Hello"},
+                  ],
+                  "optillm_approach": "mcts"})
+
+        self.assertEqual(resp.status_code, 200, resp.get_data(as_text=True))
+        self.assertEqual(CAPTURED['system_prompt'], "Answer in French.")
+        self.assertEqual(CAPTURED['initial_query'], "User: Hello")
+
+
+if __name__ == '__main__':
+    unittest.main()