fix(provider): stabilize Codex prompt cache routing (#5540)

tests/agent/test_runner_fallback.py

@@ -384,12 +384,16 @@ class TestFallbackOnPrimaryError:
             messages=[{"role": "user", "content": "hi"}],
             model="gpt-5.6",
             max_tokens=10_000,
-            provider_context=ProviderCallContext(context_window_tokens=50_000),
+            provider_context=ProviderCallContext(
+                context_window_tokens=50_000,
+                session_id="webui:cache-test",
+            ),
         )
 
         primary_context = primary.context_calls[0]
         assert primary_context is not None
         assert primary_context.context_window_tokens == 200_000
+        assert primary_context.session_id == "webui:cache-test"
         assert resolve_compact_threshold(
             primary_context.context_window_tokens,
             10_000,

tests/agent/test_runner_runtime_identity.py

@@ -8,6 +8,7 @@ from nanobot.providers.base import (
     GenerationSettings,
     LLMProvider,
     LLMResponse,
+    ProviderCallContext,
     ToolCallRequest,
 )
 from nanobot.utils.llm_runtime import LLMRuntime
@@ -22,6 +23,7 @@ async def test_active_run_keeps_provider_captured_at_admission() -> None:
     first_calls = 0
     second_calls = 0
     request_temperatures: list[float] = []
+    request_session_ids: list[str | None] = []
     selected_runtime = LLMRuntime.capture(
         first_provider,
         "captured-model",
@@ -33,6 +35,9 @@ async def test_active_run_keeps_provider_captured_at_admission() -> None:
         nonlocal first_calls, selected_runtime
         first_calls += 1
         request_temperatures.append(kwargs["temperature"])
+        provider_context = kwargs["provider_context"]
+        assert isinstance(provider_context, ProviderCallContext)
+        request_session_ids.append(provider_context.session_id)
         selected_runtime = LLMRuntime.capture(
             second_provider,
             "future-model",
@@ -63,9 +68,11 @@ async def test_active_run_keeps_provider_captured_at_admission() -> None:
         runtime=selected_runtime,
         max_iterations=2,
         max_tool_result_chars=AgentDefaults().max_tool_result_chars,
+        session_key="webui:cache-test",
     ))
 
     assert first_calls == 2
     assert second_calls == 0
     assert request_temperatures == [0.2, 0.2]
+    assert request_session_ids == ["webui:cache-test", "webui:cache-test"]
     assert selected_runtime.provider is second_provider

tests/providers/test_conversation_state.py

@@ -9,6 +9,7 @@ import pytest
 from nanobot.providers.base import (
     LLMProvider,
     LLMResponse,
+    ProviderCallContext,
     ProviderConversationState,
     ToolCallRequest,
 )
@@ -289,3 +290,23 @@ def test_independent_request_exposes_context_without_capability_check() -> None:
     assert provider_context.conversation_state is None
     assert provider_context.context_window_tokens == 200_000
     provider.supports_native_compaction.assert_not_called()
+
+
+def test_independent_request_exposes_session_id_without_token_budget() -> None:
+    provider = _provider(compact=False)
+    messages = [{"role": "user", "content": "hello"}]
+    controller = ProviderConversationStateController(
+        provider=provider,
+        model="gpt-5.6",
+        messages=messages,
+        session_id="webui:cache-test",
+    )
+
+    provider_context = controller.prepare_request(
+        messages,
+        context_window_tokens=None,
+    )
+
+    assert provider_context == ProviderCallContext(
+        session_id="webui:cache-test",
+    )

tests/providers/test_openai_codex_provider.py

@@ -260,8 +260,8 @@ async def test_codex_request_uses_configured_proxy(monkeypatch) -> None:
 
 
 @pytest.mark.asyncio
-async def test_codex_prompt_cache_key_uses_stable_conversation_prefix(monkeypatch) -> None:
-    bodies: list[dict] = []
+async def test_codex_omits_prompt_cache_key_without_session_id(monkeypatch) -> None:
+    bodies: list[dict[str, Any]] = []
 
     _mock_codex_token(monkeypatch)
 
@@ -289,25 +289,40 @@ async def test_codex_prompt_cache_key_uses_stable_conversation_prefix(monkeypatc
             {"role": "assistant", "content": "first answer"},
         ],
     )
-    await provider.chat(
-        [
-            {"role": "system", "content": "You are nanobot."},
-            {"role": "user", "content": "first request"},
-            {"role": "assistant", "content": "first answer"},
-            {"role": "user", "content": "follow up"},
-        ],
-    )
-    await provider.chat(
-        [
-            {"role": "system", "content": "You are nanobot."},
-            {"role": "user", "content": "different request"},
-            {"role": "assistant", "content": "first answer"},
-        ],
-    )
+
+    assert "prompt_cache_key" not in bodies[0]
+    assert "service_tier" not in bodies[0]
+
+
+@pytest.mark.asyncio
+async def test_codex_prompt_cache_key_prefers_stable_session_id(monkeypatch) -> None:
+    bodies: list[dict[str, Any]] = []
+    _mock_codex_token(monkeypatch)
+
+    async def fake_request(_url, _headers, body, **_kwargs):
+        bodies.append(body)
+        return provider_base.LLMResponse(content="ok")
+
+    monkeypatch.setattr("nanobot.providers.openai_codex_provider._request_codex", fake_request)
+    provider = OpenAICodexProvider()
+
+    for session_id, first_request in (
+        ("session-a", "first request"),
+        ("session-a", "different visible prefix"),
+        ("session-b", "first request"),
+    ):
+        await provider.chat(
+            [
+                {"role": "system", "content": "You are nanobot."},
+                {"role": "user", "content": first_request},
+            ],
+            provider_context=provider_base.ProviderCallContext(
+                session_id=session_id,
+            ),
+        )
 
     assert bodies[0]["prompt_cache_key"] == bodies[1]["prompt_cache_key"]
     assert bodies[0]["prompt_cache_key"] != bodies[2]["prompt_cache_key"]
-    assert all("service_tier" not in body for body in bodies)
 
 
 @pytest.mark.asyncio

tests/providers/test_provider_retry.py

@@ -431,13 +431,17 @@ async def test_image_retry_discards_provider_state_with_images(
 
     response = await provider.chat_with_retry(
         messages=messages,
-        provider_context=ProviderCallContext(conversation_state=state),
+        provider_context=ProviderCallContext(
+            conversation_state=state,
+            session_id="webui:cache-test",
+        ),
     )
 
     assert response.content == "ok, no image"
     retry_context = provider.contexts[-1]
     assert isinstance(retry_context, ProviderCallContext)
     assert retry_context.conversation_state is None
+    assert retry_context.session_id == "webui:cache-test"
     public_content = messages[0]["content"]
     if isinstance(public_content, list):
         assert all(block.get("type") != "image_url" for block in public_content)