fix(auxiliary): propagate OpenCode session header

agent/auxiliary_client.py

@@ -2921,7 +2921,9 @@ def _try_anthropic(explicit_api_key: str = None) -> Tuple[Optional[Any], Optiona
 
 
 _MAIN_RUNTIME_FIELDS = ("provider", "model", "base_url", "api_key", "api_mode", "auth_mode")
-_MAIN_RUNTIME_CONTEXT_FIELDS = _MAIN_RUNTIME_FIELDS + ("requested_provider",)
+_MAIN_RUNTIME_CONTEXT_FIELDS = _MAIN_RUNTIME_FIELDS + (
+    "requested_provider", "session_id", "cache_scope",
+)
 
 
 def _normalize_main_runtime(main_runtime: Optional[Dict[str, Any]]) -> Dict[str, Any]:
@@ -7408,6 +7410,7 @@ def call_llm(
     prior_progress_hook = getattr(_aux_progress, "hook", None)
     try:
         with (
+            scoped_runtime_main(main_runtime),
             aux_progress_hook(
                 prior_progress_hook
                 if callable(prior_progress_hook)
@@ -7697,12 +7700,13 @@ async def async_call_llm(
     if semaphore is not None:
         await semaphore.acquire()
     try:
-        return await _async_call_llm_impl(
-            task=task, provider=provider, model=model, base_url=base_url, api_key=api_key,
-            main_runtime=main_runtime, messages=messages, temperature=temperature,
-            max_tokens=max_tokens, tools=tools, timeout=timeout, extra_body=extra_body,
-            reasoning_config=reasoning_config, route_info=route_info,
-        )
+        with scoped_runtime_main(main_runtime):
+            return await _async_call_llm_impl(
+                task=task, provider=provider, model=model, base_url=base_url, api_key=api_key,
+                main_runtime=main_runtime, messages=messages, temperature=temperature,
+                max_tokens=max_tokens, tools=tools, timeout=timeout, extra_body=extra_body,
+                reasoning_config=reasoning_config, route_info=route_info,
+            )
     finally:
         if semaphore is not None:
             semaphore.release()

run_agent.py

@@ -505,7 +505,10 @@ class AIAgent(
 
     def _current_main_runtime(self) -> Dict[str, str]:
         """Return the live main runtime for session-scoped auxiliary routing."""
-        return {key: getattr(self, key, "") or "" for key in ("model", "provider", "base_url", "api_key", "api_mode", "auth_mode")}
+        return {
+            key: getattr(self, key, "") or ""
+            for key in ("model", "provider", "base_url", "api_key", "api_mode", "auth_mode", "session_id")
+        }
 
     _check_compression_model_feasibility = _forward("agent.conversation_compression", "check_compression_model_feasibility")
     _replay_compression_warning = _forward("agent.conversation_compression", "replay_compression_warning")

tests/agent/test_opencode_session_affinity.py

@@ -2,6 +2,8 @@
 
 from __future__ import annotations
 
+from types import SimpleNamespace
+
 import pytest
 
 from agent import auxiliary_client as aux
@@ -60,3 +62,36 @@ def test_auxiliary_calls_share_the_main_turn_session_key():
         assert "x-opencode-session" not in (other.get("extra_headers") or {})
     finally:
         aux._RUNTIME_MAIN_CONTEXT.reset(token)
+
+
+def test_auxiliary_call_uses_explicit_main_runtime_session(monkeypatch):
+    captured = {}
+
+    class Completions:
+        def create(self, **kwargs):
+            captured.update(kwargs)
+            return SimpleNamespace(
+                choices=[SimpleNamespace(message=SimpleNamespace(content="ok"))],
+                model="glm-5",
+            )
+
+    client = SimpleNamespace(
+        chat=SimpleNamespace(completions=Completions()),
+        base_url="https://opencode.ai/zen/v1",
+    )
+    monkeypatch.setattr(
+        aux,
+        "_resolve_task_provider_model",
+        lambda *_args, **_kwargs: (
+            "opencode-zen", "glm-5", "https://opencode.ai/zen/v1", "test-key", None,
+        ),
+    )
+    monkeypatch.setattr(aux, "_get_cached_client", lambda *_args, **_kwargs: (client, "glm-5"))
+
+    aux.call_llm(
+        task="title_generation",
+        messages=_MSGS,
+        main_runtime={"session_id": "sess-explicit-2"},
+    )
+
+    assert captured["extra_headers"]["x-opencode-session"] == "sess-explicit-2"