fix(curator): forward credential pool from runtime resolution

agent/curator.py

@@ -1850,6 +1850,8 @@ def _run_llm_review(prompt: str) -> Dict[str, Any]:
     _base_url = None
     _api_mode = None
     _resolved_provider = None
+    _credential_pool = None
+    _request_overrides = None
     _model_name = ""
     try:
         from hermes_cli.config import load_config
@@ -1867,6 +1869,8 @@ def _run_llm_review(prompt: str) -> Dict[str, Any]:
         _base_url = _rp.get("base_url")
         _api_mode = _rp.get("api_mode")
         _resolved_provider = _rp.get("provider") or _provider
+        _credential_pool = _rp.get("credential_pool")
+        _request_overrides = _rp.get("request_overrides")
     except Exception as e:
         logger.debug("Curator provider resolution failed: %s", e, exc_info=True)
 
@@ -1881,6 +1885,8 @@ def _run_llm_review(prompt: str) -> Dict[str, Any]:
             api_key=_api_key,
             base_url=_base_url,
             api_mode=_api_mode,
+            credential_pool=_credential_pool,
+            request_overrides=_request_overrides,
             # Umbrella-building over a large skill collection is worth a
             # high iteration ceiling — the pass typically takes 50-100
             # API calls against hundreds of candidate skills. The

tests/agent/test_curator.py

@@ -1273,3 +1273,54 @@ def run_conversation(self, user_message=None, **kwargs):
         "'background_review' — the skill_manage background-review write "
         "guard would not fire (GH-47688 regression)"
     )
+
+
+def test_review_fork_forwards_runtime_pool_and_overrides(curator_env, monkeypatch):
+    """Curator must pass credential_pool + request_overrides from resolve_runtime_provider."""
+    curator = curator_env["curator"]
+    import importlib
+    importlib.reload(curator)
+
+    fake_pool = object()
+    fake_overrides = {"extra_body": {"store": False}}
+    captured = {}
+
+    def _fake_resolve_runtime_provider(**kwargs):
+        return {
+            "provider": "custom",
+            "api_key": "pool-token",
+            "base_url": "https://hyper.charm.land/v1",
+            "api_mode": "chat_completions",
+            "credential_pool": fake_pool,
+            "request_overrides": fake_overrides,
+        }
+
+    class _StubAgent:
+        def __init__(self, *args, **kwargs):
+            captured["kwargs"] = kwargs
+            self._memory_write_origin = "assistant_tool"
+            self._memory_nudge_interval = 0
+            self._skill_nudge_interval = 0
+            self._session_messages = []
+
+        def run_conversation(self, user_message=None, **kwargs):
+            return {"final_response": "ok"}
+
+        def close(self):
+            pass
+
+    monkeypatch.setattr(
+        "hermes_cli.config.load_config",
+        lambda: {"model": {"provider": "custom:hyper-charm", "default": "glm-5.2"}},
+    )
+    monkeypatch.setattr(
+        "hermes_cli.runtime_provider.resolve_runtime_provider",
+        _fake_resolve_runtime_provider,
+    )
+    monkeypatch.setattr("run_agent.AIAgent", _StubAgent)
+
+    meta = curator._run_llm_review("review prompt")
+
+    assert meta.get("error") is None, meta.get("error")
+    assert captured["kwargs"]["credential_pool"] is fake_pool
+    assert captured["kwargs"]["request_overrides"] == fake_overrides