fix(memory): use primary model for background memory/skill review

run_agent.py

@@ -2116,12 +2116,33 @@ def _run_review():
                 with open(_os.devnull, "w") as _devnull, \
                      contextlib.redirect_stdout(_devnull), \
                      contextlib.redirect_stderr(_devnull):
+                    # Use the primary model for background review, not the
+                    # current turn's model which may be a cheap/weak model
+                    # from smart routing that lacks tool-calling ability.
+                    _review_model = self.model
+                    _review_provider = self.provider
+                    _review_kwargs = {}
+                    try:
+                        from hermes_cli.config import load_config as _load_review_cfg
+                        _rcfg = _load_review_cfg()
+                        _model_cfg = _rcfg.get("model", {}) or {}
+                        if isinstance(_model_cfg, dict) and _model_cfg.get("default"):
+                            _review_model = _model_cfg["default"]
+                            _review_provider = _model_cfg.get("provider") or self.provider
+                            if _model_cfg.get("base_url"):
+                                _review_kwargs["base_url"] = _model_cfg["base_url"]
+                            if _model_cfg.get("api_key"):
+                                _review_kwargs["api_key"] = _model_cfg["api_key"]
+                    except Exception:
+                        pass  # Fall back to current model
+
                     review_agent = AIAgent(
-                        model=self.model,
+                        model=_review_model,
                         max_iterations=8,
                         quiet_mode=True,
                         platform=self.platform,
-                        provider=self.provider,
+                        provider=_review_provider,
+                        **_review_kwargs,
                     )
                     review_agent._memory_store = self._memory_store
                     review_agent._memory_enabled = self._memory_enabled

tests/run_agent/test_background_review_model.py

@@ -0,0 +1,130 @@
+"""Tests for _spawn_background_review model selection.
+
+When smart model routing is active, the current turn's model may be a cheap/
+weak model that lacks tool-calling ability. The background review agent should
+use the primary (strong) model from config.yaml instead.
+"""
+
+from unittest.mock import MagicMock, patch
+
+import pytest
+
+from run_agent import AIAgent
+
+
+def _make_tool_defs(*names):
+    return [
+        {
+            "type": "function",
+            "function": {
+                "name": n,
+                "description": f"{n} tool",
+                "parameters": {"type": "object", "properties": {}},
+            },
+        }
+        for n in names
+    ]
+
+
+@pytest.fixture()
+def cheap_agent():
+    """Agent whose self.model is a cheap routing model."""
+    with (
+        patch(
+            "run_agent.get_tool_definitions",
+            return_value=_make_tool_defs("web_search", "memory"),
+        ),
+        patch("run_agent.check_toolset_requirements", return_value={}),
+        patch("run_agent.OpenAI"),
+    ):
+        a = AIAgent(
+            api_key="test-key-1234567890",
+            model="local-cheap-9b",
+            provider="custom",
+            quiet_mode=True,
+            skip_context_files=True,
+            skip_memory=True,
+        )
+        a.client = MagicMock()
+        a._memory_enabled = True
+        a._user_profile_enabled = True
+        a._memory_store = MagicMock()
+        return a
+
+
+class TestBackgroundReviewModelSelection:
+    """Verify background review reads the primary model from config."""
+
+    def test_review_uses_config_primary_model(self, cheap_agent, monkeypatch):
+        """When config has model.default, review agent uses it (not self.model)."""
+        created_agents = []
+
+        original_init = AIAgent.__init__
+
+        def tracking_init(self, *args, **kwargs):
+            original_init(self, *args, **kwargs)
+            created_agents.append({"model": kwargs.get("model"), "provider": kwargs.get("provider")})
+
+        monkeypatch.setattr(
+            "hermes_cli.config.load_config",
+            lambda: {
+                "model": {
+                    "default": "strong-cloud-70b",
+                    "provider": "openrouter",
+                    "base_url": "https://openrouter.ai/api/v1",
+                    "api_key": "sk-strong-key",
+                }
+            },
+        )
+
+        # Capture AIAgent constructor calls during _spawn_background_review
+        with patch.object(AIAgent, "__init__", tracking_init):
+            with patch.object(AIAgent, "run_conversation", return_value=None):
+                cheap_agent._spawn_background_review(
+                    messages_snapshot=[{"role": "user", "content": "test"}],
+                    review_memory=True,
+                )
+
+                # Wait for the background thread to complete
+                import threading
+                for t in threading.enumerate():
+                    if t.name != "MainThread" and t.is_alive():
+                        t.join(timeout=5)
+
+        # The review agent should have been created with the config's primary model
+        assert len(created_agents) >= 1
+        review = created_agents[-1]
+        assert review["model"] == "strong-cloud-70b"
+        assert review["provider"] == "openrouter"
+
+    def test_review_falls_back_to_self_model_on_config_error(self, cheap_agent, monkeypatch):
+        """When config loading fails, review agent uses self.model as fallback."""
+        created_agents = []
+
+        original_init = AIAgent.__init__
+
+        def tracking_init(self, *args, **kwargs):
+            original_init(self, *args, **kwargs)
+            created_agents.append({"model": kwargs.get("model"), "provider": kwargs.get("provider")})
+
+        monkeypatch.setattr(
+            "hermes_cli.config.load_config",
+            lambda: (_ for _ in ()).throw(RuntimeError("config broken")),
+        )
+
+        with patch.object(AIAgent, "__init__", tracking_init):
+            with patch.object(AIAgent, "run_conversation", return_value=None):
+                cheap_agent._spawn_background_review(
+                    messages_snapshot=[{"role": "user", "content": "test"}],
+                    review_memory=True,
+                )
+
+                import threading
+                for t in threading.enumerate():
+                    if t.name != "MainThread" and t.is_alive():
+                        t.join(timeout=5)
+
+        assert len(created_agents) >= 1
+        review = created_agents[-1]
+        # Falls back to cheap_agent's own model
+        assert review["model"] == "local-cheap-9b"