fix(commandcode): forward DeepSeek reasoning controls through the wire

CommandCode fronts DeepSeek with vendor-prefixed ids
(deepseek/deepseek-v4-flash). DeepSeek V4+ defaults to thinking mode
when the thinking field is omitted, so /reasoning none changed the
Hermes session state but not the actual request -- the turn sat in
reflecting.../brainstorming... for minutes (#95232). Strip the vendor
prefix for DeepSeek-family ids and delegate to the native DeepSeek
profile's build_api_kwargs_extras (extra_body.thinking +
reasoning_effort mapping); other CommandCode model families keep the
base no-op behavior. The prior no-op tests codified the bug and are
rewritten to pin the new contract.

plugins/model-providers/commandcode/__init__.py

@@ -48,6 +48,32 @@ class CommandCodeAnthropicProfile(CommandCodeProfile):
         all_models = super().fetch_models(api_key=api_key, base_url=base_url, timeout=timeout)
         return None if all_models is None else [m for m in all_models if m.startswith("claude-")]
 
+    def build_api_kwargs_extras(
+        self, *, reasoning_config: dict | None = None, model: str | None = None, **context
+    ) -> tuple[dict, dict]:
+        """Apply the native DeepSeek reasoning controls for DeepSeek-family ids.
+
+        CommandCode fronts DeepSeek with vendor-prefixed ids
+        (``deepseek/deepseek-v4-flash``). DeepSeek V4+ defaults to thinking
+        mode when the ``thinking`` field is omitted, so without an explicit
+        wire control a Hermes ``/reasoning none`` changes the session state
+        but not the actual request — the turn sits in
+        ``reflecting.../brainstorming...`` for minutes (#95232). Strip the
+        vendor prefix and delegate to the native DeepSeek profile's logic
+        (extra_body.thinking + reasoning_effort mapping); other CommandCode
+        model families keep the base no-op behavior.
+        """
+        m = (model or "").strip()
+        if not m.lower().startswith("deepseek/") or len(m) <= len("deepseek/"):
+            return {}, {}
+        from plugins.model_providers.deepseek import deepseek as _deepseek_profile
+
+        return _deepseek_profile.build_api_kwargs_extras(
+            reasoning_config=reasoning_config,
+            model=m.split("/", 1)[1],
+            **context,
+        )
+
 
 commandcode = CommandCodeProfile(
     name="commandcode", aliases=("commandcode-chat",), api_mode="chat_completions",

tests/plugins/model_providers/test_commandcode_profile.py

@@ -86,28 +86,64 @@ class TestCommandCodeProfileIdentity:
 
 
 class TestCommandCodeProfileNoThinkingInterference:
-    """Chat completions profile is a no-op for thinking config — it delegates
-    to the underlying model's provider (DeepSeek, Qwen, etc.) for wire format.
+    """Reasoning wire controls for the chat-completions profile.
+
+    DeepSeek-family ids get the native DeepSeek controls (DeepSeek V4+
+    defaults to thinking when the field is omitted, so an explicit wire
+    control is required for ``/reasoning none`` to reach the request,
+    #95232); every other CommandCode model family keeps the base no-op.
     """
 
-    def test_passthrough_no_reasoning_config(self, commandcode_profile):
+    def test_deepseek_disabled_reasoning_sends_thinking_disabled(
+        self, commandcode_profile
+    ):
         extra_body, top_level = commandcode_profile.build_api_kwargs_extras(
-            reasoning_config=None, model="deepseek/deepseek-v4-pro"
+            reasoning_config={"enabled": False},
+            model="deepseek/deepseek-v4-flash",
         )
-        # Chat completions profile doesn't inject thinking params — that's
-        # the DeepSeek provider's job when routed through DeepSeek's own profile.
-        # When routed through CommandCode, the underlying model API handles it.
-        assert isinstance(extra_body, dict)
-        assert isinstance(top_level, dict)
-        # Default ProviderProfile returns ({}, {}).
+        assert extra_body.get("thinking") == {"type": "disabled"}
+        assert top_level == {}
 
-    def test_passthrough_with_reasoning_config(self, commandcode_profile):
+    def test_deepseek_enabled_reasoning_maps_effort(self, commandcode_profile):
         extra_body, top_level = commandcode_profile.build_api_kwargs_extras(
             reasoning_config={"enabled": True, "effort": "high"},
             model="deepseek/deepseek-v4-pro",
         )
-        assert isinstance(extra_body, dict)
-        assert isinstance(top_level, dict)
+        assert extra_body.get("thinking") == {"type": "enabled"}
+        assert top_level.get("reasoning_effort") == "high"
+
+    def test_deepseek_no_config_defaults_to_enabled(self, commandcode_profile):
+        # Matches DeepSeek's API default, applied explicitly so the field is
+        # never omitted for thinking-capable models.
+        extra_body, _ = commandcode_profile.build_api_kwargs_extras(
+            reasoning_config=None, model="deepseek/deepseek-v4-flash"
+        )
+        assert extra_body.get("thinking") == {"type": "enabled"}
+
+    def test_deepseek_v3_stays_noop(self, commandcode_profile):
+        extra_body, top_level = commandcode_profile.build_api_kwargs_extras(
+            reasoning_config={"enabled": False},
+            model="deepseek/deepseek-v3",
+        )
+        assert extra_body == {}
+        assert top_level == {}
+
+    def test_passthrough_non_deepseek_family(self, commandcode_profile):
+        extra_body, top_level = commandcode_profile.build_api_kwargs_extras(
+            reasoning_config={"enabled": True, "effort": "high"},
+            model="Qwen/Qwen3.7-Max",
+        )
+        assert extra_body == {}
+        assert top_level == {}
+
+    def test_passthrough_no_reasoning_config_non_deepseek(
+        self, commandcode_profile
+    ):
+        extra_body, top_level = commandcode_profile.build_api_kwargs_extras(
+            reasoning_config=None, model="gpt-5.5"
+        )
+        assert extra_body == {}
+        assert top_level == {}
 
 
 # ── Anthropic Messages profile ────────────────────────────────────────────────