test(gemini): update stale passthrough assertion in test_gemini_provider

test_strip_vendor_prefix asserted the OLD (buggy) behavior where a self-prefix
survived to the native API. Now asserts the prefix is stripped, matching the
adapter/normalizer fix and the method's own name.

agent/gemini_native_adapter.py

@@ -33,6 +33,39 @@ logger = logging.getLogger(__name__)
 
 DEFAULT_GEMINI_BASE_URL = "https://generativelanguage.googleapis.com/v1beta"
 
+# Self prefixes that may ride along on a model id when a user copies an
+# aggregator-style slug (``google/gemini-2.0-flash``) into config.yaml, or
+# when an upstream resolver leaves the provider prefix attached.  Google's
+# native ``models/{model}:generateContent`` endpoint expects a bare model id
+# (``gemini-2.0-flash``); a self-prefixed value produces a malformed URL like
+# ``models/google/gemini-2.0-flash:generateContent`` and a 404.  We strip only
+# prefixes that name THIS provider so genuine slash-bearing ids are untouched.
+_GEMINI_SELF_PREFIXES = ("google/", "gemini/")
+
+
+def bare_gemini_model_id(model: str) -> str:
+    """Strip a leading ``google/`` or ``gemini/`` self prefix from a model id.
+
+    Only the provider's own prefix is removed, case-insensitively.  Any other
+    slash-bearing value (or a bare id) passes through unchanged so we never
+    mangle a legitimately namespaced model name.
+
+    Examples::
+
+        >>> bare_gemini_model_id("google/gemini-2.0-flash")
+        'gemini-2.0-flash'
+        >>> bare_gemini_model_id("gemini/gemini-3-pro-preview")
+        'gemini-3-pro-preview'
+        >>> bare_gemini_model_id("gemini-2.5-flash")
+        'gemini-2.5-flash'
+    """
+    name = (model or "").strip()
+    lowered = name.lower()
+    for prefix in _GEMINI_SELF_PREFIXES:
+        if lowered.startswith(prefix):
+            return name[len(prefix):].strip() or name
+    return name
+
 
 def is_native_gemini_base_url(base_url: str) -> bool:
     """Return True when the endpoint speaks Gemini's native REST API."""
@@ -895,6 +928,7 @@ class GeminiNativeClient:
             thinking_config=thinking_config,
         )
 
+        model = bare_gemini_model_id(model)
         if stream:
             return self._stream_completion(model=model, request=request, timeout=timeout)
 

hermes_cli/model_normalize.py

@@ -83,8 +83,9 @@ _STRIP_VENDOR_ONLY_PROVIDERS: frozenset[str] = frozenset({
 })
 
 # Providers whose native naming is authoritative -- pass through unchanged.
+# HuggingFace ids are legitimately ``org/model`` (e.g. ``Qwen/Qwen3.5-397B``),
+# so we must NOT strip the leading segment for it.
 _AUTHORITATIVE_NATIVE_PROVIDERS: frozenset[str] = frozenset({
-    "gemini",
     "huggingface",
 })
 
@@ -103,6 +104,13 @@ _MATCHING_PREFIX_STRIP_PROVIDERS: frozenset[str] = frozenset({
     "arcee",
     "ollama-cloud",
     "custom",
+    # Native Google (Gemini) and xAI APIs take bare model ids.  A self-prefixed
+    # value like ``google/gemini-2.0-flash`` or ``xai/grok-4`` — easily pasted
+    # from an aggregator slug into config.yaml — must have the matching prefix
+    # stripped before the native call, or the request 400/404s.  HuggingFace is
+    # deliberately NOT here: its ids are legitimately ``org/model``.
+    "gemini",
+    "xai",
 })
 
 # Providers whose APIs require lowercase model IDs.  Xiaomi's

tests/agent/test_gemini_native_adapter.py

@@ -198,6 +198,66 @@ def test_native_client_uses_x_goog_api_key_and_native_models_endpoint(monkeypatc
     assert response.choices[0].message.content == "hello"
 
 
+@pytest.mark.parametrize("model_in,expected", [
+    ("google/gemini-2.0-flash", "gemini-2.0-flash"),
+    ("gemini/gemini-3-pro-preview", "gemini-3-pro-preview"),
+    ("Google/Gemini-2.5-Pro", "Gemini-2.5-Pro"),   # prefix match is case-insensitive; model casing preserved
+    ("gemini-2.5-flash", "gemini-2.5-flash"),       # bare id unchanged
+    ("models/gemini-x", "models/gemini-x"),         # non-self prefix unchanged
+    ("tunedModels/my-tune", "tunedModels/my-tune"),  # legit Gemini namespaced id unchanged
+    ("", ""),
+])
+def test_bare_gemini_model_id_strips_only_self_prefix(model_in, expected):
+    from agent.gemini_native_adapter import bare_gemini_model_id
+
+    assert bare_gemini_model_id(model_in) == expected
+
+
+def test_native_client_strips_self_prefix_from_model_url(monkeypatch):
+    """A self-prefixed model id must resolve to a bare native resource path.
+
+    Port of openclaw/openclaw#88781.  Before the fix, ``google/gemini-2.0-flash``
+    built ``models/google/gemini-2.0-flash:generateContent`` — an invalid
+    Google resource name that 404s.  The adapter now strips the matching
+    self-prefix at URL construction so the request reaches the real model.
+    """
+    from agent.gemini_native_adapter import GeminiNativeClient
+
+    recorded = {}
+
+    class DummyHTTP:
+        def post(self, url, json=None, headers=None, timeout=None):
+            recorded["url"] = url
+            return DummyResponse(
+                payload={
+                    "candidates": [
+                        {"content": {"parts": [{"text": "ok"}]}, "finishReason": "STOP"}
+                    ],
+                    "usageMetadata": {
+                        "promptTokenCount": 1,
+                        "candidatesTokenCount": 1,
+                        "totalTokenCount": 2,
+                    },
+                }
+            )
+
+        def close(self):
+            return None
+
+    monkeypatch.setattr("agent.gemini_native_adapter.httpx.Client", lambda *a, **k: DummyHTTP())
+
+    client = GeminiNativeClient(api_key="AIza-test", base_url="https://generativelanguage.googleapis.com/v1beta")
+    client.chat.completions.create(
+        model="google/gemini-2.0-flash",
+        messages=[{"role": "user", "content": "Hello"}],
+    )
+
+    assert recorded["url"] == (
+        "https://generativelanguage.googleapis.com/v1beta/models/gemini-2.0-flash:generateContent"
+    )
+    assert "models/google/" not in recorded["url"]
+
+
 def test_native_http_error_keeps_status_and_retry_after():
     from agent.gemini_native_adapter import gemini_http_error
 

tests/hermes_cli/test_gemini_provider.py

@@ -143,7 +143,11 @@ class TestGeminiModelNormalization:
         assert normalize_model_for_provider("gemini-2.5-flash", "gemini") == "gemini-2.5-flash"
 
     def test_strip_vendor_prefix(self):
-        assert normalize_model_for_provider("google/gemini-2.5-flash", "gemini") == "google/gemini-2.5-flash"
+        # Native Gemini takes a bare model id; a self-prefix (``google/`` or
+        # ``gemini/``) must be stripped before the generateContent call, or the
+        # request 404s on a malformed ``models/google/...`` resource path.
+        assert normalize_model_for_provider("google/gemini-2.5-flash", "gemini") == "gemini-2.5-flash"
+        assert normalize_model_for_provider("gemini/gemini-2.5-flash", "gemini") == "gemini-2.5-flash"
 
     def test_gemma_vendor_detection(self):
         assert detect_vendor("gemma-4-31b-it") == "google"

tests/hermes_cli/test_model_normalize.py

@@ -167,10 +167,19 @@ class TestAggregatorProviders:
 class TestIssue6211NativeProviderPrefixNormalization:
     @pytest.mark.parametrize("model,target_provider,expected", [
         ("zai/glm-5.1", "zai", "glm-5.1"),
-        ("google/gemini-2.5-pro", "gemini", "google/gemini-2.5-pro"),
+        # Native Gemini/xAI now strip the matching self-prefix before the
+        # native API call (port of openclaw/openclaw#88781).  Sending
+        # ``google/gemini-2.5-pro`` to generativelanguage.googleapis.com builds
+        # a malformed ``models/google/...`` resource path and 404s.
+        ("google/gemini-2.5-pro", "gemini", "gemini-2.5-pro"),
+        ("gemini/gemini-2.5-pro", "gemini", "gemini-2.5-pro"),
+        ("xai/grok-4-fast-reasoning", "xai", "grok-4-fast-reasoning"),
         ("moonshot/kimi-k2.5", "kimi-coding", "kimi-k2.5"),
         ("anthropic/claude-sonnet-4.6", "openrouter", "anthropic/claude-sonnet-4.6"),
+        # HuggingFace ids are legitimately ``org/model`` — never stripped.
         ("Qwen/Qwen3.5-397B-A17B", "huggingface", "Qwen/Qwen3.5-397B-A17B"),
+        # Cross-vendor prefixes on a native provider stay untouched.
+        ("openai/gpt-5.4", "xai", "openai/gpt-5.4"),
         ("modal/zai-org/GLM-5-FP8", "custom", "modal/zai-org/GLM-5-FP8"),
     ])
     def test_native_provider_prefixes_are_only_stripped_on_matching_provider(