feat(providers): route gemini through the native AI Studio API

- add a native Gemini adapter over generateContent/streamGenerateContent
- switch the built-in gemini provider off the OpenAI-compatible endpoint
- preserve thought signatures and native functionResponse replay
- route auxiliary Gemini clients through the same adapter
- add focused unit coverage plus native-provider integration checks

tests/agent/test_gemini_native_adapter.py

@@ -0,0 +1,212 @@
+"""Tests for the native Google AI Studio Gemini adapter."""
+
+from __future__ import annotations
+
+import json
+from types import SimpleNamespace
+
+import pytest
+
+
+class DummyResponse:
+    def __init__(self, status_code=200, payload=None, headers=None, text=None):
+        self.status_code = status_code
+        self._payload = payload or {}
+        self.headers = headers or {}
+        self.text = text if text is not None else json.dumps(self._payload)
+
+    def json(self):
+        return self._payload
+
+
+def test_build_native_request_preserves_thought_signature_on_tool_replay():
+    from agent.gemini_native_adapter import build_gemini_request
+
+    request = build_gemini_request(
+        messages=[
+            {"role": "system", "content": "Be helpful."},
+            {
+                "role": "assistant",
+                "content": "",
+                "tool_calls": [
+                    {
+                        "id": "call_1",
+                        "type": "function",
+                        "function": {
+                            "name": "get_weather",
+                            "arguments": '{"city": "Paris"}',
+                        },
+                        "extra_content": {
+                            "google": {"thought_signature": "sig-123"}
+                        },
+                    }
+                ],
+            },
+        ],
+        tools=[],
+        tool_choice=None,
+    )
+
+    parts = request["contents"][0]["parts"]
+    assert parts[0]["functionCall"]["name"] == "get_weather"
+    assert parts[0]["thoughtSignature"] == "sig-123"
+
+
+def test_build_native_request_uses_original_function_name_for_tool_result():
+    from agent.gemini_native_adapter import build_gemini_request
+
+    request = build_gemini_request(
+        messages=[
+            {
+                "role": "assistant",
+                "content": "",
+                "tool_calls": [
+                    {
+                        "id": "call_1",
+                        "type": "function",
+                        "function": {
+                            "name": "get_weather",
+                            "arguments": '{"city": "Paris"}',
+                        },
+                    }
+                ],
+            },
+            {
+                "role": "tool",
+                "tool_call_id": "call_1",
+                "content": '{"forecast": "sunny"}',
+            },
+        ],
+        tools=[],
+        tool_choice=None,
+    )
+
+    tool_response = request["contents"][1]["parts"][0]["functionResponse"]
+    assert tool_response["name"] == "get_weather"
+
+
+def test_translate_native_response_surfaces_reasoning_and_tool_calls():
+    from agent.gemini_native_adapter import translate_gemini_response
+
+    payload = {
+        "candidates": [
+            {
+                "content": {
+                    "parts": [
+                        {"thought": True, "text": "thinking..."},
+                        {"functionCall": {"name": "search", "args": {"q": "hermes"}}},
+                    ]
+                },
+                "finishReason": "STOP",
+            }
+        ],
+        "usageMetadata": {
+            "promptTokenCount": 10,
+            "candidatesTokenCount": 5,
+            "totalTokenCount": 15,
+        },
+    }
+
+    response = translate_gemini_response(payload, model="gemini-2.5-flash")
+    choice = response.choices[0]
+    assert choice.finish_reason == "tool_calls"
+    assert choice.message.reasoning == "thinking..."
+    assert choice.message.tool_calls[0].function.name == "search"
+    assert json.loads(choice.message.tool_calls[0].function.arguments) == {"q": "hermes"}
+
+
+def test_native_client_uses_x_goog_api_key_and_native_models_endpoint(monkeypatch):
+    from agent.gemini_native_adapter import GeminiNativeClient
+
+    recorded = {}
+
+    class DummyHTTP:
+        def post(self, url, json=None, headers=None, timeout=None):
+            recorded["url"] = url
+            recorded["json"] = json
+            recorded["headers"] = headers
+            return DummyResponse(
+                payload={
+                    "candidates": [
+                        {
+                            "content": {"parts": [{"text": "hello"}]},
+                            "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")
+    response = client.chat.completions.create(
+        model="gemini-2.5-flash",
+        messages=[{"role": "user", "content": "Hello"}],
+    )
+
+    assert recorded["url"] == "https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash:generateContent"
+    assert recorded["headers"]["x-goog-api-key"] == "AIza-test"
+    assert "Authorization" not in recorded["headers"]
+    assert response.choices[0].message.content == "hello"
+
+
+def test_native_http_error_keeps_status_and_retry_after():
+    from agent.gemini_native_adapter import gemini_http_error
+
+    response = DummyResponse(
+        status_code=429,
+        headers={"Retry-After": "17"},
+        payload={
+            "error": {
+                "code": 429,
+                "message": "quota exhausted",
+                "status": "RESOURCE_EXHAUSTED",
+                "details": [
+                    {
+                        "@type": "type.googleapis.com/google.rpc.ErrorInfo",
+                        "reason": "RESOURCE_EXHAUSTED",
+                        "metadata": {"service": "generativelanguage.googleapis.com"},
+                    }
+                ],
+            }
+        },
+    )
+
+    err = gemini_http_error(response)
+    assert getattr(err, "status_code", None) == 429
+    assert getattr(err, "retry_after", None) == 17.0
+    assert "quota exhausted" in str(err)
+
+
+def test_stream_event_translation_emits_tool_call_delta_with_stable_index():
+    from agent.gemini_native_adapter import translate_stream_event
+
+    tool_call_indices = {}
+    event = {
+        "candidates": [
+            {
+                "content": {
+                    "parts": [
+                        {"functionCall": {"name": "search", "args": {"q": "abc"}}}
+                    ]
+                },
+                "finishReason": "STOP",
+            }
+        ]
+    }
+
+    first = translate_stream_event(event, model="gemini-2.5-flash", tool_call_indices=tool_call_indices)
+    second = translate_stream_event(event, model="gemini-2.5-flash", tool_call_indices=tool_call_indices)
+
+    assert first[0].choices[0].delta.tool_calls[0].index == 0
+    assert second[0].choices[0].delta.tool_calls[0].index == 0
+    assert first[0].choices[0].delta.tool_calls[0].id == second[0].choices[0].delta.tool_calls[0].id
+    assert first[-1].choices[0].finish_reason == "tool_calls"