refactor(provider): preserve extra_content verbatim for Gemini thought_signature round-trip

Replace the flatten/unflatten approach (merging extra_content.google.*
into provider_specific_fields then reconstructing) with direct pass-through:
parse extra_content as-is, store on ToolCallRequest.extra_content, serialize
back untouched.  This is lossless, requires no hardcoded field names, and
covers all three parsing branches (str, dict, SDK object) plus streaming.

nanobot/providers/base.py

@@ -16,6 +16,7 @@ class ToolCallRequest:
     id: str
     name: str
     arguments: dict[str, Any]
+    extra_content: dict[str, Any] | None = None
     provider_specific_fields: dict[str, Any] | None = None
     function_provider_specific_fields: dict[str, Any] | None = None
 
@@ -29,22 +30,10 @@ class ToolCallRequest:
                 "arguments": json.dumps(self.arguments, ensure_ascii=False),
             },
         }
+        if self.extra_content:
+            tool_call["extra_content"] = self.extra_content
         if self.provider_specific_fields:
-            # Gemini OpenAI compatibility expects thought signatures in extra_content.google.
+            tool_call["provider_specific_fields"] = self.provider_specific_fields
-            if "thought_signature" in self.provider_specific_fields:
-                tool_call["extra_content"] = {
-                    "google": {
-                        "thought_signature": self.provider_specific_fields["thought_signature"],
-                    }
-                }
-                other_fields = {
-                    k: v for k, v in self.provider_specific_fields.items()
-                    if k != "thought_signature"
-                }
-                if other_fields:
-                    tool_call["provider_specific_fields"] = other_fields
-            else:
-                tool_call["provider_specific_fields"] = self.provider_specific_fields
         if self.function_provider_specific_fields:
             tool_call["function"]["provider_specific_fields"] = self.function_provider_specific_fields
         return tool_call

nanobot/providers/openai_compat_provider.py

@@ -19,42 +19,13 @@ if TYPE_CHECKING:
     from nanobot.providers.registry import ProviderSpec
 
 _ALLOWED_MSG_KEYS = frozenset({
-    "role", "content", "tool_calls", "tool_call_id", "name", "reasoning_content",
+    "role", "content", "tool_calls", "tool_call_id", "name",
+    "reasoning_content", "extra_content",
 })
 _ALNUM = string.ascii_letters + string.digits
 
-
+_STANDARD_TC_KEYS = frozenset({"id", "type", "index", "function"})
-def _get_attr_or_item(obj: Any, key: str, default: Any = None) -> Any:
+_STANDARD_FN_KEYS = frozenset({"name", "arguments"})
-    """Read an attribute or dict key from provider SDK objects."""
-    if obj is None:
-        return default
-    if isinstance(obj, dict):
-        return obj.get(key, default)
-    return getattr(obj, key, default)
-
-
-def _coerce_dict(value: Any) -> dict[str, Any] | None:
-    """Return a shallow dict if the value looks mapping-like."""
-    if isinstance(value, dict):
-        return dict(value)
-    return None
-
-
-def _extract_tool_call_fields(tc: Any) -> tuple[dict[str, Any] | None, dict[str, Any] | None]:
-    """Extract provider-specific metadata from a tool call object."""
-    provider_specific_fields = _coerce_dict(_get_attr_or_item(tc, "provider_specific_fields"))
-    extra_content = _coerce_dict(_get_attr_or_item(tc, "extra_content"))
-    google_content = _coerce_dict(_get_attr_or_item(extra_content, "google")) if extra_content else None
-    if google_content:
-        provider_specific_fields = {
-            **(provider_specific_fields or {}),
-            **google_content,
-        }
-    function = _get_attr_or_item(tc, "function")
-    function_provider_specific_fields = _coerce_dict(
-        _get_attr_or_item(function, "provider_specific_fields")
-    )
-    return provider_specific_fields, function_provider_specific_fields
 
 
 def _short_tool_id() -> str:
@@ -62,6 +33,62 @@ def _short_tool_id() -> str:
     return "".join(secrets.choice(_ALNUM) for _ in range(9))
 
 
+def _get(obj: Any, key: str) -> Any:
+    """Get a value from dict or object attribute, returning None if absent."""
+    if isinstance(obj, dict):
+        return obj.get(key)
+    return getattr(obj, key, None)
+
+
+def _coerce_dict(value: Any) -> dict[str, Any] | None:
+    """Try to coerce *value* to a dict; return None if not possible or empty."""
+    if value is None:
+        return None
+    if isinstance(value, dict):
+        return value if value else None
+    model_dump = getattr(value, "model_dump", None)
+    if callable(model_dump):
+        dumped = model_dump()
+        if isinstance(dumped, dict) and dumped:
+            return dumped
+    return None
+
+
+def _extract_tc_extras(tc: Any) -> tuple[
+    dict[str, Any] | None,
+    dict[str, Any] | None,
+    dict[str, Any] | None,
+]:
+    """Extract (extra_content, provider_specific_fields, fn_provider_specific_fields).
+
+    Works for both SDK objects and dicts.  Captures Gemini ``extra_content``
+    verbatim and any non-standard keys on the tool-call / function.
+    """
+    extra_content = _coerce_dict(_get(tc, "extra_content"))
+
+    tc_dict = _coerce_dict(tc)
+    prov = None
+    fn_prov = None
+    if tc_dict is not None:
+        leftover = {k: v for k, v in tc_dict.items()
+                    if k not in _STANDARD_TC_KEYS and k != "extra_content" and v is not None}
+        if leftover:
+            prov = leftover
+        fn = _coerce_dict(tc_dict.get("function"))
+        if fn is not None:
+            fn_leftover = {k: v for k, v in fn.items()
+                          if k not in _STANDARD_FN_KEYS and v is not None}
+            if fn_leftover:
+                fn_prov = fn_leftover
+    else:
+        prov = _coerce_dict(_get(tc, "provider_specific_fields"))
+        fn_obj = _get(tc, "function")
+        if fn_obj is not None:
+            fn_prov = _coerce_dict(_get(fn_obj, "provider_specific_fields"))
+
+    return extra_content, prov, fn_prov
+
+
 class OpenAICompatProvider(LLMProvider):
     """Unified provider for all OpenAI-compatible APIs.
 
@@ -332,10 +359,14 @@ class OpenAICompatProvider(LLMProvider):
                 args = fn.get("arguments", {})
                 if isinstance(args, str):
                     args = json_repair.loads(args)
+                ec, prov, fn_prov = _extract_tc_extras(tc)
                 parsed_tool_calls.append(ToolCallRequest(
                     id=_short_tool_id(),
                     name=str(fn.get("name") or ""),
                     arguments=args if isinstance(args, dict) else {},
+                    extra_content=ec,
+                    provider_specific_fields=prov,
+                    function_provider_specific_fields=fn_prov,
                 ))
 
             return LLMResponse(
@@ -366,17 +397,17 @@ class OpenAICompatProvider(LLMProvider):
 
         tool_calls = []
         for tc in raw_tool_calls:
-            function = _get_attr_or_item(tc, "function")
+            args = tc.function.arguments
-            args = _get_attr_or_item(function, "arguments")
             if isinstance(args, str):
                 args = json_repair.loads(args)
-            provider_specific_fields, function_provider_specific_fields = _extract_tool_call_fields(tc)
+            ec, prov, fn_prov = _extract_tc_extras(tc)
             tool_calls.append(ToolCallRequest(
                 id=_short_tool_id(),
-                name=_get_attr_or_item(function, "name", ""),
+                name=tc.function.name,
                 arguments=args,
-                provider_specific_fields=provider_specific_fields,
+                extra_content=ec,
-                function_provider_specific_fields=function_provider_specific_fields,
+                provider_specific_fields=prov,
+                function_provider_specific_fields=fn_prov,
             ))
 
         return LLMResponse(
@@ -390,10 +421,36 @@ class OpenAICompatProvider(LLMProvider):
     @classmethod
     def _parse_chunks(cls, chunks: list[Any]) -> LLMResponse:
         content_parts: list[str] = []
-        tc_bufs: dict[int, dict[str, str]] = {}
+        tc_bufs: dict[int, dict[str, Any]] = {}
         finish_reason = "stop"
         usage: dict[str, int] = {}
 
+        def _accum_tc(tc: Any, idx_hint: int) -> None:
+            """Accumulate one streaming tool-call delta into *tc_bufs*."""
+            tc_index: int = _get(tc, "index") if _get(tc, "index") is not None else idx_hint
+            buf = tc_bufs.setdefault(tc_index, {
+                "id": "", "name": "", "arguments": "",
+                "extra_content": None, "prov": None, "fn_prov": None,
+            })
+            tc_id = _get(tc, "id")
+            if tc_id:
+                buf["id"] = str(tc_id)
+            fn = _get(tc, "function")
+            if fn is not None:
+                fn_name = _get(fn, "name")
+                if fn_name:
+                    buf["name"] = str(fn_name)
+                fn_args = _get(fn, "arguments")
+                if fn_args:
+                    buf["arguments"] += str(fn_args)
+            ec, prov, fn_prov = _extract_tc_extras(tc)
+            if ec:
+                buf["extra_content"] = ec
+            if prov:
+                buf["prov"] = prov
+            if fn_prov:
+                buf["fn_prov"] = fn_prov
+
         for chunk in chunks:
             if isinstance(chunk, str):
                 content_parts.append(chunk)
@@ -418,16 +475,7 @@ class OpenAICompatProvider(LLMProvider):
                 if text:
                     content_parts.append(text)
                 for idx, tc in enumerate(delta.get("tool_calls") or []):
-                    tc_map = cls._maybe_mapping(tc) or {}
+                    _accum_tc(tc, idx)
-                    tc_index = tc_map.get("index", idx)
-                    buf = tc_bufs.setdefault(tc_index, {"id": "", "name": "", "arguments": ""})
-                    if tc_map.get("id"):
-                        buf["id"] = str(tc_map["id"])
-                    fn = cls._maybe_mapping(tc_map.get("function")) or {}
-                    if fn.get("name"):
-                        buf["name"] = str(fn["name"])
-                    if fn.get("arguments"):
-                        buf["arguments"] += str(fn["arguments"])
                 usage = cls._extract_usage(chunk_map) or usage
                 continue
 
@@ -441,34 +489,7 @@ class OpenAICompatProvider(LLMProvider):
             if delta and delta.content:
                 content_parts.append(delta.content)
             for tc in (delta.tool_calls or []) if delta else []:
-                idx = _get_attr_or_item(tc, "index")
+                _accum_tc(tc, getattr(tc, "index", 0))
-                if idx is None:
-                    continue
-                buf = tc_bufs.setdefault(
-                    idx,
-                    {
-                        "id": "",
-                        "name": "",
-                        "arguments": "",
-                        "provider_specific_fields": None,
-                        "function_provider_specific_fields": None,
-                    },
-                )
-                tc_id = _get_attr_or_item(tc, "id")
-                if tc_id:
-                    buf["id"] = tc_id
-                function = _get_attr_or_item(tc, "function")
-                function_name = _get_attr_or_item(function, "name")
-                if function_name:
-                    buf["name"] = function_name
-                arguments = _get_attr_or_item(function, "arguments")
-                if arguments:
-                    buf["arguments"] += arguments
-                provider_specific_fields, function_provider_specific_fields = _extract_tool_call_fields(tc)
-                if provider_specific_fields:
-                    buf["provider_specific_fields"] = provider_specific_fields
-                if function_provider_specific_fields:
-                    buf["function_provider_specific_fields"] = function_provider_specific_fields
 
         return LLMResponse(
             content="".join(content_parts) or None,
@@ -477,8 +498,9 @@ class OpenAICompatProvider(LLMProvider):
                     id=b["id"] or _short_tool_id(),
                     name=b["name"],
                     arguments=json_repair.loads(b["arguments"]) if b["arguments"] else {},
-                    provider_specific_fields=b["provider_specific_fields"],
+                    extra_content=b.get("extra_content"),
-                    function_provider_specific_fields=b["function_provider_specific_fields"],
+                    provider_specific_fields=b.get("prov"),
+                    function_provider_specific_fields=b.get("fn_prov"),
                 )
                 for b in tc_bufs.values()
             ],

tests/agent/test_gemini_thought_signature.py

@@ -1,19 +1,200 @@
+"""Tests for Gemini thought_signature round-trip through extra_content.
+
+The Gemini OpenAI-compatibility API returns tool calls with an extra_content
+field: ``{"google": {"thought_signature": "..."}}``.  This MUST survive the
+parse → serialize round-trip so the model can continue reasoning.
+"""
+
 from types import SimpleNamespace
+from unittest.mock import patch
 
 from nanobot.providers.base import ToolCallRequest
+from nanobot.providers.openai_compat_provider import OpenAICompatProvider
 
 
-def test_tool_call_request_serializes_provider_fields() -> None:
+GEMINI_EXTRA = {"google": {"thought_signature": "sig-abc-123"}}
-    tool_call = ToolCallRequest(
+
+
+# ── ToolCallRequest serialization ──────────────────────────────────────
+
+def test_tool_call_request_serializes_extra_content() -> None:
+    tc = ToolCallRequest(
         id="abc123xyz",
         name="read_file",
         arguments={"path": "todo.md"},
-        provider_specific_fields={"thought_signature": "signed-token"},
+        extra_content=GEMINI_EXTRA,
+    )
+
+    payload = tc.to_openai_tool_call()
+
+    assert payload["extra_content"] == GEMINI_EXTRA
+    assert payload["function"]["arguments"] == '{"path": "todo.md"}'
+
+
+def test_tool_call_request_serializes_provider_fields() -> None:
+    tc = ToolCallRequest(
+        id="abc123xyz",
+        name="read_file",
+        arguments={"path": "todo.md"},
+        provider_specific_fields={"custom_key": "custom_val"},
         function_provider_specific_fields={"inner": "value"},
     )
 
-    message = tool_call.to_openai_tool_call()
+    payload = tc.to_openai_tool_call()
 
-    assert message["extra_content"] == {"google": {"thought_signature": "signed-token"}}
+    assert payload["provider_specific_fields"] == {"custom_key": "custom_val"}
-    assert message["function"]["provider_specific_fields"] == {"inner": "value"}
+    assert payload["function"]["provider_specific_fields"] == {"inner": "value"}
-    assert message["function"]["arguments"] == '{"path": "todo.md"}'
+
+
+def test_tool_call_request_omits_absent_extras() -> None:
+    tc = ToolCallRequest(id="x", name="fn", arguments={})
+    payload = tc.to_openai_tool_call()
+
+    assert "extra_content" not in payload
+    assert "provider_specific_fields" not in payload
+    assert "provider_specific_fields" not in payload["function"]
+
+
+# ── _parse: SDK-object branch ──────────────────────────────────────────
+
+def _make_sdk_response_with_extra_content():
+    """Simulate a Gemini response via the OpenAI SDK (SimpleNamespace)."""
+    fn = SimpleNamespace(name="get_weather", arguments='{"city":"Tokyo"}')
+    tc = SimpleNamespace(
+        id="call_1",
+        index=0,
+        type="function",
+        function=fn,
+        extra_content=GEMINI_EXTRA,
+    )
+    msg = SimpleNamespace(
+        content=None,
+        tool_calls=[tc],
+        reasoning_content=None,
+    )
+    choice = SimpleNamespace(message=msg, finish_reason="tool_calls")
+    usage = SimpleNamespace(prompt_tokens=10, completion_tokens=5, total_tokens=15)
+    return SimpleNamespace(choices=[choice], usage=usage)
+
+
+def test_parse_sdk_object_preserves_extra_content() -> None:
+    with patch("nanobot.providers.openai_compat_provider.AsyncOpenAI"):
+        provider = OpenAICompatProvider()
+
+    result = provider._parse(_make_sdk_response_with_extra_content())
+
+    assert len(result.tool_calls) == 1
+    tc = result.tool_calls[0]
+    assert tc.name == "get_weather"
+    assert tc.extra_content == GEMINI_EXTRA
+
+    payload = tc.to_openai_tool_call()
+    assert payload["extra_content"] == GEMINI_EXTRA
+
+
+# ── _parse: dict/mapping branch ───────────────────────────────────────
+
+def test_parse_dict_preserves_extra_content() -> None:
+    with patch("nanobot.providers.openai_compat_provider.AsyncOpenAI"):
+        provider = OpenAICompatProvider()
+
+    response_dict = {
+        "choices": [{
+            "message": {
+                "content": None,
+                "tool_calls": [{
+                    "id": "call_1",
+                    "type": "function",
+                    "function": {"name": "get_weather", "arguments": '{"city":"Tokyo"}'},
+                    "extra_content": GEMINI_EXTRA,
+                }],
+            },
+            "finish_reason": "tool_calls",
+        }],
+        "usage": {"prompt_tokens": 10, "completion_tokens": 5, "total_tokens": 15},
+    }
+
+    result = provider._parse(response_dict)
+
+    assert len(result.tool_calls) == 1
+    tc = result.tool_calls[0]
+    assert tc.name == "get_weather"
+    assert tc.extra_content == GEMINI_EXTRA
+
+    payload = tc.to_openai_tool_call()
+    assert payload["extra_content"] == GEMINI_EXTRA
+
+
+# ── _parse_chunks: streaming round-trip ───────────────────────────────
+
+def test_parse_chunks_sdk_preserves_extra_content() -> None:
+    fn_delta = SimpleNamespace(name="get_weather", arguments='{"city":"Tokyo"}')
+    tc_delta = SimpleNamespace(
+        id="call_1",
+        index=0,
+        function=fn_delta,
+        extra_content=GEMINI_EXTRA,
+    )
+    delta = SimpleNamespace(content=None, tool_calls=[tc_delta])
+    choice = SimpleNamespace(finish_reason="tool_calls", delta=delta)
+    chunk = SimpleNamespace(choices=[choice], usage=None)
+
+    result = OpenAICompatProvider._parse_chunks([chunk])
+
+    assert len(result.tool_calls) == 1
+    tc = result.tool_calls[0]
+    assert tc.extra_content == GEMINI_EXTRA
+
+    payload = tc.to_openai_tool_call()
+    assert payload["extra_content"] == GEMINI_EXTRA
+
+
+def test_parse_chunks_dict_preserves_extra_content() -> None:
+    chunk = {
+        "choices": [{
+            "finish_reason": "tool_calls",
+            "delta": {
+                "content": None,
+                "tool_calls": [{
+                    "index": 0,
+                    "id": "call_1",
+                    "function": {"name": "get_weather", "arguments": '{"city":"Tokyo"}'},
+                    "extra_content": GEMINI_EXTRA,
+                }],
+            },
+        }],
+    }
+
+    result = OpenAICompatProvider._parse_chunks([chunk])
+
+    assert len(result.tool_calls) == 1
+    tc = result.tool_calls[0]
+    assert tc.extra_content == GEMINI_EXTRA
+
+    payload = tc.to_openai_tool_call()
+    assert payload["extra_content"] == GEMINI_EXTRA
+
+
+# ── Model switching: stale extras shouldn't break other providers ─────
+
+def test_stale_extra_content_in_tool_calls_survives_sanitize() -> None:
+    """When switching from Gemini to OpenAI, extra_content inside tool_calls
+    should survive message sanitization (it lives inside the tool_call dict,
+    not at message level, so it bypasses _ALLOWED_MSG_KEYS filtering)."""
+    with patch("nanobot.providers.openai_compat_provider.AsyncOpenAI"):
+        provider = OpenAICompatProvider()
+
+    messages = [{
+        "role": "assistant",
+        "content": None,
+        "tool_calls": [{
+            "id": "call_1",
+            "type": "function",
+            "function": {"name": "fn", "arguments": "{}"},
+            "extra_content": GEMINI_EXTRA,
+        }],
+    }]
+
+    sanitized = provider._sanitize_messages(messages)
+
+    assert sanitized[0]["tool_calls"][0]["extra_content"] == GEMINI_EXTRA

tests/providers/test_litellm_kwargs.py

@@ -30,7 +30,7 @@ def _fake_chat_response(content: str = "ok") -> SimpleNamespace:
 
 
 def _fake_tool_call_response() -> SimpleNamespace:
-    """Build a minimal chat response that includes Gemini-style provider fields."""
+    """Build a minimal chat response that includes Gemini-style extra_content."""
     function = SimpleNamespace(
         name="exec",
         arguments='{"cmd":"ls"}',
@@ -39,6 +39,7 @@ def _fake_tool_call_response() -> SimpleNamespace:
     tool_call = SimpleNamespace(
         id="call_123",
         index=0,
+        type="function",
         function=function,
         extra_content={"google": {"thought_signature": "signed-token"}},
     )
@@ -134,8 +135,8 @@ async def test_standard_provider_passes_model_through() -> None:
 
 
 @pytest.mark.asyncio
-async def test_openai_compat_preserves_provider_specific_fields_on_tool_calls() -> None:
+async def test_openai_compat_preserves_extra_content_on_tool_calls() -> None:
-    """Gemini thought signatures must survive parsing so they can be sent back."""
+    """Gemini extra_content (thought signatures) must survive parse→serialize round-trip."""
     mock_create = AsyncMock(return_value=_fake_tool_call_response())
     spec = find_by_name("gemini")
 
@@ -156,7 +157,7 @@ async def test_openai_compat_preserves_provider_specific_fields_on_tool_calls()
 
     assert len(result.tool_calls) == 1
     tool_call = result.tool_calls[0]
-    assert tool_call.provider_specific_fields == {"thought_signature": "signed-token"}
+    assert tool_call.extra_content == {"google": {"thought_signature": "signed-token"}}
     assert tool_call.function_provider_specific_fields == {"inner": "value"}
 
     serialized = tool_call.to_openai_tool_call()