fix: support gemini-3.1-flash-lite-preview by preserving thought_signature in tool calls (#568)

Gemini 3.1+ thinking models include a thought_signature field in functionCall
parts. When reconstructing conversation history for subsequent turns, this
signature must be preserved or the API returns 400 INVALID_ARGUMENT.

- Add optional thought_signature field to LLMToolCall
- Capture thought_signature from Gemini response parts
- Pass thought_signature back when reconstructing multi-turn history
- Add gemini-3.1-flash-lite-preview to the LLM provider test matrix

hindsight-api-slim/hindsight_api/engine/providers/gemini_llm.py

@@ -470,9 +470,11 @@ class GeminiLLM(LLMInterface):
                         fn_name = fn.get("name", "")
                         fn_args_str = fn.get("arguments", "{}")
                         fn_args = parse_llm_json(fn_args_str)
-                        parts.append(
-                            genai_types.Part(function_call=genai_types.FunctionCall(name=fn_name, args=fn_args))
-                        )
+                        thought_signature = tc.get("thought_signature")
+                        fc_kwargs: dict[str, Any] = {"name": fn_name, "args": fn_args}
+                        if thought_signature:
+                            fc_kwargs["thought_signature"] = thought_signature
+                        parts.append(genai_types.Part(function_call=genai_types.FunctionCall(**fc_kwargs)))
                     gemini_contents.append(genai_types.Content(role="model", parts=parts))
                 else:
                     gemini_contents.append(genai_types.Content(role="model", parts=[genai_types.Part(text=content)]))
@@ -545,11 +547,13 @@ class GeminiLLM(LLMInterface):
                                 content = part.text
                             if hasattr(part, "function_call") and part.function_call:
                                 fc = part.function_call
+                                thought_signature = getattr(fc, "thought_signature", None)
                                 tool_calls.append(
                                     LLMToolCall(
                                         id=f"gemini_{len(tool_calls)}",
                                         name=fc.name,
                                         arguments=dict(fc.args) if fc.args else {},
+                                        thought_signature=thought_signature,
                                     )
                                 )
 

hindsight-api-slim/hindsight_api/engine/reflect/agent.py

@@ -895,7 +895,7 @@ async def run_reflect_agent(
 
 def _tool_call_to_dict(tc: "LLMToolCall") -> dict[str, Any]:
     """Convert LLMToolCall to OpenAI message format."""
-    return {
+    d: dict[str, Any] = {
         "id": tc.id,
         "type": "function",
         "function": {
@@ -903,6 +903,9 @@ def _tool_call_to_dict(tc: "LLMToolCall") -> dict[str, Any]:
             "arguments": json.dumps(tc.arguments),
         },
     }
+    if tc.thought_signature is not None:
+        d["thought_signature"] = tc.thought_signature
+    return d
 
 
 async def _process_done_tool(

hindsight-api-slim/hindsight_api/engine/response_models.py

@@ -20,6 +20,10 @@ class LLMToolCall(BaseModel):
     id: str = Field(description="Unique identifier for this tool call")
     name: str = Field(description="Name of the tool to call")
     arguments: dict[str, Any] = Field(description="Arguments to pass to the tool")
+    thought_signature: str | None = Field(
+        default=None,
+        description="Opaque token required by Gemini 3.1+ thinking models to preserve thought context across turns",
+    )
 
 
 class LLMToolCallResult(BaseModel):

hindsight-api-slim/tests/test_llm_provider.py

@@ -35,6 +35,7 @@ MODEL_MATRIX = [
     ("gemini", "gemini-2.5-flash"),
     ("gemini", "gemini-2.5-flash-lite"),
     ("gemini", "gemini-3-pro-preview"),
+    ("gemini", "gemini-3.1-flash-lite-preview"),
     # Ollama models (local)
     ("ollama", "gemma3:12b"),
     ("ollama", "gemma3:1b"),