Merge origin/main into fix/structured-retry-classification-main

Made-with: Cursor

tests/providers/test_litellm_kwargs.py

@@ -226,7 +226,39 @@ def test_openai_model_passthrough() -> None:
     assert provider.get_default_model() == "gpt-4o"
 
 
-def test_openai_compat_strips_message_level_reasoning_fields() -> None:
+def test_openai_compat_supports_temperature_matches_reasoning_model_rules() -> None:
+    assert OpenAICompatProvider._supports_temperature("gpt-4o") is True
+    assert OpenAICompatProvider._supports_temperature("gpt-5-chat") is False
+    assert OpenAICompatProvider._supports_temperature("o3-mini") is False
+    assert OpenAICompatProvider._supports_temperature("gpt-4o", reasoning_effort="medium") is False
+
+
+def test_openai_compat_build_kwargs_uses_gpt5_safe_parameters() -> None:
+    spec = find_by_name("openai")
+    with patch("nanobot.providers.openai_compat_provider.AsyncOpenAI"):
+        provider = OpenAICompatProvider(
+            api_key="sk-test-key",
+            default_model="gpt-5-chat",
+            spec=spec,
+        )
+
+    kwargs = provider._build_kwargs(
+        messages=[{"role": "user", "content": "hello"}],
+        tools=None,
+        model="gpt-5-chat",
+        max_tokens=4096,
+        temperature=0.7,
+        reasoning_effort=None,
+        tool_choice=None,
+    )
+
+    assert kwargs["model"] == "gpt-5-chat"
+    assert kwargs["max_completion_tokens"] == 4096
+    assert "max_tokens" not in kwargs
+    assert "temperature" not in kwargs
+
+
+def test_openai_compat_preserves_message_level_reasoning_fields() -> None:
     with patch("nanobot.providers.openai_compat_provider.AsyncOpenAI"):
         provider = OpenAICompatProvider()
 
@@ -247,8 +279,8 @@ def test_openai_compat_strips_message_level_reasoning_fields() -> None:
         }
     ])
 
-    assert "reasoning_content" not in sanitized[0]
-    assert "extra_content" not in sanitized[0]
+    assert sanitized[0]["reasoning_content"] == "hidden"
+    assert sanitized[0]["extra_content"] == {"debug": True}
     assert sanitized[0]["tool_calls"][0]["extra_content"] == {"google": {"thought_signature": "sig"}}
 
 
@@ -275,3 +307,54 @@ async def test_openai_compat_stream_watchdog_returns_error_on_stall(monkeypatch)
     assert result.finish_reason == "error"
     assert result.content is not None
     assert "stream stalled" in result.content
+
+
+# ---------------------------------------------------------------------------
+# Provider-specific thinking parameters (extra_body)
+# ---------------------------------------------------------------------------
+
+def _build_kwargs_for(provider_name: str, model: str, reasoning_effort=None):
+    spec = find_by_name(provider_name)
+    with patch("nanobot.providers.openai_compat_provider.AsyncOpenAI"):
+        p = OpenAICompatProvider(api_key="k", default_model=model, spec=spec)
+    return p._build_kwargs(
+        messages=[{"role": "user", "content": "hi"}],
+        tools=None, model=model, max_tokens=1024, temperature=0.7,
+        reasoning_effort=reasoning_effort, tool_choice=None,
+    )
+
+
+def test_dashscope_thinking_enabled_with_reasoning_effort() -> None:
+    kw = _build_kwargs_for("dashscope", "qwen3-plus", reasoning_effort="medium")
+    assert kw["extra_body"] == {"enable_thinking": True}
+
+
+def test_dashscope_thinking_disabled_for_minimal() -> None:
+    kw = _build_kwargs_for("dashscope", "qwen3-plus", reasoning_effort="minimal")
+    assert kw["extra_body"] == {"enable_thinking": False}
+
+
+def test_dashscope_no_extra_body_when_reasoning_effort_none() -> None:
+    kw = _build_kwargs_for("dashscope", "qwen-turbo", reasoning_effort=None)
+    assert "extra_body" not in kw
+
+
+def test_volcengine_thinking_enabled() -> None:
+    kw = _build_kwargs_for("volcengine", "doubao-seed-2-0-pro", reasoning_effort="high")
+    assert kw["extra_body"] == {"thinking": {"type": "enabled"}}
+
+
+def test_byteplus_thinking_disabled_for_minimal() -> None:
+    kw = _build_kwargs_for("byteplus", "doubao-seed-2-0-pro", reasoning_effort="minimal")
+    assert kw["extra_body"] == {"thinking": {"type": "disabled"}}
+
+
+def test_byteplus_no_extra_body_when_reasoning_effort_none() -> None:
+    kw = _build_kwargs_for("byteplus", "doubao-seed-2-0-pro", reasoning_effort=None)
+    assert "extra_body" not in kw
+
+
+def test_openai_no_thinking_extra_body() -> None:
+    """Non-thinking providers should never get extra_body for thinking."""
+    kw = _build_kwargs_for("openai", "gpt-4o", reasoning_effort="medium")
+    assert "extra_body" not in kw

tests/providers/test_prompt_cache_markers.py

@@ -0,0 +1,87 @@
+from __future__ import annotations
+
+from typing import Any
+
+from nanobot.providers.anthropic_provider import AnthropicProvider
+from nanobot.providers.openai_compat_provider import OpenAICompatProvider
+
+
+def _openai_tools(*names: str) -> list[dict[str, Any]]:
+    return [
+        {
+            "type": "function",
+            "function": {
+                "name": name,
+                "description": f"{name} tool",
+                "parameters": {"type": "object", "properties": {}},
+            },
+        }
+        for name in names
+    ]
+
+
+def _anthropic_tools(*names: str) -> list[dict[str, Any]]:
+    return [
+        {
+            "name": name,
+            "description": f"{name} tool",
+            "input_schema": {"type": "object", "properties": {}},
+        }
+        for name in names
+    ]
+
+
+def _marked_openai_tool_names(tools: list[dict[str, Any]] | None) -> list[str]:
+    if not tools:
+        return []
+    marked: list[str] = []
+    for tool in tools:
+        if "cache_control" in tool:
+            marked.append((tool.get("function") or {}).get("name", ""))
+    return marked
+
+
+def _marked_anthropic_tool_names(tools: list[dict[str, Any]] | None) -> list[str]:
+    if not tools:
+        return []
+    return [tool.get("name", "") for tool in tools if "cache_control" in tool]
+
+
+def test_openai_compat_marks_builtin_boundary_and_tail_tool() -> None:
+    messages = [
+        {"role": "system", "content": "system"},
+        {"role": "assistant", "content": "assistant"},
+        {"role": "user", "content": "user"},
+    ]
+    _, marked_tools = OpenAICompatProvider._apply_cache_control(
+        messages,
+        _openai_tools("read_file", "write_file", "mcp_fs_ls", "mcp_git_status"),
+    )
+    assert _marked_openai_tool_names(marked_tools) == ["write_file", "mcp_git_status"]
+
+
+def test_anthropic_marks_builtin_boundary_and_tail_tool() -> None:
+    messages = [
+        {"role": "user", "content": "u1"},
+        {"role": "assistant", "content": "a1"},
+        {"role": "user", "content": "u2"},
+    ]
+    _, _, marked_tools = AnthropicProvider._apply_cache_control(
+        "system",
+        messages,
+        _anthropic_tools("read_file", "write_file", "mcp_fs_ls", "mcp_git_status"),
+    )
+    assert _marked_anthropic_tool_names(marked_tools) == ["write_file", "mcp_git_status"]
+
+
+def test_openai_compat_marks_only_tail_without_mcp() -> None:
+    messages = [
+        {"role": "system", "content": "system"},
+        {"role": "assistant", "content": "assistant"},
+        {"role": "user", "content": "user"},
+    ]
+    _, marked_tools = OpenAICompatProvider._apply_cache_control(
+        messages,
+        _openai_tools("read_file", "write_file"),
+    )
+    assert _marked_openai_tool_names(marked_tools) == ["write_file"]