Skip to content

armgit / strix

Public Imported from GitHub
0

Open-source AI penetration testing tool to find and fix your app’s vulnerabilities.

fix(inputs): send reasoning_effort as configured; no route-specific handling

Ahmed Allam8ab5e39c0612
diff --git a/strix/core/execution.py b/strix/core/execution.py
index 5d4f0a32..69a5cf87 100644
--- a/strix/core/execution.py
+++ b/strix/core/execution.py
@@ -125,20 +125,6 @@ _TRANSIENT_MODEL_RETRY_BASE_DELAY_S = 2.0
_TRANSIENT_MODEL_RETRY_MAX_DELAY_S = 90.0
-_TOOLS_WITH_REASONING_EFFORT = "Function tools with reasoning_effort are not supported"
-_TOOLS_WITH_REASONING_EFFORT_HINT = (
- "This model takes reasoning_effort together with tools on the Responses API only: "
- "set STRIX_API_TYPE=responses, or STRIX_REASONING_EFFORT=none to stay on chat completions."
-)
-
-
-def _failure_text(exc: BaseException) -> str:
- text = request_log.failure_text(exc)
- if _model_error_status_code(exc) == 400 and _TOOLS_WITH_REASONING_EFFORT in text:
- text = f"{text} {_TOOLS_WITH_REASONING_EFFORT_HINT}"
- return text
-
-
def _model_error_status_code(exc: BaseException) -> int | None:
code = getattr(exc, "status_code", None)
return code if isinstance(code, int) else None
@@ -707,7 +693,7 @@ async def _run_cycle_parked(
raise
except Exception as exc:
logger.exception("error escaped the run cycle for %s; parking as failed", agent_id)
- await coordinator.set_status(agent_id, "failed", error=_failure_text(exc))
+ await coordinator.set_status(agent_id, "failed", error=request_log.failure_text(exc))
await notify_parent_on_terminal(coordinator, agent_id, "failed")
return None
@@ -868,7 +854,9 @@ async def _run_cycle( # noqa: PLR0912, PLR0915
await _salvage_stream_to_session(session, pre_run_items, stream, agent_id)
if isinstance(exc, ProviderRefusalError):
logger.warning("agent %s refused by the model provider: %s", agent_id, exc)
- await coordinator.set_status(agent_id, "failed", error=_failure_text(exc))
+ await coordinator.set_status(
+ agent_id, "failed", error=request_log.failure_text(exc)
+ )
await notify_parent_on_terminal(coordinator, agent_id, "failed")
return None
if isinstance(exc, MaxTurnsExceeded):
@@ -882,7 +870,7 @@ async def _run_cycle( # noqa: PLR0912, PLR0915
# non-interactive agent's task: a child that dies still owes its parent a
# report, and the parent would otherwise wait out its timeout on a message
# the dead child can no longer send.
- await coordinator.set_status(agent_id, status, error=_failure_text(exc))
+ await coordinator.set_status(agent_id, status, error=request_log.failure_text(exc))
await notify_parent_on_terminal(coordinator, agent_id, status)
if not interactive:
raise
diff --git a/strix/core/inputs.py b/strix/core/inputs.py
index f99e7a28..c83b874e 100644
--- a/strix/core/inputs.py
+++ b/strix/core/inputs.py
@@ -28,7 +28,7 @@ logger = logging.getLogger(__name__)
if TYPE_CHECKING:
- from strix.config.settings import ApiType, ReasoningEffort
+ from strix.config.settings import ReasoningEffort
def _accepts_required_tool_choice(model_name: str | None) -> bool:
@@ -257,9 +257,7 @@ def make_model_settings(
prompt_cache: bool = True,
extra_headers: dict[str, str] | None = None,
has_tools: bool = True,
- api_type: ApiType | None = None,
) -> ModelSettings:
- """``api_type`` is the resolved SDK-native OpenAI route, when known."""
headers = _request_headers(model_name, extra_headers)
model_settings = ModelSettings(
parallel_tool_calls=False if has_tools else None,
@@ -268,13 +266,12 @@ def make_model_settings(
extra_args=request_timeout_extra_args(request_timeout),
extra_headers=headers,
)
- if reasoning_effort is not None and model_supports_reasoning(model_name):
- if reasoning_effort != "none":
- model_settings = model_settings.resolve(_reasoning_settings(reasoning_effort))
- elif _explicit_none_required(model_name, api_type):
- model_settings = model_settings.resolve(
- ModelSettings(reasoning=Reasoning(effort="none"))
- )
+ if (
+ reasoning_effort is not None
+ and reasoning_effort != "none"
+ and model_supports_reasoning(model_name)
+ ):
+ model_settings = model_settings.resolve(_reasoning_settings(reasoning_effort))
if force_required_tool_choice and _accepts_required_tool_choice(model_name):
model_settings = model_settings.resolve(ModelSettings(tool_choice="required"))
@@ -288,14 +285,6 @@ def make_model_settings(
return model_settings
-def _explicit_none_required(model_name: str, api_type: ApiType | None) -> bool:
- """OpenAI's chat completions reason at a default effort when the field is
- absent, and the newer reasoning models reject function tools at any effort
- but ``none``, so ``none`` has to be sent, not left out. LiteLLM's own route
- maps the field per provider and is left alone."""
- return api_type == "chat_completions" and not routes_through_litellm(model_name)
-
-
def _request_headers(
model_name: str, extra_headers: dict[str, str] | None
) -> dict[str, str] | None:
diff --git a/strix/core/runner.py b/strix/core/runner.py
index 4e58faad..17a45ca0 100644
--- a/strix/core/runner.py
+++ b/strix/core/runner.py
@@ -377,7 +377,6 @@ async def run_strix_scan(
request_timeout=settings.llm.timeout,
prompt_cache=settings.llm.prompt_cache,
extra_headers=settings.llm.extra_headers,
- api_type="chat_completions" if chat_completions_tools else "responses",
)
run_config = RunConfig(
model=resolved_model,
diff --git a/tests/test_execution_transient_retry.py b/tests/test_execution_transient_retry.py
index d53a25d2..d81e4e70 100644
--- a/tests/test_execution_transient_retry.py
+++ b/tests/test_execution_transient_retry.py
@@ -88,23 +88,6 @@ def test_client_errors_are_not_transient() -> None:
assert execution._is_transient_model_error(ValueError("nope")) is False
-def test_tools_with_reasoning_effort_rejection_carries_a_route_hint() -> None:
- rejected = BadRequestError(
- "Error code: 400 - {'error': {'message': \"Function tools with reasoning_effort are "
- "not supported for gpt-5.6-sol in /v1/chat/completions. To use function tools, use "
- "/v1/responses or set reasoning_effort to 'none'.\", 'param': 'reasoning_effort'}}",
- response=httpx.Response(400, request=_request()),
- body=None,
- )
- text = execution._failure_text(rejected)
- assert text.startswith("Error code: 400")
- assert "STRIX_API_TYPE=responses" in text
- assert "STRIX_REASONING_EFFORT=none" in text
-
- other = BadRequestError("bad", response=httpx.Response(400, request=_request()), body=None)
- assert execution._failure_text(other) == "bad"
-
-
class _FakeStream:
def __init__(self, exc: BaseException | None = None) -> None:
self._exc = exc
diff --git a/tests/test_inputs.py b/tests/test_inputs.py
index 01d7f89a..2bdde73a 100644
--- a/tests/test_inputs.py
+++ b/tests/test_inputs.py
@@ -432,31 +432,8 @@ def test_user_headers_override_openrouter_attribution() -> None:
assert headers["HTTP-Referer"] == "https://strix.ai"
-def test_reasoning_effort_none_sent_explicitly_on_chat_completions() -> None:
- # Absent the field, OpenAI's chat completions reason at a default effort,
- # which the newer models reject together with function tools.
- for model in ("gpt-daybreak-blue-latest", "gpt-5.6-sol", "openai/gpt-5.4"):
- chat = make_model_settings("none", model_name=model, api_type="chat_completions")
- assert chat.reasoning is not None, model
- assert chat.reasoning.effort == "none"
-
-
-def test_reasoning_effort_none_omitted_off_chat_completions() -> None:
- assert (
- make_model_settings("none", model_name="gpt-5.6-sol", api_type="responses").reasoning
- is None
- )
- assert make_model_settings("none", model_name="gpt-5.6-sol", api_type=None).reasoning is None
- litellm_route = make_model_settings(
- "none", model_name="litellm/gpt-5.6-sol", api_type="chat_completions"
- )
- assert litellm_route.reasoning is None
-
-
-def test_reasoning_effort_sent_as_configured_on_every_route() -> None:
- for api_type in ("chat_completions", "responses", None):
- settings = make_model_settings(
- "high", model_name="gpt-daybreak-blue-latest", api_type=api_type
- )
- assert settings.reasoning is not None, api_type
- assert settings.reasoning.effort == "high"
+def test_reasoning_effort_sent_as_configured_and_none_omitted() -> None:
+ assert make_model_settings("none", model_name="gpt-5.6-sol").reasoning is None
+ settings = make_model_settings("high", model_name="gpt-daybreak-blue-latest")
+ assert settings.reasoning is not None
+ assert settings.reasoning.effort == "high"