diff --git a/libs/langchain_v1/langchain/agents/structured_output.py b/libs/langchain_v1/langchain/agents/structured_output.py
index d3c73cdb5aa1b..f4e34dbce49df 100644
--- a/libs/langchain_v1/langchain/agents/structured_output.py
+++ b/libs/langchain_v1/langchain/agents/structured_output.py
@@ -192,7 +192,7 @@ def __init__(
 class ToolStrategy(Generic[SchemaT]):
     """Use a tool calling strategy for model responses."""
 
-    schema: type[SchemaT]
+    schema: type[SchemaT] | dict[str, Any]
     """Schema for the tool calls."""
 
     schema_specs: list[_SchemaSpec[SchemaT]]
@@ -218,7 +218,7 @@ class ToolStrategy(Generic[SchemaT]):
 
     def __init__(
         self,
-        schema: type[SchemaT],
+        schema: type[SchemaT] | dict[str, Any],
         *,
         tool_message_content: str | None = None,
         handle_errors: bool
diff --git a/libs/langchain_v1/pyproject.toml b/libs/langchain_v1/pyproject.toml
index a5493093b71cd..cddc495a1b6eb 100644
--- a/libs/langchain_v1/pyproject.toml
+++ b/libs/langchain_v1/pyproject.toml
@@ -139,7 +139,6 @@ ignore-var-parameters = true  # ignore missing documentation for *args and **kwa
     "ARG", # Arguments, needs to fix
 ]
 "tests/unit_tests/agents/test_return_direct_spec.py" = ["F821"]
-"tests/unit_tests/agents/test_responses_spec.py" = ["F821"]
 "tests/unit_tests/agents/test_responses.py" = ["F821"]
 "tests/unit_tests/agents/test_react_agent.py" = ["ALL"]
 
diff --git a/libs/langchain_v1/tests/unit_tests/agents/test_responses_spec.py b/libs/langchain_v1/tests/unit_tests/agents/test_responses_spec.py
index 38d3da9f3aceb..041df1f9dc787 100644
--- a/libs/langchain_v1/tests/unit_tests/agents/test_responses_spec.py
+++ b/libs/langchain_v1/tests/unit_tests/agents/test_responses_spec.py
@@ -1,16 +1,34 @@
 from __future__ import annotations
 
+import os
+from typing import (
+    TYPE_CHECKING,
+    Any,
+)
+from unittest.mock import MagicMock
+
+import httpx
 import pytest
+from langchain_core.messages import HumanMessage
+from langchain_core.tools import tool
+from pydantic import BaseModel, create_model
+
+from langchain.agents import create_agent
+from langchain.agents.structured_output import (
+    ToolStrategy,
+)
+from tests.unit_tests.agents.utils import BaseSchema, load_spec
+
+if TYPE_CHECKING:
+    from collections.abc import Callable
 
-# Skip this test since langgraph.prebuilt.responses is not available
-pytest.skip("langgraph.prebuilt.responses not available", allow_module_level=True)
 
 try:
     from langchain_openai import ChatOpenAI
 except ImportError:
     skip_openai_integration_tests = True
 else:
-    skip_openai_integration_tests = False
+    skip_openai_integration_tests = "OPENAI_API_KEY" not in os.environ
 
 AGENT_PROMPT = "You are an HR assistant."
 
@@ -30,8 +48,8 @@ class AssertionByInvocation(BaseSchema):
 
 class TestCase(BaseSchema):
     name: str
-    response_format: Union[Dict[str, Any], List[Dict[str, Any]]]
-    assertions_by_invocation: List[AssertionByInvocation]
+    response_format: dict[str, Any] | list[dict[str, Any]]
+    assertions_by_invocation: list[AssertionByInvocation]
 
 
 class Employee(BaseModel):
@@ -49,12 +67,12 @@ class Employee(BaseModel):
 TEST_CASES = load_spec("responses", as_model=TestCase)
 
 
-def _make_tool(fn, *, name: str, description: str):
+def _make_tool(fn: Callable[..., str | None], *, name: str, description: str) -> dict[str, Any]:
     mock = MagicMock(side_effect=lambda *, name: fn(name=name))
     input_model = create_model(f"{name}_input", name=(str, ...))
 
     @tool(name, description=description, args_schema=input_model)
-    def _wrapped(name: str):
+    def _wrapped(name: str) -> Any:
         return mock(name=name)
 
     return {"tool": _wrapped, "mock": mock}
@@ -106,7 +124,7 @@ def get_employee_department(*, name: str) -> str | None:
 
     for assertion in case.assertions_by_invocation:
 
-        def on_request(request: httpx.Request) -> None:
+        def on_request(_request: httpx.Request) -> None:
             nonlocal llm_request_count
             llm_request_count += 1
 
@@ -123,7 +141,7 @@ def on_request(request: httpx.Request) -> None:
         agent = create_agent(
             model,
             tools=[role_tool["tool"], dept_tool["tool"]],
-            prompt=AGENT_PROMPT,
+            system_prompt=AGENT_PROMPT,
             response_format=tool_output,
         )
 
diff --git a/libs/langchain_v1/tests/unit_tests/agents/utils.py b/libs/langchain_v1/tests/unit_tests/agents/utils.py
index 21a9d438af556..7adccaf368b7b 100644
--- a/libs/langchain_v1/tests/unit_tests/agents/utils.py
+++ b/libs/langchain_v1/tests/unit_tests/agents/utils.py
@@ -1,5 +1,6 @@
 import json
 from pathlib import Path
+from typing import TypeVar
 
 from pydantic import BaseModel, ConfigDict
 from pydantic.alias_generators import to_camel
@@ -13,7 +14,10 @@ class BaseSchema(BaseModel):
     )
 
 
-def load_spec(spec_name: str, as_model: type[BaseModel]) -> list[BaseModel]:
+_T = TypeVar("_T", bound=BaseModel)
+
+
+def load_spec(spec_name: str, as_model: type[_T]) -> list[_T]:
     with (Path(__file__).parent / "specifications" / f"{spec_name}.json").open(
         "r", encoding="utf-8"
     ) as f: