Allowing change of max turns in agent as tool

src/agents/agent.py

@@ -384,6 +384,7 @@ def as_tool(
         custom_output_extractor: Callable[[RunResult], Awaitable[str]] | None = None,
         is_enabled: bool
         | Callable[[RunContextWrapper[Any], AgentBase[Any]], MaybeAwaitable[bool]] = True,
+        max_turns: int | None = None,
     ) -> Tool:
         """Transform this agent into a tool, callable by other agents.
 
@@ -402,6 +403,8 @@ def as_tool(
             is_enabled: Whether the tool is enabled. Can be a bool or a callable that takes the run
                 context and agent and returns whether the tool is enabled. Disabled tools are hidden
                 from the LLM at runtime.
+            max_turns: The maximum number of turns the agent can take when running as a tool.
+                If not provided, the default value will be used.
         """
 
         @function_tool(
@@ -413,9 +416,7 @@ async def run_agent(context: RunContextWrapper, input: str) -> str:
             from .run import Runner
 
             output = await Runner.run(
-                starting_agent=self,
-                input=input,
-                context=context.context,
+                starting_agent=self, input=input, context=context.context, max_turns=max_turns
             )
             if custom_output_extractor:
                 return await custom_output_extractor(output)

src/agents/run.py

@@ -236,7 +236,7 @@ async def run(
         input: str | list[TResponseInputItem],
         *,
         context: TContext | None = None,
-        max_turns: int = DEFAULT_MAX_TURNS,
+        max_turns: int | None = DEFAULT_MAX_TURNS,
         hooks: RunHooks[TContext] | None = None,
         run_config: RunConfig | None = None,
         previous_response_id: str | None = None,
@@ -277,6 +277,8 @@ async def run(
             agent. Agents may perform handoffs, so we don't know the specific type of the output.
         """
         runner = DEFAULT_AGENT_RUNNER
+        if max_turns is None:
+            max_turns = DEFAULT_MAX_TURNS
         return await runner.run(
             starting_agent,
             input,

tests/test_agent_as_tool_max_turns.py

@@ -0,0 +1,84 @@
+from __future__ import annotations
+
+import json
+
+import pytest
+
+from agents import Agent, MaxTurnsExceeded, Runner
+from agents.items import TResponseOutputItem
+from agents.run import DEFAULT_MAX_TURNS
+
+from .fake_model import FakeModel
+from .test_responses import get_function_tool, get_function_tool_call, get_text_message
+
+
+@pytest.mark.asyncio
+async def test_runner_run_max_turns_none_defaults_to_constant():
+    model = FakeModel()
+    agent = Agent(
+        name="test_runner_max_turns_none",
+        model=model,
+        tools=[get_function_tool("tool", "ok")],
+    )
+
+    # Prepare 11 turns (DEFAULT_MAX_TURNS is 10) to ensure exceeding default.
+    func_output = json.dumps({"a": "b"})
+    turns: list[list[TResponseOutputItem] | Exception] = []
+    for i in range(1, DEFAULT_MAX_TURNS + 1):
+        turns.append([get_text_message(str(i)), get_function_tool_call("tool", func_output)])
+    model.add_multiple_turn_outputs(turns)
+
+    # Passing None should make Runner default to DEFAULT_MAX_TURNS (10), so 11th turn exceeds.
+    with pytest.raises(MaxTurnsExceeded):
+        await Runner.run(agent, input="go", max_turns=None)
+
+
+@pytest.mark.asyncio
+async def test_agent_as_tool_forwards_max_turns():
+    # Inner agent will exceed when limited to 1 turn.
+    inner_model = FakeModel()
+    inner_agent = Agent(
+        name="inner",
+        model=inner_model,
+        tools=[get_function_tool("some_function", "ok")],
+    )
+
+    # Make inner agent require more than 1 turn.
+    func_output = json.dumps({"x": 1})
+    inner_turns: list[list[TResponseOutputItem] | Exception] = [
+        [get_text_message("t1"), get_function_tool_call("some_function", func_output)],
+        [get_text_message("t2"), get_function_tool_call("some_function", func_output)],
+    ]
+    inner_model.add_multiple_turn_outputs(inner_turns)
+
+    # Wrap inner agent as a tool with max_turns=1.
+    wrapped_tool = inner_agent.as_tool(
+        tool_name="inner_tool",
+        tool_description="runs inner agent",
+        max_turns=1,
+    )
+
+    # Orchestrator will call the wrapped tool twice, causing inner to exceed its max_turns.
+    outer_model = FakeModel()
+    orchestrator = Agent(
+        name="orchestrator",
+        model=outer_model,
+        tools=[wrapped_tool],
+    )
+
+    # Outer model asks to call the tool once;
+    # exceeding happens inside the tool call when inner runs.
+    outer_turns: list[list[TResponseOutputItem] | Exception] = [
+        [get_function_tool_call("inner_tool")],
+    ]
+    outer_model.add_multiple_turn_outputs(outer_turns)
+
+    # Since tool default error handling returns a string on error,
+    # the run should not raise here.
+    result = await Runner.run(orchestrator, input="start")
+
+    # The tool call error should be surfaced as a message back to the model;
+    # ensure we have some output.
+    # We don't assert exact message text to avoid brittleness;
+    # just ensure the run completed with items.
+    assert len(result.new_items) >= 1