cr

libs/langchain_v1/langchain/agents/factory.py

@@ -610,6 +610,28 @@ def create_agent(  # noqa: PLR0915
     if tools is None:
         tools = []
 
+    # Expand nested middleware: if a middleware declares additional middleware,
+    # insert them immediately after, recursively (parent -> children -> ...).
+    def _expand_middleware_sequence(
+        seq: "Sequence[AgentMiddleware[Any, Any]]",
+        _seen: set[int] | None = None,
+    ) -> list[AgentMiddleware[Any, Any]]:
+        expanded: list[AgentMiddleware[Any, Any]] = []
+        seen = _seen if _seen is not None else set()
+        for m in seq:
+            mid = id(m)
+            if mid in seen:
+                # Skip already expanded instances to avoid cycles/duplicates
+                continue
+            seen.add(mid)
+            expanded.append(m)
+            children = getattr(m, "middleware", []) or []
+            if children:
+                expanded.extend(_expand_middleware_sequence(children, seen))
+        return expanded
+
+    middleware = tuple(_expand_middleware_sequence(middleware))
+
     # Convert response format and setup structured output tools
     # Raw schemas are wrapped in AutoStrategy to preserve auto-detection intent.
     # AutoStrategy is converted to ToolStrategy upfront to calculate tools during agent creation,

libs/langchain_v1/langchain/agents/middleware/types.py

@@ -201,6 +201,14 @@ class AgentMiddleware(Generic[StateT, ContextT]):
     tools: list[BaseTool]
     """Additional tools registered by the middleware."""
 
+    # Allow middleware to declare additional middleware to be inserted
+    # immediately after this middleware. Factored recursively, the final order
+    # is parent -> children -> grandchildren, etc.
+    # This enables composition patterns where a middleware can bundle and
+    # expose other middleware without users having to pass them explicitly.
+    middleware: list["AgentMiddleware[Any, Any]"]
+    """Additional middleware registered by this middleware (optional)."""
+
     @property
     def name(self) -> str:
         """The name of the middleware instance.

libs/langchain_v1/tests/unit_tests/agents/middleware/test_nested_middleware.py

@@ -0,0 +1,45 @@
+from typing import Any
+
+from langchain.agents.factory import create_agent
+from langchain.agents.middleware.types import AgentMiddleware, AgentState
+from langchain_core.messages import SystemMessage
+
+from ..model import FakeToolCallingModel
+
+
+class _AppendMiddleware(AgentMiddleware):
+    def __init__(self, label: str, children: list[AgentMiddleware] | None = None) -> None:
+        # No tools registered by default
+        self.tools = []
+        # Optional nested middleware
+        self.middleware = children or []
+        self._label = label
+
+    def before_model(self, state: AgentState, runtime) -> dict[str, Any] | None:  # type: ignore[override]
+        return {"messages": [SystemMessage(self._label)]}
+
+
+def test_nested_middleware_ordering() -> None:
+    # Build nested chain: X -> Y -> Z
+    z = _AppendMiddleware("Z")
+    y = _AppendMiddleware("Y", [z])
+    x = _AppendMiddleware("X", [y])
+
+    # Siblings before and after X
+    a = _AppendMiddleware("A")
+    b = _AppendMiddleware("B")
+
+    agent = create_agent(
+        model=FakeToolCallingModel(),
+        tools=[],
+        system_prompt=None,
+        middleware=[a, x, b],
+    )
+
+    result = agent.invoke({"messages": [{"role": "user", "content": "hi"}]})
+    # The FakeToolCallingModel joins message contents with '-'
+    final_ai = result["messages"][-1]
+    content: str = final_ai.content  # type: ignore[assignment]
+
+    # Ensure correct in-order appearance: A -> X -> Y -> Z -> B
+    assert content.index("A") < content.index("X") < content.index("Y") < content.index("Z") < content.index("B")