diff --git a/EvoScientist/EvoScientist.py b/EvoScientist/EvoScientist.py
index acfc09a..cfdadc8 100644
--- a/EvoScientist/EvoScientist.py
+++ b/EvoScientist/EvoScientist.py
@@ -23,7 +23,7 @@ from .backends import CustomSandboxBackend, MergedReadOnlyBackend
 from .config import get_effective_config, apply_config_to_env
 from .llm import get_chat_model
 from .mcp import load_mcp_tools
-from .middleware import create_skills_middleware, create_memory_middleware
+from .middleware import create_memory_middleware
 from .prompts import RESEARCHER_INSTRUCTIONS, get_system_prompt
 from .utils import load_subagents
 from .tools import tavily_search, think_tool, skill_manager
@@ -136,6 +136,7 @@ def _build_base_kwargs(base_backend, base_middleware):
         subagents=subs,
         middleware=base_middleware,
         system_prompt=SYSTEM_PROMPT,
+        skills=["/skills/"],
     )
 
 
@@ -176,6 +177,7 @@ def load_mcp_and_build_kwargs(base_backend, base_middleware):
         subagents=subs,
         middleware=base_middleware,
         system_prompt=SYSTEM_PROMPT,
+        skills=["/skills/"],
     )
 
 
@@ -185,7 +187,6 @@ prompt_refs = {
 
 base_middleware = [
     create_memory_middleware(MEMORY_DIR, extraction_model=chat_model),
-    create_skills_middleware(backend),
 ]
 
 # Default agent (no checkpointer) — used by langgraph dev / LangSmith / notebooks.
@@ -253,7 +254,6 @@ def create_cli_agent(workspace_dir: str | None = None, checkpointer=None):
 
     mw = [
         create_memory_middleware(MEMORY_DIR, extraction_model=chat_model),
-        create_skills_middleware(be),
     ]
 
     # Re-load MCP tools from current config (picks up /mcp add changes)
diff --git a/EvoScientist/__init__.py b/EvoScientist/__init__.py
index c581660..ef2d113 100644
--- a/EvoScientist/__init__.py
+++ b/EvoScientist/__init__.py
@@ -28,8 +28,6 @@ _EXPORTS: dict[str, tuple[str, str]] = {
     "MODELS": (".llm", "MODELS"),
     "list_models": (".llm", "list_models"),
     "DEFAULT_MODEL": (".llm", "DEFAULT_MODEL"),
-    # Middleware
-    "create_skills_middleware": (".middleware", "create_skills_middleware"),
     # Prompts
     "get_system_prompt": (".prompts", "get_system_prompt"),
     "RESEARCHER_INSTRUCTIONS": (".prompts", "RESEARCHER_INSTRUCTIONS"),
diff --git a/EvoScientist/middleware.py b/EvoScientist/middleware.py
index 1924f22..0ecac63 100644
--- a/EvoScientist/middleware.py
+++ b/EvoScientist/middleware.py
@@ -5,7 +5,6 @@ from __future__ import annotations
 from typing import TYPE_CHECKING
 
 from deepagents.backends import FilesystemBackend
-from deepagents.middleware.skills import SkillsMiddleware
 
 from .memory import EvoMemoryMiddleware
 from .paths import MEMORY_DIR as _DEFAULT_MEMORY_DIR
@@ -14,27 +13,6 @@ if TYPE_CHECKING:
     from langchain.chat_models import BaseChatModel
 
 
-def create_skills_middleware(
-    composite_backend,
-) -> SkillsMiddleware:
-    """Create a SkillsMiddleware that loads skills.
-
-    Uses the CompositeBackend directly so that skill paths in the system
-    prompt match the ``/skills/`` route (e.g. ``/skills/find-skills/SKILL.md``).
-
-    Args:
-        composite_backend: The CompositeBackend that routes ``/skills/`` to
-            the MergedReadOnlyBackend.
-
-    Returns:
-        Configured SkillsMiddleware instance
-    """
-    return SkillsMiddleware(
-        backend=composite_backend,
-        sources=["/skills/"],
-    )
-
-
 def create_memory_middleware(
     memory_dir: str = str(_DEFAULT_MEMORY_DIR),
     extraction_model: BaseChatModel | None = None,