diff --git a/backend/packages/harness/deerflow/persistence/engine.py b/backend/packages/harness/deerflow/persistence/engine.py
index 127ee73f3..9aa58d7f3 100644
--- a/backend/packages/harness/deerflow/persistence/engine.py
+++ b/backend/packages/harness/deerflow/persistence/engine.py
@@ -10,6 +10,7 @@ None and fall back to in-memory implementations.
 
 from __future__ import annotations
 
+import asyncio
 import json
 import logging
 
@@ -97,7 +98,11 @@ async def init_engine(
 
         from sqlalchemy import event
 
-        os.makedirs(sqlite_dir or ".", exist_ok=True)
+        # Offload the directory creation: ``init_engine`` runs on the FastAPI
+        # lifespan event loop, and a sync ``os.makedirs`` (a stat + mkdir
+        # syscall) blocks it during startup. Mirrors the #1912 fix for the
+        # checkpointer's ``ensure_sqlite_parent_dir``.
+        await asyncio.to_thread(os.makedirs, sqlite_dir or ".", exist_ok=True)
         _engine = create_async_engine(url, echo=echo, json_serializer=_json_serializer)
 
         # Enable WAL on every new connection. SQLite PRAGMA settings are
diff --git a/backend/tests/blocking_io/test_persistence_engine_sqlite.py b/backend/tests/blocking_io/test_persistence_engine_sqlite.py
new file mode 100644
index 000000000..f7cc3daa8
--- /dev/null
+++ b/backend/tests/blocking_io/test_persistence_engine_sqlite.py
@@ -0,0 +1,64 @@
+"""Regression test: persistence-engine sqlite dir setup must run off the loop.
+
+Anchors the production offload in `persistence/engine.py:init_engine`, where the
+SQLite data directory is created with `os.makedirs`. `init_engine` runs on the
+FastAPI lifespan event loop, so a sync `os.makedirs` (a stat + mkdir syscall)
+there blocks startup — the same class of bug fixed for the checkpointer's
+`ensure_sqlite_parent_dir` in #1912 (see `test_sqlite_lifespan.py`).
+
+This invokes the production `init_engine(backend="sqlite", ...)` under the strict
+Blockbuster context with a `sqlite_dir` that does not yet exist, so `os.makedirs`
+actually runs. The async engine/session machinery is mocked out so the only host
+filesystem operation under test is the directory creation; if it regresses to run
+directly on the event loop, Blockbuster raises `BlockingError` and this fails.
+"""
+
+from __future__ import annotations
+
+from pathlib import Path
+from unittest.mock import AsyncMock, MagicMock, patch
+
+import pytest
+
+# Pre-import so `init_engine`'s lazy ``import deerflow.persistence.models`` is a
+# cached no-op rather than a file read under the strict gate.
+import deerflow.persistence.models  # noqa: E402,F401
+from deerflow.persistence import engine as engine_mod  # noqa: E402
+
+pytestmark = pytest.mark.asyncio
+
+
+def _noop_listens_for(*_args, **_kwargs):
+    """Decorator factory that registers nothing (mock engine has no real events)."""
+
+    def _decorator(fn):
+        return fn
+
+    return _decorator
+
+
+async def test_init_engine_sqlite_dir_setup_does_not_block_event_loop(tmp_path: Path) -> None:
+    data_dir = tmp_path / "newsubdir"  # does not exist yet -> os.makedirs runs
+    db_file = data_dir / "app.db"
+
+    mock_conn = AsyncMock()
+    begin_ctx = AsyncMock()
+    begin_ctx.__aenter__.return_value = mock_conn
+    begin_ctx.__aexit__.return_value = False
+    mock_engine = MagicMock()
+    mock_engine.begin.return_value = begin_ctx
+    mock_engine.dispose = AsyncMock()
+
+    with (
+        patch.object(engine_mod, "create_async_engine", return_value=mock_engine),
+        patch.object(engine_mod, "async_sessionmaker", return_value=MagicMock()),
+        patch("sqlalchemy.event.listens_for", _noop_listens_for),
+    ):
+        await engine_mod.init_engine(
+            backend="sqlite",
+            url=f"sqlite+aiosqlite:///{db_file}",
+            sqlite_dir=str(data_dir),
+        )
+        assert data_dir.exists()
+
+    await engine_mod.close_engine()