diff --git a/CHANGELOG.md b/CHANGELOG.md
index 5f6cbbab70..d1ad3499b2 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -1402,6 +1402,12 @@ As of 2026-10-21, the milestone includes **829 merged pull requests**.
 
 ### Fixed
 
+- **channels:** IM replies that present files no longer resolve those files
+  on the event loop. For every artifact, both reply paths (`runs.wait` and
+  streaming) resolved the virtual path (`realpath`), resolved the outputs
+  root and checked the file with `stat` inline, stalling other channels and
+  requests on network-backed storage. They now do it in a worker. ([#7594])
+  
 - **channels:** A streamed Feishu reply no longer rewrites the same thread
   bindings on every chunk. Each outbound to the running card stored all of
   the reply's aliases again (about five `set_thread_id` writes per chunk: a
@@ -13196,3 +13202,4 @@ with **180 merged pull requests** since the first 2.0 milestone tag.
 [#7588]: https://github.com/bytedance/deer-flow/pull/7588
 [#7591]: https://github.com/bytedance/deer-flow/pull/7591
 [#7593]: https://github.com/bytedance/deer-flow/pull/7593
+[#7594]: https://github.com/bytedance/deer-flow/pull/7594
diff --git a/CHANGELOG_zh.md b/CHANGELOG_zh.md
index 8ce3376e8c..69084adcbb 100644
--- a/CHANGELOG_zh.md
+++ b/CHANGELOG_zh.md
@@ -1116,6 +1116,10 @@
 
 ### 修复
 
+- **渠道：** 展示文件的 IM 回复不再在事件循环上解析这些文件。此前两条回复路径（`runs.wait`
+  和流式）都会对每个产物内联解析虚拟路径（`realpath`）、解析输出根目录并用 `stat` 检查文件，
+  在网络存储上会让其他渠道和请求停顿。现在这些工作在工作线程中进行。([#7594])
+  
 - **渠道：** 飞书流式回复不再在每个分块上重写相同的线程绑定。此前每次向运行中卡片发送消息都会
   重新保存该回复的全部别名（每个分块约五次 `set_thread_id`：每次都是在存储锁下完整重写
   `store.json`，或一次数据库 upsert），而实际只有 `updated_at` 变化。现在非最终分块对每个
@@ -10519,9 +10523,9 @@ DeerFlow 2.0 是围绕"超级智能体"框架的彻底重写，核心包含子
 [#7530]: https://github.com/bytedance/deer-flow/pull/7530
 [#7557]: https://github.com/bytedance/deer-flow/pull/7557
 [#7560]: https://github.com/bytedance/deer-flow/pull/7560
-[#7580]: https://github.com/bytedance/deer-flow/pull/7580
 [#7576]: https://github.com/bytedance/deer-flow/pull/7576
 [#7580]: https://github.com/bytedance/deer-flow/pull/7580
 [#7588]: https://github.com/bytedance/deer-flow/pull/7588
 [#7591]: https://github.com/bytedance/deer-flow/pull/7591
 [#7593]: https://github.com/bytedance/deer-flow/pull/7593
+[#7594]: https://github.com/bytedance/deer-flow/pull/7594
diff --git a/backend/app/channels/AGENTS.md b/backend/app/channels/AGENTS.md
index 6ecd64dfc5..b15c773019 100644
--- a/backend/app/channels/AGENTS.md
+++ b/backend/app/channels/AGENTS.md
@@ -60,7 +60,7 @@ The `channel_agent_name` cache reloads after restart before a resumed turn route
 - `FeishuChannel._receive_single_file(...)` / `DingTalkChannel._receive_single_file(...)` — normalize filenames and claim/write through `write_upload_file_no_symlink(..., exclusive=True)`, retrying `FileExistsError` under the channel lock; the final basename drives agent-visible paths and sandbox sync
 - `sandbox_files.py` — non-mounted Feishu/DingTalk syncs acquire unique execution holders, drain blocking `update_file` workers across repeated cancellation, and release only after the last sandbox operation, so a parallel run cannot close the shared client mid-upload. Both channels request provider release after the final holder exits, parking standalone temporary sandboxes.
 - `_ingest_inbound_files(...)` and the underlying `ensure_uploads_dir` / `get_uploads_dir` — owner-scoped via the same kwarg
-- `_resolve_attachments` / `_prepare_artifact_delivery` — resolve output artifacts from the bound owner's bucket through `app.gateway.path_utils.resolve_outputs_confined_path`, the same outputs-only rule the artifact editor uses, so a sibling `uploads/`/`workspace/` path or a symlink planted in `outputs/` is skipped with a warning
+- `_resolve_attachments` / `_prepare_artifact_delivery` — resolve output artifacts from the bound owner's bucket through `app.gateway.path_utils.resolve_outputs_confined_path`, the same outputs-only rule the artifact editor uses, so a sibling `uploads/`/`workspace/` path or a symlink planted in `outputs/` is skipped with a warning; both reply paths run it via `to_thread` (`blocking_io/test_channel_artifact_delivery.py`)
 The cached value is reused for both the blocking (`runs.wait`) and streaming (`_handle_streaming_chat`) paths, so uploads and artifact delivery always target the same bucket even if a channel returns a rewritten `InboundMessage` from `receive_file`. The bucket id matches the memory bucket resolved by `_resolve_memory_user_id` (both normalize through `make_safe_user_id`).
 
 **Configuration** (`config.yaml` -> `channels`):
diff --git a/backend/app/channels/manager.py b/backend/app/channels/manager.py
index 829e0d293c..cd1c7741cd 100644
--- a/backend/app/channels/manager.py
+++ b/backend/app/channels/manager.py
@@ -1416,7 +1416,11 @@ def _prepare_artifact_delivery(
     *,
     user_id: str | None = None,
 ) -> tuple[str, list[ResolvedAttachment]]:
-    """Resolve attachments and append filename fallbacks to the text response."""
+    """Resolve attachments and append filename fallbacks to the text response.
+
+    Blocking: every artifact is resolved through ``realpath`` and checked with
+    ``stat``, so the async reply paths call this through ``asyncio.to_thread``.
+    """
     attachments: list[ResolvedAttachment] = []
     if not artifacts:
         return response_text, attachments
@@ -3472,7 +3476,7 @@ class ChannelManager:
         # Reuse the storage owner cached at the top of _handle_chat so uploads and
         # artifact delivery always resolve to the same bucket, even if a future
         # channel.receive_file returns a rewritten InboundMessage.
-        response_text, attachments = _prepare_artifact_delivery(thread_id, response_text, artifacts, user_id=storage_user_id)
+        response_text, attachments = await asyncio.to_thread(_prepare_artifact_delivery, thread_id, response_text, artifacts, user_id=storage_user_id)
 
         if not response_text:
             if attachments:
@@ -3646,7 +3650,7 @@ class ChannelManager:
             # Reuse the storage owner resolved by _handle_chat so artifact delivery
             # matches the upload bucket and we avoid re-running _safe_user_id_for_run
             # (and its possible filesystem touch) on the streaming-error path.
-            response_text, attachments = _prepare_artifact_delivery(thread_id, response_text, artifacts, user_id=storage_user_id)
+            response_text, attachments = await asyncio.to_thread(_prepare_artifact_delivery, thread_id, response_text, artifacts, user_id=storage_user_id)
             if step_limit_text:
                 # The last text is mid-task narration: say the run stopped, in
                 # place of the generic failure, so the user knows to continue.
diff --git a/backend/tests/blocking_io/test_channel_artifact_delivery.py b/backend/tests/blocking_io/test_channel_artifact_delivery.py
new file mode 100644
index 0000000000..180fa6a646
--- /dev/null
+++ b/backend/tests/blocking_io/test_channel_artifact_delivery.py
@@ -0,0 +1,122 @@
+"""Regression anchors: resolving reply artifacts must not block the event loop.
+
+When a run presents files, ``ChannelManager`` turns each virtual
+``/mnt/user-data/outputs/...`` path into a host path before delivering the
+reply: ``resolve_outputs_confined_path`` resolves it (``realpath``) and checks it
+against the resolved outputs root, then the file is checked with ``is_file`` and
+``stat``. Both reply paths (``runs.wait`` and ``runs.stream``) run that in a
+worker through ``asyncio.to_thread``.
+
+Each case drives the real manager with a real thread outputs directory under a
+temporary ``DEER_FLOW_HOME`` and asserts the outbound reply carries the
+resolved attachment, so an anchor cannot pass because the artifact was never
+resolved.
+"""
+
+from __future__ import annotations
+
+import asyncio
+from pathlib import Path
+from types import SimpleNamespace
+from unittest.mock import AsyncMock, MagicMock
+
+import pytest
+import pytest_asyncio
+from blockbuster import BlockingError
+
+from app.channels import manager as mgr
+from app.channels.message_bus import InboundMessage, MessageBus, OutboundMessage
+from app.channels.store import JsonChannelStore
+from deerflow.config import paths as paths_mod
+
+pytestmark = pytest.mark.asyncio
+
+_THREAD_ID = "thread-artifacts"
+_ARTIFACT = "/mnt/user-data/outputs/report.txt"
+
+
+def _final_state() -> dict:
+    return {
+        "messages": [
+            {"type": "human", "content": "make a report"},
+            {"type": "ai", "content": "", "tool_calls": [{"name": "present_files", "args": {"filepaths": [_ARTIFACT]}}]},
+            {"type": "tool", "name": "present_files", "content": "ok"},
+            {"type": "ai", "content": "Here is the report."},
+        ],
+        "artifacts": [_ARTIFACT],
+    }
+
+
+def _write_artifact(user_id: str) -> None:
+    outputs = paths_mod.get_paths().sandbox_outputs_dir(_THREAD_ID, user_id=user_id)
+    outputs.mkdir(parents=True, exist_ok=True)
+    (outputs / "report.txt").write_text("report", encoding="utf-8")
+
+
+@pytest_asyncio.fixture
+async def home(tmp_path: Path, monkeypatch: pytest.MonkeyPatch) -> Path:
+    monkeypatch.setenv("DEER_FLOW_HOME", str(tmp_path))
+    monkeypatch.setattr(paths_mod, "_paths", None)
+    storage_user_id = await asyncio.to_thread(mgr._safe_user_id_for_run, "user-1")
+    await asyncio.to_thread(_write_artifact, storage_user_id)
+    return tmp_path
+
+
+async def _deliver(home: Path, channel_name: str, client: MagicMock) -> OutboundMessage:
+    bus = MessageBus()
+    manager = mgr.ChannelManager(bus=bus, store=JsonChannelStore(path=home / "store.json"))
+    manager._client = client
+    finals: list[OutboundMessage] = []
+
+    async def capture(msg: OutboundMessage) -> None:
+        if msg.is_final:
+            finals.append(msg)
+
+    bus.subscribe_outbound(capture)
+    await manager.start()
+    try:
+        await bus.publish_inbound(InboundMessage(channel_name=channel_name, chat_id="chat-1", user_id="user-1", text="make a report", thread_ts="source-1"))
+        async with asyncio.timeout(5):
+            while not finals:
+                await asyncio.sleep(0.01)
+    finally:
+        await manager.stop()
+    return finals[0]
+
+
+def _client() -> MagicMock:
+    client = MagicMock()
+    client.threads.create = AsyncMock(return_value={"thread_id": _THREAD_ID})
+    client.threads.get = AsyncMock(return_value={"thread_id": _THREAD_ID})
+    client.threads.update = AsyncMock(return_value={"thread_id": _THREAD_ID})
+    return client
+
+
+async def test_wait_reply_resolves_artifacts_off_the_loop(home: Path) -> None:
+    client = _client()
+    client.runs.wait = AsyncMock(return_value=_final_state())
+
+    reply = await _deliver(home, "slack", client)
+
+    assert [attachment.filename for attachment in reply.attachments] == ["report.txt"]
+    assert reply.attachments[0].size == len("report")
+
+
+async def test_streamed_reply_resolves_artifacts_off_the_loop(home: Path) -> None:
+    async def stream(*_args, **_kwargs):
+        yield SimpleNamespace(event="values", data=_final_state())
+
+    client = _client()
+    client.runs.stream = MagicMock(side_effect=stream)
+
+    reply = await _deliver(home, "telegram", client)
+
+    assert [attachment.filename for attachment in reply.attachments] == ["report.txt"]
+    assert reply.attachments[0].size == len("report")
+
+
+async def test_blocking_probe_inline_artifact_delivery_actually_trips_the_gate(home: Path) -> None:
+    """Meta-check: the same resolution inline on the loop is caught by the gate."""
+    storage_user_id = await asyncio.to_thread(mgr._safe_user_id_for_run, "user-1")
+    with pytest.raises(BlockingError):
+        mgr._prepare_artifact_delivery(_THREAD_ID, "Here is the report.", [_ARTIFACT], user_id=storage_user_id)