fix(code): keep TODO and edit tools expanded (#4704)

TODO updates and file edits now remain expanded in the dcode TUI.

---

Keep `write_todos` and `edit_file` calls outside collapsed tool-group
summaries so their user-facing details remain visible in live and
restored transcripts.

Made by [Open
SWE](https://openswe.vercel.app/agents/019f5d2e-6138-71e1-91e5-95ba16811dad)

---------

Co-authored-by: open-swe[bot] <open-swe@users.noreply.github.com>
Co-authored-by: Mason Daugherty <mason@langchain.dev>
Co-authored-by: Mason Daugherty <github@mdrxy.com>

libs/code/deepagents_code/app.py

@@ -160,6 +160,16 @@ single thread imports it re-entrantly, but can trip CPython's per-module import
 deadlock detector when two threads cold-import overlapping modules.
 """
 
+_TOOL_GROUP_EXCLUSIONS = frozenset({"ask_user", "edit_file", "write_todos"})
+"""Tools that stay expanded instead of collapsing into step summaries.
+
+Each surfaces user-facing content worth keeping visible on its own — an
+interactive prompt (`ask_user`), a diff (`edit_file`), or a todo list
+(`write_todos`) — so it renders standalone and acts as a boundary between
+adjacent tool groups. Add a tool here only when its collapsed one-line
+summary would hide something the user needs to see.
+"""
+
 _MESSAGE_TIMESTAMP_FOOTER_CLASS = "message-timestamp-footer"
 """CSS class applied to individual message timestamp footer widgets."""
 
@@ -12128,12 +12138,16 @@ class DeepAgentsApp(App):
         # Eagerly fold tool calls into a single live summary so they are
         # collapsed from the moment they start, rather than rendering verbose
         # then snapping shut. A groupable tool joins (or opens) the current
-        # step's group; a diff folds into it; anything else is a step boundary
-        # that closes the group.
+        # step's group; a diff from a groupable tool folds into it; anything
+        # else is a step boundary that closes the group.
         is_groupable_tool = (
-            isinstance(widget, ToolCallMessage) and widget.tool_name != "ask_user"
+            isinstance(widget, ToolCallMessage)
+            and widget.tool_name not in _TOOL_GROUP_EXCLUSIONS
+        )
+        is_groupable_diff = (
+            isinstance(widget, DiffMessage)
+            and widget._tool_name not in _TOOL_GROUP_EXCLUSIONS
         )
-        is_diff = isinstance(widget, DiffMessage)
 
         # Store message data for virtualization
         message_data = MessageData.from_widget(widget)
@@ -12150,7 +12164,7 @@ class DeepAgentsApp(App):
         # folding it into the group hides it on the next frame — bouncing the
         # bottom-anchored transcript on every tool call.
         with self.batch_update():
-            if not (is_groupable_tool or is_diff):
+            if not (is_groupable_tool or is_groupable_diff):
                 self._close_active_tool_group()
                 # Re-derive groups for any tools mounted outside this path
                 # (resumed history), which carry no live group.
@@ -12176,7 +12190,7 @@ class DeepAgentsApp(App):
             ):
                 if is_groupable_tool:
                     self._active_tool_group.add_member(widget)
-                elif is_diff:
+                elif is_groupable_diff:
                     self._active_tool_group.add_collapsible(widget)
 
         self._schedule_message_height_measurement(message_data.id)
@@ -12341,7 +12355,7 @@ class DeepAgentsApp(App):
                     continue  # footers are transparent to grouping
                 if isinstance(child, ToolCallMessage):
                     groupable = (
-                        child.tool_name != "ask_user"
+                        child.tool_name not in _TOOL_GROUP_EXCLUSIONS
                         and child.is_success
                         and not child.has_class("-grouped")
                     )
@@ -12354,8 +12368,13 @@ class DeepAgentsApp(App):
                     run_collapsible.append(child)
                     continue
                 if isinstance(child, DiffMessage):
-                    # A diff belongs to the tool above it; never starts a run.
-                    if run_anchor is not None:
+                    # A diff belongs to the tool above it and never starts a
+                    # run: normally it folds into the open run, but a diff from
+                    # an excluded tool (e.g. edit_file) stays standalone and
+                    # ends the run so the edit and its diff remain visible.
+                    if child._tool_name in _TOOL_GROUP_EXCLUSIONS:
+                        await flush()
+                    elif run_anchor is not None:
                         run_collapsible.append(child)
                     continue
                 # Assistant text, notices, an existing summary, etc. end the run.

libs/code/deepagents_code/tui/textual_adapter.py

@@ -1114,7 +1114,11 @@ async def execute_task_textual(
                                 pending_text_by_namespace[ns_key] = ""
                             if record.diff:
                                 await adapter._mount_message(
-                                    DiffMessage(record.diff, record.display_path)
+                                    DiffMessage(
+                                        record.diff,
+                                        record.display_path,
+                                        tool_name=record.tool_name,
+                                    )
                                 )
 
                         # Reshow spinner only when all in-flight tools have

libs/code/deepagents_code/tui/widgets/message_store.py

@@ -159,6 +159,9 @@ class MessageData:
     diff_file_path: str | None = None
     """File path associated with the diff (DIFF messages only)."""
 
+    diff_tool_name: str | None = None
+    """Name of the file tool that produced the diff (DIFF messages only)."""
+
     # SKILL message fields - only populated for SKILL messages
     skill_name: str | None = None
     """Name of the skill that was invoked."""
@@ -280,6 +283,7 @@ class MessageData:
                 return DiffMessage(
                     self.content,
                     file_path=self.diff_file_path or "",
+                    tool_name=self.diff_tool_name,
                     id=self.id,
                 )
 
@@ -384,6 +388,7 @@ class MessageData:
                 content=widget._diff_content,
                 id=widget_id,
                 diff_file_path=widget._file_path,
+                diff_tool_name=widget._tool_name,
             )
 
         if isinstance(widget, SummarizationMessage):

libs/code/deepagents_code/tui/widgets/messages.py

@@ -3232,17 +3232,26 @@ class DiffMessage(Static):
     """
     """Diff syntax coloring per theme: additions, removals, muted context."""
 
-    def __init__(self, diff_content: str, file_path: str = "", **kwargs: Any) -> None:
+    def __init__(
+        self,
+        diff_content: str,
+        file_path: str = "",
+        *,
+        tool_name: str | None = None,
+        **kwargs: Any,
+    ) -> None:
         """Initialize a diff message.
 
         Args:
             diff_content: The unified diff content
             file_path: Path to the file being modified
+            tool_name: Name of the file tool that produced the diff
             **kwargs: Additional arguments passed to parent
         """
         super().__init__(**kwargs)
         self._diff_content = diff_content
         self._file_path = file_path
+        self._tool_name = tool_name
 
     def compose(self) -> ComposeResult:
         """Compose the diff message layout.

libs/code/tests/unit_tests/test_app.py

@@ -9001,6 +9001,7 @@ class TestMessageTimestampFooters:
 
         async with app.run_test() as pilot:
             await pilot.pause()
+            monkeypatch.setattr(app, "_check_hydration_below_needed", lambda: None)
             # Shrink the window so a small load archives messages above the
             # visible range, mirroring a long thread scrolled to the bottom.
             monkeypatch.setattr(app._message_store, "WINDOW_SIZE", 2)
@@ -24272,6 +24273,92 @@ class TestToolGroupCollapse:
             assert isinstance(rendered, Content)
             assert "Read 1 file, ran 1 shell command" in rendered.plain
 
+    @pytest.mark.parametrize("tool_name", ["ask_user", "edit_file", "write_todos"])
+    async def test_regroup_leaves_excluded_tools_expanded(self, tool_name: str) -> None:
+        """Excluded tools stay visible and split adjacent tool groups."""
+        from deepagents_code.tui.widgets.messages import ToolGroupSummary
+
+        app = DeepAgentsApp(agent=MagicMock(), thread_id="t-excluded-history")
+        app._load_thread_history = AsyncMock()  # ty: ignore
+        async with app.run_test() as pilot:
+            messages = app.query_one("#messages", Container)
+            await messages.remove_children()
+            before, excluded, after = await self._mount_tools(
+                pilot,
+                messages,
+                [
+                    ("before", "read_file", {"file_path": "a.py"}, "success"),
+                    ("excluded", tool_name, {}, "success"),
+                    ("after", "execute", {"command": "ls"}, "success"),
+                ],
+            )
+
+            await app._regroup_completed_tools()
+            await pilot.pause()
+
+            assert len(list(app.query(ToolGroupSummary))) == 2
+            assert before.display is False
+            assert excluded.display is True
+            assert not excluded.has_class("-grouped")
+            assert after.display is False
+
+    async def test_regroup_leaves_edit_diff_outside_later_tool_group(self) -> None:
+        """An edit diff arriving after a parallel read stays expanded."""
+        from deepagents_code.tui.widgets.messages import DiffMessage, ToolGroupSummary
+
+        app = DeepAgentsApp(agent=MagicMock(), thread_id="t-edit-diff-history")
+        app._load_thread_history = AsyncMock()  # ty: ignore
+        async with app.run_test() as pilot:
+            messages = app.query_one("#messages", Container)
+            await messages.remove_children()
+            edit, read = await self._mount_tools(
+                pilot,
+                messages,
+                [
+                    ("edit", "edit_file", {"file_path": "a.py"}, "success"),
+                    ("read", "read_file", {"file_path": "b.py"}, "success"),
+                ],
+            )
+            diff = DiffMessage("-old\n+new", "a.py", tool_name="edit_file")
+            await messages.mount(diff)
+            await pilot.pause()
+
+            await app._regroup_completed_tools()
+            await pilot.pause()
+
+            assert len(list(app.query(ToolGroupSummary))) == 1
+            assert edit.display is True
+            assert read.display is False
+            assert diff.display is True
+            assert not diff.has_class("-grouped")
+
+    async def test_regroup_leaves_consecutive_excluded_tools_expanded(self) -> None:
+        """Two adjacent excluded tools stay expanded with no summary between them."""
+        from deepagents_code.tui.widgets.messages import ToolGroupSummary
+
+        app = DeepAgentsApp(agent=MagicMock(), thread_id="t-excluded-adjacent")
+        app._load_thread_history = AsyncMock()  # ty: ignore
+        async with app.run_test() as pilot:
+            messages = app.query_one("#messages", Container)
+            await messages.remove_children()
+            edit, todos = await self._mount_tools(
+                pilot,
+                messages,
+                [
+                    ("edit", "edit_file", {"file_path": "a.py"}, "success"),
+                    ("todos", "write_todos", {}, "success"),
+                ],
+            )
+
+            await app._regroup_completed_tools()
+            await pilot.pause()
+
+            assert len(list(app.query(ToolGroupSummary))) == 0
+            assert edit.display is True
+            assert todos.display is True
+            assert not edit.has_class("-grouped")
+            assert not todos.has_class("-grouped")
+
     async def test_regroup_treats_timestamp_footer_as_transparent(self) -> None:
         """A timestamp footer between two tools does not split the run.
 
@@ -24455,6 +24542,79 @@ class TestToolGroupCollapse:
             assert tool.display is False
             await pilot.pause()
 
+    @pytest.mark.parametrize("tool_name", ["ask_user", "edit_file", "write_todos"])
+    async def test_mount_leaves_excluded_tools_expanded(self, tool_name: str) -> None:
+        """Excluded tools mount standalone instead of opening a live group."""
+        from deepagents_code.tui.widgets.messages import ToolCallMessage
+
+        app = DeepAgentsApp(agent=MagicMock(), thread_id="t-excluded-live")
+        app._load_thread_history = AsyncMock()  # ty: ignore
+        async with app.run_test():
+            messages = app.query_one("#messages", Container)
+            await messages.remove_children()
+
+            tool = ToolCallMessage(tool_name, {})
+            await app._mount_message(tool)
+
+            assert app._active_tool_group is None
+            assert tool.display is True
+            assert not tool.has_class("-grouped")
+
+    async def test_mount_leaves_edit_diff_outside_parallel_read_group(self) -> None:
+        """A late edit diff does not join the active parallel read group."""
+        from deepagents_code.tui.widgets.messages import DiffMessage, ToolGroupSummary
+
+        app = DeepAgentsApp(agent=MagicMock(), thread_id="t-edit-diff-live")
+        app._load_thread_history = AsyncMock()  # ty: ignore
+        async with app.run_test():
+            messages = app.query_one("#messages", Container)
+            await messages.remove_children()
+
+            edit = ToolCallMessage("edit_file", {"file_path": "a.py"})
+            read = ToolCallMessage("read_file", {"file_path": "b.py"})
+            diff = DiffMessage("-old\n+new", "a.py", tool_name="edit_file")
+            await app._mount_message(edit)
+            await app._mount_message(read)
+            await app._mount_message(diff)
+
+            assert len(list(app.query(ToolGroupSummary))) == 1
+            assert edit.display is True
+            assert read.display is False
+            assert diff.display is True
+            assert not diff.has_class("-grouped")
+            assert app._active_tool_group is None
+
+    @pytest.mark.parametrize("tool_name", ["ask_user", "edit_file", "write_todos"])
+    async def test_mount_excluded_tool_closes_open_live_group(
+        self, tool_name: str
+    ) -> None:
+        """An excluded tool closes the live group; the next tool opens a new one."""
+        from deepagents_code.tui.widgets.messages import ToolCallMessage
+
+        app = DeepAgentsApp(agent=MagicMock(), thread_id="t-excluded-boundary")
+        app._load_thread_history = AsyncMock()  # ty: ignore
+        async with app.run_test():
+            messages = app.query_one("#messages", Container)
+            await messages.remove_children()
+
+            first = ToolCallMessage("execute", {"command": "ls"})
+            excluded = ToolCallMessage(tool_name, {})
+            later = ToolCallMessage("read_file", {"file_path": "a.py"})
+
+            await app._mount_message(first)
+            opened = app._active_tool_group
+            assert opened is not None  # groupable tool opened a live group
+
+            await app._mount_message(excluded)
+            assert app._active_tool_group is None  # excluded tool closed it
+            assert excluded.display is True
+            assert not excluded.has_class("-grouped")
+
+            await app._mount_message(later)
+            # A fresh groupable tool opens a new group, not the closed one.
+            assert app._active_tool_group is not None
+            assert app._active_tool_group is not opened
+
     async def test_group_survives_idle_after_completion(self) -> None:
         """A folded group stays mounted across completion, idle, and a boundary.
 

libs/code/tests/unit_tests/test_transcript_virtualization.py

@@ -200,7 +200,8 @@ class TestScrollDrivenHydration:
             # Archive the newest rows below the window (the state after the user
             # has scrolled up and older history was mounted in their place).
             monkeypatch.setattr(app._message_store, "WINDOW_SIZE", 3)
-            monkeypatch.setattr(app._message_store, "HYDRATE_BUFFER", 20)
+            monkeypatch.setattr(app._message_store, "HYDRATE_BUFFER", 2)
+            monkeypatch.setattr(app, "_check_hydration_needed", lambda: None)
             messages = app.query_one("#messages", Container)
             await app._prune_messages_below_window(messages)
             await pilot.pause()

libs/code/tests/unit_tests/tui/widgets/test_message_store.py

@@ -178,13 +178,19 @@ class TestMessageData:
     def test_diff_message_roundtrip(self):
         """Test DiffMessage serialization and deserialization."""
         diff_content = "--- a/file.py\n+++ b/file.py\n@@ -1 +1 @@\n-old\n+new"
-        original = DiffMessage(diff_content, file_path="src/file.py", id="test-diff-1")
+        original = DiffMessage(
+            diff_content,
+            file_path="src/file.py",
+            tool_name="edit_file",
+            id="test-diff-1",
+        )
 
         # Serialize
         data = MessageData.from_widget(original)
         assert data.type == MessageType.DIFF
         assert data.content == diff_content
         assert data.diff_file_path == "src/file.py"
+        assert data.diff_tool_name == "edit_file"
         assert data.id == "test-diff-1"
 
         # Deserialize
@@ -192,6 +198,7 @@ class TestMessageData:
         assert isinstance(restored, DiffMessage)
         assert restored._diff_content == diff_content
         assert restored._file_path == "src/file.py"
+        assert restored._tool_name == "edit_file"
         assert restored.id == "test-diff-1"
 
     def test_summarization_message_roundtrip(self):