fix(setup-agent): keep a failed discard from replacing the publish outcome (#7564)

* fix(setup-agent): keep a failed discard from replacing the publish outcome

* test(setup-agent): pin the post-publish discard and assert the real temp name

---------

Co-authored-by: Willem Jiang <willem.jiang@gmail.com>

backend/packages/harness/deerflow/tools/builtins/setup_agent_tool.py

@@ -14,6 +14,7 @@ from deerflow.knowledge_scope import canonicalize_knowledge_scope
 from deerflow.persistence.agents import get_agent_store
 from deerflow.runtime.user_context import resolve_runtime_user_id
 from deerflow.tools.types import Runtime
+from deerflow.utils.atomic_write import cleanup_failed_atomic_write
 
 logger = logging.getLogger(__name__)
 
@@ -48,7 +49,9 @@ def _publish_text(target: Path, text: str) -> None:
         temporary.replace(target)
     finally:
         if temporary is not None:
-            temporary.unlink(missing_ok=True)
+            # The publish outcome is already decided; a temp we cannot discard
+            # must not replace it with the cleanup error.
+            cleanup_failed_atomic_write(temporary)
 
 
 @tool(parse_docstring=True)

backend/tests/test_setup_agent_atomic_cleanup.py

@@ -0,0 +1,78 @@
+"""A failed discard must not replace the outcome of an agent file publish.
+
+``_publish_text`` writes an agent's ``SOUL.md``/config through a same-directory
+temporary. On Windows a scanner or a peer holding the handle can make the
+removal fail after a successful ``os.replace`` — or while another error is
+already propagating — and the cleanup error used to win.
+"""
+
+from __future__ import annotations
+
+from pathlib import Path
+
+import pytest
+
+
+@pytest.fixture
+def real_unlink():
+    return Path.unlink
+
+
+def _fail_unlink_on(monkeypatch, predicate, real_unlink):
+    def unlink(self, *args, **kwargs):
+        if predicate(self):
+            raise PermissionError("temporary unlink failed")
+        return real_unlink(self, *args, **kwargs)
+
+    monkeypatch.setattr(Path, "unlink", unlink)
+
+
+def test_publish_text_keeps_the_replace_error_when_cleanup_fails(tmp_path, monkeypatch, real_unlink):
+    from deerflow.tools.builtins import setup_agent_tool as module
+
+    target = tmp_path / "SOUL.md"
+    target.write_text("original", encoding="utf-8")
+
+    def _failing_replace(self, *args, **kwargs):
+        raise OSError("replace failed")
+
+    monkeypatch.setattr(Path, "replace", _failing_replace)
+    _fail_unlink_on(monkeypatch, lambda path: path.suffix == ".tmp", real_unlink)
+
+    with pytest.raises(OSError, match="replace failed"):
+        module._publish_text(target, "replacement")
+
+    assert target.read_text(encoding="utf-8") == "original"
+
+
+def test_publish_text_still_succeeds_and_leaves_no_temporary(tmp_path, caplog):
+    from deerflow.tools.builtins import setup_agent_tool as module
+
+    target = tmp_path / "SOUL.md"
+    target.write_text("original", encoding="utf-8")
+    caplog.clear()
+
+    with caplog.at_level("WARNING", logger="deerflow.utils.atomic_write"):
+        module._publish_text(target, "replacement")
+
+    assert target.read_text(encoding="utf-8") == "replacement"
+    # The staged name comes from ``NamedTemporaryFile`` (``tmpXXXXXXXX.tmp``), so
+    # assert the whole directory: nothing leaked and nothing else was created.
+    assert sorted(p.name for p in tmp_path.iterdir()) == ["SOUL.md"]
+    assert caplog.text == ""
+
+
+def test_publish_text_survives_a_discard_that_fails_after_publishing(tmp_path, monkeypatch, caplog, real_unlink):
+    """Removal can fail after a successful replace; the publish must still stand."""
+    from deerflow.tools.builtins import setup_agent_tool as module
+
+    target = tmp_path / "SOUL.md"
+    target.write_text("original", encoding="utf-8")
+    caplog.clear()
+    _fail_unlink_on(monkeypatch, lambda path: path.suffix == ".tmp", real_unlink)
+
+    with caplog.at_level("WARNING", logger="deerflow.utils.atomic_write"):
+        module._publish_text(target, "replacement")
+
+    assert target.read_text(encoding="utf-8") == "replacement"
+    assert "Failed to remove temporary output file" in caplog.text