feat: updated file management for main agent

surfsense_backend/tests/unit/middleware/test_filesystem_backends.py

@@ -36,11 +36,18 @@ def test_backend_resolver_returns_multi_root_backend_for_single_root(tmp_path: P
 def test_backend_resolver_uses_cloud_mode_by_default():
     resolver = build_backend_resolver(FilesystemSelection())
     backend = resolver(_RuntimeStub())
-    # StateBackend class name check keeps this test decoupled
-    # from internal deepagents runtime class identity.
+    # When no search_space_id is provided we fall back to StateBackend so
+    # sub-agents / tests without DB access still work.
     assert backend.__class__.__name__ == "StateBackend"
 
 
+def test_backend_resolver_uses_kb_postgres_in_cloud_with_search_space():
+    resolver = build_backend_resolver(FilesystemSelection(), search_space_id=42)
+    backend = resolver(_RuntimeStub())
+    assert backend.__class__.__name__ == "KBPostgresBackend"
+    assert backend.search_space_id == 42
+
+
 def test_backend_resolver_returns_multi_root_backend_for_multiple_roots(tmp_path: Path):
     root_one = tmp_path / "resume"
     root_two = tmp_path / "notes"

surfsense_backend/tests/unit/middleware/test_filesystem_middleware.py

@@ -0,0 +1,204 @@
+"""Unit tests for the SurfSense filesystem middleware new behaviors.
+
+Covers:
+* cloud cwd defaults to ``/documents`` and relative paths resolve under it
+* cloud writes outside ``/documents/`` are rejected unless basename starts
+  with ``temp_``
+* cloud writes/edits to the anonymous document are rejected (read-only)
+* helper methods on the middleware (``_resolve_relative``,
+  ``_check_cloud_write_namespace``, ``_default_cwd``)
+
+These tests use ``__new__`` to bypass the heavy ``__init__`` and exercise
+the helper methods directly so the test surface stays narrow and fast.
+"""
+
+from __future__ import annotations
+
+from types import SimpleNamespace
+
+import pytest
+
+from app.agents.new_chat.filesystem_selection import FilesystemMode
+from app.agents.new_chat.middleware.filesystem import (
+    SurfSenseFilesystemMiddleware,
+    _build_filesystem_system_prompt,
+    _build_tool_descriptions,
+)
+
+pytestmark = pytest.mark.unit
+
+
+def _make_middleware(mode: FilesystemMode = FilesystemMode.CLOUD):
+    middleware = SurfSenseFilesystemMiddleware.__new__(SurfSenseFilesystemMiddleware)
+    middleware._filesystem_mode = mode
+    return middleware
+
+
+def _runtime(state: dict | None = None) -> SimpleNamespace:
+    return SimpleNamespace(state=state or {})
+
+
+class TestCloudCwdDefaults:
+    def test_default_cwd_in_cloud_is_documents_root(self):
+        m = _make_middleware()
+        assert m._default_cwd() == "/documents"
+
+    def test_default_cwd_in_desktop_is_root(self):
+        m = _make_middleware(FilesystemMode.DESKTOP_LOCAL_FOLDER)
+        assert m._default_cwd() == "/"
+
+    def test_current_cwd_uses_state_when_set(self):
+        m = _make_middleware()
+        runtime = _runtime({"cwd": "/documents/notes"})
+        assert m._current_cwd(runtime) == "/documents/notes"
+
+    def test_current_cwd_falls_back_to_default(self):
+        m = _make_middleware()
+        runtime = _runtime({})
+        assert m._current_cwd(runtime) == "/documents"
+
+    def test_current_cwd_ignores_invalid(self):
+        m = _make_middleware()
+        runtime = _runtime({"cwd": "not-absolute"})
+        assert m._current_cwd(runtime) == "/documents"
+
+
+class TestRelativePathResolution:
+    def test_relative_path_resolves_against_cwd(self):
+        m = _make_middleware()
+        runtime = _runtime({"cwd": "/documents/projects"})
+        assert (
+            m._resolve_relative("notes.md", runtime) == "/documents/projects/notes.md"
+        )
+
+    def test_relative_path_with_dotdot(self):
+        m = _make_middleware()
+        runtime = _runtime({"cwd": "/documents/a/b"})
+        assert m._resolve_relative("../c.md", runtime) == "/documents/a/c.md"
+
+    def test_absolute_path_is_kept(self):
+        m = _make_middleware()
+        runtime = _runtime({"cwd": "/documents"})
+        assert m._resolve_relative("/other/x.md", runtime) == "/other/x.md"
+
+    def test_empty_path_returns_cwd(self):
+        m = _make_middleware()
+        runtime = _runtime({"cwd": "/documents/projects"})
+        assert m._resolve_relative("", runtime) == "/documents/projects"
+
+
+class TestCloudWriteNamespacePolicy:
+    def test_documents_path_allowed(self):
+        m = _make_middleware()
+        runtime = _runtime()
+        assert m._check_cloud_write_namespace("/documents/foo.md", runtime) is None
+
+    def test_documents_root_allowed(self):
+        m = _make_middleware()
+        runtime = _runtime()
+        assert m._check_cloud_write_namespace("/documents", runtime) is None
+
+    def test_temp_basename_anywhere_allowed(self):
+        m = _make_middleware()
+        runtime = _runtime()
+        assert m._check_cloud_write_namespace("/temp_scratch.md", runtime) is None
+        assert m._check_cloud_write_namespace("/foo/temp_x.md", runtime) is None
+        assert m._check_cloud_write_namespace("/documents/temp_x.md", runtime) is None
+
+    def test_other_paths_rejected(self):
+        m = _make_middleware()
+        runtime = _runtime()
+        err = m._check_cloud_write_namespace("/foo/bar.md", runtime)
+        assert err is not None
+        assert "must target /documents" in err
+
+    def test_anon_doc_path_is_read_only(self):
+        m = _make_middleware()
+        runtime = _runtime(
+            {
+                "kb_anon_doc": {
+                    "path": "/documents/uploaded.xml",
+                    "title": "uploaded",
+                    "content": "",
+                    "chunks": [],
+                }
+            }
+        )
+        err = m._check_cloud_write_namespace("/documents/uploaded.xml", runtime)
+        assert err is not None
+        assert "read-only" in err
+
+    def test_desktop_mode_skips_namespace_policy(self):
+        m = _make_middleware(FilesystemMode.DESKTOP_LOCAL_FOLDER)
+        runtime = _runtime()
+        assert m._check_cloud_write_namespace("/random/path.md", runtime) is None
+
+
+class TestModeSpecificPrompts:
+    """The prompt and tool descriptions must only describe the active mode.
+
+    Cross-mode noise wastes tokens and confuses the model with rules it
+    cannot use this session.
+    """
+
+    def test_cloud_prompt_omits_desktop_section(self):
+        prompt = _build_filesystem_system_prompt(
+            FilesystemMode.CLOUD, sandbox_available=False
+        )
+        assert "Local Folder Mode" not in prompt
+        assert "mount-prefixed" not in prompt
+        assert "Persistence Rules" in prompt
+        assert "/documents" in prompt
+        assert "temp_" in prompt
+
+    def test_desktop_prompt_omits_cloud_persistence_rules(self):
+        prompt = _build_filesystem_system_prompt(
+            FilesystemMode.DESKTOP_LOCAL_FOLDER, sandbox_available=False
+        )
+        assert "Persistence Rules" not in prompt
+        assert "Workspace Tree" not in prompt
+        assert "<priority_documents>" not in prompt
+        assert "Local Folder Mode" in prompt
+        assert "mount-prefixed" in prompt
+
+    def test_cloud_tool_descs_omit_desktop_phrases(self):
+        descs = _build_tool_descriptions(FilesystemMode.CLOUD)
+        for name in (
+            "write_file",
+            "edit_file",
+            "move_file",
+            "mkdir",
+            "list_tree",
+            "grep",
+        ):
+            text = descs[name]
+            assert "Desktop" not in text, f"{name} leaks desktop hints"
+            assert "Cloud mode:" not in text, f"{name} qualifies a cloud-only desc"
+
+    def test_desktop_tool_descs_omit_cloud_phrases(self):
+        descs = _build_tool_descriptions(FilesystemMode.DESKTOP_LOCAL_FOLDER)
+        for name in (
+            "write_file",
+            "edit_file",
+            "move_file",
+            "mkdir",
+            "list_tree",
+            "grep",
+        ):
+            text = descs[name]
+            assert "Cloud" not in text, f"{name} leaks cloud hints"
+            assert "/documents/" not in text, f"{name} mentions cloud namespace"
+            assert "temp_" not in text, f"{name} mentions cloud temp_ semantics"
+
+    def test_sandbox_addendum_appended_when_available(self):
+        prompt = _build_filesystem_system_prompt(
+            FilesystemMode.CLOUD, sandbox_available=True
+        )
+        assert "execute_code" in prompt
+        assert "Code Execution" in prompt
+
+    def test_sandbox_addendum_absent_when_unavailable(self):
+        prompt = _build_filesystem_system_prompt(
+            FilesystemMode.CLOUD, sandbox_available=False
+        )
+        assert "execute_code" not in prompt

surfsense_backend/tests/unit/middleware/test_filesystem_verification.py

@@ -11,25 +11,6 @@ from app.agents.new_chat.middleware.multi_root_local_folder_backend import (
 pytestmark = pytest.mark.unit
 
 
-class _BackendWithRawRead:
-    def __init__(self, content: str) -> None:
-        self._content = content
-
-    def read(self, file_path: str, offset: int = 0, limit: int = 200000) -> str:
-        del file_path, offset, limit
-        return "     1\tline1\n     2\tline2"
-
-    async def aread(self, file_path: str, offset: int = 0, limit: int = 200000) -> str:
-        return self.read(file_path, offset, limit)
-
-    def read_raw(self, file_path: str) -> str:
-        del file_path
-        return self._content
-
-    async def aread_raw(self, file_path: str) -> str:
-        return self.read_raw(file_path)
-
-
 class _RuntimeNoSuggestedPath:
     state = {"file_operation_contract": {}}
 
@@ -39,40 +20,19 @@ class _RuntimeWithSuggestedPath:
         self.state = {"file_operation_contract": {"suggested_path": suggested_path}}
 
 
-def test_verify_written_content_prefers_raw_sync() -> None:
-    middleware = SurfSenseFilesystemMiddleware.__new__(SurfSenseFilesystemMiddleware)
-    expected = "line1\nline2"
-    backend = _BackendWithRawRead(expected)
-
-    verify_error = middleware._verify_written_content_sync(
-        backend=backend,
-        path="/note.md",
-        expected_content=expected,
-    )
-
-    assert verify_error is None
-
-
-def test_contract_suggested_path_falls_back_to_notes_md() -> None:
+def test_contract_suggested_path_falls_back_to_documents_notes_md() -> None:
     middleware = SurfSenseFilesystemMiddleware.__new__(SurfSenseFilesystemMiddleware)
     middleware._filesystem_mode = FilesystemMode.CLOUD
     suggested = middleware._get_contract_suggested_path(_RuntimeNoSuggestedPath())  # type: ignore[arg-type]
-    assert suggested == "/notes.md"
+    # Cloud default cwd is /documents so the fallback lands in the KB.
+    assert suggested == "/documents/notes.md"
 
 
-@pytest.mark.asyncio
-async def test_verify_written_content_prefers_raw_async() -> None:
+def test_contract_suggested_path_falls_back_to_root_notes_md_in_desktop() -> None:
     middleware = SurfSenseFilesystemMiddleware.__new__(SurfSenseFilesystemMiddleware)
-    expected = "line1\nline2"
-    backend = _BackendWithRawRead(expected)
-
-    verify_error = await middleware._verify_written_content_async(
-        backend=backend,
-        path="/note.md",
-        expected_content=expected,
-    )
-
-    assert verify_error is None
+    middleware._filesystem_mode = FilesystemMode.DESKTOP_LOCAL_FOLDER
+    suggested = middleware._get_contract_suggested_path(_RuntimeNoSuggestedPath())  # type: ignore[arg-type]
+    assert suggested == "/notes.md"
 
 
 def test_normalize_local_mount_path_prefixes_default_mount(tmp_path: Path) -> None:

surfsense_backend/tests/unit/middleware/test_knowledge_search.py

@@ -5,10 +5,10 @@ import json
 import pytest
 from langchain_core.messages import AIMessage, HumanMessage
 
+from app.agents.new_chat.document_xml import build_document_xml as _build_document_xml
 from app.agents.new_chat.middleware.knowledge_search import (
     KBSearchPlan,
     KnowledgeBaseSearchMiddleware,
-    _build_document_xml,
     _normalize_optional_date_range,
     _parse_kb_search_plan_response,
     _render_recent_conversation,
@@ -248,17 +248,10 @@ class TestKnowledgeBaseSearchMiddlewarePlanner:
             captured.update(kwargs)
             return []
 
-        async def fake_build_scoped_filesystem(**kwargs):
-            return {}, {}
-
         monkeypatch.setattr(
             "app.agents.new_chat.middleware.knowledge_search.search_knowledge_base",
             fake_search_knowledge_base,
         )
-        monkeypatch.setattr(
-            "app.agents.new_chat.middleware.knowledge_search.build_scoped_filesystem",
-            fake_build_scoped_filesystem,
-        )
 
         llm = FakeLLM(
             json.dumps(
@@ -298,17 +291,10 @@ class TestKnowledgeBaseSearchMiddlewarePlanner:
             captured.update(kwargs)
             return []
 
-        async def fake_build_scoped_filesystem(**kwargs):
-            return {}, {}
-
         monkeypatch.setattr(
             "app.agents.new_chat.middleware.knowledge_search.search_knowledge_base",
             fake_search_knowledge_base,
         )
-        monkeypatch.setattr(
-            "app.agents.new_chat.middleware.knowledge_search.build_scoped_filesystem",
-            fake_build_scoped_filesystem,
-        )
 
         middleware = KnowledgeBaseSearchMiddleware(
             llm=FakeLLM("not json"),
@@ -334,17 +320,10 @@ class TestKnowledgeBaseSearchMiddlewarePlanner:
             captured.update(kwargs)
             return []
 
-        async def fake_build_scoped_filesystem(**kwargs):
-            return {}, {}
-
         monkeypatch.setattr(
             "app.agents.new_chat.middleware.knowledge_search.search_knowledge_base",
             fake_search_knowledge_base,
         )
-        monkeypatch.setattr(
-            "app.agents.new_chat.middleware.knowledge_search.build_scoped_filesystem",
-            fake_build_scoped_filesystem,
-        )
 
         middleware = KnowledgeBaseSearchMiddleware(
             llm=FakeLLM(
@@ -386,9 +365,6 @@ class TestKnowledgeBaseSearchMiddlewarePlanner:
             search_called = True
             return []
 
-        async def fake_build_scoped_filesystem(**kwargs):
-            return {}, {}
-
         monkeypatch.setattr(
             "app.agents.new_chat.middleware.knowledge_search.browse_recent_documents",
             fake_browse_recent_documents,
@@ -397,10 +373,6 @@ class TestKnowledgeBaseSearchMiddlewarePlanner:
             "app.agents.new_chat.middleware.knowledge_search.search_knowledge_base",
             fake_search_knowledge_base,
         )
-        monkeypatch.setattr(
-            "app.agents.new_chat.middleware.knowledge_search.build_scoped_filesystem",
-            fake_build_scoped_filesystem,
-        )
 
         llm = FakeLLM(
             json.dumps(
@@ -440,9 +412,6 @@ class TestKnowledgeBaseSearchMiddlewarePlanner:
             search_captured.update(kwargs)
             return []
 
-        async def fake_build_scoped_filesystem(**kwargs):
-            return {}, {}
-
         monkeypatch.setattr(
             "app.agents.new_chat.middleware.knowledge_search.browse_recent_documents",
             fake_browse_recent_documents,
@@ -451,10 +420,6 @@ class TestKnowledgeBaseSearchMiddlewarePlanner:
             "app.agents.new_chat.middleware.knowledge_search.search_knowledge_base",
             fake_search_knowledge_base,
         )
-        monkeypatch.setattr(
-            "app.agents.new_chat.middleware.knowledge_search.build_scoped_filesystem",
-            fake_build_scoped_filesystem,
-        )
 
         llm = FakeLLM(
             json.dumps(