fix: #2962 normalize sandbox paths and add Windows CI

.github/workflows/tests.yml

@@ -104,6 +104,33 @@ jobs:
         if: steps.changes.outputs.run != 'true'
         run: echo "Skipping tests for non-code changes."
 
+  tests-windows:
+    runs-on: windows-latest
+    env:
+      OPENAI_API_KEY: fake-for-tests
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd
+      - name: Detect code changes
+        id: changes
+        shell: bash
+        run: ./.github/scripts/detect-changes.sh code "${{ github.event.pull_request.base.sha || github.event.before }}" "${{ github.sha }}"
+      - name: Setup uv
+        if: steps.changes.outputs.run == 'true'
+        uses: astral-sh/setup-uv@cec208311dfd045dd5311c1add060b2062131d57
+        with:
+          enable-cache: true
+          python-version: "3.13"
+      - name: Install dependencies
+        if: steps.changes.outputs.run == 'true'
+        run: uv sync --all-extras --all-packages --group dev
+      - name: Run tests
+        if: steps.changes.outputs.run == 'true'
+        run: uv run pytest
+      - name: Skip tests
+        if: steps.changes.outputs.run != 'true'
+        run: echo "Skipping tests for non-code changes."
+
   build-docs:
     runs-on: ubuntu-latest
     env:

src/agents/extensions/sandbox/blaxel/mounts.py

@@ -31,6 +31,7 @@
 from ....sandbox.materialization import MaterializedFile
 from ....sandbox.session.base_sandbox_session import BaseSandboxSession
 from ....sandbox.types import FileMode, Permissions
+from ....sandbox.workspace_paths import sandbox_path_str
 
 logger = logging.getLogger(__name__)
 
@@ -81,7 +82,7 @@ async def activate(
         _assert_blaxel_session(session)
         _ = base_dir
         mount_path = mount._resolve_mount_path(session, dest)
-        config = _build_mount_config(mount, mount_path=str(mount_path))
+        config = _build_mount_config(mount, mount_path=mount_path.as_posix())
         await _mount_bucket(session, config)
         return []
 
@@ -95,7 +96,7 @@ async def deactivate(
         _assert_blaxel_session(session)
         _ = base_dir
         mount_path = mount._resolve_mount_path(session, dest)
-        await _unmount_bucket(session, str(mount_path))
+        await _unmount_bucket(session, mount_path.as_posix())
 
     async def teardown_for_snapshot(
         self,
@@ -105,7 +106,7 @@ async def teardown_for_snapshot(
     ) -> None:
         _assert_blaxel_session(session)
         _ = mount
-        await _unmount_bucket(session, str(path))
+        await _unmount_bucket(session, sandbox_path_str(path))
 
     async def restore_after_snapshot(
         self,
@@ -114,7 +115,7 @@ async def restore_after_snapshot(
         path: Path,
     ) -> None:
         _assert_blaxel_session(session)
-        config = _build_mount_config(mount, mount_path=str(path))
+        config = _build_mount_config(mount, mount_path=sandbox_path_str(path))
         await _mount_bucket(session, config)
 
     def build_docker_volume_driver_config(
@@ -606,7 +607,9 @@ def _resolve_config(
                 message="BlaxelDriveMountStrategy requires a BlaxelDriveMount entry",
                 context={"mount_type": mount.type},
             )
-        mount_path = mount.drive_mount_path or str(mount._resolve_mount_path(session, dest))
+        mount_path = mount.drive_mount_path or sandbox_path_str(
+            mount._resolve_mount_path(session, dest)
+        )
         return BlaxelDriveMountConfig(
             drive_name=mount.drive_name,
             mount_path=mount_path,
@@ -619,7 +622,7 @@ def _effective_mount_path(mount: Mount, fallback: Path) -> str:
         """Return the actual mount path, preferring ``drive_mount_path`` over the manifest path."""
         if isinstance(mount, BlaxelDriveMount) and mount.drive_mount_path:
             return mount.drive_mount_path
-        return str(fallback)
+        return sandbox_path_str(fallback)
 
     @staticmethod
     def _resolve_config_from_source(mount: Mount, mount_path: str) -> BlaxelDriveMountConfig:

src/agents/extensions/sandbox/blaxel/sandbox.py

@@ -65,6 +65,7 @@
     retry_async,
 )
 from ....sandbox.util.tar_utils import UnsafeTarMemberError, validate_tar_bytes
+from ....sandbox.workspace_paths import coerce_posix_path, posix_path_as_path, sandbox_path_str
 
 DEFAULT_BLAXEL_WORKSPACE_ROOT = "/workspace"
 logger = logging.getLogger(__name__)
@@ -319,7 +320,7 @@ async def start(self) -> None:
         # Ensure workspace root exists before BaseSandboxSession.start() materializes
         # the manifest.  Blaxel base images run as root and do not ship a pre-created
         # workspace directory.
-        root = self.state.manifest.root
+        root = sandbox_path_str(self.state.manifest.root)
         try:
             await self._sandbox.process.exec(
                 {
@@ -368,7 +369,7 @@ async def mkdir(
         if path == Path("/"):
             return
         try:
-            await self._sandbox.fs.mkdir(str(path))
+            await self._sandbox.fs.mkdir(sandbox_path_str(path))
         except Exception as e:
             raise WorkspaceArchiveWriteError(
                 path=path,
@@ -377,14 +378,14 @@ async def mkdir(
             ) from e
 
     async def read(self, path: Path | str, *, user: str | User | None = None) -> io.IOBase:
-        path = Path(path)
+        error_path = posix_path_as_path(coerce_posix_path(path))
         if user is not None:
             workspace_path = await self._check_read_with_exec(path, user=user)
         else:
             workspace_path = await self._validate_path_access(path)
 
         try:
-            data: Any = await self._sandbox.fs.read_binary(str(workspace_path))
+            data: Any = await self._sandbox.fs.read_binary(sandbox_path_str(workspace_path))
             if isinstance(data, str):
                 data = data.encode("utf-8")
             return io.BytesIO(bytes(data))
@@ -397,8 +398,8 @@ async def read(self, path: Path | str, *, user: str | User | None = None) -> io.
                     status = first_arg.get("status")
             error_str = str(e).lower()
             if status == 404 or "not found" in error_str or "no such file" in error_str:
-                raise WorkspaceReadNotFoundError(path=path, cause=e) from e
-            raise WorkspaceArchiveReadError(path=path, cause=e) from e
+                raise WorkspaceReadNotFoundError(path=error_path, cause=e) from e
+            raise WorkspaceArchiveReadError(path=error_path, cause=e) from e
 
     async def write(
         self,
@@ -407,19 +408,19 @@ async def write(
         *,
         user: str | User | None = None,
     ) -> None:
-        path = Path(path)
+        error_path = posix_path_as_path(coerce_posix_path(path))
         if user is not None:
             await self._check_write_with_exec(path, user=user)
 
         payload = data.read()
         if isinstance(payload, str):
             payload = payload.encode("utf-8")
         if not isinstance(payload, bytes | bytearray):
-            raise WorkspaceWriteTypeError(path=path, actual_type=type(payload).__name__)
+            raise WorkspaceWriteTypeError(path=error_path, actual_type=type(payload).__name__)
 
         workspace_path = await self._validate_path_access(path, for_write=True)
         try:
-            await self._sandbox.fs.write_binary(str(workspace_path), bytes(payload))
+            await self._sandbox.fs.write_binary(sandbox_path_str(workspace_path), bytes(payload))
         except Exception as e:
             raise WorkspaceArchiveWriteError(path=workspace_path, cause=e) from e
 
@@ -525,11 +526,11 @@ def _tar_exclude_args(self) -> list[str]:
         )
     )
     async def persist_workspace(self) -> io.IOBase:
-        root = Path(self.state.manifest.root)
+        root = self._workspace_root_path()
         tar_path = f"/tmp/bl-persist-{self.state.session_id.hex}.tar"
         excludes = " ".join(self._tar_exclude_args())
         tar_cmd = (
-            f"tar {excludes} -C {shlex.quote(str(root))} -cf {shlex.quote(tar_path)} ."
+            f"tar {excludes} -C {shlex.quote(root.as_posix())} -cf {shlex.quote(tar_path)} ."
         ).strip()
 
         unmounted_mounts: list[tuple[Mount, Path]] = []
@@ -596,7 +597,7 @@ async def persist_workspace(self) -> io.IOBase:
         return io.BytesIO(raw)
 
     async def hydrate_workspace(self, data: io.IOBase) -> None:
-        root = self.state.manifest.root
+        root = self._workspace_root_path()
         tar_path = f"/tmp/bl-hydrate-{self.state.session_id.hex}.tar"
         payload = data.read()
         if isinstance(payload, str):
@@ -608,7 +609,7 @@ async def hydrate_workspace(self, data: io.IOBase) -> None:
             validate_tar_bytes(bytes(payload))
         except UnsafeTarMemberError as e:
             raise WorkspaceArchiveWriteError(
-                path=Path(root),
+                path=root,
                 context={
                     "reason": "unsafe_or_invalid_tar",
                     "member": e.member,
@@ -623,12 +624,12 @@ async def hydrate_workspace(self, data: io.IOBase) -> None:
             result = await self._exec_internal(
                 "sh",
                 "-c",
-                f"tar -C {shlex.quote(root)} -xf {shlex.quote(tar_path)}",
+                f"tar -C {shlex.quote(root.as_posix())} -xf {shlex.quote(tar_path)}",
                 timeout=self.state.timeouts.workspace_tar_s,
             )
             if result.exit_code != 0:
                 raise WorkspaceArchiveWriteError(
-                    path=Path(root),
+                    path=root,
                     context={
                         "reason": "tar_extract_failed",
                         "output": result.stderr.decode("utf-8", errors="replace"),
@@ -637,7 +638,7 @@ async def hydrate_workspace(self, data: io.IOBase) -> None:
         except WorkspaceArchiveWriteError:
             raise
         except Exception as e:
-            raise WorkspaceArchiveWriteError(path=Path(root), cause=e) from e
+            raise WorkspaceArchiveWriteError(path=root, cause=e) from e
         finally:
             try:
                 await self._exec_internal(