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

@@ -81,7 +81,7 @@ 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 +95,7 @@ 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,

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
 
 DEFAULT_BLAXEL_WORKSPACE_ROOT = "/workspace"
 logger = logging.getLogger(__name__)
@@ -377,7 +378,7 @@ async def mkdir(
             ) from e
 
     async def read(self, path: Path | str, *, user: str | User | None = None) -> io.IOBase:
-        path = Path(path)
+        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:
@@ -407,7 +408,7 @@ async def write(
         *,
         user: str | User | None = None,
     ) -> None:
-        path = Path(path)
+        path = posix_path_as_path(coerce_posix_path(path))
         if user is not None:
             await self._check_write_with_exec(path, user=user)
 
@@ -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(

src/agents/extensions/sandbox/cloudflare/sandbox.py

@@ -67,6 +67,7 @@
     retry_async,
 )
 from ....sandbox.util.tar_utils import UnsafeTarMemberError, validate_tar_bytes
+from ....sandbox.workspace_paths import coerce_posix_path, posix_path_as_path
 
 _DEFAULT_EXEC_TIMEOUT_S = 30.0
 _DEFAULT_REQUEST_TIMEOUT_S = 120.0
@@ -345,7 +346,9 @@ async def mount_bucket(
         mount_path: Path | str,
         options: dict[str, object],
     ) -> None:
-        workspace_path = await self._validate_path_access(mount_path, for_write=True)
+        workspace_path = await self._validate_path_access(
+            coerce_posix_path(mount_path).as_posix(), for_write=True
+        )
         http = self._session()
         url = self._url("mount")
         payload = {
@@ -389,7 +392,9 @@ async def mount_bucket(
             ) from e
 
     async def unmount_bucket(self, mount_path: Path | str) -> None:
-        workspace_path = await self._validate_path_access(mount_path, for_write=True)
+        workspace_path = await self._validate_path_access(
+            coerce_posix_path(mount_path).as_posix(), for_write=True
+        )
         http = self._session()
         url = self._url("unmount")
         payload = {"mountPath": str(workspace_path)}
@@ -501,10 +506,10 @@ def _handle_event_payload(data: str) -> None:
 
     async def _prepare_backend_workspace(self) -> None:
         try:
-            root = Path(self.state.manifest.root)
-            await self._exec_internal("mkdir", "-p", "--", str(root))
+            root = self._workspace_root_path()
+            await self._exec_internal("mkdir", "-p", "--", root.as_posix())
         except Exception as e:
-            raise WorkspaceStartError(path=Path(self.state.manifest.root), cause=e) from e
+            raise WorkspaceStartError(path=self._workspace_root_path(), cause=e) from e
 
     async def _can_reuse_restorable_snapshot_workspace(self) -> bool:
         if not self._workspace_state_preserved_on_start():
@@ -518,7 +523,7 @@ async def _can_reuse_restorable_snapshot_workspace(self) -> bool:
         return await self._can_skip_snapshot_restore_on_resume(is_running=is_running)
 
     async def _restore_snapshot_into_workspace_on_resume(self) -> None:
-        root = Path(self.state.manifest.root)
+        root = self._workspace_root_path()
         detached_mounts: list[tuple[Any, Path]] = []
         if self._restore_workspace_was_running:
             for mount_entry, mount_path in self.state.manifest.ephemeral_mount_targets():
@@ -965,7 +970,7 @@ async def pty_terminate_all(self) -> None:
             await self._terminate_pty_entry(entry)
 
     async def read(self, path: Path | str, *, user: str | User | None = None) -> io.IOBase:
-        path = Path(path)
+        path = posix_path_as_path(coerce_posix_path(path))
         if user is not None:
             await self._check_read_with_exec(path, user=user)
 
@@ -1029,7 +1034,7 @@ async def write(
         *,
         user: str | User | None = None,
     ) -> None:
-        path = Path(path)
+        path = posix_path_as_path(coerce_posix_path(path))
         if user is not None:
             await self._check_write_with_exec(path, user=user)
 
@@ -1106,7 +1111,7 @@ async def running(self) -> bool:
         or _is_transient_workspace_error(exc)
     )
     async def _persist_workspace_via_http(self) -> io.IOBase:
-        root = Path(self.state.manifest.root)
+        root = self._workspace_root_path()
         skip = self._persist_workspace_skip_relpaths()
         excludes_param = ",".join(
             rel.as_posix().removeprefix("./")
@@ -1148,7 +1153,7 @@ async def _persist_workspace_via_http(self) -> io.IOBase:
         or _is_transient_workspace_error(exc)
     )
     async def _hydrate_workspace_via_http(self, data: io.IOBase) -> None:
-        root = Path(self.state.manifest.root)
+        root = self._workspace_root_path()
         raw = data.read()
         if isinstance(raw, str):
             raw = raw.encode("utf-8")
@@ -1199,7 +1204,7 @@ async def _hydrate_workspace_via_http(self, data: io.IOBase) -> None:
             raise WorkspaceArchiveWriteError(path=root, cause=e) from e
 
     async def persist_workspace(self) -> io.IOBase:
-        root = Path(self.state.manifest.root)
+        root = self._workspace_root_path()
         unmounted_mounts: list[tuple[Any, Path]] = []
         unmount_error: WorkspaceArchiveReadError | None = None
         for mount_entry, mount_path in self.state.manifest.ephemeral_mount_targets():
@@ -1245,7 +1250,7 @@ async def persist_workspace(self) -> io.IOBase:
         return persisted
 
     async def hydrate_workspace(self, data: io.IOBase) -> None:
-        root = Path(self.state.manifest.root)
+        root = self._workspace_root_path()
         unmounted_mounts: list[tuple[Any, Path]] = []
         unmount_error: WorkspaceArchiveWriteError | None = None
         for mount_entry, mount_path in self.state.manifest.ephemeral_mount_targets():

src/agents/extensions/sandbox/daytona/sandbox.py

@@ -63,6 +63,7 @@
     retry_async,
 )
 from ....sandbox.util.tar_utils import UnsafeTarMemberError, validate_tar_bytes
+from ....sandbox.workspace_paths import coerce_posix_path, posix_path_as_path
 
 DEFAULT_DAYTONA_WORKSPACE_ROOT = "/home/daytona/workspace"
 logger = logging.getLogger(__name__)
@@ -782,7 +783,7 @@ async def _terminate_pty_entry(self, entry: _DaytonaPtySessionEntry) -> None:
             pass
 
     async def read(self, path: Path | str, *, user: str | User | None = None) -> io.IOBase:
-        path = Path(path)
+        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:
@@ -809,7 +810,7 @@ async def write(
         *,
         user: str | User | None = None,
     ) -> None:
-        path = Path(path)
+        path = posix_path_as_path(coerce_posix_path(path))
         if user is not None:
             await self._check_write_with_exec(path, user=user)
 
@@ -859,7 +860,6 @@ def _tar_exclude_args(self) -> list[str]:
         )
     )
     async def _run_persist_workspace_command(self, tar_cmd: str, tar_path: str) -> bytes:
-        root = self.state.manifest.root
         try:
             envs = await self._resolved_envs()
             result = await self._sandbox.process.exec(
@@ -869,7 +869,7 @@ async def _run_persist_workspace_command(self, tar_cmd: str, tar_path: str) -> b
             )
             if result.exit_code != 0:
                 raise WorkspaceArchiveReadError(
-                    path=Path(root),
+                    path=self._workspace_root_path(),
                     context={"reason": "tar_failed", "output": result.result or ""},
                 )
             return cast(
@@ -882,7 +882,7 @@ async def _run_persist_workspace_command(self, tar_cmd: str, tar_path: str) -> b
         except WorkspaceArchiveReadError:
             raise
         except Exception as e:
-            raise WorkspaceArchiveReadError(path=Path(root), cause=e) from e
+            raise WorkspaceArchiveReadError(path=self._workspace_root_path(), cause=e) from e
 
     async def persist_workspace(self) -> io.IOBase:
         def _error_context_summary(error: WorkspaceArchiveReadError) -> dict[str, str]:
@@ -892,11 +892,11 @@ def _error_context_summary(error: WorkspaceArchiveReadError) -> dict[str, str]:
                 summary["cause"] = str(error.cause)
             return summary
 
-        root = Path(self.state.manifest.root)
+        root = self._workspace_root_path()
         tar_path = f"/tmp/sandbox-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]] = []
@@ -964,7 +964,7 @@ def _error_context_summary(error: WorkspaceArchiveReadError) -> dict[str, str]:
         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/sandbox-hydrate-{self.state.session_id.hex}.tar"
         payload = data.read()
         if isinstance(payload, str):
@@ -976,7 +976,7 @@ 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,
@@ -994,19 +994,19 @@ async def hydrate_workspace(self, data: io.IOBase) -> None:
                 timeout=self.state.timeouts.file_upload_s,
             )
             result = await self._sandbox.process.exec(
-                f"tar -C {shlex.quote(root)} -xf {shlex.quote(tar_path)}",
+                f"tar -C {shlex.quote(root.as_posix())} -xf {shlex.quote(tar_path)}",
                 env=envs or None,
                 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.result or ""},
                 )
         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:
                 envs = await self._resolved_envs()