langchain-ai/deepagents · error · ExtensionError

Extension backend route {item.name!r} from {item.source.labe

Error message

Extension backend route {item.name!r} from {item.source.label} cannot mount {type(item.unit).__name__} in sandbox mode

What it means

Raised by validate_backend_route when sandbox mode is active and an extension attempts to mount a FilesystemBackend on a backend route. In sandbox mode extensions are forbidden from mounting real filesystem backends, which could bypass sandboxing.

Source

Thrown at libs/code/deepagents_code/extensions/hosting.py:149

    Raises:
        ExtensionError: If the route overlaps internal storage or directly
            exposes a host filesystem backend to a sandboxed agent.
    """
    if any(
        item.name.startswith(prefix) or prefix.startswith(item.name)
        for prefix in protected_routes
    ):
        msg = (
            f"Extension backend route {item.name!r} from {item.source.label} "
            "overlaps an internal route"
        )
        raise ExtensionError(msg)
    if sandbox_active and isinstance(item.unit, FilesystemBackend):
        msg = (
            f"Extension backend route {item.name!r} from {item.source.label} "
            f"cannot mount {type(item.unit).__name__} in sandbox mode"
        )
        raise ExtensionError(msg)


def bind_runtime_host_policy(
    registry: ExtensionRegistry,
    protected_routes: Collection[str],
    *,
    sandbox_active: bool = False,
) -> None:
    """Validate late routes and flag graph-bound registrations for restart."""

    def apply(kind: str, item: RegisteredUnit[Any]) -> None:
        if kind == "middleware":
            registry.require_restart()
            return
        if kind == "backend_route":
            validate_backend_route(
                item, protected_routes, sandbox_active=sandbox_active
            )

View on GitHub (pinned to a1af029e6e)

Solutions

1. Disable sandbox mode if the extension's filesystem route is intentionally required
2. Use an extension whose backend is not a FilesystemBackend (e.g. a virtual/in-memory backend)
3. Update or replace the extension with a sandbox-compatible version
4. Contact the extension author to expose a non-filesystem backend option

Example fix

// before
ext.register_backend_route("/data/", FilesystemBackend(root="/data"))

// after
ext.register_backend_route("/data/", InMemoryBackend(data))  # sandbox-safe

from deepagents_code.extensions.hosting import validate_backend_route  # policy lives here

def is_sandbox_safe(unit) -> bool:
    from langchain.backends import FilesystemBackend  # illustrative import
    return not (sandbox_active and isinstance(unit, FilesystemBackend))

def is_sandbox_compatible(backend) -> bool:
    return not isinstance(backend, FilesystemBackend)

try:
    ext.register_backend_route(prefix, backend)
except ExtensionError as exc:
    logger.error("sandbox rejected backend %s: %s", type(backend).__name__, exc)

When it happens

Trigger: Registering a FilesystemBackend (or subclass) route from an extension while the agent runs with sandbox enabled, then creating the agent (create_cli_agent) or applying the registry.

Common situations: An extension designed for non-sandboxed use installed into a sandboxed CLI agent; an extension upgrade that switched to a filesystem-backed route; users enabling sandbox mode after installing such an extension.

Related errors

• Invalid backend route prefix {prefix!r}: use lowercase path
• Backend route {prefix!r} got {type(backend).__name__}, which
• Extension backend route {item.name!r} from {item.source.labe
• enable_interpreter=True is not supported with a remote sandb
• Extension registration is closed for this session

AI-assisted analysis of langchain-ai/deepagents@a1af029e6e (2026-08-29). Data as JSON: /api/errors/0adda46491318b28. Report an issue: GitHub.