`WorkspaceStartError` on Windows

[onmyraedar]

Please read this first

• ✅ Have you read the docs?Agents SDK docs
• ✅ Have you searched for related issues? Others may have faced similar issues.

Describe the bug

I'm trying to use the Agents SDK with a Modal sandbox. I'm getting the following (running on Windows):

  File "C:\Users\Rae\Desktop\Code\.venv\Lib\site-packages\agents\sandbox\session\base_sandbox_session.py", line 766, in _validate_remote_path_access
    raise ExecNonZeroError(
agents.sandbox.errors.ExecNonZeroError: executing processes for container: executing command "\\tmp\\openai-agents\\bin\\resolve-workspace-path-f8e24896b498 \\workspace \\workspace 1" in sandbox: error finding executable "\\tmp\\openai-agents\\bin\\resolve-workspace-path-f8e24896b498" in PATH [/usr/local/bin /usr/local/sbin /usr/local/bin /usr/sbin /usr/bin /sbin /bin]: no such file or directory

The above exception was the direct cause of the following exception:

  File "C:\Users\Rae\Desktop\Code\.venv\Lib\site-packages\agents\sandbox\session\base_sandbox_session.py", line 142, in start
    raise wrapped from e
agents.sandbox.errors.WorkspaceStartError: failed to start session

Debug information

• Agents SDK version: (e.g. v0.0.3): 0.14.2
• Python version (e.g. Python 3.14): 3.12.1
• OpenAI version: 2.32.0

Repro steps

If you set remote=False in this script on Windows, you will get the error. If you set remote=True to run the script on a remote Linux machine (I was using a Modal function), you'll see that it works.

You could also set remote=False and run it on a machine that's also not Windows; I think it'll also work there.

import asyncio
from agents import Runner
from agents.sandbox import SandboxAgent
from agents.extensions.sandbox.modal import (
    ModalImageSelector,
    ModalSandboxClient,
    ModalSandboxClientOptions,
)
from agents.sandbox import SandboxRunConfig
from agents.run import RunConfig
import modal

image = modal.Image.debian_slim(python_version="3.12").uv_pip_install(
    "openai-agents[modal]"
)
app = modal.App("test-sandbox")

agent = SandboxAgent(name="test", model="gpt-5.2", instructions="You are helpful.")


async def main() -> str:
    client = ModalSandboxClient(image=ModalImageSelector.from_image(image))
    opts = ModalSandboxClientOptions(app_name="test-sandbox", timeout=60)
    result = await Runner.run(
        agent,
        "Tell me a joke.",
        run_config=RunConfig(sandbox=SandboxRunConfig(client=client, options=opts)),
    )
    return result.final_output


@app.function(image=image, secrets=[modal.Secret.from_name("openai-api-key")])
async def run_remote() -> str:
    return await main()


if __name__ == "__main__":
    remote = False

    with app.run():
        if remote:
            output = run_remote.remote()
            print(output)
        else:
            print(asyncio.run(main()))