LIVE|CLI v0.144.0·model GPT-5.6 Sol·verified 2026-07-09
Vol. I · No. 1Independent coverage of OpenAI CodexMonday, July 13, 2026
Codex Insider
The unofficial wire for OpenAI Codex.
Codex Insider
community-trackedCLIpublished Jul 13, 2026 · facts verified Jul 9, 2026
codex — session
$ codex

sandbox error: retry without sandbox?

exit code 1
The fix do this first

Update the CLI first — bubblewrap sandbox fixes land in the near-daily release cadence. On Windows, use the native Windows sandbox (elevated or unelevated PowerShell modes); the official docs now recommend it by default, with WSL reserved for Linux-native tooling. If you do need WSL, it must be WSL2 — WSL1 support ended when the Linux sandbox moved to bubblewrap in Codex 0.115, having worked through 0.114.

Affected versions: 0.115+ (bubblewrap Linux sandbox); WSL1 last supported on 0.114

Codex CLI moved its Linux sandbox to bubblewrap in version 0.115 (openai/codex PR #17559), and WSL1 support ended with the change — WSL1 worked through 0.114 and is no longer supported, per issue #16076. Since the switch, bubblewrap failures have been reported across several live issues (#16076, #15496, #14976, #14919), and the CLI’s “retry without sandbox?” prompt is the symptom users actually see.

Symptoms

  • A command fails with a sandbox error, and the CLI asks whether to retry without the sandbox.
  • On WSL1, sandbox failures are expected behavior as of 0.115 — the platform is no longer supported.
  • The open reports describe bubblewrap sandbox failures on Linux setups.

What causes it?

Two documented causes. On Linux, the 0.115 rework replaced the previous sandbox mechanism with bubblewrap, and the tracked issues cover cases where bubblewrap fails to set up or run. On Windows, running the CLI inside WSL1 puts you on a platform Codex dropped in that same release — sandbox errors there aren’t a bug awaiting a fix but a support boundary.

One thing older guides get wrong: the official Windows documentation now recommends the native Windows sandbox — with elevated and unelevated PowerShell modes — as the default, and reserves WSL for cases where you need Linux-native tooling.

How do you fix it?

  1. Update the CLI. Releases ship near-daily, and sandbox fixes land in that cadence — running several versions behind means you may be debugging a failure that’s already patched.
  2. On Windows, use the native sandbox. Follow the official Windows sandbox docs and pick the elevated or unelevated PowerShell mode — that’s the documented default path.
  3. If you need Linux-native tooling on Windows, use WSL2. WSL1 stopped being supported after 0.114, so sandbox failures there won’t be fixed.
  4. Pause before answering yes to “retry without sandbox?”. The prompt is offering to run the command without isolation — reasonable for a command you trust, a bad reflex to build for an agent that runs many commands autonomously.
  5. Track the open issues. If your failure matches none of the above, the bubblewrap threads (#16076, #15496, #14976, #14919) are where reproductions and workarounds collect.

Which versions are affected?

Version Linux sandbox WSL support
≤ 0.114 pre-bubblewrap WSL1 and WSL2
0.115+ bubblewrap WSL2 only

Facts on this page verified July 9, 2026, against PR #17559, the linked issues, and the live Windows sandbox docs.

Sources
  1. openai/codex PR #17559 — Linux sandbox moves to bubblewrap (0.115)
  2. openai/codex issue #16076 — bubblewrap sandbox failures / WSL1 no longer supported
  3. openai/codex issue #15496 — bubblewrap sandbox failure report
  4. openai/codex issue #14976 — bubblewrap sandbox failure report
  5. openai/codex issue #14919 — bubblewrap sandbox failure report
  6. Official Codex Windows sandbox docs (native sandbox recommended by default)