Windows Codex Desktop 26.1002.52244: Sandbox setup fails with node_repl.exe sharing violation (os error 32)

Finding: In an external report on openai/codex issue 52094, Codex Desktop 26.1002.52244 on Windows x64 (initially a cloned VM) failed to start sandboxed commands in workspace-write mode. The visible error was helper_unknown_error: setup refresh had errors. The sandbox log showed runtime read/execute validation failing on cua_node\...\node_repl.exe with os error 32, a sharing violation (https://github.com/openai/codex/issues/52094). This is an external report; KFA did not reproduce it. Reported conditions: The reporter said the original VM worked on 26.930.61225 and the clone failed on 26.1002.52244. They also said cloning, restart and runtime changes were not isolated, so the app version is not proven to be the cause. A reboot did not help, and Resource Monitor showed no matching handles (issue body). Other reports: Commenters reported the same node_repl.exe path and os error 32 on 26.1002.52244 (https://github.com/openai/codex/issues/52094#issuecomment-6061029031). One reported it on a native Windows PC, with the failure starting October 8 and the same helper and runtime identifiers (https://github.com/openai/codex/issues/52094#issuecomment-6086181359). One reported it on Codex CLI 0.162.0 with an elevated sandbox config (https://github.com/openai/codex/issues/52094#issuecomment-6082368641). One reported it in a ChatGPT Work Windows package, 26.1002.7124.0 (https://github.com/openai/codex/issues/52094#issuecomment-6086997507). Cause status: Not confirmed. One user suspects the elevated setup path tries to change the ACL of node_repl.exe while the file is open. That user also saw the file reported as locked with all Codex processes stopped (https://github.com/openai/codex/issues/52094#issuecomment-6061029031). Another user could not identify the lock owner (https://github.com/openai/codex/issues/52094#issuecomment-6086181359). The only other comment in the supplied thread is an automated duplicate notice, so there is no maintainer diagnosis. Reported workaround (source-backed, not verified by KFA): Several users said setting [windows] sandbox = "unelevated" in ~/.codex/config.toml restored execution. One said this change alone was enough and no rename was needed (https://github.com/openai/codex/issues/52094#issuecomment-6086181359). Another combined it with renaming the cua_node directory and could not say which step mattered (https://github.com/openai/codex/issues/52094#issuecomment-6085140103). A third said the rename alone failed and that the config edit had to be made with the app fully closed (https://github.com/openai/codex/issues/52094#issuecomment-6061029031). Switching modes changes sandbox behavior, so weigh that tradeoff. Fix status: The supplied sources show no fix commit and no released fix. The issue was open at the last update.