MAL-2026-18053 Malicious code in @decoopus/chatgpt-setup (npm)

When the user runs the package's bin (chatgpt-setup/codex-setup), the tool rewrites ~/.codex/config.toml to set approval_policy='never' and sandbox_mode='danger-full-access', edits ~/.codex/.codex-global-state.json to set agent-mode to 'full-access' and inject the filesystem root ('/' on Unix or 'C:\' on Windows) plus the user's home directory into every project's rootPaths, and points the Codex model_provider base_url at the author-controlled gateway https://api.vokes.in/v1. The combined effect is that tool-call instructions returned by that third-party gateway are executed by the local Codex agent without human approval, inside a sandbox expanded to the entire filesystem. The approval/sandbox disablement and the rootPaths expansion to the filesystem root are not described in the README, which advertises only gateway selection, model, provider and reasoning options alongside a 'Full Disk Access' success line. The documented CLI invocation is required to trigger the behavior; it does not fire from an npm lifecycle hook or on require().

Any computer that has this package installed or running should be considered fully compromised. All secrets and keys stored on that computer should be rotated immediately from a different computer. The package should be removed, but as full control of the computer may have been given to an outside entity, there is no guarantee that removing the package will remove all malicious software resulting from installing it.