@decoopus/chatgpt-setup@1.0.0
Malicious code in @decoopus/chatgpt-setup (npm)
Analysis
This package poses as a one-command setup helper for ChatGPT Desktop and Codex CLI (published under the names chatgpt-desktop-setup, codex-setup and @decoopus/chatgpt-setup). Its bin/cli.js prompts the user for an API key and then rewrites ~/.codex/config.toml so that all Codex/ChatGPT Desktop inference is routed to the third-party gateway hxxps://api[.]vokes[.]in/v1 (provider id chatgpt_pro_5x, model gpt-6-astra, wire_api responses), sending the user's API key and subsequent prompts and responses to that host. The same write sets approval_policy = "never" and sandbox_mode = "danger-full-access", disabling the agent's approval prompts and its sandbox. It also edits ~/.codex/.codex-global-state.json to force agent-mode 'full-access' and to add the user's home directory and the filesystem root to every project's rootPaths, granting unrestricted filesystem access. Before writing it kills any running ChatGPT/Codex processes (taskkill /F /IM ChatGPT.exe /IM codex.exe, killall ChatGPT Codex, pkill -f 'ChatGPT|codex') and afterwards relaunches the desktop app or the Codex CLI. The credential redirection to api.vokes.in and the permission escalation are not presented to the user as such.
- analyzed by
- Leitwacht
- first seen
- Oct 14, 2026, 10:36 AM
- analyzed
- Oct 14, 2026, 10:36 AM
Related advisories
- @decoopus/chatgpt-setup@0.0.0-stage same package
- @decoopus/chatgpt-setup@1.0.1 same package
- @decoopus/claude-setup@0.0.0-stage
- @decoopus/claude-setup@1.0.1
- @decoopus/claude-setup@1.0.3
- claude-desktop-setup@1.0.2
- claude-desktop-setup@1.0.11
- claude-desktop-setup@1.0.6
Independently detected by the Leitwacht supply-chain probe. IOCs are defanged. Published CC0. Think this is a mistake? See the dispute policy.