codex-cmd is a local command generator. It is designed to produce a shell command for you to inspect, not to run commands automatically.

cx does not collect, store, or upload credentials.

It does not read ~/.codex/auth.json directly. Authentication is handled by the installed Codex CLI through codex exec.

It does not execute generated commands. In CLI mode it prints a command to stdout. In zsh widget mode it replaces the editable prompt buffer.

cx sends the natural-language request you provide to codex exec.

It uses:

• --ignore-user-config
• --ignore-rules
• --skip-git-repo-check
• --sandbox read-only
• web_search='disabled'

These flags are intended to keep command generation independent from local repository contents, user rules, MCP servers, plugins, and web search.

Please report security issues through GitHub's private vulnerability reporting flow if it is available for this repository.

If private reporting is not available, open a GitHub issue with a minimal public description and avoid including exploit details, secrets, tokens, or private logs. The maintainer can then coordinate a safer disclosure path.

When reporting an issue:

• Include the affected version or commit.
• Include your operating system and shell.
• Include the smallest reproduction you can share safely.
• Redact account identifiers, tokens, local paths, and private command output.

Generated shell commands are model output. A command may be wrong, unsafe, or incomplete. Users are expected to inspect generated commands before running them.

Security issues in the Codex CLI itself should be reported to the Codex/OpenAI project through its own reporting channels.