codex-cli-docs
Fail
Audited by Gen Agent Trust Hub on Jul 8, 2026
Risk Level: HIGHEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill instructs the user to perform a global installation of the npm package
@openai/codexand a Homebrew caskcodex. These packages do not correspond to verified official OpenAI tools and may represent unverified or deceptive software. - [COMMAND_EXECUTION]: The documentation encourages the use of dangerous command-line flags such as
--yoloand--sandbox danger-full-access, which are explicitly described as bypassing approval prompts and security sandboxes. This advice significantly increases the risk of unintended system operations when executing code via an agent. - [EXTERNAL_DOWNLOADS]: The documentation references multiple URLs (e.g.,
developers.openai.com/codex/cli) and release pages (e.g.,github.com/openai/codex/releases) that do not follow official OpenAI documentation structures or link to non-existent resources, potentially leading users to misidentify the source and safety of the software. - [EXTERNAL_DOWNLOADS]: Fetches and runs the Model Context Protocol server-filesystem package using
npxas part of the MCP configuration instructions.
Recommendations
- AI detected serious security threats
Audit Metadata