codex
Fail
Audited by Gen Agent Trust Hub on Jul 8, 2026
Risk Level: HIGHCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the use of the
--dangerously-bypass-approvals-and-sandboxflag. This flag is explicitly described as providing 'zero OS control' in exchange for the ability to run network-bound tools like dev servers and browsers, representing a significant bypass of safety constraints. - [COMMAND_EXECUTION]: Instructions include modifying the global configuration file
~/.codex/config.tomlto settrust_level = "trusted"for specific paths. This action automates the elevation of privilege within the Codex tool's security model, eliminating interactive security prompts for designated directories. - [EXTERNAL_DOWNLOADS]: The skill provides a link to
developers.openai.comfor tool installation. As this is an official service from a well-known organization, this reference is documented as a safe external resource. - [COMMAND_EXECUTION]: The execution patterns involve shell command substitution (e.g.,
$(cat prompt.txt)) and background process management (kill,nohup), which could lead to unintended command execution if the content of input files is not strictly controlled.
Recommendations
- AI detected serious security threats
Audit Metadata