codex-redteam-mode
Fail
Audited by Gen Agent Trust Hub on Sep 16, 2026
Risk Level: HIGHPROMPT_INJECTIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPRIVILEGE_ESCALATION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill instructs cloning an external repository from an untrusted source:
https://github.com/chAng-L19/codex-redteam-mode.git.- [REMOTE_CODE_EXECUTION]: The instructions direct the execution of code downloaded from the external repository usingpython scripts/install.pyandpython scripts/validate.pywithout verification of the script content.- [PRIVILEGE_ESCALATION]: The skill explicitly commands the use ofpowershell -ExecutionPolicy Bypassto run an installation script, which is a technique used to circumvent system security policies that restrict script execution.- [PROMPT_INJECTION]: The skill defines a 'Red Team Mode' that acts as a set of instructions to override standard AI safety filters. It explicitly prompts the AI to assist in 'web exploitation', 'Active Directory exploitation', and 'evasion techniques' (e.g., evading AV signature detection), which are typical patterns for bypassing safety guidelines.- [COMMAND_EXECUTION]: The skill relies on executing multiple shell commands (git clone,python,powershell) to modify the local environment and agent configuration files (AGENTS.md,hooks.json).
Recommendations
- AI detected serious security threats
Audit Metadata