The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires cloning a repository from an unverified GitHub account (chAng-L19/codex-redteam-mode.git) that does not match the stated author's vendor identity (Aradotso).
[REMOTE_CODE_EXECUTION]: The installation instructions direct the user or agent to execute unverified Python scripts (python scripts/install.py, python scripts/validate.py) immediately after cloning the external repository.
[COMMAND_EXECUTION]: On Windows systems, the skill uses powershell -ExecutionPolicy Bypass to run scripts. This is a common technique to circumvent local security restrictions designed to prevent the execution of untrusted or malicious scripts.
[PROMPT_INJECTION]: The skill provides instructions that guide the AI to generate malicious content, specifically reverse shell payloads designed to evade Antivirus (AV) and Endpoint Detection and Response (EDR) systems, which may bypass the model's safety guardrails.
[PROMPT_INJECTION]: The skill defines a complex routing mechanism (phase -> router -> pack -> leaf) that modifies the agent's behavior into a 'red team' state, potentially overriding standard operational safety constraints.
[REMOTE_CODE_EXECUTION]: The troubleshooting and installation steps involve executing arbitrary commands such as python -c "import json; print(json.load(open('.codex/hooks.json'))['redteam'])", which could be exploited if the configuration files were maliciously modified.

codex-redteam-mode