The Agent Skills Directory

[COMMAND_EXECUTION]: The skill promotes the use of the --madmax flag, which is explicitly documented to map to the --dangerously-bypass-approvals-and-sandbox parameter in the Codex CLI, effectively disabling security constraints.
[EXTERNAL_DOWNLOADS]: Installation requires fetching the oh-my-codex and @openai/codex packages from the public NPM registry, which are then executed with broad permissions.
[REMOTE_CODE_EXECUTION]: The skill implements a lifecycle hook system that automatically executes arbitrary JavaScript code from the .omx/hooks/*.mjs directory during session events like session-start or turn-complete.
[PROMPT_INJECTION]: Includes instructions to override default agent behavior using flags like --yolo (minimal verification) and --madmax (bypass safety), which are intended to circumvent standard model safety protocols.
[PROMPT_INJECTION]: The skill orchestrates complex tasks using user input in automated pipelines such as $autopilot and $team. These ingestion points lack defined boundary markers or sanitization logic, creating a surface for indirect prompt injection that can trigger subsequent tool use or file writes.

omx