The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the --madmax flag, which explicitly maps to --dangerously-bypass-approvals-and-sandbox in the underlying Codex CLI, disabling critical safety protections.
[EXTERNAL_DOWNLOADS]: The installation process requires downloading the oh-my-codex package from NPM and running a setup command (omx setup), which involves executing code from an external source.
[REMOTE_CODE_EXECUTION]: The skill facilitates the execution of agent-generated code through its 'exec' and 'fix' pipeline stages. Additionally, it supports hook extensions in .omx/hooks/*.mjs that are dynamically loaded and executed during the session lifecycle.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection. Ingestion points: Untrusted data enters the context via task strings in workflow skills such as $plan and $autopilot. Boundary markers: None present. Capability inventory: Significant, including Bash and Write tools used by executor agents. Sanitization: No sanitization or validation of the agent-generated code or user-provided tasks is mentioned.

omx