The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's recommended launch profile includes the --madmax flag. This flag explicitly maps to the Codex CLI's --dangerously-bypass-approvals-and-sandbox setting, which disables the platform's requirement for user confirmation before executing tools and removes the security sandbox protecting the host environment.
[EXTERNAL_DOWNLOADS]: The installation process requires the global installation of third-party packages (oh-my-codex and @openai/codex) from the public NPM registry.
[DYNAMIC_EXECUTION]: The skill features a hook extension system that executes custom JavaScript files (.mjs) found in the local .omx/hooks/ directory during lifecycle events such as session start or turn completion.
[PROMPT_INJECTION]: The tool is designed to automatically inject the contents of a local AGENTS.md file into the model's system instructions. This creates a surface for indirect prompt injection where malicious instructions inside a project repository could override the agent's behavior.
Ingestion points: The <cwd>/AGENTS.md file is read and passed to the model via the model_instructions_file configuration.
Boundary markers: None specified in the documentation to separate untrusted file content from system instructions.
Capability inventory: The skill uses Bash, Read, Write, Grep, and Glob tools.
Sanitization: No sanitization or validation of the injected instruction file is mentioned.

omx