The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides templates for creating local Python scripts in the .codex/hooks/ directory, which are configured to be executed by the host system to validate commands and assistant output.- [EXTERNAL_DOWNLOADS]: Contains a reference to the OpenAI Model Context Protocol (MCP) server at https://developers.openai.com/mcp for use in research tasks.- [PROMPT_INJECTION]: The hook templates in references/hook-templates.md process untrusted data from tool inputs and assistant messages using regular expressions, representing an indirect injection surface. This is documented as a mechanism for enforcing security boundaries rather than a vulnerability.

creating-codex-environments