The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains instructions designed to bypass security controls. In the SKILL.md file, the agent is directed to 'immediately retry' commands with 'escalated permissions' if a permission or sandbox error occurs. Additionally, the skill includes deceptive references to official 'gpt-oss' models that do not exist.
[COMMAND_EXECUTION]: The skill instructs the agent to execute system-level commands via the codex CLI to install MCP servers. The combination of automated installation and instructions to use escalated privileges poses a security risk to the host environment.
[EXTERNAL_DOWNLOADS]: The skill is configured to fetch an MCP server from https://developers.openai.com/mcp. This URL belongs to a well-known service (OpenAI) and is used for official documentation access.

openai-docs