The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user or agent to install a global NPM package @openai/codex. This package name uses the scope of a well-known service but is not a verified official package from that organization for the described functionality, making it an unverifiable dependency from an external source.
[COMMAND_EXECUTION]: The skill describes and enables the use of a tool (codex) that is designed to execute arbitrary shell commands. Specifically, it promotes a full-auto mode that allows the tool to run tests and fix code failures autonomously. This autonomous command execution on the host system presents a significant capability risk if the agent's instructions or the data it processes are compromised.
[PROMPT_INJECTION]: The skill documents the ingestion of instructions from a codex.md file within a project's repository, creating an indirect prompt injection surface.
Ingestion points: Project-level codex.md file and the general codebase accessed by the tool.
Boundary markers: The documentation does not provide specific delimiters or warnings to the agent to ignore instructions embedded within the processed code files.
Capability inventory: The tool has the capability to read and write to the file system and execute shell commands via the codex CLI.
Sanitization: While the skill mentions an optional 'sandbox' configuration to restrict network and file access, its implementation is left to the user and is not a mandatory safeguard for processing instruction files.

openai-codex-cli