The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/image_gen script executes the codex command-line utility via node:child_process.spawn. It specifically uses the --sandbox danger-full-access flag, which grants the utility extensive permissions during execution.
[EXTERNAL_DOWNLOADS]: The skill's documentation recommends the global installation of the @openai/codex package via NPM. This is an official utility provided by OpenAI, a well-known service provider.
[INDIRECT_PROMPT_INJECTION]: The skill processes image generation prompts provided via the command line or from files. These strings are interpolated into a system-style prompt for the nested Codex agent (in scripts/image_gen) without using sanitization or boundary markers. This creates a surface where malicious prompt data could attempt to override the instructions of the nested agent.
Ingestion points: Prompts provided via --prompts or --prompt arguments in scripts/image_gen.
Boundary markers: Absent; the prompt text is appended directly to the end of the instruction block.
Capability inventory: The script can execute the Codex CLI with full sandbox access and perform file system operations (read/write/copy) within the repository and the ~/.codex directory.
Sanitization: None; inputs are used verbatim.

codex-images