The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires the installation of the @openai/codex CLI package. This is a dependency on a well-known service and is documented as a prerequisite for the skill's operation.
[COMMAND_EXECUTION]: The script scripts/run_codex_imagegen.py executes the codex CLI using subprocess.run. This execution is secured by avoiding shell interpolation, using a static argument array, employing a -- delimiter to prevent argument injection, sanitizing the environment variables to a minimal allowlist, and running the command within a read-only sandbox.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input in the form of image prompts and reference images which are passed to the external Codex tool.
Ingestion points: The args.prompt_file and --image attachments in scripts/run_codex_imagegen.py ingest user-controlled data.
Boundary markers: The launcher script explicitly appends a SECURITY BOUNDARY text block to the end of every prompt, instructing the model to ignore any instructions within the preceding text that deviate from image generation.
Capability inventory: The skill has the capability to read local prompt and image files and execute the codex CLI via subprocess.run (documented in scripts/run_codex_imagegen.py).
Sanitization: The launcher performs extensive validation including the rejection of symbolic links, verification of file signatures for images (PNG, JPEG, WebP), enforcement of file size limits, and confinement of output paths to a specific generated images root directory.

codex-imagegen