The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the codex CLI, invoking it with the --approve-for-me flag. This flag automates the approval of actions within the tool, which reduces user oversight and bypasses standard confirmation prompts during execution.
[EXTERNAL_DOWNLOADS]: The skill directs users to install the @openai/codex package via npm. This package resides within the scope of the openai organization, which is a recognized and trusted service provider.
[DYNAMIC_EXECUTION]: The troubleshooting guide provides a command to execute a local Python script (remove_chroma_key.py) using the uv tool. The script's location is dynamically determined using the $CODEX_HOME environment variable.
[INDIRECT_PROMPT_INJECTION]: The skill accepts user-supplied text for image design briefs and incorporates this input directly into a CLI command. This creates a potential surface where adversarial input could attempt to influence the CLI tool's behavior.
Ingestion points: User-provided image descriptions and design briefs collected in Phase 2 and used in Phase 3.
Boundary markers: The prompt text is enclosed in single quotes within the shell command to prevent unintended shell expansion.
Capability inventory: The codex CLI is used to generate images and has permissions to move files into the workspace.
Sanitization: There is no evidence of input validation or content filtering for the design briefs before they are processed by the CLI.

codex-imagegen