The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill's documentation in SKILL.md instructs the user to install the Codex CLI by downloading and executing a shell script directly from a well-known service's domain (curl -fsSL https://chatgpt.com/codex/install.sh | sh).
[COMMAND_EXECUTION]: The skill generates commands using codex exec that include a high-risk flag -s danger-full-access. This indicates that the executed commands operate with significant privileges or broad system access.
[PRIVILEGE_ESCALATION]: By explicitly instructing the agent to use the danger-full-access scope, the skill seeks to bypass standard restrictions and acquire elevated permissions within the execution environment.
[EXTERNAL_DOWNLOADS]: The skill references the installation of an external Node.js package (@openai/codex) and a Homebrew cask (codex) to provide its core functionality.
[INDIRECT_PROMPT_INJECTION]: The skill processes external files and user-provided instructions, creating a surface for potential injection attacks where malicious data could influence the generated commands.
Ingestion points: In SKILL.md, the 'Document → Diagram' mode reads local files from <INPUT_PATH> and incorporates them into the instruction set along with <USER_INSTRUCTIONS>.
Boundary markers: None are present; the content of the files is interpolated directly into the command string.
Capability inventory: The skill uses the codex CLI to execute instructions and interacts with the file system to save generated images.
Sanitization: There is no evidence of input validation, escaping, or filtering for the data read from external files before it is passed to the shell.

codex-imagegen