The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from users, such as image prompts, verbatim text to render, and constraints, and interpolates this content into strings passed to the codex exec command.
Ingestion points: User-supplied prompts and image metadata defined in the Workflow section of SKILL.md and the Write the prompt section of references/prompting.md.
Boundary markers: The skill uses single quotes to wrap instructions within the codex exec command, but it lacks explicit instructions for the agent to sanitize or escape user-provided text that might contain shell control characters or quotes.
Capability inventory: The skill possesses the capability to execute shell commands via codex exec, run Python scripts with python3 -c, and manage dependencies with pip install.
Sanitization: No evidence of input validation, filtering, or escaping is present in the provided instructions.
[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands via the Codex CLI (codex exec) and Python (python3 -c) for the purpose of generating, editing, and verifying images.
Evidence: Examples in SKILL.md show the use of codex exec --sandbox workspace-write to invoke image generation tools, and references/transparency.md includes a python3 -c command to verify alpha transparency in generated files.
[EXTERNAL_DOWNLOADS]: The skill documentation references external tools and libraries required for full functionality.
Evidence: SKILL.md directs the user to install the Codex CLI from an official OpenAI domain and suggests installing the pillow Python package for local image inspection via python3 -m pip install pillow.

imagegen