The Agent Skills Directory

[SAFE]: The skill implements secure credential handling by instructing users to set the OPENAI_API_KEY environment variable locally rather than providing it in chat, preventing accidental exposure of sensitive keys.
[SAFE]: Dependency management is handled via uv or pip to install well-known and trusted packages such as openai and pillow. No unverifiable or suspicious packages are requested.
[SAFE]: The bundled Python script (scripts/image_gen.py) performs legitimate image processing tasks. It decodes standard Base64 image data from the OpenAI API and uses the Pillow library for image downscaling. No malicious obfuscation or hidden execution patterns were detected.
[SAFE]: File operations are restricted to specific project directories (output/imagegen/ and tmp/imagegen/), with appropriate checks to prevent overwriting files unless the --force flag is explicitly used.
[SAFE]: Network activity is restricted to communication with the official OpenAI API endpoints, which is the stated primary purpose of the skill. The documentation in references/codex-network.md provides transparent information regarding sandbox network approvals.
[INDIRECT_PROMPT_INJECTION]: The skill processes natural language prompts, creating a standard surface for indirect prompt injection. However, it uses prompt augmentation to structure and delimit inputs, and all associated capabilities (API calls and file writes) are consistent with the skill's intended use case, posing minimal risk.

imagegen