openai-gpt-image
Pass
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [SAFE]: The skill promotes security best practices, such as minimizing metadata uploads, protecting personal data, and obtaining consent for likenesses. It explicitly prohibits evading safety blocks or obfuscating prompts to bypass moderation filters.
- [INDIRECT_PROMPT_INJECTION]: The skill describes a workflow for generating images from external prompts and reference images, identifying a potential injection surface. This is mitigated by documented safety guidelines and tool-level moderation settings.
- Ingestion points: User-supplied text prompts and image files referenced in SKILL.md.
- Boundary markers: Recommends using explicit delimiters like quotes for verbatim copy and indexing multiple reference images to maintain clarity.
- Capability inventory: Execution of network requests to OpenAI endpoints and file write operations to save generated assets, as seen in examples in SKILL.md.
- Sanitization: Directs the use of 'moderation=auto' and provides a detailed review checklist to be applied to final assets before delivery.
- [EXTERNAL_DOWNLOADS]: The skill instructs users to install and use the 'openai' Python package. This is a standard library from a recognized and well-known organization for the purpose of the skill.
Audit Metadata