chatgpt-image-ad
Pass
Audited by Gen Agent Trust Hub on Oct 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests user-provided prompts and reference images, creating a surface for indirect prompt injection, though the impact is limited by its specific use case for image generation.
- Ingestion points: User seed prompts are collected in
SKILL.mdand passed toscripts/generate_image.pyvia the--promptargument. Reference images are also read from the filesystem. - Boundary markers: The script
scripts/generate_image.pyautomatically appends safety suffixes (NO_CHROME_SUFFIX,SAFE_ZONE_SUFFIX,GLYPH_SAFETY_SUFFIX) to all prompts to control the generation output. - Capability inventory:
scripts/generate_image.pyperforms network operations to the Arcads API (https://external-api.arcads.ai) and writes generated files to the./generateddirectory. - Sanitization: Prompt suffixes are used to guide the external model, although the input prompt itself is not escaped or filtered for instruction-like content.
Audit Metadata