The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted data from external product URLs and images to extract product information, usage mechanics, and branding details (Step 1 of SKILL.md). This information is subsequently used to construct prompts for image generation. This workflow creates an attack surface for indirect prompt injection, where malicious instructions hidden in a product's website or image metadata could attempt to influence the agent's behavior or subvert the generation process.
Ingestion points: External product images or URLs are ingested at the beginning of the workflow in SKILL.md (Step 1).
Boundary markers: The skill lacks explicit instructions to the agent to isolate external content or to ignore any embedded directives within the ingested data.
Capability inventory: The skill has the capability to trigger image_generate and image_analysis tools, which involve external model invocation and potential financial cost.
Sanitization: There is no requirement for the agent to sanitize, escape, or validate the content retrieved from external sources before using it to define the ad concept or generate the final output.
[REMOTE_CODE_EXECUTION]: No remote code execution patterns or external script downloads were detected. The skill utilizes designated platform tools (image_generate, image_analysis, list_image_models).
[DATA_EXFILTRATION]: No data exfiltration patterns or hardcoded credentials were found. The use of product URLs is consistent with the skill's primary purpose.

generating-image-ads