The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill introduces a data ingestion surface in SKILL.md where untrusted data is collected from arbitrary URLs using firecrawl_branding_extract and firecrawl_urls_scrape. The extracted data (such as fonts, brand colors, and personality/tone strings) is then dynamically interpolated into the prompts for images_generate and other high-level workflow tools. No prompt boundary markers or sanitization mechanisms are specified in the instructions to protect against malicious text hidden on the target website that could attempt to hijack the image generation process.

image-generation