The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes content from external social media feeds and URLs. It implements a mitigation strategy by using the analyze_media tool to perform OCR and content verification on all visual artifacts (reel frames, mood boards, and PDF pages) before they are finalized, reducing the risk of processing malicious embedded instructions. Each ingestion point (social scrape, taste URLs, photos) is countered by mandatory analysis of output frames and PDF pages using analyze_media to detect and reject baked-in text.
[EXTERNAL_DOWNLOADS]: Functional data is retrieved from standard social media platforms (Instagram, TikTok, YouTube, Pinterest, Twitter) and well-known services like Google Fonts. These connections are transparent, directed at reputable services, and are strictly necessary for the skill's persona analysis and document styling.
[DATA_EXFILTRATION]: The skill utilizes upload_asset to store generated branding documents and images on a CDN for user access. This process is limited to the skill's own generated output and does not attempt to access or exfiltrate private user data, environment variables, or local system credentials. The use of rehost: true in social scraping is a vendor-native feature for media durability.

persona-builder