The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses a vulnerability surface for indirect prompt injection because it is designed to ingest and process untrusted data from external sources such as live web pages, customer reviews, and merchant-supplied ad creatives.
Ingestion points: Data enters the agent's context through tools like lexsis_pages.get, lexsis_catalog.reviews, and the reading of campaign materials.
Boundary markers: The skill's instructions mitigate this risk by requiring that external content (like reviews) remain verbatim and be placed within defined section specifications, which helps maintain structural boundaries.
Capability inventory: The skill has the ability to modify web page content and structure through the lexsis_drafts.page_patch and lexsis_pages.compile actions.
Sanitization: The skill documents that the platform's compiler sanitizes and scopes both CSS and JavaScript. Custom animations and interactive modules are executed within a restricted runtime that lacks access to dangerous browser globals like window, document, or fetch.

optimize