The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data by crawling live web pages, which presents a surface for indirect prompt injection if those pages contain adversarial content (e.g., instructions in meta tags or titles).
Ingestion points: The audit_onpage_seo tool fetches and reads HTML content from external URLs provided by the user or found in Search Console.
Boundary markers: The skill instructions do not define delimiters to isolate external page content from the agent's internal reasoning logic.
Capability inventory: The skill uses tools to list sites (list_sites), query performance (query_top_pages), and inspect URLs (inspect_url) within the Google Search Console environment.
Sanitization: There is no mention of sanitizing or escaping the fetched HTML metadata before it is processed by the agent.

onpage-audit