The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the internet via tools like scrape_page_content and fetch_url. It implements proactive mitigations by wrapping all ingested content in untrusted data fences and stripping invisible characters to prevent the agent from following malicious instructions embedded in web pages.
[EXTERNAL_DOWNLOADS]: The skill performs network operations to fetch and download web content. It utilizes an internal checkOutboundUrl guard to prevent Server-Side Request Forgery (SSRF) and includes logic to safely normalize GitHub URLs to their official raw content domains.
[SAFE]: The skill follows security best practices by enforcing strict resource constraints, including a 10MB limit for images, a 50MB limit for other files, and character limits for text returned to the agent context. Additionally, file writes are restricted to a workspace environment with a specific list of allowed file extensions.

web-scraping