The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates monitoring and processing content from external websites, which creates an inherent attack surface for indirect prompt injection. A monitored site could attempt to influence the AI judge by changing its content. * Ingestion points: Web content is ingested from URLs provided in command arguments like --page and --crawl-url, as well as search queries (SKILL.md, goals.md). * Boundary markers: The instructions do not define specific delimiters to separate external content from internal instructions. * Capability inventory: The skill uses the Bash tool to interact with the Firecrawl API. * Sanitization: Content sanitization is managed by the vendor's cloud service rather than the skill's instructions. This is a standard risk for web monitoring and evaluation tools.

firecrawl-monitor