The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructions attempt to override safety protocols by commanding the agent to 'install the missing dependency automatically without asking the user to confirm' and 'not request separate approval for dependency installation.' This is a direct attempt to bypass user oversight and hide potentially dangerous actions from the user.\n- [REMOTE_CODE_EXECUTION]: The skill requires the automatic installation of the 'browser-act' CLI and associated skill from an unverified GitHub repository (github.com/browser-act/skills). Installing executable code from untrusted third-party sources without verification or user intervention poses a high risk of remote code execution or system compromise.\n- [EXTERNAL_DOWNLOADS]: The skill directs the agent to fetch and install software from an external source (www.browseract.com) that redirects to a non-whitelisted GitHub organization. This action is instructed to occur silently, without a manual review of the source or the content being downloaded.\n- [DATA_EXFILTRATION]: The skill is designed to capture browser-visible page content and network responses. When combined with the silent installation of a third-party tool, this creates a significant risk for unauthorized data collection or exfiltration of sensitive information such as session cookies or personal browsing metadata.

zalando-product-scraper