web-scraping

Pass

Audited by Gen Agent Trust Hub on Sep 15, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to retrieve and process content from third-party websites, which constitutes an attack surface for indirect prompt injection where malicious instructions could be embedded in the scraped data.
  • Ingestion points: Web content retrieved via TrafilaturaScraper, RequestsScraper, PlaywrightScraper, and media metadata extracted from YouTube, Instagram, and TikTok via yt-dlp and instaloader (SKILL.md).
  • Boundary markers: The skill explicitly instructs the agent to use <EXTERNAL_DATA> tags to delimit untrusted content and maintain its provenance (SKILL.md).
  • Capability inventory: The skill utilizes network request libraries (requests, playwright) and performs file system operations for media downloads (yt-dlp, instaloader) (SKILL.md).
  • Sanitization: The instructions include a detailed "Untrusted content contract" that directs the agent to ignore embedded instructions, validate URLs, and avoid passing sensitive context to third parties (SKILL.md).
  • [EXTERNAL_DOWNLOADS]: The skill provides functionality and code for downloading media files and metadata from various social media platforms using the yt-dlp and instaloader libraries.
  • [COMMAND_EXECUTION]: The skill contains extensive Python code blocks for web automation and scraping, which are intended to be executed within the agent's environment to perform the requested tasks.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 15, 2026, 12:59 AM
Security Audit — agent-trust-hub — web-scraping