lead-list

Pass

Audited by Gen Agent Trust Hub on Sep 20, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from various public sources including Reddit, X/Twitter, Google Maps, and user-pasted exports, creating a surface for indirect prompt injection. However, it implements a comprehensive defense strategy documented in references/agent-security.md.
  • Ingestion points: Scraped content from Google Maps, LinkedIn Jobs, Reddit (praw), X/Twitter (snscrape), G2/Capterra/Trustpilot (scrapy), and YouTube (yt-dlp), plus user-pasted CRM exports.
  • Boundary markers: The skill explicitly requires treating all retrieved content as untrusted data rather than instructions, as stated in Rule 1 of the Agent Security reference. It instructs the agent to report injection attempts as findings rather than executing them.
  • Capability inventory: The agent has access to WebFetch, WebSearch, a local scraper suite, and the ability to write to the .agents/product-context.md file.
  • Sanitization: Rule 3 of the Agent Security reference mandates HTML escaping for all interpolated values in emitted markup to prevent XSS.
  • [COMMAND_EXECUTION]: The skill uses a 'local scraper suite' including tools like github-leads-praw and github-leads-scrapy to fetch data. These are intended functional components for sourcing lead signals.
  • [EXTERNAL_DOWNLOADS]: The skill references standard scraping libraries (e.g., snscrape, yt-dlp, instaloader) which are used to interact with public platforms. All data fetching is guided by strict compliance rules (e.g., respecting robots.txt, no login-wall bypass).
  • [SAFE]: The skill demonstrates high security maturity by explicitly flagging credential exposure in inputs, prohibiting the use of sensitive personal attributes for scoring, and providing clear instructions on handling untrusted data.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 20, 2026, 02:18 PM