NYC

cloudflare-browser-rendering

Fail

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: CRITICALPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • PROMPT_INJECTION (LOW): Indirect Prompt Injection surface detected in 'templates/ai-enhanced-scraper.ts'.
  • Ingestion points: HTML content is extracted from untrusted URLs via 'page.$eval' and subsequently passed to an LLM.
  • Boundary markers: Untrusted content is interpolated into the AI prompt without escaping or robust delimiters.
  • Capability inventory: The script transmits extracted content to the '@cf/meta/llama-3.1-8b-instruct' model.
  • Sanitization: No sanitization is performed on the extracted HTML beyond basic truncation.
  • EXTERNAL_DOWNLOADS (SAFE): The 'scripts/check-versions.sh' file uses 'npm view' to query package metadata for version updates, which is a non-executable metadata check.
  • COMMAND_EXECUTION (LOW): The skill lacks URL/HTML sanitization in its screenshot and PDF generation templates, which is a common risk factor for SSRF or local resource access in headless browser environments.
  • SAFE (INFO): The automated scanner alert for 'browser.se' is a false positive triggered by the substring in the standard Cloudflare Puppeteer API method 'browser.sessions'.
Recommendations
  • Contains 1 malicious URL(s) - DO NOT USE
Audit Metadata
Risk Level
CRITICAL
Analyzed
Feb 17, 2026, 05:51 PM