google-ai-search

Pass

Audited by Gen Agent Trust Hub on Jul 30, 2026

Risk Level: SAFECOMMAND_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructions provide commands to launch Google Chrome with the --remote-debugging-port flag and execute the retrieval script using node. These are standard developer operations for browser automation.
  • [DATA_EXFILTRATION]: The script connects to localhost via the Chrome DevTools Protocol (CDP) to interact with an active browser session. It captures search result data and saves it locally to the user's ~/Downloads directory. No external network requests to third-party domains (other than Google for the search) were identified.
  • [INDIRECT_PROMPT_INJECTION]: The skill retrieves and saves content from external web pages (Google AI Overviews). While this introduces a potential attack surface if the saved HTML is subsequently processed by another AI agent without sanitization, the skill itself only performs a write operation to the local disk.
  • Ingestion points: Captures HTML via network response interception and DOM selection from www.google.com in scripts/google-ai-search.ts.
  • Boundary markers: None. The raw HTML or text is captured as-is.
  • Capability inventory: The script has the capability to write files (fs.writeFile) to the user's Downloads directory.
  • Sanitization: No sanitization or filtering is performed on the retrieved content before it is saved to the filesystem.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 30, 2026, 01:44 PM
Security Audit — agent-trust-hub — google-ai-search