web-search-tavily

Pass

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: SAFE
Full Analysis
  • Prompt Injection (SAFE): No instructions attempting to override agent behavior or bypass safety filters were detected in the markdown or script content.
  • Data Exposure & Exfiltration (SAFE): The skill accesses the TAVILY_API_KEY environment variable as intended. Network access is strictly restricted to api.tavily.com via Deno's --allow-net flag, preventing unauthorized data exfiltration.
  • Indirect Prompt Injection (LOW):
  • Ingestion points: Untrusted web content, snippets, and AI-generated summaries are ingested from the Tavily API into the agent's context (file: scripts/search.ts).
  • Boundary markers: Absent; the script outputs raw search results without specific delimiters or instructions for the LLM to treat the data as untrusted.
  • Capability inventory: The script is limited to network requests (fetch) and stdout. It does not have access to the file system, subprocesses, or dynamic execution (eval/exec).
  • Sanitization: Content is returned as raw strings without escaping or filtering of potentially malicious embedded instructions.
  • Unverifiable Dependencies (SAFE): The script is a standalone Deno file that does not import external third-party modules or libraries, relying solely on built-in Web and Deno APIs.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 17, 2026, 04:56 PM