tavily-best-practices
Pass
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADSCREDENTIALS_UNSAFE
Full Analysis
- Indirect Prompt Injection (LOW): The skill fetches web content via the Tavily API, which introduces a surface for indirect prompt injection where malicious data from external websites could influence agent behavior. (1) Ingestion points: Content is retrieved using
search(),extract(), andcrawl()methods. (2) Boundary markers: Absent in implementation examples. (3) Capability inventory: Extracted content is passed to LLMs in frameworks like LangChain and CrewAI as seen inreferences/integrations.md. (4) Sanitization: No sanitization of web content is demonstrated in the snippets. - External Downloads (SAFE): Recommends installing legitimate, well-known packages from PyPI and npm.
- Credentials Unsafe (SAFE): Mentions storage of API keys and database URIs using standard placeholders and following documentation best practices.
Audit Metadata