tavily-best-practices
Pass
Audited by Gen Agent Trust Hub on Sep 21, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill provides documentation and code snippets for the Tavily API. All external references are to official SDKs and well-known AI frameworks. No executable code or malicious instructions were found.
- [EXTERNAL_DOWNLOADS]: The skill instructs users to install official packages from PyPI (tavily-python, langchain-tavily, pydantic-ai-slim, agno, anthropic, google-adk, mcp, crewai) and npm (@tavily/core, ai, @ai-sdk/openai, @tavily/ai-sdk). These are standard dependencies for the integration's purpose.
- [INDIRECT_PROMPT_INJECTION]: The skill documents patterns for ingesting web content into AI systems. Evidence chain: 1. Ingestion points: Tavily search, extract, and crawl results ingested in SKILL.md and references/*.md. 2. Boundary markers: Snippets in research.md and search.md suggest using structured schemas and LLM verification prompts to handle external data. 3. Capability inventory: Network requests to Tavily and LLM APIs. 4. Sanitization: Examples use JSON serialization for tool outputs. While these methods ingest untrusted web data, the skill correctly documents architectural best practices to manage this inherent risk.
Audit Metadata