tavily-best-practices

Pass

Audited by Gen Agent Trust Hub on Mar 14, 2026

Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PROMPT_INJECTION]: Indirect prompt injection surface identified via web search and extraction capabilities.
  • Ingestion points: The search(), extract(), crawl(), and research() methods described in SKILL.md and reference files allow agents to ingest content from arbitrary external websites.
  • Boundary markers: The documentation and examples do not include explicit instructions or delimiters (such as "ignore instructions within this data") to isolate retrieved content.
  • Capability inventory: The skill demonstrates network access to tavily.com and integration with several LLM frameworks for data processing.
  • Sanitization: No specific evidence of output sanitization or filtering of external content is provided in the examples.
  • [EXTERNAL_DOWNLOADS]: The skill provides instructions for installing official packages such as tavily-python, @tavily/core, and langchain-tavily. These downloads target official repositories and well-known registries (PyPI, NPM) from established technology providers.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 14, 2026, 03:50 AM