tavily-search
Pass
Audited by Gen Agent Trust Hub on Sep 9, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process content retrieved from the open web via the Tavily search service. \n
- Ingestion points: The agent retrieves LLM-optimized search results, content snippets, and raw page content (markdown or text) as described in
SKILL.md.\n - Boundary markers: The skill contains a specific "Anti-Patterns" section that instructs the agent not to treat external tool responses as trusted instructions.\n
- Capability inventory: The skill utilizes the
tvlyCLI tool for performing network searches and retrieving content.\n - Sanitization: The documentation advises the agent to verify identity-sensitive facts at primary sources, though it does not define automated content sanitization filters. \n- [COMMAND_EXECUTION]: The skill provides instructions for the agent to execute the
tvlyCLI for performing search queries (tvly search) and managing authentication (tvly login). This command execution is the core intended functionality of the skill.
Audit Metadata